C2S4 QS8 Neon GEMM/IGEMM microkernels.

- Rename existing S4C2 to C2S4 as it isa shuffle variant of the C2 dup microkernel.
- Add QC8 microkernels
- Add all sizes.  Same set of microkernels as C2 and C4, except no gemmlowp.

PiperOrigin-RevId: 408950530
diff --git a/test/qc8-gemm-minmax-fp32.cc b/test/qc8-gemm-minmax-fp32.cc
index 663259a..f82e4f7 100644
--- a/test/qc8-gemm-minmax-fp32.cc
+++ b/test/qc8-gemm-minmax-fp32.cc
@@ -23,6 +23,1830 @@
 
 
 #if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_lt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_lt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_lt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t m = 1; m <= 2; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t m = 1; m <= 2; m++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_lt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
   TEST(QC8_GEMM_MINMAX_FP32_1X8C4__NEON_MLAL_PADAL_DUP, k_eq_16) {
     TEST_REQUIRES_ARM_NEON;
     GemmMicrokernelTester()
diff --git a/test/qc8-gemm-minmax-fp32.yaml b/test/qc8-gemm-minmax-fp32.yaml
index 14156f9..b226618 100644
--- a/test/qc8-gemm-minmax-fp32.yaml
+++ b/test/qc8-gemm-minmax-fp32.yaml
@@ -3,6 +3,18 @@
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
+- name: xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal
+  init: xnn_init_qs8_minmax_neon_fp32_params
+  k-block: 16
+- name: xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal
+  init: xnn_init_qs8_minmax_neon_fp32_params
+  k-block: 16
+- name: xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal
+  init: xnn_init_qs8_minmax_neon_params
+  k-block: 16
+- name: xnn_qc8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal
+  init: xnn_init_qs8_minmax_neon_params
+  k-block: 16
 - name: xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4__neon_mlal_padal_dup
   init: xnn_init_qs8_minmax_neon_fp32_params
   k-block: 16
diff --git a/test/qc8-igemm-minmax-fp32.cc b/test/qc8-igemm-minmax-fp32.cc
index feeed40..8565cd6 100644
--- a/test/qc8-igemm-minmax-fp32.cc
+++ b/test/qc8-igemm-minmax-fp32.cc
@@ -23,6 +23,1878 @@
 
 
 #if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_gt_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_div_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(83)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 1; mz++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(83)
+          .zero_index(mz)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_gt_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_div_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(163)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 2; mz++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(163)
+          .zero_index(mz)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_minmax_neon_fp32_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_div_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(83)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t mz = 0; mz < 1; mz++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(83)
+          .zero_index(mz)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t m = 1; m <= 2; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t m = 1; m <= 2; m++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_div_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(163)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t mz = 0; mz < 2; mz++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(163)
+          .zero_index(mz)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
   TEST(QC8_IGEMM_MINMAX_FP32_1X8C4__NEON_MLAL_PADAL_DUP, k_eq_16) {
     TEST_REQUIRES_ARM_NEON;
     GemmMicrokernelTester()
diff --git a/test/qc8-igemm-minmax-fp32.yaml b/test/qc8-igemm-minmax-fp32.yaml
index 47928e1..b300995 100644
--- a/test/qc8-igemm-minmax-fp32.yaml
+++ b/test/qc8-igemm-minmax-fp32.yaml
@@ -3,6 +3,18 @@
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
+- name: xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal
+  init: xnn_init_qs8_minmax_neon_fp32_params
+  k-block: 16
+- name: xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal
+  init: xnn_init_qs8_minmax_neon_fp32_params
+  k-block: 16
+- name: xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal
+  init: xnn_init_qs8_minmax_neon_params
+  k-block: 16
+- name: xnn_qc8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal
+  init: xnn_init_qs8_minmax_neon_params
+  k-block: 16
 - name: xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4__neon_mlal_padal_dup
   init: xnn_init_qs8_minmax_neon_fp32_params
   k-block: 16
diff --git a/test/qs8-gemm-minmax-fp32.cc b/test/qs8-gemm-minmax-fp32.cc
index cdfe380..217e951 100644
--- a/test/qs8-gemm-minmax-fp32.cc
+++ b/test/qs8-gemm-minmax-fp32.cc
@@ -23,6 +23,1830 @@
 
 
 #if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_lt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_lt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_lt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t m = 1; m <= 2; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t m = 1; m <= 2; m++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_lt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
   TEST(QS8_GEMM_MINMAX_FP32_1X8C4__NEON_MLAL_PADAL_DUP, k_eq_16) {
     TEST_REQUIRES_ARM_NEON;
     GemmMicrokernelTester()
diff --git a/test/qs8-gemm-minmax-fp32.yaml b/test/qs8-gemm-minmax-fp32.yaml
index 8a30130..e14b172 100644
--- a/test/qs8-gemm-minmax-fp32.yaml
+++ b/test/qs8-gemm-minmax-fp32.yaml
@@ -3,6 +3,18 @@
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
+- name: xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_fp32_neon_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_fp32_neon_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal
+  init: xnn_init_qs8_conv_minmax_fp32_neonv8_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal
+  init: xnn_init_qs8_conv_minmax_fp32_neonv8_params
+  k-block: 16
 - name: xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4__neon_mlal_padal_dup
   init: xnn_init_qs8_conv_minmax_fp32_neon_params
   k-block: 16
diff --git a/test/qs8-gemm-minmax-rndnu.cc b/test/qs8-gemm-minmax-rndnu.cc
index fc36cfa..9537fc4 100644
--- a/test/qs8-gemm-minmax-rndnu.cc
+++ b/test/qs8-gemm-minmax-rndnu.cc
@@ -23,6 +23,7302 @@
 
 
 #if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(8)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_lt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, n_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, n_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(8)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(8)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_lt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, n_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, n_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(8)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(8)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 3; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 3; m++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_lt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, n_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 3; m++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, n_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 3; m++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(8)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(8)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_lt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, n_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, n_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(8)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_lt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, n_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, n_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, n_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_lt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, n_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, n_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, n_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 3; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 3; m++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_lt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, n_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 3; m++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, n_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, n_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 3; m++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_lt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, n_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, n_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, n_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_lt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_lt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 3; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 3; m++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_lt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 3; m++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 3; m++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_lt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_lt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, n_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, n_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, n_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_lt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, n_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, n_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, n_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 3; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 3; m++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_lt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, n_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 3; m++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, n_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, n_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 3; m++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_lt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, n_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, n_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, n_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
   TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4__NEON_MULL_PADAL_DUP, k_eq_8) {
     TEST_REQUIRES_ARM_NEON;
     GemmMicrokernelTester()
@@ -7319,1830 +14615,6 @@
 
 
 #if XNN_ARCH_ARM || XNN_ARCH_ARM64
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_eq_16) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(16)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(16)
-      .cn_stride(11)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_eq_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(16)
-      .a_stride(19)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_eq_16_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t m = 1; m <= 1; m++) {
-      for (uint32_t n = 1; n <= 8; n++) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(m)
-          .n(n)
-          .k(16)
-          .iterations(1)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t m = 1; m <= 1; m++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(m)
-        .n(8)
-        .k(16)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 1; n <= 8; n++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(n)
-        .k(16)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_lt_16) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_lt_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .a_stride(19)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_lt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 16; k++) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_gt_16) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 17; k < 32; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_gt_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 17; k < 32; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .a_stride(37)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_gt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 17; k < 32; k++) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_div_16) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 32; k <= 160; k += 16) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_div_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 32; k <= 160; k += 16) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .a_stride(163)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_div_16_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 32; k <= 160; k += 16) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, n_gt_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(8)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(8)
-          .k(k)
-          .cn_stride(11)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, n_gt_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(n)
-          .k(k)
-          .a_stride(83)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, n_gt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        for (uint32_t m = 1; m <= 1; m++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, n_div_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(8)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, n_div_8_strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(n)
-          .k(k)
-          .cn_stride(11)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, n_div_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(n)
-          .k(k)
-          .a_stride(83)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, n_div_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        for (uint32_t m = 1; m <= 1; m++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, strided_cm_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 80; k += 17) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .cm_stride(11)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, qmin) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(16)
-      .qmin(128)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, qmax) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(16)
-      .qmax(128)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, strided_cm) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(16)
-      .cm_stride(11)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
-
-
-#if XNN_ARCH_ARM || XNN_ARCH_ARM64
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_eq_16) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(16)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(16)
-      .cn_stride(11)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_eq_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(16)
-      .a_stride(19)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_eq_16_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t m = 1; m <= 2; m++) {
-      for (uint32_t n = 1; n <= 8; n++) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(m)
-          .n(n)
-          .k(16)
-          .iterations(1)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t m = 1; m <= 2; m++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(m)
-        .n(8)
-        .k(16)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 1; n <= 8; n++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(n)
-        .k(16)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_lt_16) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_lt_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .a_stride(19)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_lt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 16; k++) {
-      for (uint32_t m = 1; m <= 2; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_gt_16) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 17; k < 32; k++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_gt_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 17; k < 32; k++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .a_stride(37)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_gt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 17; k < 32; k++) {
-      for (uint32_t m = 1; m <= 2; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_div_16) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 32; k <= 160; k += 16) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_div_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 32; k <= 160; k += 16) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .a_stride(163)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_div_16_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 32; k <= 160; k += 16) {
-      for (uint32_t m = 1; m <= 2; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, n_gt_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(8)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(8)
-          .k(k)
-          .cn_stride(11)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, n_gt_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(n)
-          .k(k)
-          .a_stride(83)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, n_gt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        for (uint32_t m = 1; m <= 2; m++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, n_div_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(8)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, n_div_8_strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(n)
-          .k(k)
-          .cn_stride(11)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, n_div_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(n)
-          .k(k)
-          .a_stride(83)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, n_div_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        for (uint32_t m = 1; m <= 2; m++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, strided_cm_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 80; k += 17) {
-      for (uint32_t m = 1; m <= 2; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .cm_stride(11)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, qmin) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(16)
-      .qmin(128)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, qmax) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(16)
-      .qmax(128)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, strided_cm) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(16)
-      .cm_stride(11)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
-
-
-#if XNN_ARCH_ARM || XNN_ARCH_ARM64
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_eq_8) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(8)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(8)
-      .cn_stride(11)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_eq_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(8)
-      .a_stride(11)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_eq_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t m = 1; m <= 1; m++) {
-      for (uint32_t n = 1; n <= 8; n++) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(m)
-          .n(n)
-          .k(8)
-          .iterations(1)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_eq_8_subtile_m) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t m = 1; m <= 1; m++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(m)
-        .n(8)
-        .k(8)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_eq_8_subtile_n) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 1; n <= 8; n++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(n)
-        .k(8)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_lt_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 8; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_lt_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 8; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .a_stride(11)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_lt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 8; k++) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_gt_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 9; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_gt_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 9; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .a_stride(19)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_gt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 9; k < 16; k++) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_div_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 16; k <= 80; k += 8) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_div_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 16; k <= 80; k += 8) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .a_stride(83)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_div_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 16; k <= 80; k += 8) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, n_gt_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(8)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, n_gt_8_strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(8)
-          .k(k)
-          .cn_stride(11)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, n_gt_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(n)
-          .k(k)
-          .a_stride(43)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, n_gt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        for (uint32_t m = 1; m <= 1; m++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, n_div_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(8)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, n_div_8_strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(n)
-          .k(k)
-          .cn_stride(11)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, n_div_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(n)
-          .k(k)
-          .a_stride(43)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, n_div_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        for (uint32_t m = 1; m <= 1; m++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, strided_cm_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 40; k += 9) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .cm_stride(11)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, qmin) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(8)
-      .qmin(128)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, qmax) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(8)
-      .qmax(128)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, strided_cm) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(8)
-      .cm_stride(11)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
-
-
-#if XNN_ARCH_ARM || XNN_ARCH_ARM64
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_eq_8) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(8)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(8)
-      .cn_stride(11)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_eq_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(8)
-      .a_stride(11)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_eq_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t m = 1; m <= 2; m++) {
-      for (uint32_t n = 1; n <= 8; n++) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(m)
-          .n(n)
-          .k(8)
-          .iterations(1)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_eq_8_subtile_m) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t m = 1; m <= 2; m++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(m)
-        .n(8)
-        .k(8)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_eq_8_subtile_n) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 1; n <= 8; n++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(n)
-        .k(8)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_lt_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 8; k++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_lt_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 8; k++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .a_stride(11)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_lt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 8; k++) {
-      for (uint32_t m = 1; m <= 2; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_gt_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 9; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_gt_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 9; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .a_stride(19)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_gt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 9; k < 16; k++) {
-      for (uint32_t m = 1; m <= 2; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_div_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 16; k <= 80; k += 8) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_div_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 16; k <= 80; k += 8) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .a_stride(83)
-        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_div_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 16; k <= 80; k += 8) {
-      for (uint32_t m = 1; m <= 2; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, n_gt_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(8)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, n_gt_8_strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(8)
-          .k(k)
-          .cn_stride(11)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, n_gt_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(n)
-          .k(k)
-          .a_stride(43)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, n_gt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        for (uint32_t m = 1; m <= 2; m++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, n_div_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(8)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, n_div_8_strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(n)
-          .k(k)
-          .cn_stride(11)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, n_div_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(n)
-          .k(k)
-          .a_stride(43)
-          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, n_div_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        for (uint32_t m = 1; m <= 2; m++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, strided_cm_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 40; k += 9) {
-      for (uint32_t m = 1; m <= 2; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .cm_stride(11)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, qmin) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(8)
-      .qmin(128)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, qmax) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(8)
-      .qmax(128)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_GEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, strided_cm) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(8)
-      .cm_stride(11)
-      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
-
-
-#if XNN_ARCH_ARM || XNN_ARCH_ARM64
   TEST(QS8_GEMM_MINMAX_RNDNU_1X8C2__NEON_MULL_PADAL_DUP, k_eq_8) {
     TEST_REQUIRES_ARM_NEON;
     GemmMicrokernelTester()
diff --git a/test/qs8-gemm-minmax-rndnu.yaml b/test/qs8-gemm-minmax-rndnu.yaml
index 626e4ab..dc61434 100644
--- a/test/qs8-gemm-minmax-rndnu.yaml
+++ b/test/qs8-gemm-minmax-rndnu.yaml
@@ -3,6 +3,54 @@
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
 - name: xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4__neon_mull_padal_dup
   init: xnn_init_qs8_conv_minmax_rndnu_neon_params
   k-block: 8
@@ -51,18 +99,6 @@
 - name: xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4__neon_mlal_padal_dup
   init: xnn_init_qs8_conv_minmax_rndnu_neon_params
   k-block: 16
-- name: xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal
-  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
-  k-block: 16
-- name: xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal
-  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
-  k-block: 16
-- name: xnn_qs8_gemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal
-  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
-  k-block: 8
-- name: xnn_qs8_gemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal
-  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
-  k-block: 8
 - name: xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2__neon_mull_padal_dup
   init: xnn_init_qs8_conv_minmax_rndnu_neon_params
   k-block: 8
diff --git a/test/qs8-igemm-minmax-fp32.cc b/test/qs8-igemm-minmax-fp32.cc
index 659c844..80d01e0 100644
--- a/test/qs8-igemm-minmax-fp32.cc
+++ b/test/qs8-igemm-minmax-fp32.cc
@@ -23,6 +23,1878 @@
 
 
 #if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_gt_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, n_div_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(83)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 1; mz++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(83)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_gt_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, n_div_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(163)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 2; mz++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(163)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, n_div_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(83)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t mz = 0; mz < 1; mz++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(83)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_1X8C2S4__NEONV8_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t m = 1; m <= 2; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t m = 1; m <= 2; m++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_gt_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, n_div_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(163)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    for (uint32_t mz = 0; mz < 2; mz++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(163)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_2X8C2S4__NEONV8_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
   TEST(QS8_IGEMM_MINMAX_FP32_1X8C4__NEON_MLAL_PADAL_DUP, k_eq_16) {
     TEST_REQUIRES_ARM_NEON;
     GemmMicrokernelTester()
diff --git a/test/qs8-igemm-minmax-fp32.yaml b/test/qs8-igemm-minmax-fp32.yaml
index 5dd77a9..1024cef 100644
--- a/test/qs8-igemm-minmax-fp32.yaml
+++ b/test/qs8-igemm-minmax-fp32.yaml
@@ -3,6 +3,18 @@
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
+- name: xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_fp32_neon_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_fp32_neon_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neonv8_mlal_padal
+  init: xnn_init_qs8_conv_minmax_fp32_neonv8_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_fp32_ukernel_2x8c2s4__neonv8_mlal_padal
+  init: xnn_init_qs8_conv_minmax_fp32_neonv8_params
+  k-block: 16
 - name: xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4__neon_mlal_padal_dup
   init: xnn_init_qs8_conv_minmax_fp32_neon_params
   k-block: 16
diff --git a/test/qs8-igemm-minmax-rndnu.cc b/test/qs8-igemm-minmax-rndnu.cc
index 50e5aef..8378366 100644
--- a/test/qs8-igemm-minmax-rndnu.cc
+++ b/test/qs8-igemm-minmax-rndnu.cc
@@ -23,6 +23,7494 @@
 
 
 #if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(8)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(8)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, n_gt_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, n_div_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(43)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 1; mz++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(43)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(8)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(8)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(8)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, n_gt_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, n_div_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(83)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 2; mz++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(83)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(8)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(8)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(8)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 3; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 3; m++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 3; m++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 3; m++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, n_gt_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, n_div_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(127)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 3; mz++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(127)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(8)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(8)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(8)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, n_gt_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, n_div_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(163)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 4; mz++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(163)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(8)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, n_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, n_gt_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, n_div_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(43)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 1; mz++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(43)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, n_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, n_gt_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, n_div_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(83)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 2; mz++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(83)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 3; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 3; m++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 3; m++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, n_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 3; m++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, n_gt_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, n_div_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(127)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 3; mz++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(127)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 16; k <= 80; k += 8) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, n_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, n_gt_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, n_div_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(163)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 4; mz++) {
+      for (size_t k = 1; k <= 40; k += 9) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(163)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, n_gt_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, n_div_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(83)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 1; mz++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(83)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, n_gt_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, n_div_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(163)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 2; mz++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(163)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 3; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 3; m++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 3; m++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 3; m++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, n_gt_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, n_div_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(251)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 3; mz++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(251)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 8; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 8; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, n_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, n_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, n_div_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, n_div_8_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, n_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, n_gt_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 9; n < 16; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, n_div_8_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 16; n <= 24; n += 8) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 8; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(8)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(331)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 4; mz++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(8)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(331)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, n_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, n_gt_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, n_div_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(1)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(83)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 1; mz++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(1)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(83)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(1)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 2; m++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, n_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 2; m++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, n_gt_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, n_div_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 2; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(2)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(2)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(163)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 2; mz++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(2)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(2)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(163)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(2)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 3; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 3; m++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 3; m++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, n_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 3; m++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, n_gt_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, n_div_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 3; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(3)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(3)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(251)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 3; mz++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(3)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(3)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(251)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(3)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 32; k <= 160; k += 16) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, n_div_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, n_gt_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, n_div_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(2)
+            .sr(4)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(2)
+        .sr(4)
+        .m(4)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(331)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, zero) {
+    TEST_REQUIRES_ARM_NEON;
+    for (uint32_t mz = 0; mz < 4; mz++) {
+      for (size_t k = 1; k <= 80; k += 17) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(2)
+          .sr(4)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(331)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C2S4__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(2)
+      .sr(4)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
   TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4__NEON_MULL_PADAL_DUP, k_eq_8) {
     TEST_REQUIRES_ARM_NEON;
     GemmMicrokernelTester()
@@ -7511,1878 +14999,6 @@
 
 
 #if XNN_ARCH_ARM || XNN_ARCH_ARM64
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_eq_16) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(16)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(16)
-      .cn_stride(11)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_eq_16_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t m = 1; m <= 1; m++) {
-      for (uint32_t n = 1; n <= 8; n++) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(m)
-          .n(n)
-          .k(16)
-          .iterations(1)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t m = 1; m <= 1; m++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(m)
-        .n(8)
-        .k(16)
-        .iterations(1)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 1; n <= 8; n++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(n)
-        .k(16)
-        .iterations(1)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_lt_16) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_lt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 16; k++) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_gt_16) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 17; k < 32; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_gt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 17; k < 32; k++) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_div_16) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 32; k <= 160; k += 16) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, k_div_16_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 32; k <= 160; k += 16) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, n_gt_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(8)
-          .k(k)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(8)
-          .k(k)
-          .cn_stride(11)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, n_gt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        for (uint32_t m = 1; m <= 1; m++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, n_div_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(8)
-          .k(k)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, n_div_8_strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(n)
-          .k(k)
-          .cn_stride(11)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, n_div_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        for (uint32_t m = 1; m <= 1; m++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, small_kernel) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 80; k += 17) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .ks(3)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, small_kernel_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 80; k += 17) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .ks(3)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, n_gt_8_small_kernel) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(8)
-          .k(k)
-          .ks(3)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, n_div_8_small_kernel) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(8)
-          .k(k)
-          .ks(3)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, strided_cm_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 80; k += 17) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .cm_stride(11)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, a_offset) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 80; k += 17) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .ks(3)
-        .a_offset(83)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, zero) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t mz = 0; mz < 1; mz++) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(8)
-          .k(k)
-          .ks(3)
-          .a_offset(83)
-          .zero_index(mz)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, qmin) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(16)
-      .qmin(128)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, qmax) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(16)
-      .qmax(128)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MLAL_PADAL, strided_cm) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(16)
-      .cm_stride(11)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
-
-
-#if XNN_ARCH_ARM || XNN_ARCH_ARM64
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_eq_16) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(16)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(16)
-      .cn_stride(11)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_eq_16_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t m = 1; m <= 2; m++) {
-      for (uint32_t n = 1; n <= 8; n++) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(m)
-          .n(n)
-          .k(16)
-          .iterations(1)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_eq_16_subtile_m) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t m = 1; m <= 2; m++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(m)
-        .n(8)
-        .k(16)
-        .iterations(1)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_eq_16_subtile_n) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 1; n <= 8; n++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(n)
-        .k(16)
-        .iterations(1)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_lt_16) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_lt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 16; k++) {
-      for (uint32_t m = 1; m <= 2; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_gt_16) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 17; k < 32; k++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_gt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 17; k < 32; k++) {
-      for (uint32_t m = 1; m <= 2; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_div_16) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 32; k <= 160; k += 16) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, k_div_16_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 32; k <= 160; k += 16) {
-      for (uint32_t m = 1; m <= 2; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, n_gt_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(8)
-          .k(k)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, n_gt_8_strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(8)
-          .k(k)
-          .cn_stride(11)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, n_gt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        for (uint32_t m = 1; m <= 2; m++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, n_div_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(8)
-          .k(k)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, n_div_8_strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(n)
-          .k(k)
-          .cn_stride(11)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, n_div_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        for (uint32_t m = 1; m <= 2; m++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, small_kernel) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 80; k += 17) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .ks(3)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, small_kernel_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 80; k += 17) {
-      for (uint32_t m = 1; m <= 2; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .ks(3)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, n_gt_8_small_kernel) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(8)
-          .k(k)
-          .ks(3)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, n_div_8_small_kernel) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(8)
-          .k(k)
-          .ks(3)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, strided_cm_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 80; k += 17) {
-      for (uint32_t m = 1; m <= 2; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .cm_stride(11)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, a_offset) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 80; k += 17) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .ks(3)
-        .a_offset(163)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, zero) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t mz = 0; mz < 2; mz++) {
-      for (size_t k = 1; k <= 80; k += 17) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(8)
-          .k(k)
-          .ks(3)
-          .a_offset(163)
-          .zero_index(mz)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, qmin) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(16)
-      .qmin(128)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, qmax) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(16)
-      .qmax(128)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MLAL_PADAL, strided_cm) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(16)
-      .cm_stride(11)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
-
-
-#if XNN_ARCH_ARM || XNN_ARCH_ARM64
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_eq_8) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(8)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(8)
-      .cn_stride(11)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_eq_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t m = 1; m <= 1; m++) {
-      for (uint32_t n = 1; n <= 8; n++) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(m)
-          .n(n)
-          .k(8)
-          .iterations(1)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_eq_8_subtile_m) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t m = 1; m <= 1; m++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(m)
-        .n(8)
-        .k(8)
-        .iterations(1)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_eq_8_subtile_n) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 1; n <= 8; n++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(n)
-        .k(8)
-        .iterations(1)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_lt_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 8; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_lt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 8; k++) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_gt_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 9; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_gt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 9; k < 16; k++) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_div_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 16; k <= 80; k += 8) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, k_div_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 16; k <= 80; k += 8) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, n_gt_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(8)
-          .k(k)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, n_gt_8_strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(8)
-          .k(k)
-          .cn_stride(11)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, n_gt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        for (uint32_t m = 1; m <= 1; m++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, n_div_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(8)
-          .k(k)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, n_div_8_strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(n)
-          .k(k)
-          .cn_stride(11)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, n_div_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        for (uint32_t m = 1; m <= 1; m++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, small_kernel) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 40; k += 9) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .ks(3)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, small_kernel_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 40; k += 9) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .ks(3)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, n_gt_8_small_kernel) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(8)
-          .k(k)
-          .ks(3)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, n_div_8_small_kernel) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(8)
-          .k(k)
-          .ks(3)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, strided_cm_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 40; k += 9) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .cm_stride(11)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, a_offset) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 40; k += 9) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(1)
-        .n(8)
-        .k(k)
-        .ks(3)
-        .a_offset(43)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, zero) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t mz = 0; mz < 1; mz++) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(1)
-          .n(8)
-          .k(k)
-          .ks(3)
-          .a_offset(43)
-          .zero_index(mz)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, qmin) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(8)
-      .qmin(128)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, qmax) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(8)
-      .qmax(128)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8S4C2__NEON_MULL_PADAL, strided_cm) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(1)
-      .n(8)
-      .k(8)
-      .cm_stride(11)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
-
-
-#if XNN_ARCH_ARM || XNN_ARCH_ARM64
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_eq_8) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(8)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(8)
-      .cn_stride(11)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_eq_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t m = 1; m <= 2; m++) {
-      for (uint32_t n = 1; n <= 8; n++) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(m)
-          .n(n)
-          .k(8)
-          .iterations(1)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_eq_8_subtile_m) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t m = 1; m <= 2; m++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(m)
-        .n(8)
-        .k(8)
-        .iterations(1)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_eq_8_subtile_n) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 1; n <= 8; n++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(n)
-        .k(8)
-        .iterations(1)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_lt_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 8; k++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_lt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k < 8; k++) {
-      for (uint32_t m = 1; m <= 2; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_gt_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 9; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_gt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 9; k < 16; k++) {
-      for (uint32_t m = 1; m <= 2; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_div_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 16; k <= 80; k += 8) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, k_div_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 16; k <= 80; k += 8) {
-      for (uint32_t m = 1; m <= 2; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, n_gt_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(8)
-          .k(k)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, n_gt_8_strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(8)
-          .k(k)
-          .cn_stride(11)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, n_gt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        for (uint32_t m = 1; m <= 2; m++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, n_div_8) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(8)
-          .k(k)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, n_div_8_strided_cn) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(n)
-          .k(k)
-          .cn_stride(11)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, n_div_8_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        for (uint32_t m = 1; m <= 2; m++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, small_kernel) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 40; k += 9) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .ks(3)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, small_kernel_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 40; k += 9) {
-      for (uint32_t m = 1; m <= 2; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .ks(3)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, n_gt_8_small_kernel) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 9; n < 16; n++) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(8)
-          .k(k)
-          .ks(3)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, n_div_8_small_kernel) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t n = 16; n <= 24; n += 8) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(8)
-          .k(k)
-          .ks(3)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, strided_cm_subtile) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 40; k += 9) {
-      for (uint32_t m = 1; m <= 2; m++) {
-        for (uint32_t n = 1; n <= 8; n++) {
-          GemmMicrokernelTester()
-            .mr(2)
-            .nr(8)
-            .kr(2)
-            .sr(4)
-            .m(m)
-            .n(n)
-            .k(k)
-            .cm_stride(11)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, a_offset) {
-    TEST_REQUIRES_ARM_NEON;
-    for (size_t k = 1; k <= 40; k += 9) {
-      GemmMicrokernelTester()
-        .mr(2)
-        .nr(8)
-        .kr(2)
-        .sr(4)
-        .m(2)
-        .n(8)
-        .k(k)
-        .ks(3)
-        .a_offset(83)
-        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, zero) {
-    TEST_REQUIRES_ARM_NEON;
-    for (uint32_t mz = 0; mz < 2; mz++) {
-      for (size_t k = 1; k <= 40; k += 9) {
-        GemmMicrokernelTester()
-          .mr(2)
-          .nr(8)
-          .kr(2)
-          .sr(4)
-          .m(2)
-          .n(8)
-          .k(k)
-          .ks(3)
-          .a_offset(83)
-          .zero_index(mz)
-          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, qmin) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(8)
-      .qmin(128)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, qmax) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(8)
-      .qmax(128)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8S4C2__NEON_MULL_PADAL, strided_cm) {
-    TEST_REQUIRES_ARM_NEON;
-    GemmMicrokernelTester()
-      .mr(2)
-      .nr(8)
-      .kr(2)
-      .sr(4)
-      .m(2)
-      .n(8)
-      .k(8)
-      .cm_stride(11)
-      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
-  }
-#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
-
-
-#if XNN_ARCH_ARM || XNN_ARCH_ARM64
   TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C2__NEON_MULL_PADAL_DUP, k_eq_8) {
     TEST_REQUIRES_ARM_NEON;
     GemmMicrokernelTester()
diff --git a/test/qs8-igemm-minmax-rndnu.yaml b/test/qs8-igemm-minmax-rndnu.yaml
index 14168f3..53695f7 100644
--- a/test/qs8-igemm-minmax-rndnu.yaml
+++ b/test/qs8-igemm-minmax-rndnu.yaml
@@ -3,6 +3,54 @@
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
 - name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4__neon_mull_padal_dup
   init: xnn_init_qs8_conv_minmax_rndnu_neon_params
   k-block: 8
@@ -51,18 +99,6 @@
 - name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4__neon_mlal_padal_dup
   init: xnn_init_qs8_conv_minmax_rndnu_neon_params
   k-block: 16
-- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mlal_padal
-  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
-  k-block: 16
-- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mlal_padal
-  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
-  k-block: 16
-- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8s4c2__neon_mull_padal
-  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
-  k-block: 8
-- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8s4c2__neon_mull_padal
-  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
-  k-block: 8
 - name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2__neon_mull_padal_dup
   init: xnn_init_qs8_conv_minmax_rndnu_neon_params
   k-block: 8