QS8 C4S2 Neon GEMM/IGEMM microkernels

PiperOrigin-RevId: 409348021
diff --git a/test/qc8-gemm-minmax-fp32.cc b/test/qc8-gemm-minmax-fp32.cc
index f82e4f7..1071145 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_1X8C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_1X8C4S2__NEONV8_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__NEONV8_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__NEONV8_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__NEONV8_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__NEONV8_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_1X8C4S2__NEONV8_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__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_2X8C4S2__NEONV8_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__NEONV8_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__NEONV8_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__NEONV8_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__NEONV8_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_GEMM_MINMAX_FP32_2X8C4S2__NEONV8_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__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_1X8C2S4__NEON_MLAL_PADAL, 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 b226618..638306e 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_1x8c4s2__neon_mlal_padal
+  init: xnn_init_qs8_minmax_neon_fp32_params
+  k-block: 16
+- name: xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neon_mlal_padal
+  init: xnn_init_qs8_minmax_neon_fp32_params
+  k-block: 16
+- name: xnn_qc8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal
+  init: xnn_init_qs8_minmax_neon_params
+  k-block: 16
+- name: xnn_qc8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal
+  init: xnn_init_qs8_minmax_neon_params
+  k-block: 16
 - name: xnn_qc8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal
   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 8565cd6..cd5f354 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_1X8C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(83)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(83)
+          .zero_index(mz)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(163)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(163)
+          .zero_index(mz)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_1X8C4S2__NEONV8_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__NEONV8_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(83)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(83)
+          .zero_index(mz)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__NEONV8_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__NEONV8_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_1X8C4S2__NEONV8_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__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_2X8C4S2__NEONV8_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__NEONV8_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(163)
+        .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(163)
+          .zero_index(mz)
+          .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__NEONV8_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__NEONV8_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal, xnn_init_qs8_minmax_neon_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QC8_IGEMM_MINMAX_FP32_2X8C4S2__NEONV8_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__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_1X8C2S4__NEON_MLAL_PADAL, 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 b300995..c4d254c 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_1x8c4s2__neon_mlal_padal
+  init: xnn_init_qs8_minmax_neon_fp32_params
+  k-block: 16
+- name: xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neon_mlal_padal
+  init: xnn_init_qs8_minmax_neon_fp32_params
+  k-block: 16
+- name: xnn_qc8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal
+  init: xnn_init_qs8_minmax_neon_params
+  k-block: 16
+- name: xnn_qc8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal
+  init: xnn_init_qs8_minmax_neon_params
+  k-block: 16
 - name: xnn_qc8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal
   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 217e951..6034268 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_1X8C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_1X8C4S2__NEONV8_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEONV8_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEONV8_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEONV8_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEONV8_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEONV8_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__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_2X8C4S2__NEONV8_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEONV8_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEONV8_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEONV8_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEONV8_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEONV8_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__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_1X8C2S4__NEON_MLAL_PADAL, 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 e14b172..25fd06b 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_1x8c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_fp32_neon_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_fp32_neon_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal
+  init: xnn_init_qs8_conv_minmax_fp32_neonv8_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal
+  init: xnn_init_qs8_conv_minmax_fp32_neonv8_params
+  k-block: 16
 - name: xnn_qs8_gemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal
   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 9537fc4..9c24f56 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_1X8C4S2__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(8)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__NEON_MULL_PADAL, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(8)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__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_2X8C4S2__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(8)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__NEON_MULL_PADAL, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(8)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__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_3X8C4S2__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(8)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__NEON_MULL_PADAL, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(8)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__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_4X8C4S2__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(8)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__NEON_MULL_PADAL, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(8)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__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_1X16C4S2__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__NEON_MULL_PADAL, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__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_2X16C4S2__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__NEON_MULL_PADAL, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__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_3X16C4S2__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__NEON_MULL_PADAL, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__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_4X16C4S2__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__NEON_MULL_PADAL, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__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_1X8C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X8C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X8C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__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_3X8C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X8C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__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_4X8C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X8C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__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_1X16C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_1X16C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__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_2X16C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_2X16C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__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_3X16C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_3X16C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__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_4X16C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__NEON_MLAL_PADAL, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_GEMM_MINMAX_RNDNU_4X16C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__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_1X8C2S4__NEON_MULL_PADAL, 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 dc61434..9427b75 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_1x8c4s2__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
 - name: xnn_qs8_gemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal
   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 80d01e0..78f69ec 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_1X8C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(83)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(83)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(163)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(163)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_1X8C4S2__NEONV8_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEONV8_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(83)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(83)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEONV8_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEONV8_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_1X8C4S2__NEONV8_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__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_2X8C4S2__NEONV8_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEONV8_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(163)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(163)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEONV8_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEONV8_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_2X8C4S2__NEONV8_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_V8;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__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_1X8C2S4__NEON_MLAL_PADAL, 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 1024cef..aabba65 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_1x8c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_fp32_neon_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_fp32_neon_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_fp32_ukernel_1x8c4s2__neonv8_mlal_padal
+  init: xnn_init_qs8_conv_minmax_fp32_neonv8_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_fp32_ukernel_2x8c4s2__neonv8_mlal_padal
+  init: xnn_init_qs8_conv_minmax_fp32_neonv8_params
+  k-block: 16
 - name: xnn_qs8_igemm_minmax_fp32_ukernel_1x8c2s4__neon_mlal_padal
   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 8378366..c9e3a9f 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_1X8C4S2__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(8)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(8)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__NEON_MULL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__NEON_MULL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(43)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(43)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(8)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__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_2X8C4S2__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(8)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(8)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__NEON_MULL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__NEON_MULL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(83)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(83)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(8)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__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_3X8C4S2__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(8)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(8)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__NEON_MULL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__NEON_MULL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(127)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(127)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(8)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__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_4X8C4S2__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(8)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(8)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__NEON_MULL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__NEON_MULL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(163)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(163)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(8)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__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_1X16C4S2__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__NEON_MULL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__NEON_MULL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(43)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(43)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__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_2X16C4S2__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__NEON_MULL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__NEON_MULL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(83)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(83)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__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_3X16C4S2__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__NEON_MULL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__NEON_MULL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(127)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(127)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__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_4X16C4S2__NEON_MULL_PADAL, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__NEON_MULL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__NEON_MULL_PADAL, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__NEON_MULL_PADAL, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__NEON_MULL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__NEON_MULL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(163)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(163)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__NEON_MULL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__NEON_MULL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__NEON_MULL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__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_1X8C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(83)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(83)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X8C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__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_2X8C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(163)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(163)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X8C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__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_3X8C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(251)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(251)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X8C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__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_4X8C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(16)
+      .cn_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(8)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(8)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(8)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(11)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(11)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(8)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(8)
+        .k(k)
+        .ks(3)
+        .a_offset(331)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(8)
+          .k(k)
+          .ks(3)
+          .a_offset(331)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X8C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(8)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(8)
+      .k(16)
+      .cm_stride(11)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__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_1X16C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(1)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(83)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__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(4)
+          .sr(2)
+          .m(1)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(83)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_1X16C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(1)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__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_2X16C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(2)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(2)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(163)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__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(4)
+          .sr(2)
+          .m(2)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(163)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_2X16C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(2)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(2)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__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_3X16C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(3)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(3)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(251)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__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(4)
+          .sr(2)
+          .m(3)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(251)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_3X16C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(3)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(3)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__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_4X16C4S2__NEON_MLAL_PADAL, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__NEON_MLAL_PADAL, strided_cn) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+        .sr(2)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__NEON_MLAL_PADAL, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__NEON_MLAL_PADAL, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__NEON_MLAL_PADAL, small_kernel) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+            .sr(2)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__NEON_MLAL_PADAL, a_offset) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(2)
+        .m(4)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(331)
+        .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__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(4)
+          .sr(2)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(331)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__NEON_MLAL_PADAL, qmin) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__NEON_MLAL_PADAL, qmax) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal, xnn_init_qs8_conv_minmax_rndnu_neon_params, xnn_init_qs8_requantization_rndnu_params, xnn_qs8_requantize_rndnu);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_RNDNU_4X16C4S2__NEON_MLAL_PADAL, strided_cm) {
+    TEST_REQUIRES_ARM_NEON;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(2)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__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_1X8C2S4__NEON_MULL_PADAL, 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 53695f7..a4fe4d07 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_1x8c4s2__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 8
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal_padal
+  init: xnn_init_qs8_conv_minmax_rndnu_neon_params
+  k-block: 16
 - name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull_padal
   init: xnn_init_qs8_conv_minmax_rndnu_neon_params
   k-block: 8