NEON Dot Product implementations of QS8 FP32 c4 GEMM and IGEMM assembly microkernels

PiperOrigin-RevId: 382208248
diff --git a/test/qs8-gemm-minmax-fp32.cc b/test/qs8-gemm-minmax-fp32.cc
index 553a0e0..36a154b 100644
--- a/test/qs8-gemm-minmax-fp32.cc
+++ b/test/qs8-gemm-minmax-fp32.cc
@@ -23,6 +23,1830 @@
 
 
 #if XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, k_eq_4) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(4)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(4)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, k_eq_4_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(4)
+      .a_stride(7)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, k_eq_4_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(m)
+          .n(n)
+          .k(4)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, k_eq_4_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(m)
+        .n(16)
+        .k(4)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, k_eq_4_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(n)
+        .k(4)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, k_lt_4) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 4; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, k_lt_4_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 4; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(7)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, k_lt_4_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 4; 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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, k_gt_4) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 5; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, k_gt_4_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 5; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, k_gt_4_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 5; 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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, k_div_4) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 8; k <= 40; k += 4) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, k_div_4_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 8; k <= 40; k += 4) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(43)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, k_div_4_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 8; k <= 40; k += 4) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(4)
+            .sr(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(1)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, n_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(23)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(4)
+            .sr(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, n_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, n_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(23)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(4)
+            .sr(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k <= 20; k += 5) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(4)
+            .sr(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, qmin) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(4)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, qmax) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(4)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD32, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(4)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, k_lt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, k_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, k_div_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, k_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(1)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, n_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, n_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, n_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, qmin) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, qmax) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_1X16C4__AARCH64_NEONDOT_LD64, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, k_eq_4) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(4)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(4)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, k_eq_4_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(4)
+      .a_stride(7)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, k_eq_4_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(m)
+          .n(n)
+          .k(4)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, k_eq_4_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(m)
+        .n(16)
+        .k(4)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, k_eq_4_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(n)
+        .k(4)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, k_lt_4) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 4; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, k_lt_4_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 4; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(7)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, k_lt_4_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 4; 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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, k_gt_4) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 5; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, k_gt_4_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 5; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, k_gt_4_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 5; 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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, k_div_4) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 8; k <= 40; k += 4) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, k_div_4_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 8; k <= 40; k += 4) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(43)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, k_div_4_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 8; k <= 40; k += 4) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(4)
+            .sr(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, n_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(23)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(4)
+            .sr(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, n_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, n_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(23)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(4)
+            .sr(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k <= 20; k += 5) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(4)
+            .sr(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, qmin) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(4)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, qmax) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(4)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD32, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(4)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_lt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_div_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, n_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, n_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, qmin) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, qmax) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM64
   TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16) {
     TEST_REQUIRES_ARM_NEON_DOT;
     GemmMicrokernelTester()
@@ -478,6 +2302,462 @@
 #endif  // XNN_ARCH_ARM64
 
 
+#if XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_lt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, qmin) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, qmax) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_GEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM64
+
+
 #if XNN_ARCH_ARM || XNN_ARCH_ARM64
   TEST(QS8_GEMM_MINMAX_FP32_1X16__NEON_MLAL_LANE, k_eq_8) {
     TEST_REQUIRES_ARM_NEON;
diff --git a/test/qs8-gemm-minmax-fp32.yaml b/test/qs8-gemm-minmax-fp32.yaml
index d489b28..37cbbb0 100644
--- a/test/qs8-gemm-minmax-fp32.yaml
+++ b/test/qs8-gemm-minmax-fp32.yaml
@@ -3,9 +3,24 @@
 # 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_1x16c4__aarch64_neondot_ld32
+  init: xnn_init_qs8_conv_minmax_fp32_neonv8_params
+  k-block: 4
+- name: xnn_qs8_gemm_minmax_fp32_ukernel_1x16c4__aarch64_neondot_ld64
+  init: xnn_init_qs8_conv_minmax_fp32_neonv8_params
+  k-block: 8
+- name: xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld32
+  init: xnn_init_qs8_conv_minmax_fp32_neonv8_params
+  k-block: 4
+- name: xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64
+  init: xnn_init_qs8_conv_minmax_fp32_neonv8_params
+  k-block: 8
 - name: xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128
   init: xnn_init_qs8_conv_minmax_fp32_neonv8_params
   k-block: 16
+- name: xnn_qs8_gemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55
+  init: xnn_init_qs8_conv_minmax_fp32_neonv8_params
+  k-block: 16
 - name: xnn_qs8_gemm_minmax_fp32_ukernel_1x16__neon_mlal_lane
   init: xnn_init_qs8_conv_minmax_fp32_neon_params
   k-block: 8
diff --git a/test/qs8-gemm-minmax-gemmlowp.cc b/test/qs8-gemm-minmax-gemmlowp.cc
index e724e64..bf6466b 100644
--- a/test/qs8-gemm-minmax-gemmlowp.cc
+++ b/test/qs8-gemm-minmax-gemmlowp.cc
@@ -23,6 +23,2742 @@
 
 
 #if XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_eq_4) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(4)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(4)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_eq_4_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(4)
+      .a_stride(7)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_eq_4_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(m)
+          .n(n)
+          .k(4)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_eq_4_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(m)
+        .n(16)
+        .k(4)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_eq_4_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(n)
+        .k(4)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_lt_4) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 4; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_lt_4_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 4; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(7)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_lt_4_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 4; 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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_gt_4) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 5; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_gt_4_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 5; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_gt_4_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 5; 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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_div_4) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 8; k <= 40; k += 4) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_div_4_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 8; k <= 40; k += 4) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(43)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_div_4_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 8; k <= 40; k += 4) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(4)
+            .sr(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(1)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, n_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(23)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(4)
+            .sr(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, n_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, n_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(23)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        for (uint32_t m = 1; m <= 1; m++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(4)
+            .sr(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k <= 20; k += 5) {
+      for (uint32_t m = 1; m <= 1; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(1)
+            .nr(16)
+            .kr(4)
+            .sr(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, qmin) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(4)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, qmax) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(4)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(4)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+#endif  // XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 1; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(1)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 1; m++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_lt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_div_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(1)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(1)
+        .n(16)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(1)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, n_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, n_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(1)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(1)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, n_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(1)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, qmin) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, qmax) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(1)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(1)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+#endif  // XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_eq_4) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(4)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(4)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_eq_4_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(4)
+      .a_stride(7)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_eq_4_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(m)
+          .n(n)
+          .k(4)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_eq_4_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(m)
+        .n(16)
+        .k(4)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_eq_4_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(n)
+        .k(4)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_lt_4) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 4; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_lt_4_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 4; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(7)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_lt_4_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 4; 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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_gt_4) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 5; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_gt_4_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 5; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_gt_4_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 5; 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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_div_4) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 8; k <= 40; k += 4) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_div_4_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 8; k <= 40; k += 4) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(43)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_div_4_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 8; k <= 40; k += 4) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(4)
+            .sr(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, n_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(23)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 17; n < 32; n++) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(4)
+            .sr(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, n_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, n_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(23)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 32; n <= 48; n += 16) {
+      for (size_t k = 1; k <= 20; k += 5) {
+        for (uint32_t m = 1; m <= 4; m++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(4)
+            .sr(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k <= 20; k += 5) {
+      for (uint32_t m = 1; m <= 4; m++) {
+        for (uint32_t n = 1; n <= 16; n++) {
+          GemmMicrokernelTester()
+            .mr(4)
+            .nr(16)
+            .kr(4)
+            .sr(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, qmin) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(4)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, qmax) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(4)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(4)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+#endif  // XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .a_stride(11)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_lt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(11)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_gt_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_div_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_div_8_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(83)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(43)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, qmin) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, qmax) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+#endif  // XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_lt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, qmin) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, qmax) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+#endif  // XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM64
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_eq_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .a_stride(19)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_lt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(19)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(37)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .a_stride(163)
+        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_gt_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_div_16_strided_a) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .a_stride(83)
+          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, qmin) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, qmax) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+#endif  // XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM64
   TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16__AARCH64_NEON_MLAL_LANE_PRFM_CORTEX_A53, k_eq_8) {
     TEST_REQUIRES_ARM_NEON;
     GemmMicrokernelTester()
@@ -40150,2742 +42886,6 @@
 #endif  // XNN_ARCH_ARM64
 
 
-#if XNN_ARCH_ARM64
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_eq_4) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(1)
-      .n(16)
-      .k(4)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(1)
-      .n(16)
-      .k(4)
-      .cn_stride(19)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_eq_4_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(1)
-      .n(16)
-      .k(4)
-      .a_stride(7)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_eq_4_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t m = 1; m <= 1; m++) {
-      for (uint32_t n = 1; n <= 16; n++) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(m)
-          .n(n)
-          .k(4)
-          .iterations(1)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_eq_4_subtile_m) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t m = 1; m <= 1; m++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(m)
-        .n(16)
-        .k(4)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_eq_4_subtile_n) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 1; n <= 16; n++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(1)
-        .n(n)
-        .k(4)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_lt_4) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k < 4; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(1)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_lt_4_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k < 4; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(1)
-        .n(16)
-        .k(k)
-        .a_stride(7)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_lt_4_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k < 4; 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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_gt_4) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 5; k < 8; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(1)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_gt_4_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 5; k < 8; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(1)
-        .n(16)
-        .k(k)
-        .a_stride(11)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_gt_4_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 5; 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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_div_4) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 8; k <= 40; k += 4) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(1)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_div_4_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 8; k <= 40; k += 4) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(1)
-        .n(16)
-        .k(k)
-        .a_stride(43)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, k_div_4_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 8; k <= 40; k += 4) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 16; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(16)
-            .kr(4)
-            .sr(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, n_gt_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 17; n < 32; n++) {
-      for (size_t k = 1; k <= 20; k += 5) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(1)
-          .n(16)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, n_gt_16_strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 17; n < 32; n++) {
-      for (size_t k = 1; k <= 20; k += 5) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(1)
-          .n(16)
-          .k(k)
-          .cn_stride(19)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, n_gt_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 17; n < 32; n++) {
-      for (size_t k = 1; k <= 20; k += 5) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(1)
-          .n(n)
-          .k(k)
-          .a_stride(23)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, n_gt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 17; n < 32; n++) {
-      for (size_t k = 1; k <= 20; k += 5) {
-        for (uint32_t m = 1; m <= 1; m++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(16)
-            .kr(4)
-            .sr(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, n_div_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 32; n <= 48; n += 16) {
-      for (size_t k = 1; k <= 20; k += 5) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(1)
-          .n(16)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, n_div_16_strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 32; n <= 48; n += 16) {
-      for (size_t k = 1; k <= 20; k += 5) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(1)
-          .n(n)
-          .k(k)
-          .cn_stride(19)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, n_div_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 32; n <= 48; n += 16) {
-      for (size_t k = 1; k <= 20; k += 5) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(1)
-          .n(n)
-          .k(k)
-          .a_stride(23)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, n_div_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 32; n <= 48; n += 16) {
-      for (size_t k = 1; k <= 20; k += 5) {
-        for (uint32_t m = 1; m <= 1; m++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(16)
-            .kr(4)
-            .sr(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, strided_cm_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k <= 20; k += 5) {
-      for (uint32_t m = 1; m <= 1; m++) {
-        for (uint32_t n = 1; n <= 16; n++) {
-          GemmMicrokernelTester()
-            .mr(1)
-            .nr(16)
-            .kr(4)
-            .sr(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .cm_stride(19)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, qmin) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(1)
-      .n(16)
-      .k(4)
-      .qmin(128)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, qmax) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(1)
-      .n(16)
-      .k(4)
-      .qmax(128)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD32, strided_cm) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(1)
-      .n(16)
-      .k(4)
-      .cm_stride(19)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-#endif  // XNN_ARCH_ARM64
-
-
-#if XNN_ARCH_ARM64
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_eq_8) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(1)
-      .n(16)
-      .k(8)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(1)
-      .n(16)
-      .k(8)
-      .cn_stride(19)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_eq_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(1)
-      .n(16)
-      .k(8)
-      .a_stride(11)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t m = 1; m <= 1; m++) {
-      for (uint32_t n = 1; n <= 16; n++) {
-        GemmMicrokernelTester()
-          .mr(1)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(m)
-          .n(n)
-          .k(8)
-          .iterations(1)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile_m) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t m = 1; m <= 1; m++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(m)
-        .n(16)
-        .k(8)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile_n) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 1; n <= 16; n++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(1)
-        .n(n)
-        .k(8)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_lt_8) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k < 8; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(1)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_lt_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k < 8; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(1)
-        .n(16)
-        .k(k)
-        .a_stride(11)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_lt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_gt_8) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 9; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(1)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_gt_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 9; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(1)
-        .n(16)
-        .k(k)
-        .a_stride(19)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_gt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_div_8) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 16; k <= 80; k += 8) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(1)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_div_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 16; k <= 80; k += 8) {
-      GemmMicrokernelTester()
-        .mr(1)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(1)
-        .n(16)
-        .k(k)
-        .a_stride(83)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, k_div_8_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, n_gt_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(1)
-          .n(16)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, n_gt_16_strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(1)
-          .n(16)
-          .k(k)
-          .cn_stride(19)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, n_gt_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(1)
-          .n(n)
-          .k(k)
-          .a_stride(43)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, n_gt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, n_div_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(1)
-          .n(16)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, n_div_16_strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(1)
-          .n(n)
-          .k(k)
-          .cn_stride(19)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, n_div_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(1)
-          .n(n)
-          .k(k)
-          .a_stride(43)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, n_div_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, strided_cm_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .cm_stride(19)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, qmin) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(1)
-      .n(16)
-      .k(8)
-      .qmin(128)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, qmax) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(1)
-      .n(16)
-      .k(8)
-      .qmax(128)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X16C4__AARCH64_NEONDOT_LD64, strided_cm) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(1)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(1)
-      .n(16)
-      .k(8)
-      .cm_stride(19)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-#endif  // XNN_ARCH_ARM64
-
-
-#if XNN_ARCH_ARM64
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_eq_4) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(4)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(4)
-      .cn_stride(19)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_eq_4_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(4)
-      .a_stride(7)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_eq_4_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t m = 1; m <= 4; m++) {
-      for (uint32_t n = 1; n <= 16; n++) {
-        GemmMicrokernelTester()
-          .mr(4)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(m)
-          .n(n)
-          .k(4)
-          .iterations(1)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_eq_4_subtile_m) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t m = 1; m <= 4; m++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(m)
-        .n(16)
-        .k(4)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_eq_4_subtile_n) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 1; n <= 16; n++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(n)
-        .k(4)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_lt_4) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k < 4; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_lt_4_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k < 4; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .a_stride(7)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_lt_4_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k < 4; 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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_gt_4) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 5; k < 8; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_gt_4_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 5; k < 8; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .a_stride(11)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_gt_4_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 5; 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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_div_4) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 8; k <= 40; k += 4) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_div_4_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 8; k <= 40; k += 4) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .a_stride(43)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, k_div_4_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 8; k <= 40; k += 4) {
-      for (uint32_t m = 1; m <= 4; m++) {
-        for (uint32_t n = 1; n <= 16; n++) {
-          GemmMicrokernelTester()
-            .mr(4)
-            .nr(16)
-            .kr(4)
-            .sr(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, n_gt_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 17; n < 32; n++) {
-      for (size_t k = 1; k <= 20; k += 5) {
-        GemmMicrokernelTester()
-          .mr(4)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, n_gt_16_strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 17; n < 32; n++) {
-      for (size_t k = 1; k <= 20; k += 5) {
-        GemmMicrokernelTester()
-          .mr(4)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .cn_stride(19)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, n_gt_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 17; n < 32; n++) {
-      for (size_t k = 1; k <= 20; k += 5) {
-        GemmMicrokernelTester()
-          .mr(4)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(4)
-          .n(n)
-          .k(k)
-          .a_stride(23)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, n_gt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 17; n < 32; n++) {
-      for (size_t k = 1; k <= 20; k += 5) {
-        for (uint32_t m = 1; m <= 4; m++) {
-          GemmMicrokernelTester()
-            .mr(4)
-            .nr(16)
-            .kr(4)
-            .sr(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, n_div_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 32; n <= 48; n += 16) {
-      for (size_t k = 1; k <= 20; k += 5) {
-        GemmMicrokernelTester()
-          .mr(4)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, n_div_16_strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 32; n <= 48; n += 16) {
-      for (size_t k = 1; k <= 20; k += 5) {
-        GemmMicrokernelTester()
-          .mr(4)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(4)
-          .n(n)
-          .k(k)
-          .cn_stride(19)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, n_div_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 32; n <= 48; n += 16) {
-      for (size_t k = 1; k <= 20; k += 5) {
-        GemmMicrokernelTester()
-          .mr(4)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(4)
-          .n(n)
-          .k(k)
-          .a_stride(23)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, n_div_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 32; n <= 48; n += 16) {
-      for (size_t k = 1; k <= 20; k += 5) {
-        for (uint32_t m = 1; m <= 4; m++) {
-          GemmMicrokernelTester()
-            .mr(4)
-            .nr(16)
-            .kr(4)
-            .sr(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, strided_cm_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k <= 20; k += 5) {
-      for (uint32_t m = 1; m <= 4; m++) {
-        for (uint32_t n = 1; n <= 16; n++) {
-          GemmMicrokernelTester()
-            .mr(4)
-            .nr(16)
-            .kr(4)
-            .sr(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .cm_stride(19)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, qmin) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(4)
-      .qmin(128)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, qmax) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(4)
-      .qmax(128)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD32, strided_cm) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(4)
-      .cm_stride(19)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-#endif  // XNN_ARCH_ARM64
-
-
-#if XNN_ARCH_ARM64
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(8)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(8)
-      .cn_stride(19)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(8)
-      .a_stride(11)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t m = 1; m <= 4; m++) {
-      for (uint32_t n = 1; n <= 16; n++) {
-        GemmMicrokernelTester()
-          .mr(4)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(m)
-          .n(n)
-          .k(8)
-          .iterations(1)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile_m) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t m = 1; m <= 4; m++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(m)
-        .n(16)
-        .k(8)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile_n) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 1; n <= 16; n++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(n)
-        .k(8)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_lt_8) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k < 8; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_lt_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k < 8; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .a_stride(11)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_lt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_gt_8) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 9; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_gt_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 9; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .a_stride(19)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_gt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_div_8) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 16; k <= 80; k += 8) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_div_8_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 16; k <= 80; k += 8) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .a_stride(83)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_div_8_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16_strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .cn_stride(19)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(n)
-          .k(k)
-          .a_stride(43)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_div_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_div_16_strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(n)
-          .k(k)
-          .cn_stride(19)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_div_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(n)
-          .k(k)
-          .a_stride(43)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_div_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, strided_cm_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .cm_stride(19)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, qmin) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(8)
-      .qmin(128)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, qmax) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(8)
-      .qmax(128)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, strided_cm) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(8)
-      .cm_stride(19)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-#endif  // XNN_ARCH_ARM64
-
-
-#if XNN_ARCH_ARM64
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(16)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(16)
-      .cn_stride(19)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(16)
-      .a_stride(19)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t m = 1; m <= 4; m++) {
-      for (uint32_t n = 1; n <= 16; n++) {
-        GemmMicrokernelTester()
-          .mr(4)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(m)
-          .n(n)
-          .k(16)
-          .iterations(1)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16_subtile_m) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t m = 1; m <= 4; m++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(m)
-        .n(16)
-        .k(16)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16_subtile_n) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 1; n <= 16; n++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(n)
-        .k(16)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_lt_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_lt_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .a_stride(19)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_lt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_gt_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 17; k < 32; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_gt_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 17; k < 32; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .a_stride(37)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_gt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_div_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 32; k <= 160; k += 16) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_div_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 32; k <= 160; k += 16) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .a_stride(163)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_div_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16_strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .cn_stride(19)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(n)
-          .k(k)
-          .a_stride(83)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_div_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_div_16_strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(n)
-          .k(k)
-          .cn_stride(19)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_div_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(n)
-          .k(k)
-          .a_stride(83)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_div_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, strided_cm_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .cm_stride(19)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, qmin) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(16)
-      .qmin(128)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, qmax) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(16)
-      .qmax(128)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, strided_cm) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(16)
-      .cm_stride(19)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-#endif  // XNN_ARCH_ARM64
-
-
-#if XNN_ARCH_ARM64
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_eq_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(16)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(16)
-      .cn_stride(19)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_eq_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(16)
-      .a_stride(19)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_eq_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t m = 1; m <= 4; m++) {
-      for (uint32_t n = 1; n <= 16; n++) {
-        GemmMicrokernelTester()
-          .mr(4)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(m)
-          .n(n)
-          .k(16)
-          .iterations(1)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_eq_16_subtile_m) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t m = 1; m <= 4; m++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(m)
-        .n(16)
-        .k(16)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_eq_16_subtile_n) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 1; n <= 16; n++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(n)
-        .k(16)
-        .iterations(1)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_lt_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_lt_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .a_stride(19)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_lt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_gt_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 17; k < 32; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_gt_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 17; k < 32; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .a_stride(37)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_gt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_div_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 32; k <= 160; k += 16) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_div_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 32; k <= 160; k += 16) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .a_stride(163)
-        .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_div_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_gt_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_gt_16_strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .cn_stride(19)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_gt_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(n)
-          .k(k)
-          .a_stride(83)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_gt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_div_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_div_16_strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(n)
-          .k(k)
-          .cn_stride(19)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_div_16_strided_a) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(n)
-          .k(k)
-          .a_stride(83)
-          .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_div_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, strided_cm_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .cm_stride(19)
-            .iterations(1)
-            .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, qmin) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(16)
-      .qmin(128)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, qmax) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(16)
-      .qmax(128)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_GEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_CORTEX_A55, strided_cm) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(16)
-      .cm_stride(19)
-      .Test(xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-#endif  // XNN_ARCH_ARM64
-
-
 #if XNN_ARCH_X86 || XNN_ARCH_X86_64
   TEST(QS8_GEMM_MINMAX_GEMMLOWP_1X4C2__SSE2_LD64, k_eq_8) {
     TEST_REQUIRES_X86_SSE2;
diff --git a/test/qs8-gemm-minmax-gemmlowp.yaml b/test/qs8-gemm-minmax-gemmlowp.yaml
index a700b01..878e34a 100644
--- a/test/qs8-gemm-minmax-gemmlowp.yaml
+++ b/test/qs8-gemm-minmax-gemmlowp.yaml
@@ -3,6 +3,24 @@
 # 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_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32
+  init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
+  k-block: 4
+- name: xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64
+  init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
+  k-block: 8
+- name: xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32
+  init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
+  k-block: 4
+- name: xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64
+  init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
+  k-block: 8
+- name: xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128
+  init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
+  k-block: 16
+- name: xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55
+  init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
+  k-block: 16
 - name: xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16__aarch64_neon_mlal_lane_prfm_cortex_a53
   init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
   k-block: 8
@@ -267,24 +285,6 @@
 - name: xnn_qs8_gemm_minmax_gemmlowp_ukernel_2x8c16__aarch64_neon_mlal_padal
   init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
   k-block: 16
-- name: xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld32
-  init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
-  k-block: 4
-- name: xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x16c4__aarch64_neondot_ld64
-  init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
-  k-block: 8
-- name: xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld32
-  init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
-  k-block: 4
-- name: xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64
-  init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
-  k-block: 8
-- name: xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128
-  init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
-  k-block: 16
-- name: xnn_qs8_gemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55
-  init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
-  k-block: 16
 - name: xnn_qs8_gemm_minmax_gemmlowp_ukernel_1x4c2__sse2_ld64
   init: xnn_init_qs8_conv_minmax_gemmlowp_sse2_params
   k-block: 8
diff --git a/test/qs8-igemm-minmax-fp32.cc b/test/qs8-igemm-minmax-fp32.cc
index 8f6b855..4db5f4e 100644
--- a/test/qs8-igemm-minmax-fp32.cc
+++ b/test/qs8-igemm-minmax-fp32.cc
@@ -22,6 +22,1410 @@
 #include "gemm-microkernel-tester.h"
 
 
+#if XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, small_kernel) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_gt_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, n_div_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, a_offset) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(331)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, zero) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(331)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, qmin) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, qmax) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_CORTEX_A55, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_cortex_a55, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_div_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, n_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, small_kernel) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, n_div_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, a_offset) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(163)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, zero) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(163)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, qmin) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, qmax) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD64, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, k_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, n_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, small_kernel) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, n_div_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, a_offset) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(331)
+        .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, zero) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(331)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, qmin) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, qmax) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_FP32_4X16C4__AARCH64_NEONDOT_LD128, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_fp32_neonv8_params, xnn_init_qs8_requantization_fp32_params, xnn_qs8_requantize_fp32);
+  }
+#endif  // XNN_ARCH_ARM64
+
+
 #if XNN_ARCH_ARM || XNN_ARCH_ARM64
   TEST(QS8_IGEMM_MINMAX_FP32_1X16__NEON_MLAL_LANE, k_eq_8) {
     TEST_REQUIRES_ARM_NEON;
diff --git a/test/qs8-igemm-minmax-fp32.yaml b/test/qs8-igemm-minmax-fp32.yaml
index 24f7599..7861808 100644
--- a/test/qs8-igemm-minmax-fp32.yaml
+++ b/test/qs8-igemm-minmax-fp32.yaml
@@ -3,6 +3,15 @@
 # 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_4x16c4__aarch64_neondot_cortex_a55
+  init: xnn_init_qs8_conv_minmax_fp32_neonv8_params
+  k-block: 16
+- name: xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld64
+  init: xnn_init_qs8_conv_minmax_fp32_neonv8_params
+  k-block: 8
+- name: xnn_qs8_igemm_minmax_fp32_ukernel_4x16c4__aarch64_neondot_ld128
+  init: xnn_init_qs8_conv_minmax_fp32_neonv8_params
+  k-block: 16
 - name: xnn_qs8_igemm_minmax_fp32_ukernel_1x16__neon_mlal_lane
   init: xnn_init_qs8_conv_minmax_fp32_neon_params
   k-block: 8
diff --git a/test/qs8-igemm-minmax-gemmlowp.cc b/test/qs8-igemm-minmax-gemmlowp.cc
index 590d5e9..fb8862e 100644
--- a/test/qs8-igemm-minmax-gemmlowp.cc
+++ b/test/qs8-igemm-minmax-gemmlowp.cc
@@ -491,6 +491,942 @@
 
 
 #if XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(m)
+          .n(n)
+          .k(8)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(m)
+        .n(16)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(n)
+        .k(8)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_lt_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 8; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_lt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_gt_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 9; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_gt_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_div_8) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 16; k <= 80; k += 8) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_div_8_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, small_kernel) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_div_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, a_offset) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k <= 40; k += 9) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(163)
+        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, zero) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(163)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, qmin) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, qmax) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(8)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+#endif  // XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM64
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cn_stride(19)
+      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      for (uint32_t n = 1; n <= 16; n++) {
+        GemmMicrokernelTester()
+          .mr(4)
+          .nr(16)
+          .kr(4)
+          .sr(1)
+          .m(m)
+          .n(n)
+          .k(16)
+          .iterations(1)
+          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16_subtile_m) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t m = 1; m <= 4; m++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(m)
+        .n(16)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16_subtile_n) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (uint32_t n = 1; n <= 16; n++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(n)
+        .k(16)
+        .iterations(1)
+        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_lt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k < 16; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_lt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 17; k < 32; k++) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 32; k <= 160; k += 16) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_div_16) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_div_16_strided_cn) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(n)
+          .k(k)
+          .cn_stride(19)
+          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_div_16_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, small_kernel) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, small_kernel_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .ks(3)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_div_16_small_kernel) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, strided_cm_subtile) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+            .m(m)
+            .n(n)
+            .k(k)
+            .cm_stride(19)
+            .iterations(1)
+            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+        }
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, a_offset) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    for (size_t k = 1; k <= 80; k += 17) {
+      GemmMicrokernelTester()
+        .mr(4)
+        .nr(16)
+        .kr(4)
+        .sr(1)
+        .m(4)
+        .n(16)
+        .k(k)
+        .ks(3)
+        .a_offset(331)
+        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, zero) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    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(1)
+          .m(4)
+          .n(16)
+          .k(k)
+          .ks(3)
+          .a_offset(331)
+          .zero_index(mz)
+          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+      }
+    }
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, qmin) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmin(128)
+      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, qmax) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .qmax(128)
+      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+
+  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, strided_cm) {
+    TEST_REQUIRES_ARM_NEON_DOT;
+    GemmMicrokernelTester()
+      .mr(4)
+      .nr(16)
+      .kr(4)
+      .sr(1)
+      .m(4)
+      .n(16)
+      .k(16)
+      .cm_stride(19)
+      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
+  }
+#endif  // XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM64
   TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16__AARCH64_NEON_MLAL_LANE_PRFM_CORTEX_A53, k_eq_8) {
     TEST_REQUIRES_ARM_NEON;
     GemmMicrokernelTester()
@@ -41206,942 +42142,6 @@
 #endif  // XNN_ARCH_ARM && !XNN_PLATFORM_IOS || XNN_ARCH_ARM64
 
 
-#if XNN_ARCH_ARM64
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(8)
-      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(8)
-      .cn_stride(19)
-      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t m = 1; m <= 4; m++) {
-      for (uint32_t n = 1; n <= 16; n++) {
-        GemmMicrokernelTester()
-          .mr(4)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(m)
-          .n(n)
-          .k(8)
-          .iterations(1)
-          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile_m) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t m = 1; m <= 4; m++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(m)
-        .n(16)
-        .k(8)
-        .iterations(1)
-        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_eq_8_subtile_n) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 1; n <= 16; n++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(n)
-        .k(8)
-        .iterations(1)
-        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_lt_8) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k < 8; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_lt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_gt_8) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 9; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_gt_8_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_div_8) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 16; k <= 80; k += 8) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, k_div_8_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16_strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .cn_stride(19)
-          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_div_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_div_16_strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(n)
-          .k(k)
-          .cn_stride(19)
-          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_div_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, small_kernel) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k <= 40; k += 9) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .ks(3)
-        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, small_kernel_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .ks(3)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_gt_16_small_kernel) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .ks(3)
-          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, n_div_16_small_kernel) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .ks(3)
-          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, strided_cm_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .cm_stride(19)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, a_offset) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k <= 40; k += 9) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .ks(3)
-        .a_offset(163)
-        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, zero) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .ks(3)
-          .a_offset(163)
-          .zero_index(mz)
-          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, qmin) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(8)
-      .qmin(128)
-      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, qmax) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(8)
-      .qmax(128)
-      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD64, strided_cm) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(8)
-      .cm_stride(19)
-      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-#endif  // XNN_ARCH_ARM64
-
-
-#if XNN_ARCH_ARM64
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(16)
-      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(16)
-      .cn_stride(19)
-      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t m = 1; m <= 4; m++) {
-      for (uint32_t n = 1; n <= 16; n++) {
-        GemmMicrokernelTester()
-          .mr(4)
-          .nr(16)
-          .kr(4)
-          .sr(1)
-          .m(m)
-          .n(n)
-          .k(16)
-          .iterations(1)
-          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16_subtile_m) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t m = 1; m <= 4; m++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(m)
-        .n(16)
-        .k(16)
-        .iterations(1)
-        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_eq_16_subtile_n) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (uint32_t n = 1; n <= 16; n++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(n)
-        .k(16)
-        .iterations(1)
-        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_lt_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k < 16; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_lt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_gt_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 17; k < 32; k++) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_gt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_div_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 32; k <= 160; k += 16) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, k_div_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16_strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .cn_stride(19)
-          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_div_16) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_div_16_strided_cn) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(n)
-          .k(k)
-          .cn_stride(19)
-          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_div_16_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, small_kernel) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k <= 80; k += 17) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .ks(3)
-        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, small_kernel_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .ks(3)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_gt_16_small_kernel) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .ks(3)
-          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, n_div_16_small_kernel) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .ks(3)
-          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, strided_cm_subtile) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-            .m(m)
-            .n(n)
-            .k(k)
-            .cm_stride(19)
-            .iterations(1)
-            .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-        }
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, a_offset) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    for (size_t k = 1; k <= 80; k += 17) {
-      GemmMicrokernelTester()
-        .mr(4)
-        .nr(16)
-        .kr(4)
-        .sr(1)
-        .m(4)
-        .n(16)
-        .k(k)
-        .ks(3)
-        .a_offset(331)
-        .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, zero) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    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(1)
-          .m(4)
-          .n(16)
-          .k(k)
-          .ks(3)
-          .a_offset(331)
-          .zero_index(mz)
-          .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-      }
-    }
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, qmin) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(16)
-      .qmin(128)
-      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, qmax) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(16)
-      .qmax(128)
-      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-
-  TEST(QS8_IGEMM_MINMAX_GEMMLOWP_4X16C4__AARCH64_NEONDOT_LD128, strided_cm) {
-    TEST_REQUIRES_ARM_NEON_DOT;
-    GemmMicrokernelTester()
-      .mr(4)
-      .nr(16)
-      .kr(4)
-      .sr(1)
-      .m(4)
-      .n(16)
-      .k(16)
-      .cm_stride(19)
-      .Test(xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128, xnn_init_qs8_conv_minmax_gemmlowp_neon_params, xnn_init_qs8_requantization_gemmlowp_params, xnn_qs8_requantize_gemmlowp);
-  }
-#endif  // XNN_ARCH_ARM64
-
-
 #if XNN_ARCH_X86 || XNN_ARCH_X86_64
   TEST(QS8_IGEMM_MINMAX_GEMMLOWP_1X4C2__SSE2_LD64, k_eq_8) {
     TEST_REQUIRES_X86_SSE2;
diff --git a/test/qs8-igemm-minmax-gemmlowp.yaml b/test/qs8-igemm-minmax-gemmlowp.yaml
index 86c7eb5..348ae0e 100644
--- a/test/qs8-igemm-minmax-gemmlowp.yaml
+++ b/test/qs8-igemm-minmax-gemmlowp.yaml
@@ -6,6 +6,12 @@
 - name: xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_cortex_a55
   init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
   k-block: 16
+- name: xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64
+  init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
+  k-block: 8
+- name: xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128
+  init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
+  k-block: 16
 - name: xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16__aarch64_neon_mlal_lane_prfm_cortex_a53
   init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
   k-block: 8
@@ -267,12 +273,6 @@
 - name: xnn_qs8_igemm_minmax_gemmlowp_ukernel_8x16c4__neondot
   init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
   k-block: 8
-- name: xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld64
-  init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
-  k-block: 8
-- name: xnn_qs8_igemm_minmax_gemmlowp_ukernel_4x16c4__aarch64_neondot_ld128
-  init: xnn_init_qs8_conv_minmax_gemmlowp_neon_params
-  k-block: 16
 - name: xnn_qs8_igemm_minmax_gemmlowp_ukernel_1x4c2__sse2_ld64
   init: xnn_init_qs8_conv_minmax_gemmlowp_sse2_params
   k-block: 8