commit 4a4be4e96af1fb103b79f2838df6613e9da6bba2
author Frank Barchard <fbarchard@google.com> Mon Mar 01 15:00:28 2021 -0800
committer XNNPACK Team <xnnpack-github-robot@google.com> Mon Mar 01 15:01:14 2021 -0800
tree 94e1f15b7ad8c6b3f85f481406d2bdd2be4aed02
parent 7aa4bfd8f8fb269e71eef31d6180a004d26dec31

QS8 1x16c4 ld32 GEMM microkernel using NEON dot product

PiperOrigin-RevId: 360285164

bench/qs8-gemm.cc

diff --git a/bench/qs8-gemm.cc b/bench/qs8-gemm.cc
index 09779ad..9e740d2 100644
--- a/bench/qs8-gemm.cc
+++ b/bench/qs8-gemm.cc
@@ -516,22 +516,26 @@
 #endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
 
 #if XNN_ARCH_ARM64
+  static void qs8_gemm_4x16c4__aarch64_neondot_cortex_a55(benchmark::State& state, const char* net) {
+    GEMMBenchmark(state, xnn_qs8_gemm_minmax_ukernel_4x16c4__aarch64_neondot_cortex_a55, 4, 16, 4, 1, benchmark::utils::CheckNEONDOT);
+  }
+  static void qs8_gemm_1x16c4__aarch64_neondot_ld32(benchmark::State& state, const char* net) {
+    GEMMBenchmark(state, xnn_qs8_gemm_minmax_ukernel_1x16c4__aarch64_neondot_ld32, 1, 16, 4, 1, benchmark::utils::CheckNEONDOT);
+  }
   static void qs8_gemm_1x16c4__aarch64_neondot_ld64(benchmark::State& state, const char* net) {
     GEMMBenchmark(state, xnn_qs8_gemm_minmax_ukernel_1x16c4__aarch64_neondot_ld64, 1, 16, 4, 1, benchmark::utils::CheckNEONDOT);
   }
-  static void qs8_gemm_4x16c4__aarch64_neondot_ld64(benchmark::State& state, const char* net) {
-    GEMMBenchmark(state, xnn_qs8_gemm_minmax_ukernel_4x16c4__aarch64_neondot_ld64, 4, 16, 4, 1, benchmark::utils::CheckNEONDOT);
-  }
   static void qs8_gemm_4x16c4__aarch64_neondot_ld32(benchmark::State& state, const char* net) {
     GEMMBenchmark(state, xnn_qs8_gemm_minmax_ukernel_4x16c4__aarch64_neondot_ld32, 4, 16, 4, 1, benchmark::utils::CheckNEONDOT);
   }
-  static void qs8_gemm_4x16c4__aarch64_neondot_cortex_a55(benchmark::State& state, const char* net) {
-    GEMMBenchmark(state, xnn_qs8_gemm_minmax_ukernel_4x16c4__aarch64_neondot_cortex_a55, 4, 16, 4, 1, benchmark::utils::CheckNEONDOT);
+  static void qs8_gemm_4x16c4__aarch64_neondot_ld64(benchmark::State& state, const char* net) {
+    GEMMBenchmark(state, xnn_qs8_gemm_minmax_ukernel_4x16c4__aarch64_neondot_ld64, 4, 16, 4, 1, benchmark::utils::CheckNEONDOT);
   }
 
+  BENCHMARK_GEMM(qs8_gemm_1x16c4__aarch64_neondot_ld32)
   BENCHMARK_GEMM(qs8_gemm_1x16c4__aarch64_neondot_ld64)
-  BENCHMARK_GEMM(qs8_gemm_4x16c4__aarch64_neondot_ld64)
   BENCHMARK_GEMM(qs8_gemm_4x16c4__aarch64_neondot_ld32)
+  BENCHMARK_GEMM(qs8_gemm_4x16c4__aarch64_neondot_ld64)
   BENCHMARK_GEMM(qs8_gemm_4x16c4__aarch64_neondot_cortex_a55)
 #endif  // XNN_ARCH_ARM64