commit 14d3ce8fbce5ebdb9061628c32264e97c311fdd0
author Marat Dukhan <maratek@google.com> Fri Jul 31 16:19:15 2020 -0700
committer XNNPACK Team <xnnpack-github-robot@google.com> Fri Jul 31 16:19:58 2020 -0700
tree c33776a9a5ea1ac180c1d7b34286f496420811fa
parent 733d0be4f4bdc02daba3b676d05825f4d563cb87

Add LD64 suffix in QS8 GEMM/IGEMM microkernels

LD64 denotes that weights are loaded 64 bits at a time and sign-extended to 128
bits

PiperOrigin-RevId: 324305250

bench/qs8-gemm.cc

diff --git a/bench/qs8-gemm.cc b/bench/qs8-gemm.cc
index a6ef819..f620dc8 100644
--- a/bench/qs8-gemm.cc
+++ b/bench/qs8-gemm.cc
@@ -102,33 +102,33 @@
 
 
 #if XNN_ARCH_X86 || XNN_ARCH_X86_64
-  static void qs8_gemm_4x4c2__sse2(benchmark::State& state, const char* net) {
-    GEMMBenchmark(state, xnn_qs8_gemm_minmax_ukernel_4x4c2__sse2, 4, 4, 2, 1);
+  static void qs8_gemm_4x4c2__sse2_ld64(benchmark::State& state, const char* net) {
+    GEMMBenchmark(state, xnn_qs8_gemm_minmax_ukernel_4x4c2__sse2_ld64, 4, 4, 2, 1);
   }
-  static void qs8_gemm_4x4c2__ssse3(benchmark::State& state, const char* net) {
-    GEMMBenchmark(state, xnn_qs8_gemm_minmax_ukernel_4x4c2__ssse3, 4, 4, 2, 1);
+  static void qs8_gemm_4x4c2__ssse3_ld64(benchmark::State& state, const char* net) {
+    GEMMBenchmark(state, xnn_qs8_gemm_minmax_ukernel_4x4c2__ssse3_ld64, 4, 4, 2, 1);
   }
-  static void qs8_gemm_4x4c2__sse41(benchmark::State& state, const char* net) {
-    GEMMBenchmark(state, xnn_qs8_gemm_minmax_ukernel_4x4c2__sse41, 4, 4, 2, 1);
+  static void qs8_gemm_4x4c2__sse41_ld64(benchmark::State& state, const char* net) {
+    GEMMBenchmark(state, xnn_qs8_gemm_minmax_ukernel_4x4c2__sse41_ld64, 4, 4, 2, 1);
   }
 
-  static void qs8_gemm_2x4c8__sse2(benchmark::State& state, const char* net) {
-    GEMMBenchmark(state, xnn_qs8_gemm_minmax_ukernel_2x4c8__sse2, 2, 4, 8, 1);
+  static void qs8_gemm_2x4c8__sse2_ld64(benchmark::State& state, const char* net) {
+    GEMMBenchmark(state, xnn_qs8_gemm_minmax_ukernel_2x4c8__sse2_ld64, 2, 4, 8, 1);
   }
-  static void qs8_gemm_2x4c8__ssse3(benchmark::State& state, const char* net) {
-    GEMMBenchmark(state, xnn_qs8_gemm_minmax_ukernel_2x4c8__ssse3, 2, 4, 8, 1);
+  static void qs8_gemm_2x4c8__ssse3_ld64(benchmark::State& state, const char* net) {
+    GEMMBenchmark(state, xnn_qs8_gemm_minmax_ukernel_2x4c8__ssse3_ld64, 2, 4, 8, 1);
   }
-  static void qs8_gemm_2x4c8__sse41(benchmark::State& state, const char* net) {
-    GEMMBenchmark(state, xnn_qs8_gemm_minmax_ukernel_2x4c8__sse41, 2, 4, 8, 1);
+  static void qs8_gemm_2x4c8__sse41_ld64(benchmark::State& state, const char* net) {
+    GEMMBenchmark(state, xnn_qs8_gemm_minmax_ukernel_2x4c8__sse41_ld64, 2, 4, 8, 1);
   }
 
-  BENCHMARK_GEMM(qs8_gemm_4x4c2__sse2)
-  BENCHMARK_GEMM(qs8_gemm_4x4c2__ssse3)
-  BENCHMARK_GEMM(qs8_gemm_4x4c2__sse41)
+  BENCHMARK_GEMM(qs8_gemm_4x4c2__sse2_ld64)
+  BENCHMARK_GEMM(qs8_gemm_4x4c2__ssse3_ld64)
+  BENCHMARK_GEMM(qs8_gemm_4x4c2__sse41_ld64)
 
-  BENCHMARK_GEMM(qs8_gemm_2x4c8__sse2)
-  BENCHMARK_GEMM(qs8_gemm_2x4c8__ssse3)
-  BENCHMARK_GEMM(qs8_gemm_2x4c8__sse41)
+  BENCHMARK_GEMM(qs8_gemm_2x4c8__sse2_ld64)
+  BENCHMARK_GEMM(qs8_gemm_2x4c8__ssse3_ld64)
+  BENCHMARK_GEMM(qs8_gemm_2x4c8__sse41_ld64)
 #endif
 
 #ifndef XNNPACK_BENCHMARK_NO_MAIN