commit e349124437bfd08272ccea36cacbd64983f94ba4
author Frank Barchard <fbarchard@google.com> Fri Jun 11 14:04:57 2021 -0700
committer XNNPACK Team <xnnpack-github-robot@google.com> Fri Jun 11 14:05:37 2021 -0700
tree c6823517b057783500c844b28ebb52f8751fd3b4
parent 7228b5e709bc210ab8988be26e258ca70548facb

fp32 IGEMM 4x8 and 6x8 ld64 microkernels

PiperOrigin-RevId: 378945611

bench/f32-igemm.cc

diff --git a/bench/f32-igemm.cc b/bench/f32-igemm.cc
index 6b2abc1..a416ab8 100644
--- a/bench/f32-igemm.cc
+++ b/bench/f32-igemm.cc
@@ -354,6 +354,10 @@
     IGEMMBenchmark(state, xnn_f32_igemm_minmax_ukernel_4x8__aarch64_neonfma_cortex_a75, 4, 8, 1, 1,
       xnn_init_f32_minmax_scalar_params);
   }
+  static void f32_igemm_4x8__aarch64_neonfma_ld64(benchmark::State& state, const char* net) {
+    IGEMMBenchmark(state, xnn_f32_igemm_minmax_ukernel_4x8__aarch64_neonfma_ld64, 4, 8, 1, 1,
+      xnn_init_f32_minmax_scalar_params);
+  }
   static void f32_igemm_5x8__aarch64_neonfma_cortex_a57(benchmark::State& state, const char* net) {
     IGEMMBenchmark(state, xnn_f32_igemm_minmax_ukernel_5x8__aarch64_neonfma_cortex_a57, 5, 8, 1, 1,
       xnn_init_f32_minmax_scalar_params);
@@ -386,6 +390,10 @@
     IGEMMBenchmark(state, xnn_f32_igemm_minmax_ukernel_6x8__aarch64_neonfma_cortex_a75, 6, 8, 1, 1,
       xnn_init_f32_minmax_scalar_params);
   }
+  static void f32_igemm_6x8__aarch64_neonfma_ld64(benchmark::State& state, const char* net) {
+    IGEMMBenchmark(state, xnn_f32_igemm_minmax_ukernel_6x8__aarch64_neonfma_ld64, 6, 8, 1, 1,
+      xnn_init_f32_minmax_scalar_params);
+  }
   static void f32_igemm_1x8__neonfma_lane_ld64(benchmark::State& state, const char* net) {
     IGEMMBenchmark(state, xnn_f32_igemm_minmax_ukernel_1x8__neonfma_lane_ld64, 1, 8, 1, 1,
       xnn_init_f32_minmax_scalar_params);
@@ -424,6 +432,7 @@
   BENCHMARK_CONV(f32_igemm_4x8__aarch64_neonfma_cortex_a55)
   BENCHMARK_CONV(f32_igemm_4x8__aarch64_neonfma_cortex_a57)
   BENCHMARK_CONV(f32_igemm_4x8__aarch64_neonfma_cortex_a75)
+  BENCHMARK_CONV(f32_igemm_4x8__aarch64_neonfma_ld64)
   BENCHMARK_CONV(f32_igemm_5x8__aarch64_neonfma_cortex_a57)
   BENCHMARK_CONV(f32_igemm_5x8__aarch64_neonfma_cortex_a75)
   BENCHMARK_CONV(f32_igemm_6x8__aarch64_neonfma_cortex_a53)
@@ -431,6 +440,7 @@
   BENCHMARK_CONV(f32_igemm_6x8__aarch64_neonfma_cortex_a73)
   BENCHMARK_CONV(f32_igemm_6x8__aarch64_neonfma_cortex_a57)
   BENCHMARK_CONV(f32_igemm_6x8__aarch64_neonfma_cortex_a75)
+  BENCHMARK_CONV(f32_igemm_6x8__aarch64_neonfma_ld64)
   BENCHMARK_CONV(f32_igemm_1x8__neonfma_lane_ld64)
   BENCHMARK_CONV(f32_igemm_4x2__neonfma_lane_ld64)
   BENCHMARK_CONV(f32_igemm_4x4__neonfma_lane_ld64)