commit 387c2d171582e981f40e07aa2fef301e175d5b61
author Frank Barchard <fbarchard@google.com> Mon Dec 16 19:14:07 2019 -0800
committer XNNPACK Team <xnnpack-github-robot@google.com> Mon Dec 16 19:14:40 2019 -0800
tree 49391949a6123358019c52c642fef316083dc7fd
parent 005feb81be10e44004f7a0a4f850791a5b8fd5aa

Generate A57 micro-kernels from A75 source.

Remove A57 source generators and replace with A75 with prefetch removed.
Add missing 4x8 and 5x8 variations.

PiperOrigin-RevId: 285897229

bench/f32-gemm.cc

diff --git a/bench/f32-gemm.cc b/bench/f32-gemm.cc
index 18cdb81..642bddf 100644
--- a/bench/f32-gemm.cc
+++ b/bench/f32-gemm.cc
@@ -378,6 +378,9 @@
   static void f32_gemm_4x8__aarch64_neonfma_ld128(benchmark::State& state, const char* net) {
     GEMMBenchmark(state, xnn_f32_gemm_ukernel_4x8__aarch64_neonfma_ld128, 4, 8, 1, 1);
   }
+  static void f32_gemm_5x8__aarch64_neonfma_cortex_a57(benchmark::State& state, const char* net) {
+    GEMMBenchmark(state, xnn_f32_gemm_ukernel_5x8__aarch64_neonfma_cortex_a57, 5, 8, 1, 1);
+  }
   static void f32_gemm_5x8__aarch64_neonfma_cortex_a75(benchmark::State& state, const char* net) {
     GEMMBenchmark(state, xnn_f32_gemm_ukernel_5x8__aarch64_neonfma_cortex_a75, 5, 8, 1, 1);
   }
@@ -390,12 +393,12 @@
   static void f32_gemm_6x8__aarch64_neonfma_cortex_a53(benchmark::State& state, const char* net) {
     GEMMBenchmark(state, xnn_f32_gemm_ukernel_6x8__aarch64_neonfma_cortex_a53, 6, 8, 1, 1);
   }
-  static void f32_gemm_6x8__aarch64_neonfma_cortex_a57(benchmark::State& state, const char* net) {
-    GEMMBenchmark(state, xnn_f32_gemm_ukernel_6x8__aarch64_neonfma_cortex_a57, 6, 8, 1, 1);
-  }
   static void f32_gemm_6x8__aarch64_neonfma_cortex_a73(benchmark::State& state, const char* net) {
     GEMMBenchmark(state, xnn_f32_gemm_ukernel_6x8__aarch64_neonfma_cortex_a73, 6, 8, 1, 1);
   }
+  static void f32_gemm_6x8__aarch64_neonfma_cortex_a57(benchmark::State& state, const char* net) {
+    GEMMBenchmark(state, xnn_f32_gemm_ukernel_6x8__aarch64_neonfma_cortex_a57, 6, 8, 1, 1);
+  }
   static void f32_gemm_6x8__aarch64_neonfma_cortex_a75(benchmark::State& state, const char* net) {
     GEMMBenchmark(state, xnn_f32_gemm_ukernel_6x8__aarch64_neonfma_cortex_a75, 6, 8, 1, 1);
   }
@@ -427,10 +430,11 @@
   BENCHMARK_GEMM(f32_gemm_4x8__aarch64_neonfma_cortex_a75)
   BENCHMARK_GEMM(f32_gemm_4x8__aarch64_neonfma_ld128)
   BENCHMARK_GEMM(f32_gemm_4x8__aarch64_neonfma_ld64)
+  BENCHMARK_GEMM(f32_gemm_5x8__aarch64_neonfma_cortex_a57)
   BENCHMARK_GEMM(f32_gemm_5x8__aarch64_neonfma_cortex_a75)
   BENCHMARK_GEMM(f32_gemm_6x8__aarch64_neonfma_cortex_a53)
-  BENCHMARK_GEMM(f32_gemm_6x8__aarch64_neonfma_cortex_a57)
   BENCHMARK_GEMM(f32_gemm_6x8__aarch64_neonfma_cortex_a73)
+  BENCHMARK_GEMM(f32_gemm_6x8__aarch64_neonfma_cortex_a57)
   BENCHMARK_GEMM(f32_gemm_6x8__aarch64_neonfma_cortex_a75)
   BENCHMARK_GEMM(f32_gemm_6x8__aarch64_neonfma_ld64)
   BENCHMARK_GEMM(f32_gemm_6x8__aarch64_neonfma_ld128)