Final cleanup for avx2 isolation and consistent file names (pytorch#40)

Summary: Pull Request resolved: pytorch#40 File name changes + removal of -mavx2 compiler flag non-avx files This completes the separation of avx2 code to few files that make minimal use of c++ std lib. Reviewed By: jianyuh Differential Revision: D13330577 fbshipit-source-id: b469ebee484168800ce2d12fd2356edecbf0fa4d
pls331 · Dec 5, 2018 · 3af8fe5 · 3af8fe5
1 parent 7e1b543
commit 3af8fe5
Show file tree

Hide file tree

Showing 10 changed files with 10 additions and 17 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -29,12 +29,11 @@ set(FBGEMM_GENERIC_SRCS src/ExecuteKernel.cc
                 src/ExecuteKernelU8S8.cc
                 src/Fbgemm.cc
                 src/FbgemmFP16.cc
-                src/FbgemmFP16UKernels.cc
                 src/FbgemmI8Spmdm.cc
                 src/GenerateKernelU8S8S32ACC16.cc
-                src/GenerateKernelU8S8S32ACC16_avx512.cc
+                src/GenerateKernelU8S8S32ACC16Avx512.cc
                 src/GenerateKernelU8S8S32ACC32.cc
-                src/GenerateKernelU8S8S32ACC32_avx512.cc
+                src/GenerateKernelU8S8S32ACC32Avx512.cc
                 src/PackAMatrix.cc
                 src/PackAWithIm2Col.cc
                 src/PackBMatrix.cc
@@ -64,13 +63,14 @@ endif()
 
 #All the source files that either use avx2 instructions statically
 set(FBGEMM_AVX2_SRCS
+  src/FbgemmFP16UKernelsAvx2.cc
   src/FbgemmI8DepthwiseAvx2.cc
   src/OptimizedKernelsAvx2.cc
   src/QuantUtilsAvx2.cc
-  src/Utils_avx2.cc)
+  src/UtilsAvx2.cc)
 
 #All the source files that use avx512 instructions statically
-set(FBGEMM_AVX512_SRCS src/Utils_avx512.cc)
+set(FBGEMM_AVX512_SRCS src/UtilsAvx512.cc)
 
 set(FBGEMM_PUBLIC_HEADERS include/fbgemm/Fbgemm.h
                           include/fbgemm/OutputProcessing-inl.h
@@ -93,8 +93,6 @@ set_target_properties(fbgemm_generic fbgemm_avx2 fbgemm_avx512 PROPERTIES
       CXX_EXTENSIONS NO
       CXX_VISIBILITY_PRESET hidden)
 
-target_compile_options(fbgemm_generic PRIVATE
-  "-m64" "-mavx2" "-mfma" "-masm=intel")
 target_compile_options(fbgemm_avx2 PRIVATE
   "-m64" "-mavx2" "-mfma" "-masm=intel")
 target_compile_options(fbgemm_avx512 PRIVATE

diff --git a/include/fbgemm/Fbgemm.h b/include/fbgemm/Fbgemm.h
@@ -9,7 +9,6 @@
 /**
  * Top level include file for FBGEMM.
  */
-#include <immintrin.h>
 #include <cassert>
 #include <cmath>
 #include <limits>

diff --git a/src/FbgemmFP16.cc b/src/FbgemmFP16.cc
@@ -10,7 +10,7 @@
 #include <array>
 #include <utility>
 
-#include "FbgemmFP16UKernels.h"
+#include "FbgemmFP16UKernelsAvx2.h"
 
 using namespace std;
 

diff --git a/src/FbgemmFP16UKernels.cc → src/FbgemmFP16UKernelsAvx2.cc b/src/FbgemmFP16UKernels.cc → src/FbgemmFP16UKernelsAvx2.cc
@@ -4,7 +4,7 @@
  * This source code is licensed under the BSD-style license found in the
  * LICENSE file in the root directory of this source tree.
  */
-#include "FbgemmFP16UKernels.h"
+#include "FbgemmFP16UKernelsAvx2.h"
 
 namespace fbgemm {
 

diff --git a/src/FbgemmFP16UKernels.h → src/FbgemmFP16UKernelsAvx2.h b/src/FbgemmFP16UKernels.h → src/FbgemmFP16UKernelsAvx2.h
@@ -7,8 +7,6 @@
 #ifndef FBGEMM_UKERNELS
 #define FBGEMM_UKERNELS
 #include <cstdint>
-#include <tuple>
-#include <vector>
 #include "fbgemm/Types.h"
 
 namespace fbgemm {

diff --git a/src/GenerateKernelU8S8S32ACC16_avx512.cc → src/GenerateKernelU8S8S32ACC16Avx512.cc b/src/GenerateKernelU8S8S32ACC16_avx512.cc → src/GenerateKernelU8S8S32ACC16Avx512.cc
diff --git a/src/GenerateKernelU8S8S32ACC32_avx512.cc → src/GenerateKernelU8S8S32ACC32Avx512.cc b/src/GenerateKernelU8S8S32ACC32_avx512.cc → src/GenerateKernelU8S8S32ACC32Avx512.cc
diff --git a/src/Utils_avx2.cc → src/UtilsAvx2.cc b/src/Utils_avx2.cc → src/UtilsAvx2.cc
diff --git a/src/Utils_avx512.cc → src/UtilsAvx512.cc b/src/Utils_avx512.cc → src/UtilsAvx512.cc
diff --git a/src/codegen_fp16fp32.cc b/src/codegen_fp16fp32.cc
@@ -65,22 +65,22 @@ int main() {
 
   // open all files
   ofstream srcfile;
-  srcfile.open("FbgemmFP16UKernels.cc");
+  srcfile.open("FbgemmFP16UKernelsAvx2.cc");
   srcfile
       << "/*\n"
          " * Copyright (c) Facebook, Inc. and its affiliates.\n"
          " * All rights reserved.\n"
          " * This source code is licensed under the BSD-style license found in the\n"
          " * LICENSE file in the root directory of this source tree.\n"
          " */\n";
-  srcfile << "#include \"FbgemmFP16UKernels.h\"\n\n";
+  srcfile << "#include \"FbgemmFP16UKernelsAvx2.h\"\n\n";
   srcfile << "namespace fbgemm {\n\n";
   if (iaca) {
     srcfile << "#include \"iacaMarks.h\"\n";
   }
 
   ofstream hdrfile;
-  hdrfile.open("FbgemmFP16UKernels.h");
+  hdrfile.open("FbgemmFP16UKernelsAvx2.h");
   hdrfile
       << "/*\n"
          " * Copyright (c) Facebook, Inc. and its affiliates.\n"
@@ -92,8 +92,6 @@ int main() {
   hdrfile << "#ifndef FBGEMM_UKERNELS\n";
   hdrfile << "#define FBGEMM_UKERNELS\n";
   hdrfile << "#include <cstdint>\n";
-  hdrfile << "#include <tuple>\n";
-  hdrfile << "#include <vector>\n";
   hdrfile << "#include \"fbgemm/Types.h\"\n\n";
   hdrfile << "namespace fbgemm {\n\n";
   hdrfile << "using fp16 = float16;\n";