From accffded5deaa561c400d1f8945ccd8dab589a35 Mon Sep 17 00:00:00 2001
From: Changming Sun <chasun@microsoft.com>
Date: Wed, 1 Apr 2020 10:07:22 -0700
Subject: [PATCH] Build options for enabling AVX/AVX2/AVX512 (#3373)

1. Add build options for enabling AVX/AVX2/AVX512
2. Update eigen to a newer version, because the current one doesn't work with VC and AVX512.
---
 cgmanifest.json      |  2 +-
 cmake/CMakeLists.txt | 39 +++++++++++++++++++++++++++++++++------
 cmake/external/eigen |  2 +-
 3 files changed, 35 insertions(+), 8 deletions(-)

diff --git a/cgmanifest.json b/cgmanifest.json
index 0adbebf636..57664cb0e8 100644
--- a/cgmanifest.json
+++ b/cgmanifest.json
@@ -139,7 +139,7 @@
          "component": {
             "type": "git",
             "git": {
-               "commitHash": "efd9867ff0e8df23016ac6c9828d0d7bf8bec1b1",
+               "commitHash": "54a0a9c9dd83aad42e79238ec8f0932b2a5e7881",
                "repositoryUrl": "https://gitlab.com/libeigen/eigen.git"
             }
          }
diff --git a/cmake/CMakeLists.txt b/cmake/CMakeLists.txt
index d07a64b2f8..6a48a01a73 100644
--- a/cmake/CMakeLists.txt
+++ b/cmake/CMakeLists.txt
@@ -69,6 +69,9 @@ option(onnxruntime_USE_PREINSTALLED_EIGEN "Use pre-installed EIGEN. Need to prov
 option(onnxruntime_BUILD_BENCHMARKS "Build ONNXRuntime micro-benchmarks" OFF)
 option(onnxruntime_USE_TVM "Build tvm for code-gen" OFF)
 option(onnxruntime_BUILD_FOR_NATIVE_MACHINE "Enable this option for turning on optimization specific to this machine" OFF)
+option(onnxruntime_USE_AVX "Use AVX instructions" OFF)
+option(onnxruntime_USE_AVX2 "Use AVX2 instructions" OFF)
+option(onnxruntime_USE_AVX512 "Use AVX512 instructions" OFF)
 option(onnxruntime_USE_LLVM "Build tvm with LLVM" OFF)
 option(onnxruntime_USE_OPENMP "Build with OpenMP support" OFF)
 option(onnxruntime_BUILD_SHARED_LIB "Build a shared library" OFF)
@@ -115,7 +118,7 @@ if(NOT WIN32)
 else()
   check_cxx_compiler_flag(/d2FH4- HAS_D2FH4)
   if (HAS_D2FH4)
-    message("Enabling /d2FH4-")
+    message("Disabling /d2FH4")
     set (CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /d2FH4-")
     set (CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /d2FH4-")
   endif()
@@ -208,8 +211,25 @@ if (MSVC)
     set(protobuf_MSVC_STATIC_RUNTIME OFF CACHE BOOL "Link protobuf to static runtime libraries" FORCE)
     set(gtest_force_shared_crt ON CACHE BOOL "Use shared (DLL) run-time lib for gtest" FORCE)
   endif()
+  #A compile option to improve code generation time, mostly for the code that heavily uses eigen.
+  check_cxx_compiler_flag(/d2ReducedOptimizeHugeFunctions HAS_D2ROHF)
+  if(HAS_D2ROHF)
+    SET(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /d2ReducedOptimizeHugeFunctions")
+    SET(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO} /d2ReducedOptimizeHugeFunctions")
+  endif()
   #Always enable exception handling, even for Windows ARM
-  SET (CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /EHsc")
+  string(APPEND CMAKE_CXX_FLAGS " /EHsc")
+  string(APPEND CMAKE_C_FLAGS " /EHsc")
+  if(onnxruntime_USE_AVX)
+    string(APPEND CMAKE_CXX_FLAGS " /arch:AVX")
+    string(APPEND CMAKE_C_FLAGS " /arch:AVX")
+  elseif(onnxruntime_USE_AVX2)
+    string(APPEND CMAKE_CXX_FLAGS " /arch:AVX2")
+    string(APPEND CMAKE_C_FLAGS " /arch:AVX2")
+  elseif(onnxruntime_USE_AVX512)
+    string(APPEND CMAKE_CXX_FLAGS " /arch:AVX512")
+    string(APPEND CMAKE_C_FLAGS " /arch:AVX512")
+  endif()
   if (onnxruntime_ENABLE_LTO AND NOT onnxruntime_USE_CUDA)
     SET (CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /Gw /GL")
     SET (CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO} /Gw /GL")
@@ -234,10 +254,17 @@ if (MSVC)
   endif()
 else()
   if(onnxruntime_BUILD_FOR_NATIVE_MACHINE)
-    string(APPEND CMAKE_CXX_FLAGS_RELEASE " -march=native -mtune=native")
-    string(APPEND CMAKE_C_FLAGS_RELEASE " -march=native -mtune=native")
-    string(APPEND CMAKE_CXX_FLAGS_RELWITHDEBINFO " -march=native -mtune=native")
-    string(APPEND CMAKE_C_FLAGS_RELWITHDEBINFO " -march=native -mtune=native")
+    string(APPEND CMAKE_CXX_FLAGS " -march=native -mtune=native")
+    string(APPEND CMAKE_C_FLAGS " -march=native -mtune=native")    
+  elseif(onnxruntime_USE_AVX)
+    string(APPEND CMAKE_CXX_FLAGS " -mavx")
+    string(APPEND CMAKE_C_FLAGS " -mavx")
+  elseif(onnxruntime_USE_AVX2)
+    string(APPEND CMAKE_CXX_FLAGS " -mavx2")
+    string(APPEND CMAKE_C_FLAGS " -mavx2")
+  elseif(onnxruntime_USE_AVX512)
+    string(APPEND CMAKE_CXX_FLAGS " -mavx512f -mavx512cd -mavx512bw -mavx512dq -mavx512vl")
+    string(APPEND CMAKE_C_FLAGS " -mavx512f -mavx512cd -mavx512bw -mavx512dq -mavx512vl")
   endif()
 endif()
 
diff --git a/cmake/external/eigen b/cmake/external/eigen
index efd9867ff0..54a0a9c9dd 160000
--- a/cmake/external/eigen
+++ b/cmake/external/eigen
@@ -1 +1 @@
-Subproject commit efd9867ff0e8df23016ac6c9828d0d7bf8bec1b1
+Subproject commit 54a0a9c9dd83aad42e79238ec8f0932b2a5e7881