ARROW-14361: [C++] Add default simd level

apache · Oct 19, 2021 · 9e51f11 · 9e51f11
1 parent d9ef519
commit 9e51f11
Show file tree

Hide file tree

Showing 4 changed files with 35 additions and 10 deletions.
diff --git a/cpp/cmake_modules/DefineOptions.cmake b/cpp/cmake_modules/DefineOptions.cmake
@@ -112,11 +112,13 @@ if("${CMAKE_SOURCE_DIR}" STREQUAL "${CMAKE_CURRENT_SOURCE_DIR}")
 
   define_option_string(ARROW_SIMD_LEVEL
                        "Compile-time SIMD optimization level"
-                       "SSE4_2" # default to SSE4.2
+                       "DEFAULT" # default to SSE4_2 on x86, NEON on Arm, NONE otherwise
                        "NONE"
                        "SSE4_2"
                        "AVX2"
-                       "AVX512")
+                       "AVX512"
+                       "NEON"
+                       "DEFAULT")
 
   define_option_string(ARROW_RUNTIME_SIMD_LEVEL
                        "Max runtime SIMD optimization level"

diff --git a/cpp/cmake_modules/SetupCxxFlags.cmake b/cpp/cmake_modules/SetupCxxFlags.cmake
@@ -94,14 +94,23 @@ if(ARROW_CPU_FLAG STREQUAL "x86")
     set(ARROW_HAVE_RUNTIME_AVX512 ON)
     add_definitions(-DARROW_HAVE_RUNTIME_AVX512 -DARROW_HAVE_RUNTIME_BMI2)
   endif()
+  if(ARROW_SIMD_LEVEL STREQUAL "DEFAULT")
+    set(ARROW_SIMD_LEVEL "SSE4_2")
+  endif()
 elseif(ARROW_CPU_FLAG STREQUAL "ppc")
   # power compiler flags, gcc/clang only
   set(ARROW_ALTIVEC_FLAG "-maltivec")
   check_cxx_compiler_flag(${ARROW_ALTIVEC_FLAG} CXX_SUPPORTS_ALTIVEC)
+  if(ARROW_SIMD_LEVEL STREQUAL "DEFAULT")
+    set(ARROW_SIMD_LEVEL "NONE")
+  endif()
 elseif(ARROW_CPU_FLAG STREQUAL "armv8")
   # Arm64 compiler flags, gcc/clang only
   set(ARROW_ARMV8_ARCH_FLAG "-march=${ARROW_ARMV8_ARCH}")
   check_cxx_compiler_flag(${ARROW_ARMV8_ARCH_FLAG} CXX_SUPPORTS_ARMV8_ARCH)
+  if(ARROW_SIMD_LEVEL STREQUAL "DEFAULT")
+    set(ARROW_SIMD_LEVEL "NEON")
+  endif()
 endif()
 
 # Support C11
@@ -431,6 +440,8 @@ if(ARROW_CPU_FLAG STREQUAL "x86")
     endif()
     set(CXX_COMMON_FLAGS "${CXX_COMMON_FLAGS} ${ARROW_SSE4_2_FLAG}")
     add_definitions(-DARROW_HAVE_SSE4_2)
+  elseif(NOT ARROW_SIMD_LEVEL STREQUAL "NONE")
+    message(WARNING "ARROW_SIMD_LEVEL=${ARROW_SIMD_LEVEL} not supported by x86.")
   endif()
 endif()
 
@@ -441,7 +452,7 @@ if(ARROW_CPU_FLAG STREQUAL "ppc")
 endif()
 
 if(ARROW_CPU_FLAG STREQUAL "armv8")
-  if(NOT ARROW_SIMD_LEVEL STREQUAL "NONE")
+  if(ARROW_SIMD_LEVEL STREQUAL "NEON")
     set(ARROW_HAVE_NEON ON)
 
     if(NOT CXX_SUPPORTS_ARMV8_ARCH)
@@ -467,6 +478,8 @@ if(ARROW_CPU_FLAG STREQUAL "armv8")
         add_definitions(-DARROW_HAVE_ARMV8_CRC)
       endif()
     endif()
+  elseif(NOT ARROW_SIMD_LEVEL STREQUAL "NONE")
+    message(WARNING "ARROW_SIMD_LEVEL=${ARROW_SIMD_LEVEL} not supported by Arm.")
   endif()
 endif()
 

diff --git a/dev/tasks/tasks.yml b/dev/tasks/tasks.yml
@@ -397,8 +397,7 @@ tasks:
     template: python-wheels/github.osx.arm64.yml
     params:
       arch: arm64
-      # SSE4_2 is the default but cmake will use NEON instead
-      arrow_simd_level: "SSE4_2"
+      arrow_simd_level: "DEFAULT"
       vcpkg_version: "2021.04.30"
       python_version: "3.8"
       macos_deployment_target: "11.0"
@@ -412,8 +411,7 @@ tasks:
     template: python-wheels/github.osx.arm64.yml
     params:
       arch: arm64
-      # SSE4_2 is the default but cmake will use NEON instead
-      arrow_simd_level: "SSE4_2"
+      arrow_simd_level: "DEFAULT"
       vcpkg_version: "2021.04.30"
       python_version: "{{ python_version }}"
       macos_deployment_target: "11.0"

diff --git a/python/CMakeLists.txt b/python/CMakeLists.txt
@@ -103,9 +103,21 @@ set(BUILD_WARNING_LEVEL "PRODUCTION")
 
 # This must be synchronized with the definition in
 # cpp/cmake_modules/DefineOptions.cmake.
-set(ARROW_ARMV8_ARCH
-    "armv8-a"
-    CACHE STRING "Arm64 arch and extensions: armv8-a, armv8-a or armv8-a+crc+crypto")
+if(NOT DEFINED ARROW_SIMD_LEVEL)
+  set(ARROW_SIMD_LEVEL
+      "DEFAULT"
+      CACHE STRING "Compile time SIMD optimization level")
+endif()
+if(NOT DEFINED ARROW_RUNTIME_SIMD_LEVEL)
+  set(ARROW_RUNTIME_SIMD_LEVEL
+      "MAX"
+      CACHE STRING "Max runtime SIMD optimization level")
+endif()
+if(NOT DEFINED ARROW_ARMV8_ARCH)
+  set(ARROW_ARMV8_ARCH
+      "armv8-a"
+      CACHE STRING "Arm64 arch and extensions: armv8-a, armv8-a or armv8-a+crc+crypto")
+endif()
 include(SetupCxxFlags)
 
 # Add common flags