Skip to content

Commit 14ba68f

Browse files
authored
Update FindAVX.cmake (#404)
* make AVX_FOUND is default value to WITH AVX * let AVX_FLAG always keep -mavx flag since compiler can build binary with -mavx even CPU does not support avx.
1 parent 0c7ac3d commit 14ba68f

File tree

2 files changed

+58
-49
lines changed

2 files changed

+58
-49
lines changed

CMakeLists.txt

Lines changed: 6 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -109,11 +109,9 @@ else()
109109
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-g -O3 --use_fast_math")
110110

111111
if(WITH_AVX)
112-
if(AVX_FOUND)
113-
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xcompiler -mavx")
114-
endif(AVX_FOUND)
112+
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xcompiler ${AVX_FLAG}")
115113
else(WITH_AVX)
116-
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xcompiler -msse3")
114+
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xcompiler ${SSE3_FLAG}")
117115
endif(WITH_AVX)
118116

119117
if(WITH_DSO)
@@ -138,11 +136,11 @@ if(NOT WITH_TIMER)
138136
endif(NOT WITH_TIMER)
139137

140138
if(WITH_AVX)
141-
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${AVX_FLAGS}")
142-
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${AVX_FLAGS}")
139+
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${AVX_FLAG}")
140+
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${AVX_FLAG}")
143141
else(WITH_AVX)
144-
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -msse3")
145-
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse3")
142+
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${SSE3_FLAG}")
143+
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${SSE3_FLAG}")
146144
endif(WITH_AVX)
147145

148146
if(WITH_PYTHON)

cmake/FindAVX.cmake

Lines changed: 52 additions & 41 deletions
Original file line numberDiff line numberDiff line change
@@ -3,36 +3,55 @@
33

44
INCLUDE(CheckCXXSourceRuns)
55

6-
SET(FIND_AVX_10)
7-
SET(FIND_AVX_20)
8-
SET(AVX_FLAGS)
9-
SET(AVX_FOUND)
10-
11-
# Check AVX 2
12-
SET(CMAKE_REQUIRED_FLAGS)
136
IF(CMAKE_COMPILER_IS_GNUCC OR CMAKE_COMPILER_IS_GNUCXX OR CMAKE_CXX_COMPILER_ID MATCHES "Clang")
14-
SET(CMAKE_REQUIRED_FLAGS "-mavx2")
15-
ELSEIF(MSVC AND NOT CMAKE_CL_64) # reserve for WINDOWS
16-
SET(CMAKE_REQUIRED_FLAGS "/arch:AVX2")
7+
set(MMX_FLAG "-mmmx")
8+
set(SSE2_FLAG "-msse2")
9+
set(SSE3_FLAG "-msse3")
10+
SET(AVX_FLAG "-mavx")
11+
SET(AVX2_FLAG "-mavx2")
12+
ELSEIF(MSVC)
13+
set(MMX_FLAG "/arch:MMX")
14+
set(SSE2_FLAG "/arch:SSE2")
15+
set(SSE3_FLAG "/arch:SSE3")
16+
SET(AVX_FLAG "/arch:AVX")
17+
SET(AVX2_FLAG "/arch:AVX2")
1718
ENDIF()
1819

20+
# Check MMX
21+
set(CMAKE_REQUIRED_FLAGS ${MMX_FLAG})
1922
CHECK_CXX_SOURCE_RUNS("
20-
#include <immintrin.h>
23+
#include <mmintrin.h>
2124
int main()
2225
{
23-
__m256i a = _mm256_set_epi32 (-1, 2, -3, 4, -1, 2, -3, 4);
24-
__m256i result = _mm256_abs_epi32 (a);
26+
_mm_setzero_si64();
2527
return 0;
26-
}" FIND_AVX_20)
28+
}" MMX_FOUND)
2729

28-
# Check AVX
29-
SET(CMAKE_REQUIRED_FLAGS)
30-
IF(CMAKE_COMPILER_IS_GNUCC OR CMAKE_COMPILER_IS_GNUCXX OR CMAKE_CXX_COMPILER_ID MATCHES "Clang")
31-
SET(CMAKE_REQUIRED_FLAGS "-mavx")
32-
ELSEIF(MSVC AND NOT CMAKE_CL_64)
33-
SET(CMAKE_REQUIRED_FLAGS "/arch:AVX")
34-
endif()
30+
# Check SSE2
31+
set(CMAKE_REQUIRED_FLAGS ${SSE2_FLAG})
32+
CHECK_CXX_SOURCE_RUNS("
33+
#include <emmintrin.h>
34+
int main()
35+
{
36+
_mm_setzero_si128();
37+
return 0;
38+
}" SSE2_FOUND)
3539

40+
# Check SSE3
41+
set(CMAKE_REQUIRED_FLAGS ${SSE3_FLAG})
42+
CHECK_CXX_SOURCE_RUNS("
43+
#include <pmmintrin.h>
44+
int main()
45+
{
46+
__m128d a = _mm_set1_pd(6.28);
47+
__m128d b = _mm_set1_pd(3.14);
48+
__m128d result = _mm_addsub_pd(a, b);
49+
result = _mm_movedup_pd(result);
50+
return 0;
51+
}" SSE3_FOUND)
52+
53+
# Check AVX
54+
set(CMAKE_REQUIRED_FLAGS ${AVX_FLAG})
3655
CHECK_CXX_SOURCE_RUNS("
3756
#include <immintrin.h>
3857
int main()
@@ -41,25 +60,17 @@ int main()
4160
__m256 b = _mm256_set_ps (1.0f, 2.0f, 3.0f, 4.0f, 1.0f, 2.0f, 3.0f, 4.0f);
4261
__m256 result = _mm256_add_ps (a, b);
4362
return 0;
44-
}" FIND_AVX_10)
45-
46-
IF(${FIND_AVX_20})
47-
IF(CMAKE_COMPILER_IS_GNUCC OR CMAKE_COMPILER_IS_GNUCXX OR CMAKE_CXX_COMPILER_ID MATCHES "Clang")
48-
SET(AVX_FLAGS "${AVX_FLAGS} -mavx2")
49-
ELSEIF(MSVC)
50-
SET(AVX_FLAGS "${AVX_FLAGS} /arch:AVX2")
51-
ENDIF()
52-
ENDIF()
63+
}" AVX_FOUND)
5364

54-
IF(${FIND_AVX_10})
55-
IF(CMAKE_COMPILER_IS_GNUCC OR CMAKE_COMPILER_IS_GNUCXX OR CMAKE_CXX_COMPILER_ID MATCHES "Clang")
56-
SET(AVX_FLAGS "${AVX_FLAGS} -mavx")
57-
ELSEIF(MSVC)
58-
SET(AVX_FLAGS "${AVX_FLAGS} /arch:AVX")
59-
ENDIF()
60-
ENDIF()
65+
# Check AVX 2
66+
set(CMAKE_REQUIRED_FLAGS ${AVX2_FLAG})
67+
CHECK_CXX_SOURCE_RUNS("
68+
#include <immintrin.h>
69+
int main()
70+
{
71+
__m256i a = _mm256_set_epi32 (-1, 2, -3, 4, -1, 2, -3, 4);
72+
__m256i result = _mm256_abs_epi32 (a);
73+
return 0;
74+
}" AVX2_FOUND)
6175

62-
IF(${FIND_AVX_10})
63-
SET(AVX_FOUND TRUE)
64-
MESSAGE(STATUS "Find CPU supports ${AVX_FLAGS}.")
65-
ENDIF()
76+
mark_as_advanced(MMX_FOUND SSE2_FOUND SSE3_FOUND AVX_FOUND AVX2_FOUND)

0 commit comments

Comments
 (0)