diff options
| author | Soumith Chintala <soumith@fb.com> | 2016-12-31 02:10:57 +0300 |
|---|---|---|
| committer | Soumith Chintala <soumith@fb.com> | 2016-12-31 02:10:57 +0300 |
| commit | 87213287e51e866c5ee6c2dba21b78adbbc0b51b (patch) | |
| tree | c9f65483d89f911cd373b82f9388ec87d08ea285 | |
| parent | b19b8c741ea6b84c0ddb340bc7e9c96360105a22 (diff) | |
use __get_cpuid when available
| -rw-r--r-- | lib/TH/CMakeLists.txt | 15 | ||||
| -rw-r--r-- | lib/TH/generic/simd/simd.h | 23 |
2 files changed, 29 insertions, 9 deletions
diff --git a/lib/TH/CMakeLists.txt b/lib/TH/CMakeLists.txt index 3912bf2..25fb5d3 100644 --- a/lib/TH/CMakeLists.txt +++ b/lib/TH/CMakeLists.txt @@ -78,6 +78,21 @@ IF (CORTEXA9_FOUND) SET(CMAKE_C_FLAGS "-mcpu=cortex-a9 ${CMAKE_C_FLAGS}") ENDIF (CORTEXA9_FOUND) +INCLUDE(CheckIncludeFile) +CHECK_INCLUDE_FILE(cpuid.h HAVE_CPUID_H) +# Check for a cpuid intrinsic +IF(HAVE_CPUID_H) + CHECK_C_SOURCE_COMPILES("#include <cpuid.h> + int main() + { + unsigned int eax, ebx, ecx, edx; + return __get_cpuid(0, &eax, &ebx, &ecx, &edx); + }" HAVE_GCC_GET_CPUID) +ENDIF() +IF(HAVE_GCC_GET_CPUID) + SET(CMAKE_C_FLAGS "-DHAVE_GCC_GET_CPUID") +ENDIF(HAVE_GCC_GET_CPUID) + FIND_PACKAGE(SSE) IF(C_SSE2_FOUND) SET(CMAKE_C_FLAGS "${C_SSE2_FLAGS} -DUSE_SSE2 ${CMAKE_C_FLAGS}") diff --git a/lib/TH/generic/simd/simd.h b/lib/TH/generic/simd/simd.h index caf671e..ee53e2c 100644 --- a/lib/TH/generic/simd/simd.h +++ b/lib/TH/generic/simd/simd.h @@ -2,8 +2,10 @@ #define TH_SIMD_INC #include <stdint.h> -#ifdef _MSC_VER +#if defined(_MSC_VER) #include <intrin.h> +#elif defined(HAVE_GCC_GET_CPUID) +#include <cpuid.h> #endif // Can be found on Intel ISA Reference for CPUID @@ -90,7 +92,17 @@ static inline uint32_t detectHostSIMDExtensions() #else // x86 static inline void cpuid(uint32_t *eax, uint32_t *ebx, uint32_t *ecx, uint32_t *edx) { -#ifndef _MSC_VER +#if defined(_MSC_VER) + uint32_t cpuInfo[4]; + __cpuid(cpuInfo, *eax); + *eax = cpuInfo[0]; + *ebx = cpuInfo[1]; + *ecx = cpuInfo[2]; + *edx = cpuInfo[3]; +#elif defined(HAVE_GCC_GET_CPUID) + uint32_t level = *eax; + __get_cpuid (level, eax, ebx, ecx, edx); +#else uint32_t a = *eax, b, c, d; asm volatile ( "cpuid\n\t" : "+a"(a), "=b"(b), "=c"(c), "=d"(d) ); @@ -98,13 +110,6 @@ static inline void cpuid(uint32_t *eax, uint32_t *ebx, uint32_t *ecx, uint32_t * *ebx = b; *ecx = c; *edx = d; -#else - uint32_t cpuInfo[4]; - __cpuid(cpuInfo, *eax); - *eax = cpuInfo[0]; - *ebx = cpuInfo[1]; - *ecx = cpuInfo[2]; - *edx = cpuInfo[3]; #endif } |