diff options
| author | Chunlin <fangchunlin@huawei.com> | 2020-05-26 18:44:45 +0800 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2020-05-26 13:44:45 +0300 |
| commit | 543655176a05e2f6a54dbbadfb754416f14fb215 (patch) | |
| tree | 984af220bf412c64e358d1fc0435862fb3c5fdf6 /numpy/core | |
| parent | 3f11db40d80b6ff122f41185b0b93637bdff0e39 (diff) | |
| download | numpy-543655176a05e2f6a54dbbadfb754416f14fb215.tar.gz | |
ENH: Optimize Cpu feature detect in X86, fix for GCC on macOS (#16297)
* Fix cpu detection for X86, GCC, and macOS
* Optimize feature detection
Co-authored-by: Matti Picus <matti.picus@gmail.com>
Diffstat (limited to 'numpy/core')
| -rw-r--r-- | numpy/core/src/common/npy_cpu_features.c.src | 20 |
1 files changed, 14 insertions, 6 deletions
diff --git a/numpy/core/src/common/npy_cpu_features.c.src b/numpy/core/src/common/npy_cpu_features.c.src index 4f193a471..e676932c9 100644 --- a/numpy/core/src/common/npy_cpu_features.c.src +++ b/numpy/core/src/common/npy_cpu_features.c.src @@ -76,11 +76,13 @@ npy__cpu_getxcr0(void) #if defined(_MSC_VER) || defined (__INTEL_COMPILER) return _xgetbv(0); #elif defined(__GNUC__) || defined(__clang__) + /* named form of xgetbv not supported on OSX, so must use byte form, see: + * https://github.com/asmjit/asmjit/issues/78 + */ unsigned int eax, edx; - __asm__("xgetbv" : "=a" (eax), "=d" (edx) : "c" (0)); - return (eax | (unsigned long long)edx << 32); + __asm(".byte 0x0F, 0x01, 0xd0" : "=a"(eax), "=d"(edx) : "c"(0)); + return eax; #else - // TODO: handle other x86 compilers return 0; #endif } @@ -110,7 +112,6 @@ npy__cpu_cpuid(int reg[4], int func_id) ); #endif #else - // TODO: handle other x86 compilers reg[0] = 0; #endif } @@ -123,8 +124,15 @@ npy__cpu_init_features(void) // validate platform support int reg[] = {0, 0, 0, 0}; npy__cpu_cpuid(reg, 0); - if (reg[0] == 0) - return; + if (reg[0] == 0) { + npy__cpu_have[NPY_CPU_FEATURE_MMX] = 1; + npy__cpu_have[NPY_CPU_FEATURE_SSE] = 1; + npy__cpu_have[NPY_CPU_FEATURE_SSE2] = 1; + #ifdef NPY_CPU_AMD64 + npy__cpu_have[NPY_CPU_FEATURE_SSE3] = 1; + #endif + return; + } npy__cpu_cpuid(reg, 1); npy__cpu_have[NPY_CPU_FEATURE_MMX] = (reg[3] & (1 << 23)) != 0; |