hashirama/uvg266
1
0
Fork 0
mirror of https://github.com/ultravideo/uvg266.git synced 2024-11-24 02:24:07 +00:00
Code Issues Actions Projects Releases Wiki Activity

Fix avx detection for gcc.

Browse source 
- GCC doesn't support _xgetbv intrinsic so we have to use inline assembler.
This commit is contained in:
Ari Koivula 2014-06-18 11:50:17 +03:00
parent 3d23c7e913
commit d5a77be4b8
1 changed files with 15 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

20
src/strategyselector.c
View file

@ -256,12 +256,22 @@ static void set_hardware_flags() {
if (ecx & (1<<BIT_SSE41)) g_hardware_flags.intel_flags.sse41 = 1;
if (ecx & (1<<BIT_SSE42)) g_hardware_flags.intel_flags.sse42 = 1;
// Use _XCR_XFEATURE_ENABLED_MASK to check if _xgetbv intrinsic is
// supported by the compiler.
#ifdef _XCR_XFEATURE_ENABLED_MASK
// Check hardware and OS support for AVX.
if (ecx & (1 << BIT_OSXSAVE)) {
uint64_t xcr0 = _xgetbv(_XCR_XFEATURE_ENABLED_MASK);
uint64_t xcr0 = 0;
// Use _XCR_XFEATURE_ENABLED_MASK to check if _xgetbv intrinsic is
// supported by the compiler.
#ifdef _XCR_XFEATURE_ENABLED_MASK
xcr0 = _xgetbv(_XCR_XFEATURE_ENABLED_MASK);
#elif defined(__GNUC__)
uint32_t geax = 0;
// Apparently there are some older assemblers that don't support xgetbv,
// so we use the byte sequence for xgetbv just in case.
//__asm__("xgetbv" : "=a" (geax), "=d" (gedx) : "c" (0));
__asm__(".byte 0x0f, 0x01, 0xd0" : "=a" (geax) : "c" (0) : "edx");
// edx is spillover, but we don't care about those bits.
xcr0 = geax;
#endif
bool avx_support = ecx & (1 << BIT_AVX) || false;
bool xmm_support = xcr0 & (1 << XCR0_XMM);
bool ymm_support = xcr0 & (1 << XCR0_YMM);
@ -275,7 +285,7 @@ static void set_hardware_flags() {
get_cpuid(7, &eax, &ebx, &ecx, &edx);
if (ebx & (1 << 5)) g_hardware_flags.intel_flags.avx2 = 1;
}
#endif
fprintf(stderr, "Compiled: INTEL, flags:");
#if COMPILE_INTEL_MMX