diff options
| author | James Almer <jamrial@gmail.com> | 2014-02-22 04:54:01 +0000 |
|---|---|---|
| committer | Janne Grunau <janne-libav@jannau.net> | 2014-02-23 15:29:36 +0100 |
| commit | 1b932eb1508f550fac9e911923a0383efda53aa3 (patch) | |
| tree | 1dadb7a0429d30648b3e22a7926bf8b1b03e4ac9 /libavutil | |
| parent | 10b0161d78148f46eaffb29ea022378947eaef2c (diff) | |
| download | ffmpeg-1b932eb1508f550fac9e911923a0383efda53aa3.tar.gz | |
x86: add detection for FMA3 instruction set
Based on x264 code
Signed-off-by: James Almer <jamrial@gmail.com>
Diffstat (limited to 'libavutil')
| -rw-r--r-- | libavutil/cpu.c | 3 | ||||
| -rw-r--r-- | libavutil/cpu.h | 1 | ||||
| -rw-r--r-- | libavutil/x86/cpu.c | 5 | ||||
| -rw-r--r-- | libavutil/x86/cpu.h | 3 |
4 files changed, 11 insertions, 1 deletions
diff --git a/libavutil/cpu.c b/libavutil/cpu.c index 8c2cfb87cc..972e4eb79f 100644 --- a/libavutil/cpu.c +++ b/libavutil/cpu.c @@ -87,6 +87,7 @@ int av_parse_cpu_flags(const char *s) #define CPUFLAG_SSE42 (AV_CPU_FLAG_SSE42 | CPUFLAG_SSE4) #define CPUFLAG_AVX (AV_CPU_FLAG_AVX | CPUFLAG_SSE42) #define CPUFLAG_XOP (AV_CPU_FLAG_XOP | CPUFLAG_AVX) +#define CPUFLAG_FMA3 (AV_CPU_FLAG_FMA3 | CPUFLAG_AVX) #define CPUFLAG_FMA4 (AV_CPU_FLAG_FMA4 | CPUFLAG_AVX) #define CPUFLAG_AVX2 (AV_CPU_FLAG_AVX2 | CPUFLAG_AVX) static const AVOption cpuflags_opts[] = { @@ -107,6 +108,7 @@ int av_parse_cpu_flags(const char *s) { "sse4.2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE42 }, .unit = "flags" }, { "avx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX }, .unit = "flags" }, { "xop" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_XOP }, .unit = "flags" }, + { "fma3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA3 }, .unit = "flags" }, { "fma4" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA4 }, .unit = "flags" }, { "avx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX2 }, .unit = "flags" }, { "3dnow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOW }, .unit = "flags" }, @@ -204,6 +206,7 @@ static const struct { { AV_CPU_FLAG_SSE42, "sse4.2" }, { AV_CPU_FLAG_AVX, "avx" }, { AV_CPU_FLAG_XOP, "xop" }, + { AV_CPU_FLAG_FMA3, "fma3" }, { AV_CPU_FLAG_FMA4, "fma4" }, { AV_CPU_FLAG_3DNOW, "3dnow" }, { AV_CPU_FLAG_3DNOWEXT, "3dnowext" }, diff --git a/libavutil/cpu.h b/libavutil/cpu.h index 29036e3941..934b3be6a2 100644 --- a/libavutil/cpu.h +++ b/libavutil/cpu.h @@ -49,6 +49,7 @@ #define AV_CPU_FLAG_FMA4 0x0800 ///< Bulldozer FMA4 functions #define AV_CPU_FLAG_CMOV 0x1000 ///< i686 cmov #define AV_CPU_FLAG_AVX2 0x8000 ///< AVX2 functions: requires OS support even if YMM registers aren't used +#define AV_CPU_FLAG_FMA3 0x10000 ///< Haswell FMA3 functions #define AV_CPU_FLAG_ALTIVEC 0x0001 ///< standard diff --git a/libavutil/x86/cpu.c b/libavutil/x86/cpu.c index 0e06d5deb2..95359eeb7b 100644 --- a/libavutil/x86/cpu.c +++ b/libavutil/x86/cpu.c @@ -131,8 +131,11 @@ int ff_get_cpu_flags_x86(void) if ((ecx & 0x18000000) == 0x18000000) { /* Check for OS support */ xgetbv(0, eax, edx); - if ((eax & 0x6) == 0x6) + if ((eax & 0x6) == 0x6) { rval |= AV_CPU_FLAG_AVX; + if (ecx & 0x00001000) + rval |= AV_CPU_FLAG_FMA3; + } } #if HAVE_AVX2 if (max_std_level >= 7) { diff --git a/libavutil/x86/cpu.h b/libavutil/x86/cpu.h index 40daf445c7..50da30e389 100644 --- a/libavutil/x86/cpu.h +++ b/libavutil/x86/cpu.h @@ -38,6 +38,7 @@ #define X86_SSE42(flags) CPUEXT(flags, SSE42) #define X86_AVX(flags) CPUEXT(flags, AVX) #define X86_XOP(flags) CPUEXT(flags, XOP) +#define X86_FMA3(flags) CPUEXT(flags, FMA3) #define X86_FMA4(flags) CPUEXT(flags, FMA4) #define X86_AVX2(flags) CPUEXT(flags, AVX2) @@ -53,6 +54,7 @@ #define EXTERNAL_SSE42(flags) CPUEXT_SUFFIX(flags, _EXTERNAL, SSE42) #define EXTERNAL_AVX(flags) CPUEXT_SUFFIX(flags, _EXTERNAL, AVX) #define EXTERNAL_XOP(flags) CPUEXT_SUFFIX(flags, _EXTERNAL, XOP) +#define EXTERNAL_FMA3(flags) CPUEXT_SUFFIX(flags, _EXTERNAL, FMA3) #define EXTERNAL_FMA4(flags) CPUEXT_SUFFIX(flags, _EXTERNAL, FMA4) #define EXTERNAL_AVX2(flags) CPUEXT_SUFFIX(flags, _EXTERNAL, AVX2) @@ -68,6 +70,7 @@ #define INLINE_SSE42(flags) CPUEXT_SUFFIX(flags, _INLINE, SSE42) #define INLINE_AVX(flags) CPUEXT_SUFFIX(flags, _INLINE, AVX) #define INLINE_XOP(flags) CPUEXT_SUFFIX(flags, _INLINE, XOP) +#define INLINE_FMA3(flags) CPUEXT_SUFFIX(flags, _INLINE, FMA3) #define INLINE_FMA4(flags) CPUEXT_SUFFIX(flags, _INLINE, FMA4) #define INLINE_AVX2(flags) CPUEXT_SUFFIX(flags, _INLINE, AVX2) |