RepoMirrors
/
ffmpeg
mirror of https://git.ffmpeg.org/ffmpeg.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

aarch64: Add CPU feature flags for SVE and SVE2 

Add code for detecting the feature on Linux and Windows.

Signed-off-by: Martin Storsjö <martin@martin.st>
This commit is contained in:
Martin Storsjö 2023-10-12 23:31:33 +03:00
parent e6e56fd7a7
commit e6eabb7ce7
6 changed files with 30 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
libavutil/aarch64/cpu.c
View File

@ -25,6 +25,8 @@
#include <sys/auxv.h> #include <sys/auxv.h>
#define HWCAP_AARCH64_ASIMDDP (1 << 20) #define HWCAP_AARCH64_ASIMDDP (1 << 20)
#define HWCAP_AARCH64_SVE (1 << 22)
#define HWCAP2_AARCH64_SVE2 (1 << 1)
#define HWCAP2_AARCH64_I8MM (1 << 13) #define HWCAP2_AARCH64_I8MM (1 << 13)
static int detect_flags(void) static int detect_flags(void)
@ -36,6 +38,10 @@ static int detect_flags(void)
if (hwcap & HWCAP_AARCH64_ASIMDDP) if (hwcap & HWCAP_AARCH64_ASIMDDP)
flags |= AV_CPU_FLAG_DOTPROD; flags |= AV_CPU_FLAG_DOTPROD;
if (hwcap & HWCAP_AARCH64_SVE)
flags |= AV_CPU_FLAG_SVE;
if (hwcap2 & HWCAP2_AARCH64_SVE2)
flags |= AV_CPU_FLAG_SVE2;
if (hwcap2 & HWCAP2_AARCH64_I8MM) if (hwcap2 & HWCAP2_AARCH64_I8MM)
flags |= AV_CPU_FLAG_I8MM; flags |= AV_CPU_FLAG_I8MM;
@ -119,6 +125,14 @@ static int detect_flags(void)
* regular I8MM is available. */ * regular I8MM is available. */
if (IsProcessorFeaturePresent(PF_ARM_SVE_I8MM_INSTRUCTIONS_AVAILABLE)) if (IsProcessorFeaturePresent(PF_ARM_SVE_I8MM_INSTRUCTIONS_AVAILABLE))
flags |= AV_CPU_FLAG_I8MM; flags |= AV_CPU_FLAG_I8MM;
#endif
#ifdef PF_ARM_SVE_INSTRUCTIONS_AVAILABLE
if (IsProcessorFeaturePresent(PF_ARM_SVE_INSTRUCTIONS_AVAILABLE))
flags |= AV_CPU_FLAG_SVE;
#endif
#ifdef PF_ARM_SVE2_INSTRUCTIONS_AVAILABLE
if (IsProcessorFeaturePresent(PF_ARM_SVE2_INSTRUCTIONS_AVAILABLE))
flags |= AV_CPU_FLAG_SVE2;
#endif #endif
return flags; return flags;
} }
@ -142,6 +156,12 @@ int ff_get_cpu_flags_aarch64(void)
#ifdef __ARM_FEATURE_MATMUL_INT8 #ifdef __ARM_FEATURE_MATMUL_INT8
flags |= AV_CPU_FLAG_I8MM; flags |= AV_CPU_FLAG_I8MM;
#endif #endif
#ifdef __ARM_FEATURE_SVE
flags |= AV_CPU_FLAG_SVE;
#endif
#ifdef __ARM_FEATURE_SVE2
flags |= AV_CPU_FLAG_SVE2;
#endif
flags |= detect_flags(); flags |= detect_flags();

2
libavutil/aarch64/cpu.h
View File

@ -27,5 +27,7 @@
#define have_vfp(flags) CPUEXT(flags, VFP) #define have_vfp(flags) CPUEXT(flags, VFP)
#define have_dotprod(flags) CPUEXT(flags, DOTPROD) #define have_dotprod(flags) CPUEXT(flags, DOTPROD)
#define have_i8mm(flags) CPUEXT(flags, I8MM) #define have_i8mm(flags) CPUEXT(flags, I8MM)
#define have_sve(flags) CPUEXT(flags, SVE)
#define have_sve2(flags) CPUEXT(flags, SVE2)
#endif /* AVUTIL_AARCH64_CPU_H */ #endif /* AVUTIL_AARCH64_CPU_H */

2
libavutil/cpu.c
View File

@ -180,6 +180,8 @@ int av_parse_cpu_caps(unsigned *flags, const char *s)
{ "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" }, { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
{ "dotprod", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_DOTPROD }, .unit = "flags" }, { "dotprod", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_DOTPROD }, .unit = "flags" },
{ "i8mm", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_I8MM }, .unit = "flags" }, { "i8mm", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_I8MM }, .unit = "flags" },
{ "sve", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SVE }, .unit = "flags" },
{ "sve2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SVE2 }, .unit = "flags" },
#elif ARCH_MIPS #elif ARCH_MIPS
{ "mmi", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMI }, .unit = "flags" }, { "mmi", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMI }, .unit = "flags" },
{ "msa", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MSA }, .unit = "flags" }, { "msa", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MSA }, .unit = "flags" },

2
libavutil/cpu.h
View File

@ -72,6 +72,8 @@
#define AV_CPU_FLAG_VFP_VM (1 << 7) ///< VFPv2 vector mode, deprecated in ARMv7-A and unavailable in various CPUs implementations #define AV_CPU_FLAG_VFP_VM (1 << 7) ///< VFPv2 vector mode, deprecated in ARMv7-A and unavailable in various CPUs implementations
#define AV_CPU_FLAG_DOTPROD (1 << 8) #define AV_CPU_FLAG_DOTPROD (1 << 8)
#define AV_CPU_FLAG_I8MM (1 << 9) #define AV_CPU_FLAG_I8MM (1 << 9)
#define AV_CPU_FLAG_SVE (1 <<10)
#define AV_CPU_FLAG_SVE2 (1 <<11)
#define AV_CPU_FLAG_SETEND (1 <<16) #define AV_CPU_FLAG_SETEND (1 <<16)
#define AV_CPU_FLAG_MMI (1 << 0) #define AV_CPU_FLAG_MMI (1 << 0)

2
libavutil/tests/cpu.c
View File

@ -40,6 +40,8 @@ static const struct {
{ AV_CPU_FLAG_VFP, "vfp" }, { AV_CPU_FLAG_VFP, "vfp" },
{ AV_CPU_FLAG_DOTPROD, "dotprod" }, { AV_CPU_FLAG_DOTPROD, "dotprod" },
{ AV_CPU_FLAG_I8MM, "i8mm" }, { AV_CPU_FLAG_I8MM, "i8mm" },
{ AV_CPU_FLAG_SVE, "sve" },
{ AV_CPU_FLAG_SVE2, "sve2" },
#elif ARCH_ARM #elif ARCH_ARM
{ AV_CPU_FLAG_ARMV5TE, "armv5te" }, { AV_CPU_FLAG_ARMV5TE, "armv5te" },
{ AV_CPU_FLAG_ARMV6, "armv6" }, { AV_CPU_FLAG_ARMV6, "armv6" },

2
tests/checkasm/checkasm.c
View File

@ -305,6 +305,8 @@ static const struct {
{ "NEON", "neon", AV_CPU_FLAG_NEON }, { "NEON", "neon", AV_CPU_FLAG_NEON },
{ "DOTPROD", "dotprod", AV_CPU_FLAG_DOTPROD }, { "DOTPROD", "dotprod", AV_CPU_FLAG_DOTPROD },
{ "I8MM", "i8mm", AV_CPU_FLAG_I8MM }, { "I8MM", "i8mm", AV_CPU_FLAG_I8MM },
{ "SVE", "sve", AV_CPU_FLAG_SVE },
{ "SVE2", "sve2", AV_CPU_FLAG_SVE2 },
#elif ARCH_ARM #elif ARCH_ARM
{ "ARMV5TE", "armv5te", AV_CPU_FLAG_ARMV5TE }, { "ARMV5TE", "armv5te", AV_CPU_FLAG_ARMV5TE },
{ "ARMV6", "armv6", AV_CPU_FLAG_ARMV6 }, { "ARMV6", "armv6", AV_CPU_FLAG_ARMV6 },