diff options
Diffstat (limited to 'patches/source/glibc/glibc-2.15_avx2.diff')
| -rw-r--r-- | patches/source/glibc/glibc-2.15_avx2.diff | 171 |
1 files changed, 171 insertions, 0 deletions
diff --git a/patches/source/glibc/glibc-2.15_avx2.diff b/patches/source/glibc/glibc-2.15_avx2.diff new file mode 100644 index 000000000..1b1b88be9 --- /dev/null +++ b/patches/source/glibc/glibc-2.15_avx2.diff @@ -0,0 +1,171 @@ +From: Ulrich Drepper <drepper@gmail.com> +Date: Thu, 26 Jan 2012 14:45:54 +0000 (-0500) +Subject: Really fix AVX tests +X-Git-Tag: glibc-2.16-tps~1052 +X-Git-Url: http://sources.redhat.com/git/gitweb.cgi?p=glibc.git;a=commitdiff_plain;h=08cf777f9e7f6d826658a99c7d77a359f73a45bf + +Really fix AVX tests + +There is no problem with strcmp, it doesn't use the YMM registers. +The math routines might since gcc perhaps generates such code. +Introduce bit_YMM_USBALE and use it in the math routines. +--- + +--- sysdeps/x86_64/fpu/multiarch/e_atan2.c ++++ sysdeps/x86_64/fpu/multiarch/e_atan2.c +@@ -14,7 +14,7 @@ extern double __ieee754_atan2_fma4 (double, double); + + libm_ifunc (__ieee754_atan2, + HAS_FMA4 ? __ieee754_atan2_fma4 +- : (HAS_AVX ? __ieee754_atan2_avx : __ieee754_atan2_sse2)); ++ : (HAS_YMM_USABLE ? __ieee754_atan2_avx : __ieee754_atan2_sse2)); + strong_alias (__ieee754_atan2, __atan2_finite) + + # define __ieee754_atan2 __ieee754_atan2_sse2 +--- sysdeps/x86_64/fpu/multiarch/e_exp.c ++++ sysdeps/x86_64/fpu/multiarch/e_exp.c +@@ -14,7 +14,7 @@ extern double __ieee754_exp_fma4 (double); + + libm_ifunc (__ieee754_exp, + HAS_FMA4 ? __ieee754_exp_fma4 +- : (HAS_AVX ? __ieee754_exp_avx : __ieee754_exp_sse2)); ++ : (HAS_YMM_USABLE ? __ieee754_exp_avx : __ieee754_exp_sse2)); + strong_alias (__ieee754_exp, __exp_finite) + + # define __ieee754_exp __ieee754_exp_sse2 +--- sysdeps/x86_64/fpu/multiarch/e_log.c ++++ sysdeps/x86_64/fpu/multiarch/e_log.c +@@ -14,7 +14,7 @@ extern double __ieee754_log_fma4 (double); + + libm_ifunc (__ieee754_log, + HAS_FMA4 ? __ieee754_log_fma4 +- : (HAS_AVX ? __ieee754_log_avx ++ : (HAS_YMM_USABLE ? __ieee754_log_avx + : __ieee754_log_sse2)); + strong_alias (__ieee754_log, __log_finite) + +--- sysdeps/x86_64/fpu/multiarch/s_atan.c ++++ sysdeps/x86_64/fpu/multiarch/s_atan.c +@@ -12,7 +12,8 @@ extern double __atan_fma4 (double); + # define __atan_fma4 ((void *) 0) + # endif + +-libm_ifunc (atan, HAS_FMA4 ? __atan_fma4 : HAS_AVX ? __atan_avx : __atan_sse2); ++libm_ifunc (atan, (HAS_FMA4 ? __atan_fma4 : ++ HAS_YMM_USABLE ? __atan_avx : __atan_sse2)); + + # define atan __atan_sse2 + #endif +--- sysdeps/x86_64/fpu/multiarch/s_sin.c ++++ sysdeps/x86_64/fpu/multiarch/s_sin.c +@@ -17,10 +17,12 @@ extern double __sin_fma4 (double); + # define __sin_fma4 ((void *) 0) + # endif + +-libm_ifunc (__cos, HAS_FMA4 ? __cos_fma4 : HAS_AVX ? __cos_avx : __cos_sse2); ++libm_ifunc (__cos, (HAS_FMA4 ? __cos_fma4 : ++ HAS_YMM_USABLE ? __cos_avx : __cos_sse2)); + weak_alias (__cos, cos) + +-libm_ifunc (__sin, HAS_FMA4 ? __sin_fma4 : HAS_AVX ? __sin_avx : __sin_sse2); ++libm_ifunc (__sin, (HAS_FMA4 ? __sin_fma4 : ++ HAS_YMM_USABLE ? __sin_avx : __sin_sse2)); + weak_alias (__sin, sin) + + # define __cos __cos_sse2 +--- sysdeps/x86_64/fpu/multiarch/s_tan.c ++++ sysdeps/x86_64/fpu/multiarch/s_tan.c +@@ -12,7 +12,8 @@ extern double __tan_fma4 (double); + # define __tan_fma4 ((void *) 0) + # endif + +-libm_ifunc (tan, HAS_FMA4 ? __tan_fma4 : HAS_AVX ? __tan_avx : __tan_sse2); ++libm_ifunc (tan, (HAS_FMA4 ? __tan_fma4 : ++ HAS_YMM_USABLE ? __tan_avx : __tan_sse2)); + + # define tan __tan_sse2 + #endif +--- sysdeps/x86_64/multiarch/init-arch.c ++++ sysdeps/x86_64/multiarch/init-arch.c +@@ -147,13 +147,13 @@ __init_cpu_features (void) + if (__cpu_features.cpuid[COMMON_CPUID_INDEX_1].ecx & bit_AVX) + { + /* Reset the AVX bit in case OSXSAVE is disabled. */ +- if ((__cpu_features.cpuid[COMMON_CPUID_INDEX_1].ecx & bit_OSXSAVE) == 0 +- || ({ unsigned int xcrlow; +- unsigned int xcrhigh; +- asm ("xgetbv" +- : "=a" (xcrlow), "=d" (xcrhigh) : "c" (0)); +- (xcrlow & 6) != 6; })) +- __cpu_features.cpuid[COMMON_CPUID_INDEX_1].ecx &= ~bit_AVX; ++ if ((__cpu_features.cpuid[COMMON_CPUID_INDEX_1].ecx & bit_OSXSAVE) != 0 ++ && ({ unsigned int xcrlow; ++ unsigned int xcrhigh; ++ asm ("xgetbv" ++ : "=a" (xcrlow), "=d" (xcrhigh) : "c" (0)); ++ (xcrlow & 6) == 6; })) ++ __cpu_features.feature[index_YMM_Usable] |= bit_YMM_Usable; + } + + __cpu_features.family = family; +--- sysdeps/x86_64/multiarch/init-arch.h ++++ sysdeps/x86_64/multiarch/init-arch.h +@@ -22,6 +22,7 @@ + #define bit_Prefer_SSE_for_memop (1 << 3) + #define bit_Fast_Unaligned_Load (1 << 4) + #define bit_Prefer_PMINUB_for_stringop (1 << 5) ++#define bit_YMM_Usable (1 << 6) + + #define bit_SSE2 (1 << 26) + #define bit_SSSE3 (1 << 9) +@@ -49,6 +50,7 @@ + # define index_Prefer_SSE_for_memop FEATURE_INDEX_1*FEATURE_SIZE + # define index_Fast_Unaligned_Load FEATURE_INDEX_1*FEATURE_SIZE + # define index_Prefer_PMINUB_for_stringop FEATURE_INDEX_1*FEATURE_SIZE ++# define index_YMM_Usable FEATURE_INDEX_1*FEATURE_SIZE + + #else /* __ASSEMBLER__ */ + +@@ -93,7 +95,7 @@ extern struct cpu_features + + + extern void __init_cpu_features (void) attribute_hidden; +-#define INIT_ARCH()\ ++# define INIT_ARCH() \ + do \ + if (__cpu_features.kind == arch_kind_unknown) \ + __init_cpu_features (); \ +@@ -126,23 +128,21 @@ extern const struct cpu_features *__get_cpu_features (void) + # define index_Slow_BSF FEATURE_INDEX_1 + # define index_Prefer_SSE_for_memop FEATURE_INDEX_1 + # define index_Fast_Unaligned_Load FEATURE_INDEX_1 ++# define index_YMM_Usable FEATURE_INDEX_1 + +-#define HAS_ARCH_FEATURE(idx, bit) \ +- ((__get_cpu_features ()->feature[idx] & (bit)) != 0) ++# define HAS_ARCH_FEATURE(name) \ ++ ((__get_cpu_features ()->feature[index_##name] & (bit_##name)) != 0) + +-#define HAS_FAST_REP_STRING \ +- HAS_ARCH_FEATURE (index_Fast_Rep_String, bit_Fast_Rep_String) ++# define HAS_FAST_REP_STRING HAS_ARCH_FEATURE (Fast_Rep_String) + +-#define HAS_FAST_COPY_BACKWARD \ +- HAS_ARCH_FEATURE (index_Fast_Copy_Backward, bit_Fast_Copy_Backward) ++# define HAS_FAST_COPY_BACKWARD HAS_ARCH_FEATURE (Fast_Copy_Backward) + +-#define HAS_SLOW_BSF \ +- HAS_ARCH_FEATURE (index_Slow_BSF, bit_Slow_BSF) ++# define HAS_SLOW_BSF HAS_ARCH_FEATURE (Slow_BSF) + +-#define HAS_PREFER_SSE_FOR_MEMOP \ +- HAS_ARCH_FEATURE (index_Prefer_SSE_for_memop, bit_Prefer_SSE_for_memop) ++# define HAS_PREFER_SSE_FOR_MEMOP HAS_ARCH_FEATURE (Prefer_SSE_for_memop) + +-#define HAS_FAST_UNALIGNED_LOAD \ +- HAS_ARCH_FEATURE (index_Fast_Unaligned_Load, bit_Fast_Unaligned_Load) ++# define HAS_FAST_UNALIGNED_LOAD HAS_ARCH_FEATURE (Fast_Unaligned_Load) ++ ++# define HAS_YMM_USABLE HAS_ARCH_FEATURE (YMM_Usable) + + #endif /* __ASSEMBLER__ */ |