Contents of /trunk/glibc/patches/glibc-2.15-avx.patch
Parent Directory | Revision Log
Revision 1806 -
(show annotations)
(download)
Tue Jun 12 12:33:27 2012 UTC (12 years, 3 months ago) by niro
File size: 7516 byte(s)
Tue Jun 12 12:33:27 2012 UTC (12 years, 3 months ago) by niro
File size: 7516 byte(s)
-added patches for 2.15-r1
1 | diff --git a/sysdeps/x86_64/fpu/multiarch/e_atan2.c b/sysdeps/x86_64/fpu/multiarch/e_atan2.c |
2 | index 6867c6e..3a615fc 100644 |
3 | --- a/sysdeps/x86_64/fpu/multiarch/e_atan2.c |
4 | +++ b/sysdeps/x86_64/fpu/multiarch/e_atan2.c |
5 | @@ -14,7 +14,7 @@ extern double __ieee754_atan2_fma4 (double, double); |
6 | |
7 | libm_ifunc (__ieee754_atan2, |
8 | HAS_FMA4 ? __ieee754_atan2_fma4 |
9 | - : (HAS_AVX ? __ieee754_atan2_avx : __ieee754_atan2_sse2)); |
10 | + : (HAS_YMM_USABLE ? __ieee754_atan2_avx : __ieee754_atan2_sse2)); |
11 | strong_alias (__ieee754_atan2, __atan2_finite) |
12 | |
13 | # define __ieee754_atan2 __ieee754_atan2_sse2 |
14 | diff --git a/sysdeps/x86_64/fpu/multiarch/e_exp.c b/sysdeps/x86_64/fpu/multiarch/e_exp.c |
15 | index 3c65028..7b2320a 100644 |
16 | --- a/sysdeps/x86_64/fpu/multiarch/e_exp.c |
17 | +++ b/sysdeps/x86_64/fpu/multiarch/e_exp.c |
18 | @@ -14,7 +14,7 @@ extern double __ieee754_exp_fma4 (double); |
19 | |
20 | libm_ifunc (__ieee754_exp, |
21 | HAS_FMA4 ? __ieee754_exp_fma4 |
22 | - : (HAS_AVX ? __ieee754_exp_avx : __ieee754_exp_sse2)); |
23 | + : (HAS_YMM_USABLE ? __ieee754_exp_avx : __ieee754_exp_sse2)); |
24 | strong_alias (__ieee754_exp, __exp_finite) |
25 | |
26 | # define __ieee754_exp __ieee754_exp_sse2 |
27 | diff --git a/sysdeps/x86_64/fpu/multiarch/e_log.c b/sysdeps/x86_64/fpu/multiarch/e_log.c |
28 | index 3b468d0..ab277d6 100644 |
29 | --- a/sysdeps/x86_64/fpu/multiarch/e_log.c |
30 | +++ b/sysdeps/x86_64/fpu/multiarch/e_log.c |
31 | @@ -14,7 +14,7 @@ extern double __ieee754_log_fma4 (double); |
32 | |
33 | libm_ifunc (__ieee754_log, |
34 | HAS_FMA4 ? __ieee754_log_fma4 |
35 | - : (HAS_AVX ? __ieee754_log_avx |
36 | + : (HAS_YMM_USABLE ? __ieee754_log_avx |
37 | : __ieee754_log_sse2)); |
38 | strong_alias (__ieee754_log, __log_finite) |
39 | |
40 | diff --git a/sysdeps/x86_64/fpu/multiarch/s_atan.c b/sysdeps/x86_64/fpu/multiarch/s_atan.c |
41 | index 3160201..78c7e09 100644 |
42 | --- a/sysdeps/x86_64/fpu/multiarch/s_atan.c |
43 | +++ b/sysdeps/x86_64/fpu/multiarch/s_atan.c |
44 | @@ -12,7 +12,8 @@ extern double __atan_fma4 (double); |
45 | # define __atan_fma4 ((void *) 0) |
46 | # endif |
47 | |
48 | -libm_ifunc (atan, HAS_FMA4 ? __atan_fma4 : HAS_AVX ? __atan_avx : __atan_sse2); |
49 | +libm_ifunc (atan, (HAS_FMA4 ? __atan_fma4 : |
50 | + HAS_YMM_USABLE ? __atan_avx : __atan_sse2)); |
51 | |
52 | # define atan __atan_sse2 |
53 | #endif |
54 | diff --git a/sysdeps/x86_64/fpu/multiarch/s_sin.c b/sysdeps/x86_64/fpu/multiarch/s_sin.c |
55 | index 1ba9dbc..417acd0 100644 |
56 | --- a/sysdeps/x86_64/fpu/multiarch/s_sin.c |
57 | +++ b/sysdeps/x86_64/fpu/multiarch/s_sin.c |
58 | @@ -17,10 +17,12 @@ extern double __sin_fma4 (double); |
59 | # define __sin_fma4 ((void *) 0) |
60 | # endif |
61 | |
62 | -libm_ifunc (__cos, HAS_FMA4 ? __cos_fma4 : HAS_AVX ? __cos_avx : __cos_sse2); |
63 | +libm_ifunc (__cos, (HAS_FMA4 ? __cos_fma4 : |
64 | + HAS_YMM_USABLE ? __cos_avx : __cos_sse2)); |
65 | weak_alias (__cos, cos) |
66 | |
67 | -libm_ifunc (__sin, HAS_FMA4 ? __sin_fma4 : HAS_AVX ? __sin_avx : __sin_sse2); |
68 | +libm_ifunc (__sin, (HAS_FMA4 ? __sin_fma4 : |
69 | + HAS_YMM_USABLE ? __sin_avx : __sin_sse2)); |
70 | weak_alias (__sin, sin) |
71 | |
72 | # define __cos __cos_sse2 |
73 | diff --git a/sysdeps/x86_64/fpu/multiarch/s_tan.c b/sysdeps/x86_64/fpu/multiarch/s_tan.c |
74 | index 8f6601e..3047155 100644 |
75 | --- a/sysdeps/x86_64/fpu/multiarch/s_tan.c |
76 | +++ b/sysdeps/x86_64/fpu/multiarch/s_tan.c |
77 | @@ -12,7 +12,8 @@ extern double __tan_fma4 (double); |
78 | # define __tan_fma4 ((void *) 0) |
79 | # endif |
80 | |
81 | -libm_ifunc (tan, HAS_FMA4 ? __tan_fma4 : HAS_AVX ? __tan_avx : __tan_sse2); |
82 | +libm_ifunc (tan, (HAS_FMA4 ? __tan_fma4 : |
83 | + HAS_YMM_USABLE ? __tan_avx : __tan_sse2)); |
84 | |
85 | # define tan __tan_sse2 |
86 | #endif |
87 | diff --git a/sysdeps/x86_64/multiarch/init-arch.c b/sysdeps/x86_64/multiarch/init-arch.c |
88 | index 65b0ee9..76d146c 100644 |
89 | --- a/sysdeps/x86_64/multiarch/init-arch.c |
90 | +++ b/sysdeps/x86_64/multiarch/init-arch.c |
91 | @@ -1,6 +1,6 @@ |
92 | /* Initialize CPU feature data. |
93 | This file is part of the GNU C Library. |
94 | - Copyright (C) 2008, 2009, 2010, 2011 Free Software Foundation, Inc. |
95 | + Copyright (C) 2008, 2009, 2010, 2011, 2012 Free Software Foundation, Inc. |
96 | Contributed by Ulrich Drepper <drepper@redhat.com>. |
97 | |
98 | The GNU C Library is free software; you can redistribute it and/or |
99 | @@ -144,6 +144,18 @@ __init_cpu_features (void) |
100 | else |
101 | kind = arch_kind_other; |
102 | |
103 | + if (__cpu_features.cpuid[COMMON_CPUID_INDEX_1].ecx & bit_AVX) |
104 | + { |
105 | + /* Reset the AVX bit in case OSXSAVE is disabled. */ |
106 | + if ((__cpu_features.cpuid[COMMON_CPUID_INDEX_1].ecx & bit_OSXSAVE) != 0 |
107 | + && ({ unsigned int xcrlow; |
108 | + unsigned int xcrhigh; |
109 | + asm ("xgetbv" |
110 | + : "=a" (xcrlow), "=d" (xcrhigh) : "c" (0)); |
111 | + (xcrlow & 6) == 6; })) |
112 | + __cpu_features.feature[index_YMM_Usable] |= bit_YMM_Usable; |
113 | + } |
114 | + |
115 | __cpu_features.family = family; |
116 | __cpu_features.model = model; |
117 | atomic_write_barrier (); |
118 | diff --git a/sysdeps/x86_64/multiarch/init-arch.h b/sysdeps/x86_64/multiarch/init-arch.h |
119 | index 2a1df39..2dc75ab 100644 |
120 | --- a/sysdeps/x86_64/multiarch/init-arch.h |
121 | +++ b/sysdeps/x86_64/multiarch/init-arch.h |
122 | @@ -1,5 +1,5 @@ |
123 | /* This file is part of the GNU C Library. |
124 | - Copyright (C) 2008, 2009, 2010, 2011 Free Software Foundation, Inc. |
125 | + Copyright (C) 2008, 2009, 2010, 2011, 2012 Free Software Foundation, Inc. |
126 | |
127 | The GNU C Library is free software; you can redistribute it and/or |
128 | modify it under the terms of the GNU Lesser General Public |
129 | @@ -22,11 +22,13 @@ |
130 | #define bit_Prefer_SSE_for_memop (1 << 3) |
131 | #define bit_Fast_Unaligned_Load (1 << 4) |
132 | #define bit_Prefer_PMINUB_for_stringop (1 << 5) |
133 | +#define bit_YMM_Usable (1 << 6) |
134 | |
135 | #define bit_SSE2 (1 << 26) |
136 | #define bit_SSSE3 (1 << 9) |
137 | #define bit_SSE4_1 (1 << 19) |
138 | #define bit_SSE4_2 (1 << 20) |
139 | +#define bit_OSXSAVE (1 << 27) |
140 | #define bit_AVX (1 << 28) |
141 | #define bit_POPCOUNT (1 << 23) |
142 | #define bit_FMA (1 << 12) |
143 | @@ -48,6 +50,7 @@ |
144 | # define index_Prefer_SSE_for_memop FEATURE_INDEX_1*FEATURE_SIZE |
145 | # define index_Fast_Unaligned_Load FEATURE_INDEX_1*FEATURE_SIZE |
146 | # define index_Prefer_PMINUB_for_stringop FEATURE_INDEX_1*FEATURE_SIZE |
147 | +# define index_YMM_Usable FEATURE_INDEX_1*FEATURE_SIZE |
148 | |
149 | #else /* __ASSEMBLER__ */ |
150 | |
151 | @@ -92,7 +95,7 @@ extern struct cpu_features |
152 | |
153 | |
154 | extern void __init_cpu_features (void) attribute_hidden; |
155 | -#define INIT_ARCH()\ |
156 | +# define INIT_ARCH() \ |
157 | do \ |
158 | if (__cpu_features.kind == arch_kind_unknown) \ |
159 | __init_cpu_features (); \ |
160 | @@ -125,23 +128,21 @@ extern const struct cpu_features *__get_cpu_features (void) |
161 | # define index_Slow_BSF FEATURE_INDEX_1 |
162 | # define index_Prefer_SSE_for_memop FEATURE_INDEX_1 |
163 | # define index_Fast_Unaligned_Load FEATURE_INDEX_1 |
164 | +# define index_YMM_Usable FEATURE_INDEX_1 |
165 | |
166 | -#define HAS_ARCH_FEATURE(idx, bit) \ |
167 | - ((__get_cpu_features ()->feature[idx] & (bit)) != 0) |
168 | +# define HAS_ARCH_FEATURE(name) \ |
169 | + ((__get_cpu_features ()->feature[index_##name] & (bit_##name)) != 0) |
170 | |
171 | -#define HAS_FAST_REP_STRING \ |
172 | - HAS_ARCH_FEATURE (index_Fast_Rep_String, bit_Fast_Rep_String) |
173 | +# define HAS_FAST_REP_STRING HAS_ARCH_FEATURE (Fast_Rep_String) |
174 | |
175 | -#define HAS_FAST_COPY_BACKWARD \ |
176 | - HAS_ARCH_FEATURE (index_Fast_Copy_Backward, bit_Fast_Copy_Backward) |
177 | +# define HAS_FAST_COPY_BACKWARD HAS_ARCH_FEATURE (Fast_Copy_Backward) |
178 | |
179 | -#define HAS_SLOW_BSF \ |
180 | - HAS_ARCH_FEATURE (index_Slow_BSF, bit_Slow_BSF) |
181 | +# define HAS_SLOW_BSF HAS_ARCH_FEATURE (Slow_BSF) |
182 | |
183 | -#define HAS_PREFER_SSE_FOR_MEMOP \ |
184 | - HAS_ARCH_FEATURE (index_Prefer_SSE_for_memop, bit_Prefer_SSE_for_memop) |
185 | +# define HAS_PREFER_SSE_FOR_MEMOP HAS_ARCH_FEATURE (Prefer_SSE_for_memop) |
186 | |
187 | -#define HAS_FAST_UNALIGNED_LOAD \ |
188 | - HAS_ARCH_FEATURE (index_Fast_Unaligned_Load, bit_Fast_Unaligned_Load) |
189 | +# define HAS_FAST_UNALIGNED_LOAD HAS_ARCH_FEATURE (Fast_Unaligned_Load) |
190 | + |
191 | +# define HAS_YMM_USABLE HAS_ARCH_FEATURE (YMM_Usable) |
192 | |
193 | #endif /* __ASSEMBLER__ */ |
194 | -- |
195 | 1.7.9 |