Mercurial > libavutil.hg
annotate x86/intreadwrite.h @ 998:bbe201fb5a98 libavutil
Use attribute force_align_arg_pointer only on x86_32.
author | cehoyos |
---|---|
date | Sat, 17 Jul 2010 16:35:01 +0000 |
parents | 4d9ad0ed07d0 |
children |
rev | line source |
---|---|
813
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
1 /* |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
2 * Copyright (c) 2010 Alexander Strange <astrange@ithinksw.com> |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
3 * |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
4 * This file is part of FFmpeg. |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
5 * |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
6 * FFmpeg is free software; you can redistribute it and/or |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
7 * modify it under the terms of the GNU Lesser General Public |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
8 * License as published by the Free Software Foundation; either |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
9 * version 2.1 of the License, or (at your option) any later version. |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
10 * |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
11 * FFmpeg is distributed in the hope that it will be useful, |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
14 * Lesser General Public License for more details. |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
15 * |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
16 * You should have received a copy of the GNU Lesser General Public |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
17 * License along with FFmpeg; if not, write to the Free Software |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
19 */ |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
20 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
21 #ifndef AVUTIL_X86_INTREADWRITE_H |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
22 #define AVUTIL_X86_INTREADWRITE_H |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
23 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
24 #include <stdint.h> |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
25 #include "config.h" |
873
4d9ad0ed07d0
Replace many includes of libavutil/common.h with what is actually needed
mru
parents:
862
diff
changeset
|
26 #include "libavutil/attributes.h" |
813
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
27 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
28 #if HAVE_MMX |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
29 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
30 #if !HAVE_FAST_64BIT && defined(__MMX__) |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
31 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
32 #define AV_COPY64 AV_COPY64 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
33 static av_always_inline void AV_COPY64(void *d, const void *s) |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
34 { |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
35 __asm__("movq %1, %%mm0 \n\t" |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
36 "movq %%mm0, %0 \n\t" |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
37 : "=m"(*(uint64_t*)d) |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
38 : "m" (*(const uint64_t*)s) |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
39 : "mm0"); |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
40 } |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
41 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
42 #define AV_SWAP64 AV_SWAP64 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
43 static av_always_inline void AV_SWAP64(void *a, void *b) |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
44 { |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
45 __asm__("movq %1, %%mm0 \n\t" |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
46 "movq %0, %%mm1 \n\t" |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
47 "movq %%mm0, %0 \n\t" |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
48 "movq %%mm1, %1 \n\t" |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
49 : "+m"(*(uint64_t*)a), "+m"(*(uint64_t*)b) |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
50 ::"mm0", "mm1"); |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
51 } |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
52 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
53 #define AV_ZERO64 AV_ZERO64 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
54 static av_always_inline void AV_ZERO64(void *d) |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
55 { |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
56 __asm__("pxor %%mm0, %%mm0 \n\t" |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
57 "movq %%mm0, %0 \n\t" |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
58 : "=m"(*(uint64_t*)d) |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
59 :: "mm0"); |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
60 } |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
61 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
62 #endif /* !HAVE_FAST_64BIT && defined(__MMX__) */ |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
63 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
64 #ifdef __SSE__ |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
65 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
66 #define AV_COPY128 AV_COPY128 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
67 static av_always_inline void AV_COPY128(void *d, const void *s) |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
68 { |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
69 struct v {uint64_t v[2];}; |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
70 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
71 __asm__("movaps %1, %%xmm0 \n\t" |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
72 "movaps %%xmm0, %0 \n\t" |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
73 : "=m"(*(struct v*)d) |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
74 : "m" (*(const struct v*)s) |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
75 : "xmm0"); |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
76 } |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
77 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
78 #endif /* __SSE__ */ |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
79 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
80 #ifdef __SSE2__ |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
81 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
82 #define AV_ZERO128 AV_ZERO128 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
83 static av_always_inline void AV_ZERO128(void *d) |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
84 { |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
85 struct v {uint64_t v[2];}; |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
86 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
87 __asm__("pxor %%xmm0, %%xmm0 \n\t" |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
88 "movdqa %%xmm0, %0 \n\t" |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
89 : "=m"(*(struct v*)d) |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
90 :: "xmm0"); |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
91 } |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
92 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
93 #endif /* __SSE2__ */ |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
94 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
95 #endif /* HAVE_MMX */ |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
96 |
7c4f89909560
Add macros for 64- and 128-bit write-combining optimization to intreadwrite.h.
astrange
parents:
diff
changeset
|
97 #endif /* AVUTIL_X86_INTREADWRITE_H */ |