Mercurial > libavcodec.hg
annotate ppc/dsputil_ppc.c @ 8621:72e109759617 libavcodec
Alpha: add ff_ prefix to idct functions
author | mru |
---|---|
date | Sun, 18 Jan 2009 05:52:17 +0000 |
parents | 68e959302527 |
children | 13b6cdf7895e |
rev | line source |
---|---|
828
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
748
diff
changeset
|
1 /* |
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
748
diff
changeset
|
2 * Copyright (c) 2002 Brian Foley |
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
748
diff
changeset
|
3 * Copyright (c) 2002 Dieter Shirley |
1949
66215baae7b9
hadamard8_diff8x8 in AltiVec, the 16bits edition by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
1879
diff
changeset
|
4 * Copyright (c) 2003-2004 Romain Dolbeau <romain@dolbeau.org> |
828
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
748
diff
changeset
|
5 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3581
diff
changeset
|
6 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3581
diff
changeset
|
7 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3581
diff
changeset
|
8 * FFmpeg is free software; you can redistribute it and/or |
828
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
748
diff
changeset
|
9 * modify it under the terms of the GNU Lesser General Public |
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
748
diff
changeset
|
10 * License as published by the Free Software Foundation; either |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3581
diff
changeset
|
11 * version 2.1 of the License, or (at your option) any later version. |
828
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
748
diff
changeset
|
12 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3581
diff
changeset
|
13 * FFmpeg is distributed in the hope that it will be useful, |
828
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
748
diff
changeset
|
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
748
diff
changeset
|
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
748
diff
changeset
|
16 * Lesser General Public License for more details. |
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
748
diff
changeset
|
17 * |
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
748
diff
changeset
|
18 * You should have received a copy of the GNU Lesser General Public |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3581
diff
changeset
|
19 * License along with FFmpeg; if not, write to the Free Software |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
2979
diff
changeset
|
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
828
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
748
diff
changeset
|
21 */ |
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
748
diff
changeset
|
22 |
6763 | 23 #include "libavcodec/dsputil.h" |
638
0012f75c92bb
altivec build tidyup patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
diff
changeset
|
24 |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
25 #include "dsputil_ppc.h" |
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
26 |
8590 | 27 #if HAVE_ALTIVEC |
638
0012f75c92bb
altivec build tidyup patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
diff
changeset
|
28 #include "dsputil_altivec.h" |
0012f75c92bb
altivec build tidyup patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
diff
changeset
|
29 |
8250 | 30 void fdct_altivec(int16_t *block); |
31 void gmc1_altivec(uint8_t *dst, uint8_t *src, int stride, int h, | |
32 int x16, int y16, int rounder); | |
33 void idct_put_altivec(uint8_t *dest, int line_size, int16_t *block); | |
34 void idct_add_altivec(uint8_t *dest, int line_size, int16_t *block); | |
3547 | 35 |
36 void dsputil_h264_init_ppc(DSPContext* c, AVCodecContext *avctx); | |
3532 | 37 |
3546
5f97ba9a4eaa
Almost cosmetic changes in dsputil_init_ppc and vorbis_inverse_coupling_altivec:
lu_zero
parents:
3542
diff
changeset
|
38 void dsputil_init_altivec(DSPContext* c, AVCodecContext *avctx); |
3542 | 39 void vc1dsp_init_altivec(DSPContext* c, AVCodecContext *avctx); |
3547 | 40 void snow_init_altivec(DSPContext* c, AVCodecContext *avctx); |
3581 | 41 void float_init_altivec(DSPContext* c, AVCodecContext *avctx); |
4838
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
4521
diff
changeset
|
42 void int_init_altivec(DSPContext* c, AVCodecContext *avctx); |
3532 | 43 |
44 #endif | |
3223 | 45 |
4197 | 46 int mm_flags = 0; |
894
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
47 |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
981
diff
changeset
|
48 int mm_support(void) |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
981
diff
changeset
|
49 { |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
981
diff
changeset
|
50 int result = 0; |
8590 | 51 #if HAVE_ALTIVEC |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
981
diff
changeset
|
52 if (has_altivec()) { |
8104
0d108ec85620
Remove duplicated MM_* macros for CPU capabilities from dsputil.h.
rathann
parents:
8031
diff
changeset
|
53 result |= FF_MM_ALTIVEC; |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
981
diff
changeset
|
54 } |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
981
diff
changeset
|
55 #endif /* result */ |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
981
diff
changeset
|
56 return result; |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
981
diff
changeset
|
57 } |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
981
diff
changeset
|
58 |
8590 | 59 #if CONFIG_POWERPC_PERF |
1352
e8ff4783f188
1) remove TBL support in PPC performance. It's much more useful to use the
michaelni
parents:
1340
diff
changeset
|
60 unsigned long long perfdata[POWERPC_NUM_PMC_ENABLED][powerpc_perf_total][powerpc_data_total]; |
1024
9cc1031e1864
More AltiVec MC functions patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1015
diff
changeset
|
61 /* list below must match enum in dsputil_ppc.h */ |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
62 static unsigned char* perfname[] = { |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
63 "ff_fft_calc_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
64 "gmc1_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
65 "dct_unquantize_h263_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
66 "fdct_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
67 "idct_add_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
68 "idct_put_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
69 "put_pixels16_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
70 "avg_pixels16_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
71 "avg_pixels8_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
72 "put_pixels8_xy2_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
73 "put_no_rnd_pixels8_xy2_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
74 "put_pixels16_xy2_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
75 "put_no_rnd_pixels16_xy2_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
76 "hadamard8_diff8x8_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
77 "hadamard8_diff16_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
78 "avg_pixels8_xy2_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
79 "clear_blocks_dcbz32_ppc", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
80 "clear_blocks_dcbz128_ppc", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
81 "put_h264_chroma_mc8_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
82 "avg_h264_chroma_mc8_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
83 "put_h264_qpel16_h_lowpass_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
84 "avg_h264_qpel16_h_lowpass_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
85 "put_h264_qpel16_v_lowpass_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
86 "avg_h264_qpel16_v_lowpass_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
87 "put_h264_qpel16_hv_lowpass_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
88 "avg_h264_qpel16_hv_lowpass_altivec", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
89 "" |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
90 }; |
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
91 #include <stdio.h> |
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
92 #endif |
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
93 |
8590 | 94 #if CONFIG_POWERPC_PERF |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
95 void powerpc_display_perf_report(void) |
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
96 { |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
97 int i, j; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
98 av_log(NULL, AV_LOG_INFO, "PowerPC performance report\n Values are from the PMC registers, and represent whatever the registers are set to record.\n"); |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
99 for(i = 0 ; i < powerpc_perf_total ; i++) { |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
100 for (j = 0; j < POWERPC_NUM_PMC_ENABLED ; j++) { |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
101 if (perfdata[j][i][powerpc_data_num] != (unsigned long long)0) |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
102 av_log(NULL, AV_LOG_INFO, |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
103 " Function \"%s\" (pmc%d):\n\tmin: %"PRIu64"\n\tmax: %"PRIu64"\n\tavg: %1.2lf (%"PRIu64")\n", |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
104 perfname[i], |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
105 j+1, |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
106 perfdata[j][i][powerpc_data_min], |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
107 perfdata[j][i][powerpc_data_max], |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
108 (double)perfdata[j][i][powerpc_data_sum] / |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
109 (double)perfdata[j][i][powerpc_data_num], |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
110 perfdata[j][i][powerpc_data_num]); |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
111 } |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
112 } |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
113 } |
4521
891590781d9e
rename POWERPC_PERFORMANCE_REPORT to CONFIG_POWERPC_PERF
mru
parents:
4227
diff
changeset
|
114 #endif /* CONFIG_POWERPC_PERF */ |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
115 |
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
116 /* ***** WARNING ***** WARNING ***** WARNING ***** */ |
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
117 /* |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
118 clear_blocks_dcbz32_ppc will not work properly on PowerPC processors with a |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
119 cache line size not equal to 32 bytes. |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
120 Fortunately all processor used by Apple up to at least the 7450 (aka second |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
121 generation G4) use 32 bytes cache line. |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
122 This is due to the use of the 'dcbz' instruction. It simply clear to zero a |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
123 single cache line, so you need to know the cache line size to use it ! |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
124 It's absurd, but it's fast... |
1334
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
125 |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
126 update 24/06/2003 : Apple released yesterday the G5, with a PPC970. cache line |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
127 size: 128 bytes. Oups. |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
128 The semantic of dcbz was changed, it always clear 32 bytes. so the function |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
129 below will work, but will be slow. So I fixed check_dcbz_effect to use dcbzl, |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
130 which is defined to clear a cache line (as dcbz before). So we still can |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
131 distinguish, and use dcbz (32 bytes) or dcbzl (one cache line) as required. |
1334
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
132 |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
133 see <http://developer.apple.com/technotes/tn/tn2087.html> |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
134 and <http://developer.apple.com/technotes/tn/tn2086.html> |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
135 */ |
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
136 void clear_blocks_dcbz32_ppc(DCTELEM *blocks) |
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
137 { |
1352
e8ff4783f188
1) remove TBL support in PPC performance. It's much more useful to use the
michaelni
parents:
1340
diff
changeset
|
138 POWERPC_PERF_DECLARE(powerpc_clear_blocks_dcbz32, 1); |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
139 register int misal = ((unsigned long)blocks & 0x00000010); |
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
140 register int i = 0; |
1352
e8ff4783f188
1) remove TBL support in PPC performance. It's much more useful to use the
michaelni
parents:
1340
diff
changeset
|
141 POWERPC_PERF_START_COUNT(powerpc_clear_blocks_dcbz32, 1); |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
142 #if 1 |
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
143 if (misal) { |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
144 ((unsigned long*)blocks)[0] = 0L; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
145 ((unsigned long*)blocks)[1] = 0L; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
146 ((unsigned long*)blocks)[2] = 0L; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
147 ((unsigned long*)blocks)[3] = 0L; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
148 i += 16; |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
149 } |
2294
fac626a2b73b
missaliged clear_blocks() and h264 not complied but referenced fix patch by (Roine Gustafsson <roine at users dot sourceforge dot net>) and me
michael
parents:
2236
diff
changeset
|
150 for ( ; i < sizeof(DCTELEM)*6*64-31 ; i += 32) { |
8031 | 151 __asm__ volatile("dcbz %0,%1" : : "b" (blocks), "r" (i) : "memory"); |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
152 } |
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
153 if (misal) { |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
154 ((unsigned long*)blocks)[188] = 0L; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
155 ((unsigned long*)blocks)[189] = 0L; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
156 ((unsigned long*)blocks)[190] = 0L; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
157 ((unsigned long*)blocks)[191] = 0L; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
158 i += 16; |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
159 } |
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
160 #else |
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
161 memset(blocks, 0, sizeof(DCTELEM)*6*64); |
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
162 #endif |
1352
e8ff4783f188
1) remove TBL support in PPC performance. It's much more useful to use the
michaelni
parents:
1340
diff
changeset
|
163 POWERPC_PERF_STOP_COUNT(powerpc_clear_blocks_dcbz32, 1); |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
164 } |
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
165 |
1334
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
166 /* same as above, when dcbzl clear a whole 128B cache line |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
167 i.e. the PPC970 aka G5 */ |
8590 | 168 #if HAVE_DCBZL |
1334
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
169 void clear_blocks_dcbz128_ppc(DCTELEM *blocks) |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
170 { |
1352
e8ff4783f188
1) remove TBL support in PPC performance. It's much more useful to use the
michaelni
parents:
1340
diff
changeset
|
171 POWERPC_PERF_DECLARE(powerpc_clear_blocks_dcbz128, 1); |
1334
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
172 register int misal = ((unsigned long)blocks & 0x0000007f); |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
173 register int i = 0; |
1352
e8ff4783f188
1) remove TBL support in PPC performance. It's much more useful to use the
michaelni
parents:
1340
diff
changeset
|
174 POWERPC_PERF_START_COUNT(powerpc_clear_blocks_dcbz128, 1); |
1334
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
175 #if 1 |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
176 if (misal) { |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
177 // we could probably also optimize this case, |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
178 // but there's not much point as the machines |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
179 // aren't available yet (2003-06-26) |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
180 memset(blocks, 0, sizeof(DCTELEM)*6*64); |
1334
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
181 } |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
182 else |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
183 for ( ; i < sizeof(DCTELEM)*6*64 ; i += 128) { |
8031 | 184 __asm__ volatile("dcbzl %0,%1" : : "b" (blocks), "r" (i) : "memory"); |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
185 } |
1334
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
186 #else |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
187 memset(blocks, 0, sizeof(DCTELEM)*6*64); |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
188 #endif |
1352
e8ff4783f188
1) remove TBL support in PPC performance. It's much more useful to use the
michaelni
parents:
1340
diff
changeset
|
189 POWERPC_PERF_STOP_COUNT(powerpc_clear_blocks_dcbz128, 1); |
1334
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
190 } |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
191 #else |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
192 void clear_blocks_dcbz128_ppc(DCTELEM *blocks) |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
193 { |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
194 memset(blocks, 0, sizeof(DCTELEM)*6*64); |
1334
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
195 } |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
196 #endif |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
197 |
8590 | 198 #if HAVE_DCBZL |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
199 /* check dcbz report how many bytes are set to 0 by dcbz */ |
1334
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
200 /* update 24/06/2003 : replace dcbz by dcbzl to get |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
201 the intended effect (Apple "fixed" dcbz) |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
202 unfortunately this cannot be used unless the assembler |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
203 knows about dcbzl ... */ |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
204 long check_dcbzl_effect(void) |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
205 { |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
206 register char *fakedata = av_malloc(1024); |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
207 register char *fakedata_middle; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
208 register long zero = 0; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
209 register long i = 0; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
210 long count = 0; |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
211 |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
212 if (!fakedata) { |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
213 return 0L; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
214 } |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
215 |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
216 fakedata_middle = (fakedata + 512); |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
217 |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
218 memset(fakedata, 0xFF, 1024); |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
219 |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
220 /* below the constraint "b" seems to mean "Address base register" |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
221 in gcc-3.3 / RS/6000 speaks. seems to avoid using r0, so.... */ |
8031 | 222 __asm__ volatile("dcbzl %0, %1" : : "b" (fakedata_middle), "r" (zero)); |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
223 |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
224 for (i = 0; i < 1024 ; i ++) { |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
225 if (fakedata[i] == (char)0) |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
226 count++; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
227 } |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
228 |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
229 av_free(fakedata); |
2967 | 230 |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
231 return count; |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
232 } |
1334
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
233 #else |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
234 long check_dcbzl_effect(void) |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
235 { |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
236 return 0; |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
237 } |
80c46c310a91
PPC970 patch + cpu-specific tuning support by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1092
diff
changeset
|
238 #endif |
1015
35cf2f4a0f8c
PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1009
diff
changeset
|
239 |
4003 | 240 static void prefetch_ppc(void *mem, int stride, int h) |
241 { | |
242 register const uint8_t *p = mem; | |
243 do { | |
8031 | 244 __asm__ volatile ("dcbt 0,%0" : : "r" (p)); |
4003 | 245 p+= stride; |
246 } while(--h); | |
247 } | |
248 | |
1092 | 249 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx) |
638
0012f75c92bb
altivec build tidyup patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
diff
changeset
|
250 { |
5749 | 251 // Common optimizations whether AltiVec is available or not |
4003 | 252 c->prefetch = prefetch_ppc; |
3546
5f97ba9a4eaa
Almost cosmetic changes in dsputil_init_ppc and vorbis_inverse_coupling_altivec:
lu_zero
parents:
3542
diff
changeset
|
253 switch (check_dcbzl_effect()) { |
5f97ba9a4eaa
Almost cosmetic changes in dsputil_init_ppc and vorbis_inverse_coupling_altivec:
lu_zero
parents:
3542
diff
changeset
|
254 case 32: |
5f97ba9a4eaa
Almost cosmetic changes in dsputil_init_ppc and vorbis_inverse_coupling_altivec:
lu_zero
parents:
3542
diff
changeset
|
255 c->clear_blocks = clear_blocks_dcbz32_ppc; |
5f97ba9a4eaa
Almost cosmetic changes in dsputil_init_ppc and vorbis_inverse_coupling_altivec:
lu_zero
parents:
3542
diff
changeset
|
256 break; |
5f97ba9a4eaa
Almost cosmetic changes in dsputil_init_ppc and vorbis_inverse_coupling_altivec:
lu_zero
parents:
3542
diff
changeset
|
257 case 128: |
5f97ba9a4eaa
Almost cosmetic changes in dsputil_init_ppc and vorbis_inverse_coupling_altivec:
lu_zero
parents:
3542
diff
changeset
|
258 c->clear_blocks = clear_blocks_dcbz128_ppc; |
5f97ba9a4eaa
Almost cosmetic changes in dsputil_init_ppc and vorbis_inverse_coupling_altivec:
lu_zero
parents:
3542
diff
changeset
|
259 break; |
5f97ba9a4eaa
Almost cosmetic changes in dsputil_init_ppc and vorbis_inverse_coupling_altivec:
lu_zero
parents:
3542
diff
changeset
|
260 default: |
5f97ba9a4eaa
Almost cosmetic changes in dsputil_init_ppc and vorbis_inverse_coupling_altivec:
lu_zero
parents:
3542
diff
changeset
|
261 break; |
5f97ba9a4eaa
Almost cosmetic changes in dsputil_init_ppc and vorbis_inverse_coupling_altivec:
lu_zero
parents:
3542
diff
changeset
|
262 } |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
2068
diff
changeset
|
263 |
8590 | 264 #if HAVE_ALTIVEC |
8596
68e959302527
replace all occurrence of ENABLE_ by the corresponding CONFIG_, HAVE_ or ARCH_
aurel
parents:
8590
diff
changeset
|
265 if(CONFIG_H264_DECODER) dsputil_h264_init_ppc(c, avctx); |
2967 | 266 |
638
0012f75c92bb
altivec build tidyup patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
diff
changeset
|
267 if (has_altivec()) { |
8104
0d108ec85620
Remove duplicated MM_* macros for CPU capabilities from dsputil.h.
rathann
parents:
8031
diff
changeset
|
268 mm_flags |= FF_MM_ALTIVEC; |
2967 | 269 |
3547 | 270 dsputil_init_altivec(c, avctx); |
8596
68e959302527
replace all occurrence of ENABLE_ by the corresponding CONFIG_, HAVE_ or ARCH_
aurel
parents:
8590
diff
changeset
|
271 if(CONFIG_SNOW_DECODER) snow_init_altivec(c, avctx); |
68e959302527
replace all occurrence of ENABLE_ by the corresponding CONFIG_, HAVE_ or ARCH_
aurel
parents:
8590
diff
changeset
|
272 if(CONFIG_VC1_DECODER || CONFIG_WMV3_DECODER) |
4227
ef1d382309e5
Conditionally compile some of the AltiVec optimizations.
diego
parents:
4197
diff
changeset
|
273 vc1dsp_init_altivec(c, avctx); |
3581 | 274 float_init_altivec(c, avctx); |
4838
eeac11145c4e
ssd_int8_vs_int16_altivec, not completely benchmarkedwith svq1
lu_zero
parents:
4521
diff
changeset
|
275 int_init_altivec(c, avctx); |
2979 | 276 c->gmc1 = gmc1_altivec; |
1092 | 277 |
8590 | 278 #if CONFIG_ENCODERS |
2979 | 279 if (avctx->dct_algo == FF_DCT_AUTO || |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
280 avctx->dct_algo == FF_DCT_ALTIVEC) { |
2979 | 281 c->fdct = fdct_altivec; |
282 } | |
1578
6a4cfc5f9f96
AltiVec optimized fdct patch by (James Klicman <james at klicman dot org>)
michael
parents:
1511
diff
changeset
|
283 #endif //CONFIG_ENCODERS |
6a4cfc5f9f96
AltiVec optimized fdct patch by (James Klicman <james at klicman dot org>)
michael
parents:
1511
diff
changeset
|
284 |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
285 if (avctx->lowres==0) { |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
286 if ((avctx->idct_algo == FF_IDCT_AUTO) || |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
287 (avctx->idct_algo == FF_IDCT_ALTIVEC)) { |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
288 c->idct_put = idct_put_altivec; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
289 c->idct_add = idct_add_altivec; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
290 c->idct_permutation_type = FF_TRANSPOSE_IDCT_PERM; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
291 } |
3546
5f97ba9a4eaa
Almost cosmetic changes in dsputil_init_ppc and vorbis_inverse_coupling_altivec:
lu_zero
parents:
3542
diff
changeset
|
292 } |
2967 | 293 |
8590 | 294 #if CONFIG_POWERPC_PERF |
1009
3b7cc8e4b83f
AltiVec perf (take 2), plus a couple AltiVec functions by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
995
diff
changeset
|
295 { |
7333
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
296 int i, j; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
297 for (i = 0 ; i < powerpc_perf_total ; i++) { |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
298 for (j = 0; j < POWERPC_NUM_PMC_ENABLED ; j++) { |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
299 perfdata[j][i][powerpc_data_min] = 0xFFFFFFFFFFFFFFFFULL; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
300 perfdata[j][i][powerpc_data_max] = 0x0000000000000000ULL; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
301 perfdata[j][i][powerpc_data_sum] = 0x0000000000000000ULL; |
a8a79f5385f6
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents:
6763
diff
changeset
|
302 perfdata[j][i][powerpc_data_num] = 0x0000000000000000ULL; |
7334 | 303 } |
304 } | |
1009
3b7cc8e4b83f
AltiVec perf (take 2), plus a couple AltiVec functions by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
995
diff
changeset
|
305 } |
4521
891590781d9e
rename POWERPC_PERFORMANCE_REPORT to CONFIG_POWERPC_PERF
mru
parents:
4227
diff
changeset
|
306 #endif /* CONFIG_POWERPC_PERF */ |
3957
b6f6bf155661
Non Altivec optimizations already present at the top
lu_zero
parents:
3949
diff
changeset
|
307 } |
1024
9cc1031e1864
More AltiVec MC functions patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents:
1015
diff
changeset
|
308 #endif /* HAVE_ALTIVEC */ |
638
0012f75c92bb
altivec build tidyup patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
diff
changeset
|
309 } |