Mercurial > libavcodec.hg
annotate ppc/h264_altivec.c @ 7246:c9dc58950620 libavcodec
10l, forgot to replace a / 256 for r14173.
Fixes MS ADPCM regression test.
author | reimar |
---|---|
date | Sat, 12 Jul 2008 07:28:45 +0000 |
parents | f7cbb7733146 |
children | a8a79f5385f6 |
rev | line source |
---|---|
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
1 /* |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
2 * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org> |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
3 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3667
diff
changeset
|
4 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3667
diff
changeset
|
5 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3667
diff
changeset
|
6 * FFmpeg is free software; you can redistribute it and/or |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
7 * modify it under the terms of the GNU Lesser General Public |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
8 * 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:
3667
diff
changeset
|
9 * version 2.1 of the License, or (at your option) any later version. |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
10 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3667
diff
changeset
|
11 * FFmpeg is distributed in the hope that it will be useful, |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
14 * Lesser General Public License for more details. |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
15 * |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
16 * 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:
3667
diff
changeset
|
17 * 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
|
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
19 */ |
2967 | 20 |
6763 | 21 #include "libavcodec/dsputil.h" |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
22 |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
23 #include "gcc_fixes.h" |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
24 |
5750
09f99af1db40
Sanitize altivec code so it can be built with runtime check properly
lu_zero
parents:
5586
diff
changeset
|
25 #include "dsputil_ppc.h" |
6077 | 26 #include "dsputil_altivec.h" |
5750
09f99af1db40
Sanitize altivec code so it can be built with runtime check properly
lu_zero
parents:
5586
diff
changeset
|
27 #include "util_altivec.h" |
4260
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
28 #include "types_altivec.h" |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
29 |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
30 #define PUT_OP_U8_ALTIVEC(d, s, dst) d = s |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
31 #define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s) |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
32 |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
33 #define OP_U8_ALTIVEC PUT_OP_U8_ALTIVEC |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
34 #define PREFIX_h264_chroma_mc8_altivec put_h264_chroma_mc8_altivec |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
35 #define PREFIX_h264_chroma_mc8_num altivec_put_h264_chroma_mc8_num |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
36 #define PREFIX_h264_qpel16_h_lowpass_altivec put_h264_qpel16_h_lowpass_altivec |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
37 #define PREFIX_h264_qpel16_h_lowpass_num altivec_put_h264_qpel16_h_lowpass_num |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
38 #define PREFIX_h264_qpel16_v_lowpass_altivec put_h264_qpel16_v_lowpass_altivec |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
39 #define PREFIX_h264_qpel16_v_lowpass_num altivec_put_h264_qpel16_v_lowpass_num |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
40 #define PREFIX_h264_qpel16_hv_lowpass_altivec put_h264_qpel16_hv_lowpass_altivec |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
41 #define PREFIX_h264_qpel16_hv_lowpass_num altivec_put_h264_qpel16_hv_lowpass_num |
3577 | 42 #include "h264_template_altivec.c" |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
43 #undef OP_U8_ALTIVEC |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
44 #undef PREFIX_h264_chroma_mc8_altivec |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
45 #undef PREFIX_h264_chroma_mc8_num |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
46 #undef PREFIX_h264_qpel16_h_lowpass_altivec |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
47 #undef PREFIX_h264_qpel16_h_lowpass_num |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
48 #undef PREFIX_h264_qpel16_v_lowpass_altivec |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
49 #undef PREFIX_h264_qpel16_v_lowpass_num |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
50 #undef PREFIX_h264_qpel16_hv_lowpass_altivec |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
51 #undef PREFIX_h264_qpel16_hv_lowpass_num |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
52 |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
53 #define OP_U8_ALTIVEC AVG_OP_U8_ALTIVEC |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
54 #define PREFIX_h264_chroma_mc8_altivec avg_h264_chroma_mc8_altivec |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
55 #define PREFIX_h264_chroma_mc8_num altivec_avg_h264_chroma_mc8_num |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
56 #define PREFIX_h264_qpel16_h_lowpass_altivec avg_h264_qpel16_h_lowpass_altivec |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
57 #define PREFIX_h264_qpel16_h_lowpass_num altivec_avg_h264_qpel16_h_lowpass_num |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
58 #define PREFIX_h264_qpel16_v_lowpass_altivec avg_h264_qpel16_v_lowpass_altivec |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
59 #define PREFIX_h264_qpel16_v_lowpass_num altivec_avg_h264_qpel16_v_lowpass_num |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
60 #define PREFIX_h264_qpel16_hv_lowpass_altivec avg_h264_qpel16_hv_lowpass_altivec |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
61 #define PREFIX_h264_qpel16_hv_lowpass_num altivec_avg_h264_qpel16_hv_lowpass_num |
3577 | 62 #include "h264_template_altivec.c" |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
63 #undef OP_U8_ALTIVEC |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
64 #undef PREFIX_h264_chroma_mc8_altivec |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
65 #undef PREFIX_h264_chroma_mc8_num |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
66 #undef PREFIX_h264_qpel16_h_lowpass_altivec |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
67 #undef PREFIX_h264_qpel16_h_lowpass_num |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
68 #undef PREFIX_h264_qpel16_v_lowpass_altivec |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
69 #undef PREFIX_h264_qpel16_v_lowpass_num |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
70 #undef PREFIX_h264_qpel16_hv_lowpass_altivec |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
71 #undef PREFIX_h264_qpel16_hv_lowpass_num |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
72 |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
73 #define H264_MC(OPNAME, SIZE, CODETYPE) \ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
74 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## CODETYPE (uint8_t *dst, uint8_t *src, int stride){\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
75 OPNAME ## pixels ## SIZE ## _ ## CODETYPE(dst, src, stride, SIZE);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
76 }\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
77 \ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
78 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){ \ |
3544
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
79 DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\ |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
80 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
81 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
82 }\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
83 \ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
84 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
85 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(dst, src, stride, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
86 }\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
87 \ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
88 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ |
3544
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
89 DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\ |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
90 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
91 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+1, half, stride, stride, SIZE);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
92 }\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
93 \ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
94 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ |
3544
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
95 DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\ |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
96 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
97 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
98 }\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
99 \ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
100 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
101 OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(dst, src, stride, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
102 }\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
103 \ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
104 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ |
3544
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
105 DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\ |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
106 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
107 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+stride, half, stride, stride, SIZE);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
108 }\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
109 \ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
110 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ |
3544
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
111 DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\ |
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
112 DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\ |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
113 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
114 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
115 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
116 }\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
117 \ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
118 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ |
3544
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
119 DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\ |
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
120 DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\ |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
121 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
122 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
123 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
124 }\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
125 \ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
126 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ |
3544
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
127 DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\ |
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
128 DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\ |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
129 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
130 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
131 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
132 }\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
133 \ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
134 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ |
3544
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
135 DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\ |
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
136 DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\ |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
137 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
138 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
139 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
140 }\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
141 \ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
142 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ |
3544
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
143 DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\ |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
144 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(dst, tmp, src, stride, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
145 }\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
146 \ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
147 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ |
3544
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
148 DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\ |
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
149 DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\ |
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
150 DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\ |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
151 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
152 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
153 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
154 }\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
155 \ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
156 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ |
3544
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
157 DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\ |
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
158 DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\ |
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
159 DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\ |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
160 put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
161 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
162 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
163 }\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
164 \ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
165 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ |
3544
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
166 DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\ |
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
167 DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\ |
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
168 DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\ |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
169 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
170 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
171 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
172 }\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
173 \ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
174 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ |
3544
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
175 DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\ |
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
176 DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\ |
8bb61d9a2c40
avoid alignment hacks, luckly gcc does the right thing on arches different from x86
lu_zero
parents:
3337
diff
changeset
|
177 DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\ |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
178 put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
179 put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
180 OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
181 }\ |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
182 |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
183 /* this code assume that stride % 16 == 0 */ |
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
184 void put_no_rnd_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride, int h, int x, int y) { |
5019
41cabe79ba25
use macro Use DECLARE_ALIGNED_16 to align stack-allocated variables
gpoirier
parents:
5010
diff
changeset
|
185 DECLARE_ALIGNED_16(signed int, ABCD[4]) = |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
186 {((8 - x) * (8 - y)), |
5586 | 187 ((x) * (8 - y)), |
188 ((8 - x) * (y)), | |
189 ((x) * (y))}; | |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
190 register int i; |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
191 vec_u8_t fperm; |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
192 const vec_s32_t vABCD = vec_ld(0, ABCD); |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
193 const vec_s16_t vA = vec_splat((vec_s16_t)vABCD, 1); |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
194 const vec_s16_t vB = vec_splat((vec_s16_t)vABCD, 3); |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
195 const vec_s16_t vC = vec_splat((vec_s16_t)vABCD, 5); |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
196 const vec_s16_t vD = vec_splat((vec_s16_t)vABCD, 7); |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
197 LOAD_ZERO; |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
198 const vec_s16_t v28ss = vec_sub(vec_sl(vec_splat_s16(1),vec_splat_u16(5)),vec_splat_s16(4)); |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
199 const vec_u16_t v6us = vec_splat_u16(6); |
5586 | 200 register int loadSecond = (((unsigned long)src) % 16) <= 7 ? 0 : 1; |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
201 register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0; |
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
202 |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
203 vec_u8_t vsrcAuc, vsrcBuc, vsrcperm0, vsrcperm1; |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
204 vec_u8_t vsrc0uc, vsrc1uc; |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
205 vec_s16_t vsrc0ssH, vsrc1ssH; |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
206 vec_u8_t vsrcCuc, vsrc2uc, vsrc3uc; |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
207 vec_s16_t vsrc2ssH, vsrc3ssH, psum; |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
208 vec_u8_t vdst, ppsum, fsum; |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
209 |
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
210 if (((unsigned long)dst) % 16 == 0) { |
5586 | 211 fperm = (vec_u8_t)AVV(0x10, 0x11, 0x12, 0x13, |
212 0x14, 0x15, 0x16, 0x17, | |
213 0x08, 0x09, 0x0A, 0x0B, | |
214 0x0C, 0x0D, 0x0E, 0x0F); | |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
215 } else { |
5586 | 216 fperm = (vec_u8_t)AVV(0x00, 0x01, 0x02, 0x03, |
217 0x04, 0x05, 0x06, 0x07, | |
218 0x18, 0x19, 0x1A, 0x1B, | |
219 0x1C, 0x1D, 0x1E, 0x1F); | |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
220 } |
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
221 |
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
222 vsrcAuc = vec_ld(0, src); |
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
223 |
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
224 if (loadSecond) |
5586 | 225 vsrcBuc = vec_ld(16, src); |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
226 vsrcperm0 = vec_lvsl(0, src); |
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
227 vsrcperm1 = vec_lvsl(1, src); |
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
228 |
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
229 vsrc0uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm0); |
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
230 if (reallyBadAlign) |
5586 | 231 vsrc1uc = vsrcBuc; |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
232 else |
5586 | 233 vsrc1uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm1); |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
234 |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
235 vsrc0ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc0uc); |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
236 vsrc1ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc1uc); |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
237 |
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
238 if (!loadSecond) {// -> !reallyBadAlign |
5586 | 239 for (i = 0 ; i < h ; i++) { |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
240 |
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
241 |
5586 | 242 vsrcCuc = vec_ld(stride + 0, src); |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
243 |
5586 | 244 vsrc2uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0); |
245 vsrc3uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1); | |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
246 |
5586 | 247 vsrc2ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc2uc); |
248 vsrc3ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc3uc); | |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
249 |
5586 | 250 psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0)); |
251 psum = vec_mladd(vB, vsrc1ssH, psum); | |
252 psum = vec_mladd(vC, vsrc2ssH, psum); | |
253 psum = vec_mladd(vD, vsrc3ssH, psum); | |
254 psum = vec_add(v28ss, psum); | |
255 psum = vec_sra(psum, v6us); | |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
256 |
5586 | 257 vdst = vec_ld(0, dst); |
258 ppsum = (vec_u8_t)vec_packsu(psum, psum); | |
259 fsum = vec_perm(vdst, ppsum, fperm); | |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
260 |
5586 | 261 vec_st(fsum, 0, dst); |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
262 |
5586 | 263 vsrc0ssH = vsrc2ssH; |
264 vsrc1ssH = vsrc3ssH; | |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
265 |
5586 | 266 dst += stride; |
267 src += stride; | |
268 } | |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
269 } else { |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
270 vec_u8_t vsrcDuc; |
5586 | 271 for (i = 0 ; i < h ; i++) { |
272 vsrcCuc = vec_ld(stride + 0, src); | |
273 vsrcDuc = vec_ld(stride + 16, src); | |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
274 |
5586 | 275 vsrc2uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0); |
276 if (reallyBadAlign) | |
277 vsrc3uc = vsrcDuc; | |
278 else | |
279 vsrc3uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1); | |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
280 |
5586 | 281 vsrc2ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc2uc); |
282 vsrc3ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc3uc); | |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
283 |
5586 | 284 psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0)); |
285 psum = vec_mladd(vB, vsrc1ssH, psum); | |
286 psum = vec_mladd(vC, vsrc2ssH, psum); | |
287 psum = vec_mladd(vD, vsrc3ssH, psum); | |
288 psum = vec_add(v28ss, psum); | |
289 psum = vec_sr(psum, v6us); | |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
290 |
5586 | 291 vdst = vec_ld(0, dst); |
292 ppsum = (vec_u8_t)vec_pack(psum, psum); | |
293 fsum = vec_perm(vdst, ppsum, fperm); | |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
294 |
5586 | 295 vec_st(fsum, 0, dst); |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
296 |
5586 | 297 vsrc0ssH = vsrc2ssH; |
298 vsrc1ssH = vsrc3ssH; | |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
299 |
5586 | 300 dst += stride; |
301 src += stride; | |
302 } | |
3667
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
303 } |
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
304 } |
56e60411527c
AltiVec version of put_no_rnd_h264_chroma_pixels_tab[0] (slightly changed version of put_h264_chroma_pixels_tab[0])
kostya
parents:
3659
diff
changeset
|
305 |
3337
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
306 static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1, |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
307 const uint8_t * src2, int dst_stride, |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
308 int src_stride1, int h) |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
309 { |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
310 int i; |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
311 vec_u8_t a, b, d, tmp1, tmp2, mask, mask_, edges, align; |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
312 |
3337
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
313 mask_ = vec_lvsl(0, src2); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
314 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
315 for (i = 0; i < h; i++) { |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
316 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
317 tmp1 = vec_ld(i * src_stride1, src1); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
318 mask = vec_lvsl(i * src_stride1, src1); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
319 tmp2 = vec_ld(i * src_stride1 + 15, src1); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
320 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
321 a = vec_perm(tmp1, tmp2, mask); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
322 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
323 tmp1 = vec_ld(i * 16, src2); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
324 tmp2 = vec_ld(i * 16 + 15, src2); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
325 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
326 b = vec_perm(tmp1, tmp2, mask_); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
327 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
328 tmp1 = vec_ld(0, dst); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
329 mask = vec_lvsl(0, dst); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
330 tmp2 = vec_ld(15, dst); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
331 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
332 d = vec_avg(a, b); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
333 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
334 edges = vec_perm(tmp2, tmp1, mask); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
335 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
336 align = vec_lvsr(0, dst); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
337 |
3659
dd55fb216497
Proper fix for the corner case that would have been corrected before, praise&blame to me and exg in equal shares
lu_zero
parents:
3658
diff
changeset
|
338 tmp2 = vec_perm(d, edges, align); |
3658 | 339 tmp1 = vec_perm(edges, d, align); |
3337
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
340 |
3659
dd55fb216497
Proper fix for the corner case that would have been corrected before, praise&blame to me and exg in equal shares
lu_zero
parents:
3658
diff
changeset
|
341 vec_st(tmp2, 15, dst); |
3583
562758eaf7bf
10l, thanks to Emanuele Giaquinta <exg@gentoo.org> for testing and finding the issue
lu_zero
parents:
3577
diff
changeset
|
342 vec_st(tmp1, 0 , dst); |
3337
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
343 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
344 dst += dst_stride; |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
345 } |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
346 } |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
347 |
3337
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
348 static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1, |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
349 const uint8_t * src2, int dst_stride, |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
350 int src_stride1, int h) |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
351 { |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
352 int i; |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
353 vec_u8_t a, b, d, tmp1, tmp2, mask, mask_, edges, align; |
3337
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
354 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
355 mask_ = vec_lvsl(0, src2); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
356 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
357 for (i = 0; i < h; i++) { |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
358 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
359 tmp1 = vec_ld(i * src_stride1, src1); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
360 mask = vec_lvsl(i * src_stride1, src1); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
361 tmp2 = vec_ld(i * src_stride1 + 15, src1); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
362 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
363 a = vec_perm(tmp1, tmp2, mask); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
364 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
365 tmp1 = vec_ld(i * 16, src2); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
366 tmp2 = vec_ld(i * 16 + 15, src2); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
367 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
368 b = vec_perm(tmp1, tmp2, mask_); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
369 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
370 tmp1 = vec_ld(0, dst); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
371 mask = vec_lvsl(0, dst); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
372 tmp2 = vec_ld(15, dst); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
373 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
374 d = vec_avg(vec_perm(tmp1, tmp2, mask), vec_avg(a, b)); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
375 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
376 edges = vec_perm(tmp2, tmp1, mask); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
377 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
378 align = vec_lvsr(0, dst); |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
379 |
3659
dd55fb216497
Proper fix for the corner case that would have been corrected before, praise&blame to me and exg in equal shares
lu_zero
parents:
3658
diff
changeset
|
380 tmp2 = vec_perm(d, edges, align); |
3658 | 381 tmp1 = vec_perm(edges, d, align); |
3337
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
382 |
3659
dd55fb216497
Proper fix for the corner case that would have been corrected before, praise&blame to me and exg in equal shares
lu_zero
parents:
3658
diff
changeset
|
383 vec_st(tmp2, 15, dst); |
3583
562758eaf7bf
10l, thanks to Emanuele Giaquinta <exg@gentoo.org> for testing and finding the issue
lu_zero
parents:
3577
diff
changeset
|
384 vec_st(tmp1, 0 , dst); |
3337
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
385 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
386 dst += dst_stride; |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
387 } |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
388 } |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
389 |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
390 /* Implemented but could be faster |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
391 #define put_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) put_pixels16_l2(d,s1,s2,ds,s1s,16,h) |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
392 #define avg_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) avg_pixels16_l2(d,s1,s2,ds,s1s,16,h) |
3337
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
393 */ |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
394 |
3337
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
395 H264_MC(put_, 16, altivec) |
bec1eb6d3746
put_pixels16_l2_altivec and avg_pixels16_l2_altivec
lu_zero
parents:
3089
diff
changeset
|
396 H264_MC(avg_, 16, altivec) |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
397 |
4260
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
398 |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
399 /**************************************************************************** |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
400 * IDCT transform: |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
401 ****************************************************************************/ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
402 |
5586 | 403 #define VEC_1D_DCT(vb0,vb1,vb2,vb3,va0,va1,va2,va3) \ |
404 /* 1st stage */ \ | |
405 vz0 = vec_add(vb0,vb2); /* temp[0] = Y[0] + Y[2] */ \ | |
406 vz1 = vec_sub(vb0,vb2); /* temp[1] = Y[0] - Y[2] */ \ | |
407 vz2 = vec_sra(vb1,vec_splat_u16(1)); \ | |
408 vz2 = vec_sub(vz2,vb3); /* temp[2] = Y[1].1/2 - Y[3] */ \ | |
409 vz3 = vec_sra(vb3,vec_splat_u16(1)); \ | |
410 vz3 = vec_add(vb1,vz3); /* temp[3] = Y[1] + Y[3].1/2 */ \ | |
411 /* 2nd stage: output */ \ | |
412 va0 = vec_add(vz0,vz3); /* x[0] = temp[0] + temp[3] */ \ | |
413 va1 = vec_add(vz1,vz2); /* x[1] = temp[1] + temp[2] */ \ | |
414 va2 = vec_sub(vz1,vz2); /* x[2] = temp[1] - temp[2] */ \ | |
415 va3 = vec_sub(vz0,vz3) /* x[3] = temp[0] - temp[3] */ | |
5094
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
416 |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
417 #define VEC_TRANSPOSE_4(a0,a1,a2,a3,b0,b1,b2,b3) \ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
418 b0 = vec_mergeh( a0, a0 ); \ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
419 b1 = vec_mergeh( a1, a0 ); \ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
420 b2 = vec_mergeh( a2, a0 ); \ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
421 b3 = vec_mergeh( a3, a0 ); \ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
422 a0 = vec_mergeh( b0, b2 ); \ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
423 a1 = vec_mergel( b0, b2 ); \ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
424 a2 = vec_mergeh( b1, b3 ); \ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
425 a3 = vec_mergel( b1, b3 ); \ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
426 b0 = vec_mergeh( a0, a2 ); \ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
427 b1 = vec_mergel( a0, a2 ); \ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
428 b2 = vec_mergeh( a1, a3 ); \ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
429 b3 = vec_mergel( a1, a3 ) |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
430 |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
431 #define VEC_LOAD_U8_ADD_S16_STORE_U8(va) \ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
432 vdst_orig = vec_ld(0, dst); \ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
433 vdst = vec_perm(vdst_orig, zero_u8v, vdst_mask); \ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
434 vdst_ss = (vec_s16_t) vec_mergeh(zero_u8v, vdst); \ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
435 va = vec_add(va, vdst_ss); \ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
436 va_u8 = vec_packsu(va, zero_s16v); \ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
437 va_u32 = vec_splat((vec_u32_t)va_u8, 0); \ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
438 vec_ste(va_u32, element, (uint32_t*)dst); |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
439 |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
440 static void ff_h264_idct_add_altivec(uint8_t *dst, DCTELEM *block, int stride) |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
441 { |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
442 vec_s16_t va0, va1, va2, va3; |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
443 vec_s16_t vz0, vz1, vz2, vz3; |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
444 vec_s16_t vtmp0, vtmp1, vtmp2, vtmp3; |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
445 vec_u8_t va_u8; |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
446 vec_u32_t va_u32; |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
447 vec_s16_t vdst_ss; |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
448 const vec_u16_t v6us = vec_splat_u16(6); |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
449 vec_u8_t vdst, vdst_orig; |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
450 vec_u8_t vdst_mask = vec_lvsl(0, dst); |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
451 int element = ((unsigned long)dst & 0xf) >> 2; |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
452 LOAD_ZERO; |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
453 |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
454 block[0] += 32; /* add 32 as a DC-level for rounding */ |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
455 |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
456 vtmp0 = vec_ld(0,block); |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
457 vtmp1 = vec_sld(vtmp0, vtmp0, 8); |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
458 vtmp2 = vec_ld(16,block); |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
459 vtmp3 = vec_sld(vtmp2, vtmp2, 8); |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
460 |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
461 VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3); |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
462 VEC_TRANSPOSE_4(va0,va1,va2,va3,vtmp0,vtmp1,vtmp2,vtmp3); |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
463 VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3); |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
464 |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
465 va0 = vec_sra(va0,v6us); |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
466 va1 = vec_sra(va1,v6us); |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
467 va2 = vec_sra(va2,v6us); |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
468 va3 = vec_sra(va3,v6us); |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
469 |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
470 VEC_LOAD_U8_ADD_S16_STORE_U8(va0); |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
471 dst += stride; |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
472 VEC_LOAD_U8_ADD_S16_STORE_U8(va1); |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
473 dst += stride; |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
474 VEC_LOAD_U8_ADD_S16_STORE_U8(va2); |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
475 dst += stride; |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
476 VEC_LOAD_U8_ADD_S16_STORE_U8(va3); |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
477 } |
ce57e3f2b2a7
h264_idct_add_altivec, based on an old patch from Mauricio Alvarez <alvarezATac.upc.edu>, polished by David Conrad <umovimusATgmail.com>
lu_zero
parents:
5019
diff
changeset
|
478 |
4260
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
479 #define IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7, d0, d1, d2, d3, d4, d5, d6, d7) {\ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
480 /* a0 = SRC(0) + SRC(4); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
481 vec_s16_t a0v = vec_add(s0, s4); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
482 /* a2 = SRC(0) - SRC(4); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
483 vec_s16_t a2v = vec_sub(s0, s4); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
484 /* a4 = (SRC(2)>>1) - SRC(6); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
485 vec_s16_t a4v = vec_sub(vec_sra(s2, onev), s6); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
486 /* a6 = (SRC(6)>>1) + SRC(2); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
487 vec_s16_t a6v = vec_add(vec_sra(s6, onev), s2); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
488 /* b0 = a0 + a6; */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
489 vec_s16_t b0v = vec_add(a0v, a6v); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
490 /* b2 = a2 + a4; */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
491 vec_s16_t b2v = vec_add(a2v, a4v); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
492 /* b4 = a2 - a4; */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
493 vec_s16_t b4v = vec_sub(a2v, a4v); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
494 /* b6 = a0 - a6; */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
495 vec_s16_t b6v = vec_sub(a0v, a6v); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
496 /* a1 = SRC(5) - SRC(3) - SRC(7) - (SRC(7)>>1); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
497 /* a1 = (SRC(5)-SRC(3)) - (SRC(7) + (SRC(7)>>1)); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
498 vec_s16_t a1v = vec_sub( vec_sub(s5, s3), vec_add(s7, vec_sra(s7, onev)) ); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
499 /* a3 = SRC(7) + SRC(1) - SRC(3) - (SRC(3)>>1); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
500 /* a3 = (SRC(7)+SRC(1)) - (SRC(3) + (SRC(3)>>1)); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
501 vec_s16_t a3v = vec_sub( vec_add(s7, s1), vec_add(s3, vec_sra(s3, onev)) );\ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
502 /* a5 = SRC(7) - SRC(1) + SRC(5) + (SRC(5)>>1); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
503 /* a5 = (SRC(7)-SRC(1)) + SRC(5) + (SRC(5)>>1); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
504 vec_s16_t a5v = vec_add( vec_sub(s7, s1), vec_add(s5, vec_sra(s5, onev)) );\ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
505 /* a7 = SRC(5)+SRC(3) + SRC(1) + (SRC(1)>>1); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
506 vec_s16_t a7v = vec_add( vec_add(s5, s3), vec_add(s1, vec_sra(s1, onev)) );\ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
507 /* b1 = (a7>>2) + a1; */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
508 vec_s16_t b1v = vec_add( vec_sra(a7v, twov), a1v); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
509 /* b3 = a3 + (a5>>2); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
510 vec_s16_t b3v = vec_add(a3v, vec_sra(a5v, twov)); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
511 /* b5 = (a3>>2) - a5; */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
512 vec_s16_t b5v = vec_sub( vec_sra(a3v, twov), a5v); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
513 /* b7 = a7 - (a1>>2); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
514 vec_s16_t b7v = vec_sub( a7v, vec_sra(a1v, twov)); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
515 /* DST(0, b0 + b7); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
516 d0 = vec_add(b0v, b7v); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
517 /* DST(1, b2 + b5); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
518 d1 = vec_add(b2v, b5v); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
519 /* DST(2, b4 + b3); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
520 d2 = vec_add(b4v, b3v); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
521 /* DST(3, b6 + b1); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
522 d3 = vec_add(b6v, b1v); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
523 /* DST(4, b6 - b1); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
524 d4 = vec_sub(b6v, b1v); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
525 /* DST(5, b4 - b3); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
526 d5 = vec_sub(b4v, b3v); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
527 /* DST(6, b2 - b5); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
528 d6 = vec_sub(b2v, b5v); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
529 /* DST(7, b0 - b7); */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
530 d7 = vec_sub(b0v, b7v); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
531 } |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
532 |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
533 #define ALTIVEC_STORE_SUM_CLIP(dest, idctv, perm_ldv, perm_stv, sel) { \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
534 /* unaligned load */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
535 vec_u8_t hv = vec_ld( 0, dest ); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
536 vec_u8_t lv = vec_ld( 7, dest ); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
537 vec_u8_t dstv = vec_perm( hv, lv, (vec_u8_t)perm_ldv ); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
538 vec_s16_t idct_sh6 = vec_sra(idctv, sixv); \ |
4294 | 539 vec_u16_t dst16 = (vec_u16_t)vec_mergeh(zero_u8v, dstv); \ |
4260
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
540 vec_s16_t idstsum = vec_adds(idct_sh6, (vec_s16_t)dst16); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
541 vec_u8_t idstsum8 = vec_packsu(zero_s16v, idstsum); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
542 vec_u8_t edgehv; \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
543 /* unaligned store */ \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
544 vec_u8_t bodyv = vec_perm( idstsum8, idstsum8, perm_stv );\ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
545 vec_u8_t edgelv = vec_perm( sel, zero_u8v, perm_stv ); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
546 lv = vec_sel( lv, bodyv, edgelv ); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
547 vec_st( lv, 7, dest ); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
548 hv = vec_ld( 0, dest ); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
549 edgehv = vec_perm( zero_u8v, sel, perm_stv ); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
550 hv = vec_sel( hv, bodyv, edgehv ); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
551 vec_st( hv, 0, dest ); \ |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
552 } |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
553 |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
554 void ff_h264_idct8_add_altivec( uint8_t *dst, DCTELEM *dct, int stride ) { |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
555 vec_s16_t s0, s1, s2, s3, s4, s5, s6, s7; |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
556 vec_s16_t d0, d1, d2, d3, d4, d5, d6, d7; |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
557 vec_s16_t idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7; |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
558 |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
559 vec_u8_t perm_ldv = vec_lvsl(0, dst); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
560 vec_u8_t perm_stv = vec_lvsr(8, dst); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
561 |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
562 const vec_u16_t onev = vec_splat_u16(1); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
563 const vec_u16_t twov = vec_splat_u16(2); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
564 const vec_u16_t sixv = vec_splat_u16(6); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
565 |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
566 const vec_u8_t sel = (vec_u8_t) AVV(0,0,0,0,0,0,0,0,-1,-1,-1,-1,-1,-1,-1,-1); |
4260
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
567 LOAD_ZERO; |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
568 |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
569 dct[0] += 32; // rounding for the >>6 at the end |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
570 |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
571 s0 = vec_ld(0x00, (int16_t*)dct); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
572 s1 = vec_ld(0x10, (int16_t*)dct); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
573 s2 = vec_ld(0x20, (int16_t*)dct); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
574 s3 = vec_ld(0x30, (int16_t*)dct); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
575 s4 = vec_ld(0x40, (int16_t*)dct); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
576 s5 = vec_ld(0x50, (int16_t*)dct); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
577 s6 = vec_ld(0x60, (int16_t*)dct); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
578 s7 = vec_ld(0x70, (int16_t*)dct); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
579 |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
580 IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7, |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
581 d0, d1, d2, d3, d4, d5, d6, d7); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
582 |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
583 TRANSPOSE8( d0, d1, d2, d3, d4, d5, d6, d7 ); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
584 |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
585 IDCT8_1D_ALTIVEC(d0, d1, d2, d3, d4, d5, d6, d7, |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
586 idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
587 |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
588 ALTIVEC_STORE_SUM_CLIP(&dst[0*stride], idct0, perm_ldv, perm_stv, sel); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
589 ALTIVEC_STORE_SUM_CLIP(&dst[1*stride], idct1, perm_ldv, perm_stv, sel); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
590 ALTIVEC_STORE_SUM_CLIP(&dst[2*stride], idct2, perm_ldv, perm_stv, sel); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
591 ALTIVEC_STORE_SUM_CLIP(&dst[3*stride], idct3, perm_ldv, perm_stv, sel); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
592 ALTIVEC_STORE_SUM_CLIP(&dst[4*stride], idct4, perm_ldv, perm_stv, sel); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
593 ALTIVEC_STORE_SUM_CLIP(&dst[5*stride], idct5, perm_ldv, perm_stv, sel); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
594 ALTIVEC_STORE_SUM_CLIP(&dst[6*stride], idct6, perm_ldv, perm_stv, sel); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
595 ALTIVEC_STORE_SUM_CLIP(&dst[7*stride], idct7, perm_ldv, perm_stv, sel); |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
596 } |
0407913ac6c6
Add IDCT8 routine in Altivec. Patch by yours truely with Linux fixes by Luca Barbato
gpoirier
parents:
4254
diff
changeset
|
597 |
5119 | 598 #define transpose4x16(r0, r1, r2, r3) { \ |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
599 register vec_u8_t r4; \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
600 register vec_u8_t r5; \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
601 register vec_u8_t r6; \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
602 register vec_u8_t r7; \ |
5119 | 603 \ |
604 r4 = vec_mergeh(r0, r2); /*0, 2 set 0*/ \ | |
605 r5 = vec_mergel(r0, r2); /*0, 2 set 1*/ \ | |
606 r6 = vec_mergeh(r1, r3); /*1, 3 set 0*/ \ | |
607 r7 = vec_mergel(r1, r3); /*1, 3 set 1*/ \ | |
608 \ | |
609 r0 = vec_mergeh(r4, r6); /*all set 0*/ \ | |
610 r1 = vec_mergel(r4, r6); /*all set 1*/ \ | |
611 r2 = vec_mergeh(r5, r7); /*all set 2*/ \ | |
612 r3 = vec_mergel(r5, r7); /*all set 3*/ \ | |
613 } | |
614 | |
5135 | 615 static inline void write16x4(uint8_t *dst, int dst_stride, |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
616 register vec_u8_t r0, register vec_u8_t r1, |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
617 register vec_u8_t r2, register vec_u8_t r3) { |
5119 | 618 DECLARE_ALIGNED_16(unsigned char, result[64]); |
619 uint32_t *src_int = (uint32_t *)result, *dst_int = (uint32_t *)dst; | |
620 int int_dst_stride = dst_stride/4; | |
621 | |
622 vec_st(r0, 0, result); | |
623 vec_st(r1, 16, result); | |
624 vec_st(r2, 32, result); | |
625 vec_st(r3, 48, result); | |
626 /* FIXME: there has to be a better way!!!! */ | |
627 *dst_int = *src_int; | |
628 *(dst_int+ int_dst_stride) = *(src_int + 1); | |
629 *(dst_int+ 2*int_dst_stride) = *(src_int + 2); | |
630 *(dst_int+ 3*int_dst_stride) = *(src_int + 3); | |
631 *(dst_int+ 4*int_dst_stride) = *(src_int + 4); | |
632 *(dst_int+ 5*int_dst_stride) = *(src_int + 5); | |
633 *(dst_int+ 6*int_dst_stride) = *(src_int + 6); | |
634 *(dst_int+ 7*int_dst_stride) = *(src_int + 7); | |
635 *(dst_int+ 8*int_dst_stride) = *(src_int + 8); | |
636 *(dst_int+ 9*int_dst_stride) = *(src_int + 9); | |
637 *(dst_int+10*int_dst_stride) = *(src_int + 10); | |
638 *(dst_int+11*int_dst_stride) = *(src_int + 11); | |
639 *(dst_int+12*int_dst_stride) = *(src_int + 12); | |
640 *(dst_int+13*int_dst_stride) = *(src_int + 13); | |
641 *(dst_int+14*int_dst_stride) = *(src_int + 14); | |
642 *(dst_int+15*int_dst_stride) = *(src_int + 15); | |
643 } | |
644 | |
645 /** \brief performs a 6x16 transpose of data in src, and stores it to dst | |
646 \todo FIXME: see if we can't spare some vec_lvsl() by them factorizing | |
647 out of unaligned_load() */ | |
648 #define readAndTranspose16x6(src, src_stride, r8, r9, r10, r11, r12, r13) {\ | |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
649 register vec_u8_t r0 = unaligned_load(0, src); \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
650 register vec_u8_t r1 = unaligned_load( src_stride, src); \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
651 register vec_u8_t r2 = unaligned_load(2* src_stride, src); \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
652 register vec_u8_t r3 = unaligned_load(3* src_stride, src); \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
653 register vec_u8_t r4 = unaligned_load(4* src_stride, src); \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
654 register vec_u8_t r5 = unaligned_load(5* src_stride, src); \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
655 register vec_u8_t r6 = unaligned_load(6* src_stride, src); \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
656 register vec_u8_t r7 = unaligned_load(7* src_stride, src); \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
657 register vec_u8_t r14 = unaligned_load(14*src_stride, src); \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
658 register vec_u8_t r15 = unaligned_load(15*src_stride, src); \ |
5119 | 659 \ |
660 r8 = unaligned_load( 8*src_stride, src); \ | |
661 r9 = unaligned_load( 9*src_stride, src); \ | |
662 r10 = unaligned_load(10*src_stride, src); \ | |
663 r11 = unaligned_load(11*src_stride, src); \ | |
664 r12 = unaligned_load(12*src_stride, src); \ | |
665 r13 = unaligned_load(13*src_stride, src); \ | |
666 \ | |
667 /*Merge first pairs*/ \ | |
668 r0 = vec_mergeh(r0, r8); /*0, 8*/ \ | |
669 r1 = vec_mergeh(r1, r9); /*1, 9*/ \ | |
670 r2 = vec_mergeh(r2, r10); /*2,10*/ \ | |
671 r3 = vec_mergeh(r3, r11); /*3,11*/ \ | |
672 r4 = vec_mergeh(r4, r12); /*4,12*/ \ | |
673 r5 = vec_mergeh(r5, r13); /*5,13*/ \ | |
674 r6 = vec_mergeh(r6, r14); /*6,14*/ \ | |
675 r7 = vec_mergeh(r7, r15); /*7,15*/ \ | |
676 \ | |
677 /*Merge second pairs*/ \ | |
678 r8 = vec_mergeh(r0, r4); /*0,4, 8,12 set 0*/ \ | |
679 r9 = vec_mergel(r0, r4); /*0,4, 8,12 set 1*/ \ | |
680 r10 = vec_mergeh(r1, r5); /*1,5, 9,13 set 0*/ \ | |
681 r11 = vec_mergel(r1, r5); /*1,5, 9,13 set 1*/ \ | |
682 r12 = vec_mergeh(r2, r6); /*2,6,10,14 set 0*/ \ | |
683 r13 = vec_mergel(r2, r6); /*2,6,10,14 set 1*/ \ | |
684 r14 = vec_mergeh(r3, r7); /*3,7,11,15 set 0*/ \ | |
685 r15 = vec_mergel(r3, r7); /*3,7,11,15 set 1*/ \ | |
686 \ | |
687 /*Third merge*/ \ | |
688 r0 = vec_mergeh(r8, r12); /*0,2,4,6,8,10,12,14 set 0*/ \ | |
689 r1 = vec_mergel(r8, r12); /*0,2,4,6,8,10,12,14 set 1*/ \ | |
690 r2 = vec_mergeh(r9, r13); /*0,2,4,6,8,10,12,14 set 2*/ \ | |
691 r4 = vec_mergeh(r10, r14); /*1,3,5,7,9,11,13,15 set 0*/ \ | |
692 r5 = vec_mergel(r10, r14); /*1,3,5,7,9,11,13,15 set 1*/ \ | |
693 r6 = vec_mergeh(r11, r15); /*1,3,5,7,9,11,13,15 set 2*/ \ | |
694 /* Don't need to compute 3 and 7*/ \ | |
695 \ | |
696 /*Final merge*/ \ | |
697 r8 = vec_mergeh(r0, r4); /*all set 0*/ \ | |
698 r9 = vec_mergel(r0, r4); /*all set 1*/ \ | |
699 r10 = vec_mergeh(r1, r5); /*all set 2*/ \ | |
700 r11 = vec_mergel(r1, r5); /*all set 3*/ \ | |
701 r12 = vec_mergeh(r2, r6); /*all set 4*/ \ | |
702 r13 = vec_mergel(r2, r6); /*all set 5*/ \ | |
703 /* Don't need to compute 14 and 15*/ \ | |
704 \ | |
705 } | |
706 | |
707 // out: o = |x-y| < a | |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
708 static inline vec_u8_t diff_lt_altivec ( register vec_u8_t x, |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
709 register vec_u8_t y, |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
710 register vec_u8_t a) { |
5119 | 711 |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
712 register vec_u8_t diff = vec_subs(x, y); |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
713 register vec_u8_t diffneg = vec_subs(y, x); |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
714 register vec_u8_t o = vec_or(diff, diffneg); /* |x-y| */ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
715 o = (vec_u8_t)vec_cmplt(o, a); |
5119 | 716 return o; |
717 } | |
718 | |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
719 static inline vec_u8_t h264_deblock_mask ( register vec_u8_t p0, |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
720 register vec_u8_t p1, |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
721 register vec_u8_t q0, |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
722 register vec_u8_t q1, |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
723 register vec_u8_t alpha, |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
724 register vec_u8_t beta) { |
5119 | 725 |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
726 register vec_u8_t mask; |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
727 register vec_u8_t tempmask; |
5119 | 728 |
729 mask = diff_lt_altivec(p0, q0, alpha); | |
730 tempmask = diff_lt_altivec(p1, p0, beta); | |
731 mask = vec_and(mask, tempmask); | |
732 tempmask = diff_lt_altivec(q1, q0, beta); | |
733 mask = vec_and(mask, tempmask); | |
734 | |
735 return mask; | |
736 } | |
737 | |
5165
c99fa49eaa80
part 2/2 of fixing Altivec-accelerated H264 luma inloop filter
gpoirier
parents:
5164
diff
changeset
|
738 // out: newp1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0) |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
739 static inline vec_u8_t h264_deblock_q1(register vec_u8_t p0, |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
740 register vec_u8_t p1, |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
741 register vec_u8_t p2, |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
742 register vec_u8_t q0, |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
743 register vec_u8_t tc0) { |
5164 | 744 |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
745 register vec_u8_t average = vec_avg(p0, q0); |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
746 register vec_u8_t temp; |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
747 register vec_u8_t uncliped; |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
748 register vec_u8_t ones; |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
749 register vec_u8_t max; |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
750 register vec_u8_t min; |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
751 register vec_u8_t newp1; |
5164 | 752 |
753 temp = vec_xor(average, p2); | |
754 average = vec_avg(average, p2); /*avg(p2, avg(p0, q0)) */ | |
755 ones = vec_splat_u8(1); | |
756 temp = vec_and(temp, ones); /*(p2^avg(p0, q0)) & 1 */ | |
757 uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */ | |
758 max = vec_adds(p1, tc0); | |
759 min = vec_subs(p1, tc0); | |
5165
c99fa49eaa80
part 2/2 of fixing Altivec-accelerated H264 luma inloop filter
gpoirier
parents:
5164
diff
changeset
|
760 newp1 = vec_max(min, uncliped); |
c99fa49eaa80
part 2/2 of fixing Altivec-accelerated H264 luma inloop filter
gpoirier
parents:
5164
diff
changeset
|
761 newp1 = vec_min(max, newp1); |
c99fa49eaa80
part 2/2 of fixing Altivec-accelerated H264 luma inloop filter
gpoirier
parents:
5164
diff
changeset
|
762 return newp1; |
5119 | 763 } |
764 | |
765 #define h264_deblock_p0_q0(p0, p1, q0, q1, tc0masked) { \ | |
766 \ | |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
767 const vec_u8_t A0v = vec_sl(vec_splat_u8(10), vec_splat_u8(4)); \ |
5119 | 768 \ |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
769 register vec_u8_t pq0bit = vec_xor(p0,q0); \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
770 register vec_u8_t q1minus; \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
771 register vec_u8_t p0minus; \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
772 register vec_u8_t stage1; \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
773 register vec_u8_t stage2; \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
774 register vec_u8_t vec160; \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
775 register vec_u8_t delta; \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
776 register vec_u8_t deltaneg; \ |
5119 | 777 \ |
5159
883a5619f52f
Use a faster way to compute 255-val: Instead of creating a vector of
gpoirier
parents:
5145
diff
changeset
|
778 q1minus = vec_nor(q1, q1); /* 255 - q1 */ \ |
5119 | 779 stage1 = vec_avg(p1, q1minus); /* (p1 - q1 + 256)>>1 */ \ |
780 stage2 = vec_sr(stage1, vec_splat_u8(1)); /* (p1 - q1 + 256)>>2 = 64 + (p1 - q1) >> 2 */ \ | |
5159
883a5619f52f
Use a faster way to compute 255-val: Instead of creating a vector of
gpoirier
parents:
5145
diff
changeset
|
781 p0minus = vec_nor(p0, p0); /* 255 - p0 */ \ |
5119 | 782 stage1 = vec_avg(q0, p0minus); /* (q0 - p0 + 256)>>1 */ \ |
783 pq0bit = vec_and(pq0bit, vec_splat_u8(1)); \ | |
784 stage2 = vec_avg(stage2, pq0bit); /* 32 + ((q0 - p0)&1 + (p1 - q1) >> 2 + 1) >> 1 */ \ | |
785 stage2 = vec_adds(stage2, stage1); /* 160 + ((p0 - q0) + (p1 - q1) >> 2 + 1) >> 1 */ \ | |
786 vec160 = vec_ld(0, &A0v); \ | |
787 deltaneg = vec_subs(vec160, stage2); /* -d */ \ | |
788 delta = vec_subs(stage2, vec160); /* d */ \ | |
789 deltaneg = vec_min(tc0masked, deltaneg); \ | |
790 delta = vec_min(tc0masked, delta); \ | |
791 p0 = vec_subs(p0, deltaneg); \ | |
792 q0 = vec_subs(q0, delta); \ | |
793 p0 = vec_adds(p0, delta); \ | |
794 q0 = vec_adds(q0, deltaneg); \ | |
795 } | |
796 | |
797 #define h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0) { \ | |
798 DECLARE_ALIGNED_16(unsigned char, temp[16]); \ | |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
799 register vec_u8_t alphavec; \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
800 register vec_u8_t betavec; \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
801 register vec_u8_t mask; \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
802 register vec_u8_t p1mask; \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
803 register vec_u8_t q1mask; \ |
5167 | 804 register vector signed char tc0vec; \ |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
805 register vec_u8_t finaltc0; \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
806 register vec_u8_t tc0masked; \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
807 register vec_u8_t newp1; \ |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
808 register vec_u8_t newq1; \ |
5119 | 809 \ |
810 temp[0] = alpha; \ | |
811 temp[1] = beta; \ | |
812 alphavec = vec_ld(0, temp); \ | |
813 betavec = vec_splat(alphavec, 0x1); \ | |
814 alphavec = vec_splat(alphavec, 0x0); \ | |
815 mask = h264_deblock_mask(p0, p1, q0, q1, alphavec, betavec); /*if in block */ \ | |
816 \ | |
817 *((int *)temp) = *((int *)tc0); \ | |
5167 | 818 tc0vec = vec_ld(0, (signed char*)temp); \ |
5119 | 819 tc0vec = vec_mergeh(tc0vec, tc0vec); \ |
820 tc0vec = vec_mergeh(tc0vec, tc0vec); \ | |
5165
c99fa49eaa80
part 2/2 of fixing Altivec-accelerated H264 luma inloop filter
gpoirier
parents:
5164
diff
changeset
|
821 mask = vec_and(mask, vec_cmpgt(tc0vec, vec_splat_s8(-1))); /* if tc0[i] >= 0 */ \ |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
822 finaltc0 = vec_and((vec_u8_t)tc0vec, mask); /* tc = tc0 */ \ |
5119 | 823 \ |
824 p1mask = diff_lt_altivec(p2, p0, betavec); \ | |
5586 | 825 p1mask = vec_and(p1mask, mask); /* if ( |p2 - p0| < beta) */ \ |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
826 tc0masked = vec_and(p1mask, (vec_u8_t)tc0vec); \ |
5119 | 827 finaltc0 = vec_sub(finaltc0, p1mask); /* tc++ */ \ |
5165
c99fa49eaa80
part 2/2 of fixing Altivec-accelerated H264 luma inloop filter
gpoirier
parents:
5164
diff
changeset
|
828 newp1 = h264_deblock_q1(p0, p1, p2, q0, tc0masked); \ |
5119 | 829 /*end if*/ \ |
830 \ | |
831 q1mask = diff_lt_altivec(q2, q0, betavec); \ | |
832 q1mask = vec_and(q1mask, mask); /* if ( |q2 - q0| < beta ) */\ | |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
833 tc0masked = vec_and(q1mask, (vec_u8_t)tc0vec); \ |
5119 | 834 finaltc0 = vec_sub(finaltc0, q1mask); /* tc++ */ \ |
5165
c99fa49eaa80
part 2/2 of fixing Altivec-accelerated H264 luma inloop filter
gpoirier
parents:
5164
diff
changeset
|
835 newq1 = h264_deblock_q1(p0, q1, q2, q0, tc0masked); \ |
5119 | 836 /*end if*/ \ |
837 \ | |
838 h264_deblock_p0_q0(p0, p1, q0, q1, finaltc0); \ | |
5165
c99fa49eaa80
part 2/2 of fixing Altivec-accelerated H264 luma inloop filter
gpoirier
parents:
5164
diff
changeset
|
839 p1 = newp1; \ |
c99fa49eaa80
part 2/2 of fixing Altivec-accelerated H264 luma inloop filter
gpoirier
parents:
5164
diff
changeset
|
840 q1 = newq1; \ |
5119 | 841 } |
842 | |
843 static void h264_v_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) { | |
844 | |
5586 | 845 if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) >= 0) { |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
846 register vec_u8_t p2 = vec_ld(-3*stride, pix); |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
847 register vec_u8_t p1 = vec_ld(-2*stride, pix); |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
848 register vec_u8_t p0 = vec_ld(-1*stride, pix); |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
849 register vec_u8_t q0 = vec_ld(0, pix); |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
850 register vec_u8_t q1 = vec_ld(stride, pix); |
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
851 register vec_u8_t q2 = vec_ld(2*stride, pix); |
5119 | 852 h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0); |
853 vec_st(p1, -2*stride, pix); | |
854 vec_st(p0, -1*stride, pix); | |
855 vec_st(q0, 0, pix); | |
856 vec_st(q1, stride, pix); | |
857 } | |
858 } | |
859 | |
860 static void h264_h_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) { | |
861 | |
5530
cd266411b11a
use shorter types vec_"type" instead of the too long vector "type"
gpoirier
parents:
5167
diff
changeset
|
862 register vec_u8_t line0, line1, line2, line3, line4, line5; |
5586 | 863 if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) < 0) |
5119 | 864 return; |
865 readAndTranspose16x6(pix-3, stride, line0, line1, line2, line3, line4, line5); | |
866 h264_loop_filter_luma_altivec(line0, line1, line2, line3, line4, line5, alpha, beta, tc0); | |
867 transpose4x16(line1, line2, line3, line4); | |
868 write16x4(pix-2, stride, line1, line2, line3, line4); | |
869 } | |
870 | |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
871 void dsputil_h264_init_ppc(DSPContext* c, AVCodecContext *avctx) { |
2967 | 872 |
5586 | 873 if (has_altivec()) { |
874 c->put_h264_chroma_pixels_tab[0] = put_h264_chroma_mc8_altivec; | |
875 c->put_no_rnd_h264_chroma_pixels_tab[0] = put_no_rnd_h264_chroma_mc8_altivec; | |
876 c->avg_h264_chroma_pixels_tab[0] = avg_h264_chroma_mc8_altivec; | |
877 c->h264_idct_add = ff_h264_idct_add_altivec; | |
878 c->h264_idct8_add = ff_h264_idct8_add_altivec; | |
879 c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_altivec; | |
880 c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_altivec; | |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
881 |
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
882 #define dspfunc(PFX, IDX, NUM) \ |
5586 | 883 c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \ |
884 c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_altivec; \ | |
885 c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_altivec; \ | |
886 c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_altivec; \ | |
887 c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_altivec; \ | |
888 c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_altivec; \ | |
889 c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_altivec; \ | |
890 c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_altivec; \ | |
891 c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_altivec; \ | |
892 c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_altivec; \ | |
893 c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_altivec; \ | |
894 c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_altivec; \ | |
895 c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_altivec; \ | |
896 c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_altivec; \ | |
897 c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_altivec; \ | |
898 c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_altivec | |
2967 | 899 |
5586 | 900 dspfunc(put_h264_qpel, 0, 16); |
901 dspfunc(avg_h264_qpel, 0, 16); | |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
902 #undef dspfunc |
5586 | 903 } |
2236
b0102ea621dd
h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)
michael
parents:
diff
changeset
|
904 } |