Mercurial > libavcodec.hg
annotate alpha/dsputil_alpha.c @ 12483:0159a19bfff7 libavcodec
aacdec: Rework channel mapping compatibility hacks.
For a PCE based configuration map the channels solely based on tags.
For an indexed configuration map the channels solely based on position.
This works with all known exotic samples including al17, elem_id0, bad_concat,
and lfe_is_sce.
author | alexc |
---|---|
date | Fri, 10 Sep 2010 18:01:48 +0000 |
parents | 9e7d38743146 |
children |
rev | line source |
---|---|
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
1 /* |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
2 * Alpha optimized DSP utils |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
3 * Copyright (c) 2002 Falk Hueffner <falk@debian.org> |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
4 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
5 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
6 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
7 * FFmpeg is free software; you can redistribute it and/or |
429 | 8 * modify it under the terms of the GNU Lesser General Public |
9 * 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:
3036
diff
changeset
|
10 * version 2.1 of the License, or (at your option) any later version. |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
11 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
12 * FFmpeg is distributed in the hope that it will be useful, |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
429 | 14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
15 * Lesser General Public License for more details. | |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
16 * |
429 | 17 * 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:
3036
diff
changeset
|
18 * License along with FFmpeg; if not, write to the Free Software |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
2967
diff
changeset
|
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
20 */ |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
21 |
6763 | 22 #include "libavcodec/dsputil.h" |
11396 | 23 #include "dsputil_alpha.h" |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
24 #include "asm.h" |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
25 |
897 | 26 void (*put_pixels_clamped_axp_p)(const DCTELEM *block, uint8_t *pixels, |
27 int line_size); | |
2967 | 28 void (*add_pixels_clamped_axp_p)(const DCTELEM *block, uint8_t *pixels, |
897 | 29 int line_size); |
509
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
30 |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
31 #if 0 |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
32 /* These functions were the base for the optimized assembler routines, |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
33 and remain here for documentation purposes. */ |
2967 | 34 static void put_pixels_clamped_mvi(const DCTELEM *block, uint8_t *pixels, |
509
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
35 int line_size) |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
36 { |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
37 int i = 8; |
509
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
38 uint64_t clampmask = zap(-1, 0xaa); /* 0x00ff00ff00ff00ff */ |
505
7a976bf93394
Ugly hack to make the assembler accept MVI instructions.
mellum
parents:
429
diff
changeset
|
39 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
40 do { |
509
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
41 uint64_t shorts0, shorts1; |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
42 |
509
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
43 shorts0 = ldq(block); |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
44 shorts0 = maxsw4(shorts0, 0); |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
45 shorts0 = minsw4(shorts0, clampmask); |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
46 stl(pkwb(shorts0), pixels); |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
47 |
509
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
48 shorts1 = ldq(block + 4); |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
49 shorts1 = maxsw4(shorts1, 0); |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
50 shorts1 = minsw4(shorts1, clampmask); |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
51 stl(pkwb(shorts1), pixels + 4); |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
52 |
509
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
53 pixels += line_size; |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
54 block += 8; |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
55 } while (--i); |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
56 } |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
57 |
2967 | 58 void add_pixels_clamped_mvi(const DCTELEM *block, uint8_t *pixels, |
509
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
59 int line_size) |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
60 { |
509
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
61 int h = 8; |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
62 /* Keep this function a leaf function by generating the constants |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
63 manually (mainly for the hack value ;-). */ |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
64 uint64_t clampmask = zap(-1, 0xaa); /* 0x00ff00ff00ff00ff */ |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
65 uint64_t signmask = zap(-1, 0x33); |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
66 signmask ^= signmask >> 1; /* 0x8000800080008000 */ |
505
7a976bf93394
Ugly hack to make the assembler accept MVI instructions.
mellum
parents:
429
diff
changeset
|
67 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
68 do { |
509
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
69 uint64_t shorts0, pix0, signs0; |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
70 uint64_t shorts1, pix1, signs1; |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
71 |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
72 shorts0 = ldq(block); |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
73 shorts1 = ldq(block + 4); |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
74 |
509
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
75 pix0 = unpkbw(ldl(pixels)); |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
76 /* Signed subword add (MMX paddw). */ |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
77 signs0 = shorts0 & signmask; |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
78 shorts0 &= ~signmask; |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
79 shorts0 += pix0; |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
80 shorts0 ^= signs0; |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
81 /* Clamp. */ |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
82 shorts0 = maxsw4(shorts0, 0); |
2967 | 83 shorts0 = minsw4(shorts0, clampmask); |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
84 |
509
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
85 /* Next 4. */ |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
86 pix1 = unpkbw(ldl(pixels + 4)); |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
87 signs1 = shorts1 & signmask; |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
88 shorts1 &= ~signmask; |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
89 shorts1 += pix1; |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
90 shorts1 ^= signs1; |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
91 shorts1 = maxsw4(shorts1, 0); |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
92 shorts1 = minsw4(shorts1, clampmask); |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
93 |
509
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
94 stl(pkwb(shorts0), pixels); |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
95 stl(pkwb(shorts1), pixels + 4); |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
96 |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
97 pixels += line_size; |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
98 block += 8; |
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
99 } while (--h); |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
100 } |
509
cab79946302f
Implement put_pixels_clamped and add_pixels_clamped in Assembler. This
mellum
parents:
505
diff
changeset
|
101 #endif |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
102 |
518 | 103 static void clear_blocks_axp(DCTELEM *blocks) { |
104 uint64_t *p = (uint64_t *) blocks; | |
105 int n = sizeof(DCTELEM) * 6 * 64; | |
106 | |
107 do { | |
108 p[0] = 0; | |
109 p[1] = 0; | |
110 p[2] = 0; | |
111 p[3] = 0; | |
112 p[4] = 0; | |
113 p[5] = 0; | |
114 p[6] = 0; | |
115 p[7] = 0; | |
116 p += 8; | |
117 n -= 8 * 8; | |
118 } while (n); | |
119 } | |
120 | |
513 | 121 static inline uint64_t avg2_no_rnd(uint64_t a, uint64_t b) |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
122 { |
513 | 123 return (a & b) + (((a ^ b) & BYTE_VEC(0xfe)) >> 1); |
124 } | |
125 | |
126 static inline uint64_t avg2(uint64_t a, uint64_t b) | |
127 { | |
2967 | 128 return (a | b) - (((a ^ b) & BYTE_VEC(0xfe)) >> 1); |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
129 } |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
130 |
546 | 131 #if 0 |
132 /* The XY2 routines basically utilize this scheme, but reuse parts in | |
133 each iteration. */ | |
513 | 134 static inline uint64_t avg4(uint64_t l1, uint64_t l2, uint64_t l3, uint64_t l4) |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
135 { |
513 | 136 uint64_t r1 = ((l1 & ~BYTE_VEC(0x03)) >> 2) |
897 | 137 + ((l2 & ~BYTE_VEC(0x03)) >> 2) |
138 + ((l3 & ~BYTE_VEC(0x03)) >> 2) | |
139 + ((l4 & ~BYTE_VEC(0x03)) >> 2); | |
513 | 140 uint64_t r2 = (( (l1 & BYTE_VEC(0x03)) |
897 | 141 + (l2 & BYTE_VEC(0x03)) |
142 + (l3 & BYTE_VEC(0x03)) | |
143 + (l4 & BYTE_VEC(0x03)) | |
144 + BYTE_VEC(0x02)) >> 2) & BYTE_VEC(0x03); | |
513 | 145 return r1 + r2; |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
146 } |
546 | 147 #endif |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
148 |
548
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
149 #define OP(LOAD, STORE) \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
150 do { \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
151 STORE(LOAD(pixels), block); \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
152 pixels += line_size; \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
153 block += line_size; \ |
513 | 154 } while (--h) |
155 | |
548
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
156 #define OP_X2(LOAD, STORE) \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
157 do { \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
158 uint64_t pix1, pix2; \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
159 \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
160 pix1 = LOAD(pixels); \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
161 pix2 = pix1 >> 8 | ((uint64_t) pixels[8] << 56); \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
162 STORE(AVG2(pix1, pix2), block); \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
163 pixels += line_size; \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
164 block += line_size; \ |
513 | 165 } while (--h) |
166 | |
548
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
167 #define OP_Y2(LOAD, STORE) \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
168 do { \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
169 uint64_t pix = LOAD(pixels); \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
170 do { \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
171 uint64_t next_pix; \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
172 \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
173 pixels += line_size; \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
174 next_pix = LOAD(pixels); \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
175 STORE(AVG2(pix, next_pix), block); \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
176 block += line_size; \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
177 pix = next_pix; \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
178 } while (--h); \ |
513 | 179 } while (0) |
180 | |
548
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
181 #define OP_XY2(LOAD, STORE) \ |
546 | 182 do { \ |
183 uint64_t pix1 = LOAD(pixels); \ | |
184 uint64_t pix2 = pix1 >> 8 | ((uint64_t) pixels[8] << 56); \ | |
185 uint64_t pix_l = (pix1 & BYTE_VEC(0x03)) \ | |
186 + (pix2 & BYTE_VEC(0x03)); \ | |
187 uint64_t pix_h = ((pix1 & ~BYTE_VEC(0x03)) >> 2) \ | |
188 + ((pix2 & ~BYTE_VEC(0x03)) >> 2); \ | |
189 \ | |
190 do { \ | |
191 uint64_t npix1, npix2; \ | |
192 uint64_t npix_l, npix_h; \ | |
193 uint64_t avg; \ | |
194 \ | |
195 pixels += line_size; \ | |
196 npix1 = LOAD(pixels); \ | |
197 npix2 = npix1 >> 8 | ((uint64_t) pixels[8] << 56); \ | |
198 npix_l = (npix1 & BYTE_VEC(0x03)) \ | |
199 + (npix2 & BYTE_VEC(0x03)); \ | |
200 npix_h = ((npix1 & ~BYTE_VEC(0x03)) >> 2) \ | |
201 + ((npix2 & ~BYTE_VEC(0x03)) >> 2); \ | |
202 avg = (((pix_l + npix_l + AVG4_ROUNDER) >> 2) & BYTE_VEC(0x03)) \ | |
203 + pix_h + npix_h; \ | |
204 STORE(avg, block); \ | |
205 \ | |
548
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
206 block += line_size; \ |
546 | 207 pix_l = npix_l; \ |
208 pix_h = npix_h; \ | |
209 } while (--h); \ | |
513 | 210 } while (0) |
211 | |
670
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
212 #define MAKE_OP(OPNAME, SUFF, OPKIND, STORE) \ |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
213 static void OPNAME ## _pixels ## SUFF ## _axp \ |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
214 (uint8_t *restrict block, const uint8_t *restrict pixels, \ |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
215 int line_size, int h) \ |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
216 { \ |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
217 if ((size_t) pixels & 0x7) { \ |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
218 OPKIND(uldq, STORE); \ |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
219 } else { \ |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
220 OPKIND(ldq, STORE); \ |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
221 } \ |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
222 } \ |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
223 \ |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
224 static void OPNAME ## _pixels16 ## SUFF ## _axp \ |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
225 (uint8_t *restrict block, const uint8_t *restrict pixels, \ |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
226 int line_size, int h) \ |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
227 { \ |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
228 OPNAME ## _pixels ## SUFF ## _axp(block, pixels, line_size, h); \ |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
229 OPNAME ## _pixels ## SUFF ## _axp(block + 8, pixels + 8, line_size, h); \ |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
230 } |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
231 |
548
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
232 #define PIXOP(OPNAME, STORE) \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
233 MAKE_OP(OPNAME, , OP, STORE) \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
234 MAKE_OP(OPNAME, _x2, OP_X2, STORE) \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
235 MAKE_OP(OPNAME, _y2, OP_Y2, STORE) \ |
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
236 MAKE_OP(OPNAME, _xy2, OP_XY2, STORE) |
513 | 237 |
238 /* Rounding primitives. */ | |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
239 #define AVG2 avg2 |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
240 #define AVG4 avg4 |
546 | 241 #define AVG4_ROUNDER BYTE_VEC(0x02) |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
242 #define STORE(l, b) stq(l, b) |
548
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
243 PIXOP(put, STORE); |
513 | 244 |
245 #undef STORE | |
246 #define STORE(l, b) stq(AVG2(l, ldq(b)), b); | |
548
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
247 PIXOP(avg, STORE); |
513 | 248 |
249 /* Not rounding primitives. */ | |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
250 #undef AVG2 |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
251 #undef AVG4 |
546 | 252 #undef AVG4_ROUNDER |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
253 #undef STORE |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
254 #define AVG2 avg2_no_rnd |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
255 #define AVG4 avg4_no_rnd |
546 | 256 #define AVG4_ROUNDER BYTE_VEC(0x01) |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
257 #define STORE(l, b) stq(l, b) |
548
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
258 PIXOP(put_no_rnd, STORE); |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
259 |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
260 #undef STORE |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
261 #define STORE(l, b) stq(AVG2(l, ldq(b)), b); |
548
3f05be811b5a
Remove support for variable BSIZE and INCR, as sub_pixels_* is no
mellum
parents:
546
diff
changeset
|
262 PIXOP(avg_no_rnd, STORE); |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
263 |
11396 | 264 static void put_pixels16_axp_asm(uint8_t *block, const uint8_t *pixels, |
265 int line_size, int h) | |
670
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
266 { |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
267 put_pixels_axp_asm(block, pixels, line_size, h); |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
268 put_pixels_axp_asm(block + 8, pixels + 8, line_size, h); |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
269 } |
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
270 |
1092 | 271 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx) |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
272 { |
856
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
273 c->put_pixels_tab[0][0] = put_pixels16_axp_asm; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
274 c->put_pixels_tab[0][1] = put_pixels16_x2_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
275 c->put_pixels_tab[0][2] = put_pixels16_y2_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
276 c->put_pixels_tab[0][3] = put_pixels16_xy2_axp; |
670
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
277 |
856
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
278 c->put_no_rnd_pixels_tab[0][0] = put_pixels16_axp_asm; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
279 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
280 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
281 c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_axp; |
670
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
282 |
856
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
283 c->avg_pixels_tab[0][0] = avg_pixels16_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
284 c->avg_pixels_tab[0][1] = avg_pixels16_x2_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
285 c->avg_pixels_tab[0][2] = avg_pixels16_y2_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
286 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_axp; |
670
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
287 |
856
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
288 c->avg_no_rnd_pixels_tab[0][0] = avg_no_rnd_pixels16_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
289 c->avg_no_rnd_pixels_tab[0][1] = avg_no_rnd_pixels16_x2_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
290 c->avg_no_rnd_pixels_tab[0][2] = avg_no_rnd_pixels16_y2_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
291 c->avg_no_rnd_pixels_tab[0][3] = avg_no_rnd_pixels16_xy2_axp; |
670
340e3ba84119
Synthesize pixels16 functions from pixels functions.
mellum
parents:
663
diff
changeset
|
292 |
856
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
293 c->put_pixels_tab[1][0] = put_pixels_axp_asm; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
294 c->put_pixels_tab[1][1] = put_pixels_x2_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
295 c->put_pixels_tab[1][2] = put_pixels_y2_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
296 c->put_pixels_tab[1][3] = put_pixels_xy2_axp; |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
297 |
856
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
298 c->put_no_rnd_pixels_tab[1][0] = put_pixels_axp_asm; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
299 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels_x2_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
300 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels_y2_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
301 c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels_xy2_axp; |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
302 |
856
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
303 c->avg_pixels_tab[1][0] = avg_pixels_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
304 c->avg_pixels_tab[1][1] = avg_pixels_x2_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
305 c->avg_pixels_tab[1][2] = avg_pixels_y2_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
306 c->avg_pixels_tab[1][3] = avg_pixels_xy2_axp; |
513 | 307 |
856
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
308 c->avg_no_rnd_pixels_tab[1][0] = avg_no_rnd_pixels_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
309 c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels_x2_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
310 c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels_y2_axp; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
311 c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels_xy2_axp; |
513 | 312 |
856
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
313 c->clear_blocks = clear_blocks_axp; |
518 | 314 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
315 /* amask clears all bits that correspond to present features. */ |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
316 if (amask(AMASK_MVI) == 0) { |
856
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
317 c->put_pixels_clamped = put_pixels_clamped_mvi_asm; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
318 c->add_pixels_clamped = add_pixels_clamped_mvi_asm; |
586 | 319 |
856
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
320 c->get_pixels = get_pixels_mvi; |
3c6df37177dd
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
744
diff
changeset
|
321 c->diff_pixels = diff_pixels_mvi; |
8625 | 322 c->sad[0] = pix_abs16x16_mvi_asm; |
1708 | 323 c->sad[1] = pix_abs8x8_mvi; |
8625 | 324 c->pix_abs[0][0] = pix_abs16x16_mvi_asm; |
1708 | 325 c->pix_abs[1][0] = pix_abs8x8_mvi; |
326 c->pix_abs[0][1] = pix_abs16x16_x2_mvi; | |
327 c->pix_abs[0][2] = pix_abs16x16_y2_mvi; | |
328 c->pix_abs[0][3] = pix_abs16x16_xy2_mvi; | |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
329 } |
897 | 330 |
331 put_pixels_clamped_axp_p = c->put_pixels_clamped; | |
332 add_pixels_clamped_axp_p = c->add_pixels_clamped; | |
2967 | 333 |
8619 | 334 if (!avctx->lowres && |
335 (avctx->idct_algo == FF_IDCT_AUTO || | |
336 avctx->idct_algo == FF_IDCT_SIMPLEALPHA)) { | |
8621 | 337 c->idct_put = ff_simple_idct_put_axp; |
338 c->idct_add = ff_simple_idct_add_axp; | |
339 c->idct = ff_simple_idct_axp; | |
8619 | 340 } |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
diff
changeset
|
341 } |