Mercurial > libavcodec.hg
annotate dsputil.h @ 4391:20944c70a3c0 libavcodec
fix segfault with darkkben.free.fr/corrupted_h264.mp4
author | michael |
---|---|
date | Tue, 23 Jan 2007 19:32:26 +0000 |
parents | f9325f2bc762 |
children | d3e389536b0a |
rev | line source |
---|---|
429 | 1 /* |
2 * DSP utils | |
3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard. | |
1739
07a484280a82
copyright year update of the files i touched and remembered, things look annoyingly unmaintained otherwise
michael
parents:
1729
diff
changeset
|
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> |
429 | 5 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
6 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
7 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
8 * FFmpeg is free software; you can redistribute it and/or |
429 | 9 * modify it under the terms of the GNU Lesser General Public |
10 * License as published by the Free Software Foundation; either | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
11 * version 2.1 of the License, or (at your option) any later version. |
429 | 12 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
13 * FFmpeg is distributed in the hope that it will be useful, |
429 | 14 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
16 * Lesser General Public License for more details. | |
17 * | |
18 * You should have received a copy of the GNU Lesser General Public | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
19 * License along with FFmpeg; if not, write to the Free Software |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
3029
diff
changeset
|
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
429 | 21 */ |
1102 | 22 |
23 /** | |
24 * @file dsputil.h | |
1106 | 25 * DSP utils. |
1213 | 26 * note, many functions in here may use MMX which trashes the FPU state, it is |
27 * absolutely necessary to call emms_c() between dsp & float/double code | |
1102 | 28 */ |
29 | |
0 | 30 #ifndef DSPUTIL_H |
31 #define DSPUTIL_H | |
32 | |
33 #include "common.h" | |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
34 #include "avcodec.h" |
0 | 35 |
1102 | 36 |
255 | 37 //#define DEBUG |
0 | 38 /* dct code */ |
39 typedef short DCTELEM; | |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
40 typedef int DWTELEM; |
0 | 41 |
474
11dbd00682fc
avoid name clash with libjpeg - added missing externs
bellard
parents:
429
diff
changeset
|
42 void fdct_ifast (DCTELEM *data); |
1571
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1567
diff
changeset
|
43 void fdct_ifast248 (DCTELEM *data); |
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
623
diff
changeset
|
44 void ff_jpeg_fdct_islow (DCTELEM *data); |
1567 | 45 void ff_fdct248_islow (DCTELEM *data); |
0 | 46 |
47 void j_rev_dct (DCTELEM *data); | |
2256 | 48 void j_rev_dct4 (DCTELEM *data); |
2257 | 49 void j_rev_dct2 (DCTELEM *data); |
2259 | 50 void j_rev_dct1 (DCTELEM *data); |
0 | 51 |
687
9abb13c21fbe
fdct_mmx -> ff_fdct_mmx (renamed to avoid namespace conflict with xvid)
arpi_esp
parents:
675
diff
changeset
|
52 void ff_fdct_mmx(DCTELEM *block); |
1565 | 53 void ff_fdct_mmx2(DCTELEM *block); |
1765
e31754bc5b65
SSE2 fdct by (Balatoni Denes <pnis at coder dot hu>)
michael
parents:
1739
diff
changeset
|
54 void ff_fdct_sse2(DCTELEM *block); |
0 | 55 |
2755 | 56 void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride); |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
57 void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride); |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3089
diff
changeset
|
58 void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3089
diff
changeset
|
59 void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride); |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
60 void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block); |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
61 void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block); |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
62 |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
63 void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1, |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
64 const float *src2, int src3, int blocksize, int step); |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
65 void ff_float_to_int16_c(int16_t *dst, const float *src, int len); |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
66 |
34 | 67 /* encoding scans */ |
1064 | 68 extern const uint8_t ff_alternate_horizontal_scan[64]; |
69 extern const uint8_t ff_alternate_vertical_scan[64]; | |
70 extern const uint8_t ff_zigzag_direct[64]; | |
1567 | 71 extern const uint8_t ff_zigzag248_direct[64]; |
190
9e0e56869d05
fix for non-mmx runtimedetect encoding bugs - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
174
diff
changeset
|
72 |
0 | 73 /* pixel operations */ |
2090 | 74 #define MAX_NEG_CROP 1024 |
0 | 75 |
76 /* temporary */ | |
4179 | 77 extern uint32_t ff_squareTbl[512]; |
4176 | 78 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP]; |
0 | 79 |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
80 /* VP3 DSP functions */ |
2693 | 81 void ff_vp3_idct_c(DCTELEM *block/* align 16*/); |
82 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); | |
83 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); | |
0 | 84 |
3245 | 85 /* 1/2^n downscaling functions from imgconvert.c */ |
86 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
87 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
88 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
89 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
90 | |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
91 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy, |
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
92 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); |
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
93 |
675 | 94 /* minimum alignment rules ;) |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
95 if u notice errors in the align stuff, need more alignment for some asm code for some cpu |
675 | 96 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ... |
97 | |
98 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible) | |
99 i (michael) didnt check them, these are just the alignents which i think could be reached easily ... | |
100 | |
101 !future video codecs might need functions with less strict alignment | |
102 */ | |
0 | 103 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
104 /* |
1064 | 105 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size); |
106 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride); | |
107 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
108 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
296 | 109 void clear_blocks_c(DCTELEM *blocks); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
110 */ |
0 | 111 |
112 /* add and put pixel (decoding) */ | |
675 | 113 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 |
1709 | 114 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4 |
1064 | 115 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h); |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
116 typedef void (*tpel_mc_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int w, int h); |
1064 | 117 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); |
1168 | 118 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y); |
2415 | 119 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset); |
3029 | 120 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset); |
0 | 121 |
984 | 122 #define DEF_OLD_QPEL(name)\ |
1064 | 123 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ |
124 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ | |
125 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
126 |
984 | 127 DEF_OLD_QPEL(qpel16_mc11_old_c) |
128 DEF_OLD_QPEL(qpel16_mc31_old_c) | |
129 DEF_OLD_QPEL(qpel16_mc12_old_c) | |
130 DEF_OLD_QPEL(qpel16_mc32_old_c) | |
131 DEF_OLD_QPEL(qpel16_mc13_old_c) | |
132 DEF_OLD_QPEL(qpel16_mc33_old_c) | |
133 DEF_OLD_QPEL(qpel8_mc11_old_c) | |
134 DEF_OLD_QPEL(qpel8_mc31_old_c) | |
135 DEF_OLD_QPEL(qpel8_mc12_old_c) | |
136 DEF_OLD_QPEL(qpel8_mc32_old_c) | |
137 DEF_OLD_QPEL(qpel8_mc13_old_c) | |
138 DEF_OLD_QPEL(qpel8_mc33_old_c) | |
651 | 139 |
140 #define CALL_2X_PIXELS(a, b, n)\ | |
141 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ | |
142 b(block , pixels , line_size, h);\ | |
143 b(block+n, pixels+n, line_size, h);\ | |
144 } | |
255 | 145 |
0 | 146 /* motion estimation */ |
1709 | 147 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2 |
148 // allthough currently h<4 is not used as functions with width <8 are not used and neither implemented | |
1708 | 149 typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size, int h)/* __attribute__ ((const))*/; |
936 | 150 |
1168 | 151 |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
152 // for snow slices |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
153 typedef struct slice_buffer_s slice_buffer; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
154 |
1102 | 155 /** |
156 * DSPContext. | |
157 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
158 typedef struct DSPContext { |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
159 /* pixel ops : interface with DCT */ |
1064 | 160 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); |
161 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); | |
162 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); | |
1984
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
163 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
1064 | 164 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
2763 | 165 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size); |
166 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size); | |
1136 | 167 /** |
168 * translational global motion compensation. | |
169 */ | |
1064 | 170 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); |
1136 | 171 /** |
172 * global motion compensation. | |
173 */ | |
1064 | 174 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy, |
2979 | 175 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
176 void (*clear_blocks)(DCTELEM *blocks/*align 16*/); |
1064 | 177 int (*pix_sum)(uint8_t * pix, int line_size); |
178 int (*pix_norm1)(uint8_t * pix, int line_size); | |
1708 | 179 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4 |
2967 | 180 |
1729 | 181 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */ |
182 me_cmp_func sse[5]; | |
183 me_cmp_func hadamard8_diff[5]; | |
184 me_cmp_func dct_sad[5]; | |
185 me_cmp_func quant_psnr[5]; | |
186 me_cmp_func bit[5]; | |
187 me_cmp_func rd[5]; | |
188 me_cmp_func vsad[5]; | |
189 me_cmp_func vsse[5]; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
190 me_cmp_func nsse[5]; |
2184 | 191 me_cmp_func w53[5]; |
192 me_cmp_func w97[5]; | |
2382 | 193 me_cmp_func dct_max[5]; |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
194 me_cmp_func dct264_sad[5]; |
936 | 195 |
1708 | 196 me_cmp_func me_pre_cmp[5]; |
197 me_cmp_func me_cmp[5]; | |
198 me_cmp_func me_sub_cmp[5]; | |
199 me_cmp_func mb_cmp[5]; | |
1729 | 200 me_cmp_func ildct_cmp[5]; //only width 16 used |
2382 | 201 me_cmp_func frame_skip_cmp[5]; //only width 8 used |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
202 |
1136 | 203 /** |
204 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
2967 | 205 * this is an array[4][4] of motion compensation funcions for 4 |
1713 | 206 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 207 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 208 * @param block destination where the result is stored |
209 * @param pixels source | |
210 * @param line_size number of bytes in a horizontal line of block | |
211 * @param h height | |
212 */ | |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
213 op_pixels_func put_pixels_tab[4][4]; |
1136 | 214 |
215 /** | |
216 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
2967 | 217 * This is an array[4][4] of motion compensation functions for 4 |
1713 | 218 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 219 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 220 * @param block destination into which the result is averaged (a+b+1)>>1 |
221 * @param pixels source | |
222 * @param line_size number of bytes in a horizontal line of block | |
223 * @param h height | |
224 */ | |
1319 | 225 op_pixels_func avg_pixels_tab[4][4]; |
1136 | 226 |
227 /** | |
228 * Halfpel motion compensation with no rounding (a+b)>>1. | |
2967 | 229 * this is an array[2][4] of motion compensation funcions for 2 |
1225 | 230 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 231 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 232 * @param block destination where the result is stored |
233 * @param pixels source | |
234 * @param line_size number of bytes in a horizontal line of block | |
235 * @param h height | |
236 */ | |
2075 | 237 op_pixels_func put_no_rnd_pixels_tab[4][4]; |
1136 | 238 |
239 /** | |
240 * Halfpel motion compensation with no rounding (a+b)>>1. | |
2967 | 241 * this is an array[2][4] of motion compensation funcions for 2 |
1225 | 242 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 243 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 244 * @param block destination into which the result is averaged (a+b)>>1 |
245 * @param pixels source | |
246 * @param line_size number of bytes in a horizontal line of block | |
247 * @param h height | |
248 */ | |
2075 | 249 op_pixels_func avg_no_rnd_pixels_tab[4][4]; |
2967 | 250 |
1864 | 251 void (*put_no_rnd_pixels_l2[2])(uint8_t *block/*align width (8 or 16)*/, const uint8_t *a/*align 1*/, const uint8_t *b/*align 1*/, int line_size, int h); |
2967 | 252 |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
253 /** |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
254 * Thirdpel motion compensation with rounding (a+b+1)>>1. |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
255 * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br> |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
256 * *pixels_tab[ xthirdpel + 4*ythirdpel ] |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
257 * @param block destination where the result is stored |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
258 * @param pixels source |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
259 * @param line_size number of bytes in a horizontal line of block |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
260 * @param h height |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
261 */ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
262 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
1319 | 263 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
264 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
265 qpel_mc_func put_qpel_pixels_tab[2][16]; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
266 qpel_mc_func avg_qpel_pixels_tab[2][16]; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
267 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16]; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
268 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; |
936 | 269 qpel_mc_func put_mspel_pixels_tab[8]; |
2967 | 270 |
1168 | 271 /** |
272 * h264 Chram MC | |
273 */ | |
274 h264_chroma_mc_func put_h264_chroma_pixels_tab[3]; | |
3663 | 275 /* This is really one func used in VC-1 decoding */ |
276 h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3]; | |
1168 | 277 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3]; |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
278 |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3010
diff
changeset
|
279 qpel_mc_func put_h264_qpel_pixels_tab[4][16]; |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3010
diff
changeset
|
280 qpel_mc_func avg_h264_qpel_pixels_tab[4][16]; |
2967 | 281 |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3746
diff
changeset
|
282 qpel_mc_func put_2tap_qpel_pixels_tab[4][16]; |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3746
diff
changeset
|
283 qpel_mc_func avg_2tap_qpel_pixels_tab[4][16]; |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3746
diff
changeset
|
284 |
2415 | 285 h264_weight_func weight_h264_pixels_tab[10]; |
286 h264_biweight_func biweight_h264_pixels_tab[10]; | |
2967 | 287 |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
288 /* AVS specific */ |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
289 qpel_mc_func put_cavs_qpel_pixels_tab[2][16]; |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
290 qpel_mc_func avg_cavs_qpel_pixels_tab[2][16]; |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
291 void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
292 void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
293 void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
294 void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
295 void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
296 |
1708 | 297 me_cmp_func pix_abs[2][4]; |
2967 | 298 |
866 | 299 /* huffyuv specific */ |
300 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); | |
936 | 301 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); |
1527 | 302 /** |
303 * subtract huffyuv's variant of median prediction | |
304 * note, this might read from src1[-1], src2[-1] | |
305 */ | |
306 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top); | |
1273 | 307 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w); |
2633 | 308 |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
309 void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
310 void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
311 void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
312 void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
313 void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta); |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
314 void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta); |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3574
diff
changeset
|
315 // h264_loop_filter_strength: simd only. the C version is inlined in h264.c |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3574
diff
changeset
|
316 void (*h264_loop_filter_strength)(int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2], |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3574
diff
changeset
|
317 int bidir, int edges, int step, int mask_mv0, int mask_mv1); |
2967 | 318 |
1644 | 319 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale); |
320 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale); | |
321 | |
2045 | 322 void (*h261_loop_filter)(uint8_t *src, int stride); |
2044
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
2024
diff
changeset
|
323 |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
324 /* assume len is a multiple of 4, and arrays are 16-byte aligned */ |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
325 void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize); |
3574 | 326 /* assume len is a multiple of 8, and arrays are 16-byte aligned */ |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
327 void (*vector_fmul)(float *dst, const float *src, int len); |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
328 void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len); |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
329 /* assume len is a multiple of 8, and src arrays are 16-byte aligned */ |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
330 void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step); |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
331 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
332 /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767] |
3660
7e1ee254a3ee
Align the input buffer in ffplay, introduce a public macro for aligned declarations
lu_zero
parents:
3656
diff
changeset
|
333 * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */ |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
334 void (*float_to_int16)(int16_t *dst, const float *src, int len); |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
335 |
1092 | 336 /* (I)DCT */ |
337 void (*fdct)(DCTELEM *block/* align 16*/); | |
1567 | 338 void (*fdct248)(DCTELEM *block/* align 16*/); |
2967 | 339 |
1324
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
340 /* IDCT really*/ |
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
341 void (*idct)(DCTELEM *block/* align 16*/); |
2967 | 342 |
1102 | 343 /** |
1104 | 344 * block -> idct -> clip to unsigned 8 bit -> dest. |
1102 | 345 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) |
1111 | 346 * @param line_size size in bytes of a horizotal line of dest |
1102 | 347 */ |
1092 | 348 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
2967 | 349 |
1102 | 350 /** |
351 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. | |
1111 | 352 * @param line_size size in bytes of a horizotal line of dest |
1102 | 353 */ |
1092 | 354 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
2967 | 355 |
1102 | 356 /** |
1104 | 357 * idct input permutation. |
1241 | 358 * several optimized IDCTs need a permutated input (relative to the normal order of the reference |
359 * IDCT) | |
360 * this permutation must be performed before the idct_put/add, note, normally this can be merged | |
361 * with the zigzag/alternate scan<br> | |
1102 | 362 * an example to avoid confusion: |
363 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...) | |
364 * - (x -> referece dct -> reference idct -> x) | |
365 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x) | |
366 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...) | |
367 */ | |
1092 | 368 uint8_t idct_permutation[64]; |
369 int idct_permutation_type; | |
370 #define FF_NO_IDCT_PERM 1 | |
371 #define FF_LIBMPEG2_IDCT_PERM 2 | |
372 #define FF_SIMPLE_IDCT_PERM 3 | |
373 #define FF_TRANSPOSE_IDCT_PERM 4 | |
2696
9699d325049d
porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents:
2693
diff
changeset
|
374 #define FF_PARTTRANS_IDCT_PERM 5 |
1092 | 375 |
1784 | 376 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale); |
377 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); | |
378 #define BASIS_SHIFT 16 | |
379 #define RECON_SHIFT 6 | |
2967 | 380 |
4268 | 381 /* h264 functions */ |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
382 void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride); |
2755 | 383 void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride); |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3089
diff
changeset
|
384 void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3089
diff
changeset
|
385 void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride); |
4279 | 386 void (*h264_dct)(DCTELEM block[4][4]); |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
387 |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
388 /* snow wavelet */ |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
389 void (*vertical_compose97i)(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width); |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
390 void (*horizontal_compose97i)(DWTELEM *b, int width); |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
391 void (*inner_add_yblock)(uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8); |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3198
diff
changeset
|
392 |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3198
diff
changeset
|
393 void (*prefetch)(void *mem, int stride, int h); |
3245 | 394 |
395 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
3526 | 396 |
397 /* vc1 functions */ | |
398 void (*vc1_inv_trans_8x8)(DCTELEM *b); | |
399 void (*vc1_inv_trans_8x4)(DCTELEM *b, int n); | |
400 void (*vc1_inv_trans_4x8)(DCTELEM *b, int n); | |
401 void (*vc1_inv_trans_4x4)(DCTELEM *b, int n); | |
4239 | 402 void (*vc1_v_overlap)(uint8_t* src, int stride); |
403 void (*vc1_h_overlap)(uint8_t* src, int stride); | |
3526 | 404 /* put 8x8 block with bicubic interpolation and quarterpel precision |
405 * last argument is actually round value instead of height | |
406 */ | |
407 op_pixels_func put_vc1_mspel_pixels_tab[16]; | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
408 } DSPContext; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
409 |
4197 | 410 void dsputil_static_init(void); |
1092 | 411 void dsputil_init(DSPContext* p, AVCodecContext *avctx); |
0 | 412 |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4279
diff
changeset
|
413 int ff_check_alignment(void); |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4279
diff
changeset
|
414 |
764 | 415 /** |
416 * permute block according to permuatation. | |
417 * @param last last non zero element in scantable order | |
418 */ | |
1064 | 419 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); |
34 | 420 |
1729 | 421 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type); |
422 | |
2979 | 423 #define BYTE_VEC32(c) ((c)*0x01010101UL) |
1264 | 424 |
425 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b) | |
426 { | |
427 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
428 } | |
429 | |
430 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b) | |
431 { | |
432 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
433 } | |
434 | |
2184 | 435 static inline int get_penalty_factor(int lambda, int lambda2, int type){ |
436 switch(type&0xFF){ | |
437 default: | |
438 case FF_CMP_SAD: | |
439 return lambda>>FF_LAMBDA_SHIFT; | |
440 case FF_CMP_DCT: | |
441 return (3*lambda)>>(FF_LAMBDA_SHIFT+1); | |
442 case FF_CMP_W53: | |
443 return (4*lambda)>>(FF_LAMBDA_SHIFT); | |
444 case FF_CMP_W97: | |
445 return (2*lambda)>>(FF_LAMBDA_SHIFT); | |
446 case FF_CMP_SATD: | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
447 case FF_CMP_DCT264: |
2184 | 448 return (2*lambda)>>FF_LAMBDA_SHIFT; |
449 case FF_CMP_RD: | |
450 case FF_CMP_PSNR: | |
451 case FF_CMP_SSE: | |
452 case FF_CMP_NSSE: | |
453 return lambda2>>FF_LAMBDA_SHIFT; | |
454 case FF_CMP_BIT: | |
455 return 1; | |
456 } | |
457 } | |
458 | |
1102 | 459 /** |
1104 | 460 * Empty mmx state. |
1102 | 461 * this must be called between any dsp function and float/double code. |
462 * for example sin(); dsp->idct_put(); emms_c(); cos() | |
463 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
464 #define emms_c() |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
465 |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
466 /* should be defined by architectures supporting |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
467 one or more MultiMedia extension */ |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
468 int mm_support(void); |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
469 |
3089 | 470 #ifdef __GNUC__ |
471 #define DECLARE_ALIGNED_16(t,v) t v __attribute__ ((aligned (16))) | |
472 #else | |
473 #define DECLARE_ALIGNED_16(t,v) __declspec(align(16)) t v | |
474 #endif | |
1974
8c5489b2cf3e
move __align16 some place where non-MMX machines can see it
melanson
parents:
1972
diff
changeset
|
475 |
62 | 476 #if defined(HAVE_MMX) |
0 | 477 |
862 | 478 #undef emms_c |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
479 |
0 | 480 #define MM_MMX 0x0001 /* standard MMX */ |
481 #define MM_3DNOW 0x0004 /* AMD 3DNOW */ | |
482 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */ | |
483 #define MM_SSE 0x0008 /* SSE functions */ | |
484 #define MM_SSE2 0x0010 /* PIV SSE2 functions */ | |
2388 | 485 #define MM_3DNOWEXT 0x0020 /* AMD 3DNowExt */ |
3279
647a677c00a4
Remove unused and unsupported Cyrix's "Extended MMX",
gpoirier
parents:
3248
diff
changeset
|
486 #define MM_SSE3 0x0040 /* Prescott SSE3 functions */ |
4311
f9325f2bc762
Add SSSE3 (Core2 aka Conroe/Merom/Woodcrester new instructions) detection
gpoirier
parents:
4281
diff
changeset
|
487 #define MM_SSSE3 0x0080 /* Conroe SSSE3 functions */ |
0 | 488 |
4197 | 489 extern int mm_flags; |
490 | |
1064 | 491 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
492 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); | |
1984
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
493 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
0 | 494 |
495 static inline void emms(void) | |
496 { | |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
497 __asm __volatile ("emms;":::"memory"); |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
498 } |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
499 |
936 | 500 |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
501 #define emms_c() \ |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
502 {\ |
4197 | 503 if (mm_flags & MM_MMX)\ |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
504 emms();\ |
0 | 505 } |
506 | |
3089 | 507 #ifdef __GNUC__ |
508 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) | |
509 #else | |
510 #define DECLARE_ALIGNED_8(t,v) __declspec(align(8)) t v | |
511 #endif | |
512 | |
2324 | 513 #define STRIDE_ALIGN 8 |
0 | 514 |
1092 | 515 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); |
516 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx); | |
1065 | 517 |
62 | 518 #elif defined(ARCH_ARMV4L) |
519 | |
520 /* This is to use 4 bytes read to the IDCT pointers for some 'zero' | |
1974
8c5489b2cf3e
move __align16 some place where non-MMX machines can see it
melanson
parents:
1972
diff
changeset
|
521 line optimizations */ |
3089 | 522 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (4))) |
2324 | 523 #define STRIDE_ALIGN 4 |
62 | 524 |
2776
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
525 #define MM_IWMMXT 0x0100 /* XScale IWMMXT */ |
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
526 |
4197 | 527 extern int mm_flags; |
528 | |
1092 | 529 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx); |
62 | 530 |
88 | 531 #elif defined(HAVE_MLIB) |
532 | |
533 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ | |
3089 | 534 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 535 #define STRIDE_ALIGN 8 |
88 | 536 |
1092 | 537 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx); |
88 | 538 |
1959
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
539 #elif defined(ARCH_SPARC) |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
540 |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
541 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ |
3089 | 542 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 543 #define STRIDE_ALIGN 8 |
1959
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
544 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx); |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
545 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
546 #elif defined(ARCH_ALPHA) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
547 |
3089 | 548 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 549 #define STRIDE_ALIGN 8 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
550 |
1092 | 551 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx); |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
552 |
623
92e99e506920
first cut at altivec support on darwin patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
612
diff
changeset
|
553 #elif defined(ARCH_POWERPC) |
92e99e506920
first cut at altivec support on darwin patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
612
diff
changeset
|
554 |
894
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
555 #define MM_ALTIVEC 0x0001 /* standard AltiVec */ |
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
556 |
4197 | 557 extern int mm_flags; |
558 | |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
559 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN) |
1653 | 560 #define pixel altivec_pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
561 #include <altivec.h> |
1653 | 562 #undef pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
563 #endif |
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
564 |
3089 | 565 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16))) |
2324 | 566 #define STRIDE_ALIGN 16 |
623
92e99e506920
first cut at altivec support on darwin patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
612
diff
changeset
|
567 |
1092 | 568 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx); |
623
92e99e506920
first cut at altivec support on darwin patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
612
diff
changeset
|
569 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
570 #elif defined(HAVE_MMI) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
571 |
3089 | 572 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16))) |
2324 | 573 #define STRIDE_ALIGN 16 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
574 |
1092 | 575 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx); |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
576 |
1259
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
577 #elif defined(ARCH_SH4) |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
578 |
3089 | 579 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 580 #define STRIDE_ALIGN 8 |
1259
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
581 |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
582 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx); |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
583 |
3728 | 584 #elif defined(ARCH_BFIN) |
585 | |
586 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) | |
587 #define STRIDE_ALIGN 8 | |
588 | |
589 void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx); | |
590 | |
0 | 591 #else |
592 | |
3089 | 593 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 594 #define STRIDE_ALIGN 8 |
0 | 595 |
596 #endif | |
597 | |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
598 /* PSNR */ |
1064 | 599 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3], |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
600 int orig_linesize[3], int coded_linesize, |
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
601 AVCodecContext *avctx); |
781 | 602 |
603 /* FFT computation */ | |
604 | |
605 /* NOTE: soon integer code will be added, so you must use the | |
606 FFTSample type */ | |
607 typedef float FFTSample; | |
608 | |
3555 | 609 struct MDCTContext; |
610 | |
781 | 611 typedef struct FFTComplex { |
612 FFTSample re, im; | |
613 } FFTComplex; | |
614 | |
615 typedef struct FFTContext { | |
616 int nbits; | |
617 int inverse; | |
618 uint16_t *revtab; | |
619 FFTComplex *exptab; | |
620 FFTComplex *exptab1; /* only used by SSE code */ | |
621 void (*fft_calc)(struct FFTContext *s, FFTComplex *z); | |
3555 | 622 void (*imdct_calc)(struct MDCTContext *s, FFTSample *output, |
623 const FFTSample *input, FFTSample *tmp); | |
781 | 624 } FFTContext; |
625 | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
626 int ff_fft_init(FFTContext *s, int nbits, int inverse); |
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
627 void ff_fft_permute(FFTContext *s, FFTComplex *z); |
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
628 void ff_fft_calc_c(FFTContext *s, FFTComplex *z); |
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
629 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z); |
3175 | 630 void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z); |
631 void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z); | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
632 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z); |
975
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
954
diff
changeset
|
633 |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
634 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z) |
781 | 635 { |
636 s->fft_calc(s, z); | |
637 } | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
638 void ff_fft_end(FFTContext *s); |
781 | 639 |
640 /* MDCT computation */ | |
641 | |
642 typedef struct MDCTContext { | |
643 int n; /* size of MDCT (i.e. number of input data * 2) */ | |
644 int nbits; /* n = 2^nbits */ | |
645 /* pre/post rotation tables */ | |
646 FFTSample *tcos; | |
647 FFTSample *tsin; | |
648 FFTContext fft; | |
649 } MDCTContext; | |
650 | |
794 | 651 int ff_mdct_init(MDCTContext *s, int nbits, int inverse); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
652 void ff_imdct_calc(MDCTContext *s, FFTSample *output, |
781 | 653 const FFTSample *input, FFTSample *tmp); |
3555 | 654 void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output, |
655 const FFTSample *input, FFTSample *tmp); | |
3746 | 656 void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output, |
657 const FFTSample *input, FFTSample *tmp); | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
658 void ff_mdct_calc(MDCTContext *s, FFTSample *out, |
781 | 659 const FFTSample *input, FFTSample *tmp); |
794 | 660 void ff_mdct_end(MDCTContext *s); |
781 | 661 |
1708 | 662 #define WARPER8_16(name8, name16)\ |
663 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
664 return name8(s, dst , src , stride, h)\ | |
665 +name8(s, dst+8 , src+8 , stride, h);\ | |
666 } | |
667 | |
668 #define WARPER8_16_SQ(name8, name16)\ | |
669 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
670 int score=0;\ | |
671 score +=name8(s, dst , src , stride, 8);\ | |
672 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
673 if(h==16){\ | |
674 dst += 8*stride;\ | |
675 src += 8*stride;\ | |
676 score +=name8(s, dst , src , stride, 8);\ | |
677 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
678 }\ | |
679 return score;\ | |
936 | 680 } |
681 | |
4240 | 682 |
683 static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
684 { | |
685 int i; | |
686 for(i=0; i<h; i++) | |
687 { | |
688 ST16(dst , LD16(src )); | |
689 dst+=dstStride; | |
690 src+=srcStride; | |
691 } | |
692 } | |
693 | |
694 static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
695 { | |
696 int i; | |
697 for(i=0; i<h; i++) | |
698 { | |
699 ST32(dst , LD32(src )); | |
700 dst+=dstStride; | |
701 src+=srcStride; | |
702 } | |
703 } | |
704 | |
705 static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
706 { | |
707 int i; | |
708 for(i=0; i<h; i++) | |
709 { | |
710 ST32(dst , LD32(src )); | |
711 ST32(dst+4 , LD32(src+4 )); | |
712 dst+=dstStride; | |
713 src+=srcStride; | |
714 } | |
715 } | |
716 | |
717 static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
718 { | |
719 int i; | |
720 for(i=0; i<h; i++) | |
721 { | |
722 ST32(dst , LD32(src )); | |
723 ST32(dst+4 , LD32(src+4 )); | |
724 dst[8]= src[8]; | |
725 dst+=dstStride; | |
726 src+=srcStride; | |
727 } | |
728 } | |
729 | |
730 static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
731 { | |
732 int i; | |
733 for(i=0; i<h; i++) | |
734 { | |
735 ST32(dst , LD32(src )); | |
736 ST32(dst+4 , LD32(src+4 )); | |
737 ST32(dst+8 , LD32(src+8 )); | |
738 ST32(dst+12, LD32(src+12)); | |
739 dst+=dstStride; | |
740 src+=srcStride; | |
741 } | |
742 } | |
743 | |
744 static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
745 { | |
746 int i; | |
747 for(i=0; i<h; i++) | |
748 { | |
749 ST32(dst , LD32(src )); | |
750 ST32(dst+4 , LD32(src+4 )); | |
751 ST32(dst+8 , LD32(src+8 )); | |
752 ST32(dst+12, LD32(src+12)); | |
753 dst[16]= src[16]; | |
754 dst+=dstStride; | |
755 src+=srcStride; | |
756 } | |
757 } | |
758 | |
1040
998d5035b15b
win32: rint() does not seem to be defined with mingw32-gcc 2.95 - do you have a better solution ?
bellard
parents:
1033
diff
changeset
|
759 #endif |