Mercurial > libavcodec.hg
annotate dsputil.h @ 1834:41c324b09fcc libavcodec
100l (field picture decoding fix)
author | michael |
---|---|
date | Thu, 26 Feb 2004 02:48:07 +0000 |
parents | 65f7bd09f37b |
children | 9a73e6dc5d68 |
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 * |
6 * This library is free software; you can redistribute it and/or | |
7 * modify it under the terms of the GNU Lesser General Public | |
8 * License as published by the Free Software Foundation; either | |
9 * version 2 of the License, or (at your option) any later version. | |
10 * | |
11 * This library is distributed in the hope that it will be useful, | |
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
14 * Lesser General Public License for more details. | |
15 * | |
16 * You should have received a copy of the GNU Lesser General Public | |
17 * License along with this library; if not, write to the Free Software | |
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
19 */ | |
1102 | 20 |
21 /** | |
22 * @file dsputil.h | |
1106 | 23 * DSP utils. |
1213 | 24 * note, many functions in here may use MMX which trashes the FPU state, it is |
25 * absolutely necessary to call emms_c() between dsp & float/double code | |
1102 | 26 */ |
27 | |
0 | 28 #ifndef DSPUTIL_H |
29 #define DSPUTIL_H | |
30 | |
31 #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
|
32 #include "avcodec.h" |
0 | 33 |
1102 | 34 |
255 | 35 //#define DEBUG |
0 | 36 /* dct code */ |
37 typedef short DCTELEM; | |
38 | |
474
11dbd00682fc
avoid name clash with libjpeg - added missing externs
bellard
parents:
429
diff
changeset
|
39 void fdct_ifast (DCTELEM *data); |
1571
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1567
diff
changeset
|
40 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
|
41 void ff_jpeg_fdct_islow (DCTELEM *data); |
1567 | 42 void ff_fdct248_islow (DCTELEM *data); |
0 | 43 |
44 void j_rev_dct (DCTELEM *data); | |
45 | |
687
9abb13c21fbe
fdct_mmx -> ff_fdct_mmx (renamed to avoid namespace conflict with xvid)
arpi_esp
parents:
675
diff
changeset
|
46 void ff_fdct_mmx(DCTELEM *block); |
1565 | 47 void ff_fdct_mmx2(DCTELEM *block); |
1765
e31754bc5b65
SSE2 fdct by (Balatoni Denes <pnis at coder dot hu>)
michael
parents:
1739
diff
changeset
|
48 void ff_fdct_sse2(DCTELEM *block); |
0 | 49 |
34 | 50 /* encoding scans */ |
1064 | 51 extern const uint8_t ff_alternate_horizontal_scan[64]; |
52 extern const uint8_t ff_alternate_vertical_scan[64]; | |
53 extern const uint8_t ff_zigzag_direct[64]; | |
1567 | 54 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
|
55 |
0 | 56 /* pixel operations */ |
57 #define MAX_NEG_CROP 384 | |
58 | |
59 /* temporary */ | |
1064 | 60 extern uint32_t squareTbl[512]; |
61 extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP]; | |
0 | 62 |
63 | |
675 | 64 /* minimum alignment rules ;) |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
65 if u notice errors in the align stuff, need more alignment for some asm code for some cpu |
675 | 66 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ... |
67 | |
68 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible) | |
69 i (michael) didnt check them, these are just the alignents which i think could be reached easily ... | |
70 | |
71 !future video codecs might need functions with less strict alignment | |
72 */ | |
0 | 73 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
74 /* |
1064 | 75 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size); |
76 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride); | |
77 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
78 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
296 | 79 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
|
80 */ |
0 | 81 |
82 /* add and put pixel (decoding) */ | |
675 | 83 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 |
1709 | 84 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4 |
1064 | 85 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
|
86 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 | 87 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); |
1168 | 88 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y); |
0 | 89 |
984 | 90 #define DEF_OLD_QPEL(name)\ |
1064 | 91 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ |
92 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ | |
93 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
|
94 |
984 | 95 DEF_OLD_QPEL(qpel16_mc11_old_c) |
96 DEF_OLD_QPEL(qpel16_mc31_old_c) | |
97 DEF_OLD_QPEL(qpel16_mc12_old_c) | |
98 DEF_OLD_QPEL(qpel16_mc32_old_c) | |
99 DEF_OLD_QPEL(qpel16_mc13_old_c) | |
100 DEF_OLD_QPEL(qpel16_mc33_old_c) | |
101 DEF_OLD_QPEL(qpel8_mc11_old_c) | |
102 DEF_OLD_QPEL(qpel8_mc31_old_c) | |
103 DEF_OLD_QPEL(qpel8_mc12_old_c) | |
104 DEF_OLD_QPEL(qpel8_mc32_old_c) | |
105 DEF_OLD_QPEL(qpel8_mc13_old_c) | |
106 DEF_OLD_QPEL(qpel8_mc33_old_c) | |
651 | 107 |
108 #define CALL_2X_PIXELS(a, b, n)\ | |
109 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ | |
110 b(block , pixels , line_size, h);\ | |
111 b(block+n, pixels+n, line_size, h);\ | |
112 } | |
255 | 113 |
0 | 114 /* motion estimation */ |
1709 | 115 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2 |
116 // allthough currently h<4 is not used as functions with width <8 are not used and neither implemented | |
1708 | 117 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 | 118 |
1168 | 119 |
1102 | 120 /** |
121 * DSPContext. | |
122 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
123 typedef struct DSPContext { |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
124 /* pixel ops : interface with DCT */ |
1064 | 125 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); |
126 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); | |
127 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); | |
128 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); | |
1136 | 129 /** |
130 * translational global motion compensation. | |
131 */ | |
1064 | 132 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); |
1136 | 133 /** |
134 * global motion compensation. | |
135 */ | |
1064 | 136 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy, |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
137 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
138 void (*clear_blocks)(DCTELEM *blocks/*align 16*/); |
1064 | 139 int (*pix_sum)(uint8_t * pix, int line_size); |
140 int (*pix_norm1)(uint8_t * pix, int line_size); | |
1708 | 141 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4 |
142 | |
1729 | 143 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */ |
144 me_cmp_func sse[5]; | |
145 me_cmp_func hadamard8_diff[5]; | |
146 me_cmp_func dct_sad[5]; | |
147 me_cmp_func quant_psnr[5]; | |
148 me_cmp_func bit[5]; | |
149 me_cmp_func rd[5]; | |
150 me_cmp_func vsad[5]; | |
151 me_cmp_func vsse[5]; | |
936 | 152 |
1708 | 153 me_cmp_func me_pre_cmp[5]; |
154 me_cmp_func me_cmp[5]; | |
155 me_cmp_func me_sub_cmp[5]; | |
156 me_cmp_func mb_cmp[5]; | |
1729 | 157 me_cmp_func ildct_cmp[5]; //only width 16 used |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
158 |
1136 | 159 /** |
160 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
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
|
161 * this is an array[4][4] of motion compensation funcions for 4 |
1713 | 162 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 163 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 164 * @param block destination where the result is stored |
165 * @param pixels source | |
166 * @param line_size number of bytes in a horizontal line of block | |
167 * @param h height | |
168 */ | |
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
|
169 op_pixels_func put_pixels_tab[4][4]; |
1136 | 170 |
171 /** | |
172 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
1320 | 173 * This is an array[4][4] of motion compensation functions for 4 |
1713 | 174 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 175 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 176 * @param block destination into which the result is averaged (a+b+1)>>1 |
177 * @param pixels source | |
178 * @param line_size number of bytes in a horizontal line of block | |
179 * @param h height | |
180 */ | |
1319 | 181 op_pixels_func avg_pixels_tab[4][4]; |
1136 | 182 |
183 /** | |
184 * Halfpel motion compensation with no rounding (a+b)>>1. | |
1225 | 185 * this is an array[2][4] of motion compensation funcions for 2 |
186 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> | |
1213 | 187 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 188 * @param block destination where the result is stored |
189 * @param pixels source | |
190 * @param line_size number of bytes in a horizontal line of block | |
191 * @param h height | |
192 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
193 op_pixels_func put_no_rnd_pixels_tab[2][4]; |
1136 | 194 |
195 /** | |
196 * Halfpel motion compensation with no rounding (a+b)>>1. | |
1225 | 197 * this is an array[2][4] of motion compensation funcions for 2 |
198 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> | |
1213 | 199 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 200 * @param block destination into which the result is averaged (a+b)>>1 |
201 * @param pixels source | |
202 * @param line_size number of bytes in a horizontal line of block | |
203 * @param h height | |
204 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
205 op_pixels_func avg_no_rnd_pixels_tab[2][4]; |
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
|
206 |
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
|
207 /** |
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
|
208 * 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
|
209 * 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
|
210 * *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
|
211 * @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
|
212 * @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
|
213 * @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
|
214 * @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
|
215 */ |
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
|
216 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
1319 | 217 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
218 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
219 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
|
220 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
|
221 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
|
222 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; |
936 | 223 qpel_mc_func put_mspel_pixels_tab[8]; |
1168 | 224 |
225 /** | |
226 * h264 Chram MC | |
227 */ | |
228 h264_chroma_mc_func put_h264_chroma_pixels_tab[3]; | |
229 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
|
230 |
1168 | 231 qpel_mc_func put_h264_qpel_pixels_tab[3][16]; |
232 qpel_mc_func avg_h264_qpel_pixels_tab[3][16]; | |
233 | |
1708 | 234 me_cmp_func pix_abs[2][4]; |
866 | 235 |
236 /* huffyuv specific */ | |
237 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); | |
936 | 238 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); |
1527 | 239 /** |
240 * subtract huffyuv's variant of median prediction | |
241 * note, this might read from src1[-1], src2[-1] | |
242 */ | |
243 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top); | |
1273 | 244 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w); |
1092 | 245 |
1644 | 246 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale); |
247 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale); | |
248 | |
1092 | 249 /* (I)DCT */ |
250 void (*fdct)(DCTELEM *block/* align 16*/); | |
1567 | 251 void (*fdct248)(DCTELEM *block/* align 16*/); |
1102 | 252 |
1324
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
253 /* IDCT really*/ |
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
254 void (*idct)(DCTELEM *block/* align 16*/); |
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
255 |
1102 | 256 /** |
1104 | 257 * block -> idct -> clip to unsigned 8 bit -> dest. |
1102 | 258 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) |
1111 | 259 * @param line_size size in bytes of a horizotal line of dest |
1102 | 260 */ |
1092 | 261 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
1102 | 262 |
263 /** | |
264 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. | |
1111 | 265 * @param line_size size in bytes of a horizotal line of dest |
1102 | 266 */ |
1092 | 267 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
1102 | 268 |
269 /** | |
1104 | 270 * idct input permutation. |
1241 | 271 * several optimized IDCTs need a permutated input (relative to the normal order of the reference |
272 * IDCT) | |
273 * this permutation must be performed before the idct_put/add, note, normally this can be merged | |
274 * with the zigzag/alternate scan<br> | |
1102 | 275 * an example to avoid confusion: |
276 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...) | |
277 * - (x -> referece dct -> reference idct -> x) | |
278 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x) | |
279 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...) | |
280 */ | |
1092 | 281 uint8_t idct_permutation[64]; |
282 int idct_permutation_type; | |
283 #define FF_NO_IDCT_PERM 1 | |
284 #define FF_LIBMPEG2_IDCT_PERM 2 | |
285 #define FF_SIMPLE_IDCT_PERM 3 | |
286 #define FF_TRANSPOSE_IDCT_PERM 4 | |
287 | |
1784 | 288 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale); |
289 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); | |
290 #define BASIS_SHIFT 16 | |
291 #define RECON_SHIFT 6 | |
292 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
293 } DSPContext; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
294 |
1201 | 295 void dsputil_static_init(void); |
1092 | 296 void dsputil_init(DSPContext* p, AVCodecContext *avctx); |
0 | 297 |
764 | 298 /** |
299 * permute block according to permuatation. | |
300 * @param last last non zero element in scantable order | |
301 */ | |
1064 | 302 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); |
34 | 303 |
1729 | 304 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type); |
305 | |
1264 | 306 #define BYTE_VEC32(c) ((c)*0x01010101UL) |
307 | |
308 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b) | |
309 { | |
310 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
311 } | |
312 | |
313 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b) | |
314 { | |
315 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
316 } | |
317 | |
1102 | 318 /** |
1104 | 319 * Empty mmx state. |
1102 | 320 * this must be called between any dsp function and float/double code. |
321 * for example sin(); dsp->idct_put(); emms_c(); cos() | |
322 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
323 #define emms_c() |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
324 |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
325 /* should be defined by architectures supporting |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
326 one or more MultiMedia extension */ |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
327 int mm_support(void); |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
328 |
62 | 329 #if defined(HAVE_MMX) |
0 | 330 |
862 | 331 #undef emms_c |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
332 |
0 | 333 #define MM_MMX 0x0001 /* standard MMX */ |
334 #define MM_3DNOW 0x0004 /* AMD 3DNOW */ | |
335 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */ | |
336 #define MM_SSE 0x0008 /* SSE functions */ | |
337 #define MM_SSE2 0x0010 /* PIV SSE2 functions */ | |
338 | |
339 extern int mm_flags; | |
340 | |
1064 | 341 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
342 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); | |
0 | 343 |
344 static inline void emms(void) | |
345 { | |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
346 __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
|
347 } |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
348 |
936 | 349 |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
350 #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
|
351 {\ |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
352 if (mm_flags & MM_MMX)\ |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
353 emms();\ |
0 | 354 } |
355 | |
356 #define __align8 __attribute__ ((aligned (8))) | |
357 | |
1092 | 358 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); |
359 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx); | |
1065 | 360 |
62 | 361 #elif defined(ARCH_ARMV4L) |
362 | |
363 /* This is to use 4 bytes read to the IDCT pointers for some 'zero' | |
364 line ptimizations */ | |
365 #define __align8 __attribute__ ((aligned (4))) | |
366 | |
1092 | 367 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx); |
62 | 368 |
88 | 369 #elif defined(HAVE_MLIB) |
370 | |
371 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ | |
372 #define __align8 __attribute__ ((aligned (8))) | |
373 | |
1092 | 374 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx); |
88 | 375 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
376 #elif defined(ARCH_ALPHA) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
377 |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
378 #define __align8 __attribute__ ((aligned (8))) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
379 |
1092 | 380 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
|
381 |
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
|
382 #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
|
383 |
894
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
384 #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
|
385 |
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
386 extern int mm_flags; |
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
387 |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
388 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN) |
1653 | 389 #define pixel altivec_pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
390 #include <altivec.h> |
1653 | 391 #undef pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
392 #endif |
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
393 |
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
|
394 #define __align8 __attribute__ ((aligned (16))) |
92e99e506920
first cut at altivec support on darwin patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
612
diff
changeset
|
395 |
1092 | 396 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
|
397 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
398 #elif defined(HAVE_MMI) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
399 |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
400 #define __align8 __attribute__ ((aligned (16))) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
401 |
1092 | 402 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
|
403 |
1259
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
404 #elif defined(ARCH_SH4) |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
405 |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
406 #define __align8 __attribute__ ((aligned (8))) |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
407 |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
408 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
|
409 |
0 | 410 #else |
411 | |
412 #define __align8 | |
413 | |
414 #endif | |
415 | |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
416 #ifdef __GNUC__ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
417 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
418 struct unaligned_64 { uint64_t l; } __attribute__((packed)); |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
419 struct unaligned_32 { uint32_t l; } __attribute__((packed)); |
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
|
420 struct unaligned_16 { uint16_t l; } __attribute__((packed)); |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
421 |
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
|
422 #define LD16(a) (((const struct unaligned_16 *) (a))->l) |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
423 #define LD32(a) (((const struct unaligned_32 *) (a))->l) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
424 #define LD64(a) (((const struct unaligned_64 *) (a))->l) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
425 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
426 #define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
427 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
428 #else /* __GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
429 |
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
|
430 #define LD16(a) (*((uint16_t*)(a))) |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
431 #define LD32(a) (*((uint32_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
432 #define LD64(a) (*((uint64_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
433 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
434 #define ST32(a, b) *((uint32_t*)(a)) = (b) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
435 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
436 #endif /* !__GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
437 |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
438 /* PSNR */ |
1064 | 439 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
|
440 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
|
441 AVCodecContext *avctx); |
781 | 442 |
443 /* FFT computation */ | |
444 | |
445 /* NOTE: soon integer code will be added, so you must use the | |
446 FFTSample type */ | |
447 typedef float FFTSample; | |
448 | |
449 typedef struct FFTComplex { | |
450 FFTSample re, im; | |
451 } FFTComplex; | |
452 | |
453 typedef struct FFTContext { | |
454 int nbits; | |
455 int inverse; | |
456 uint16_t *revtab; | |
457 FFTComplex *exptab; | |
458 FFTComplex *exptab1; /* only used by SSE code */ | |
459 void (*fft_calc)(struct FFTContext *s, FFTComplex *z); | |
460 } FFTContext; | |
461 | |
462 int fft_init(FFTContext *s, int nbits, int inverse); | |
463 void fft_permute(FFTContext *s, FFTComplex *z); | |
464 void fft_calc_c(FFTContext *s, FFTComplex *z); | |
465 void fft_calc_sse(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
|
466 void fft_calc_altivec(FFTContext *s, FFTComplex *z); |
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
954
diff
changeset
|
467 |
781 | 468 static inline void fft_calc(FFTContext *s, FFTComplex *z) |
469 { | |
470 s->fft_calc(s, z); | |
471 } | |
472 void fft_end(FFTContext *s); | |
473 | |
474 /* MDCT computation */ | |
475 | |
476 typedef struct MDCTContext { | |
477 int n; /* size of MDCT (i.e. number of input data * 2) */ | |
478 int nbits; /* n = 2^nbits */ | |
479 /* pre/post rotation tables */ | |
480 FFTSample *tcos; | |
481 FFTSample *tsin; | |
482 FFTContext fft; | |
483 } MDCTContext; | |
484 | |
794 | 485 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
|
486 void ff_imdct_calc(MDCTContext *s, FFTSample *output, |
781 | 487 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
|
488 void ff_mdct_calc(MDCTContext *s, FFTSample *out, |
781 | 489 const FFTSample *input, FFTSample *tmp); |
794 | 490 void ff_mdct_end(MDCTContext *s); |
781 | 491 |
1708 | 492 #define WARPER8_16(name8, name16)\ |
493 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
494 return name8(s, dst , src , stride, h)\ | |
495 +name8(s, dst+8 , src+8 , stride, h);\ | |
496 } | |
497 | |
498 #define WARPER8_16_SQ(name8, name16)\ | |
499 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
500 int score=0;\ | |
501 score +=name8(s, dst , src , stride, 8);\ | |
502 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
503 if(h==16){\ | |
504 dst += 8*stride;\ | |
505 src += 8*stride;\ | |
506 score +=name8(s, dst , src , stride, 8);\ | |
507 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
508 }\ | |
509 return score;\ | |
936 | 510 } |
511 | |
838
b78812db886f
lrintf detection (based upon a patch by Franois Revol <revol at free dot fr>)
michaelni
parents:
802
diff
changeset
|
512 #ifndef HAVE_LRINTF |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
513 /* XXX: add ISOC specific test to avoid specific BSD testing. */ |
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
514 /* better than nothing implementation. */ |
802 | 515 /* btw, rintf() is existing on fbsd too -- alex */ |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
516 static inline long int lrintf(float x) |
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
517 { |
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
|
518 #ifdef CONFIG_WIN32 |
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
|
519 /* XXX: incorrect, but make it compile */ |
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
|
520 return (int)(x); |
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
|
521 #else |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
522 return (int)(rint(x)); |
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
|
523 #endif |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
524 } |
0 | 525 #endif |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
526 |
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
527 #endif |