Mercurial > libavcodec.hg
annotate dsputil.h @ 2674:58fb8e1efcae libavcodec
10l (mb_type uninitalized and then changed and overwritten)
author | michael |
---|---|
date | Fri, 13 May 2005 08:17:25 +0000 |
parents | 72e6ffa1f3a5 |
children | 02925a3903b6 |
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); | |
2256 | 45 void j_rev_dct4 (DCTELEM *data); |
2257 | 46 void j_rev_dct2 (DCTELEM *data); |
2259 | 47 void j_rev_dct1 (DCTELEM *data); |
0 | 48 |
687
9abb13c21fbe
fdct_mmx -> ff_fdct_mmx (renamed to avoid namespace conflict with xvid)
arpi_esp
parents:
675
diff
changeset
|
49 void ff_fdct_mmx(DCTELEM *block); |
1565 | 50 void ff_fdct_mmx2(DCTELEM *block); |
1765
e31754bc5b65
SSE2 fdct by (Balatoni Denes <pnis at coder dot hu>)
michael
parents:
1739
diff
changeset
|
51 void ff_fdct_sse2(DCTELEM *block); |
0 | 52 |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
53 void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride); |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
54 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
|
55 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
|
56 |
34 | 57 /* encoding scans */ |
1064 | 58 extern const uint8_t ff_alternate_horizontal_scan[64]; |
59 extern const uint8_t ff_alternate_vertical_scan[64]; | |
60 extern const uint8_t ff_zigzag_direct[64]; | |
1567 | 61 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
|
62 |
0 | 63 /* pixel operations */ |
2090 | 64 #define MAX_NEG_CROP 1024 |
0 | 65 |
66 /* temporary */ | |
1064 | 67 extern uint32_t squareTbl[512]; |
68 extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP]; | |
0 | 69 |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
70 /* VP3 DSP functions */ |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
71 void vp3_dsp_init_c(void); |
1977 | 72 void vp3_idct_c(int16_t *input_data, int16_t *dequant_matrix, |
73 int coeff_count, DCTELEM *output_data); | |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
74 |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
75 void vp3_dsp_init_mmx(void); |
1977 | 76 void vp3_idct_mmx(int16_t *input_data, int16_t *dequant_matrix, |
77 int coeff_count, DCTELEM *output_data); | |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
78 |
1972 | 79 void vp3_dsp_init_sse2(void); |
1977 | 80 void vp3_idct_sse2(int16_t *input_data, int16_t *dequant_matrix, |
81 int coeff_count, DCTELEM *output_data); | |
0 | 82 |
675 | 83 /* minimum alignment rules ;) |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
84 if u notice errors in the align stuff, need more alignment for some asm code for some cpu |
675 | 85 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ... |
86 | |
87 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible) | |
88 i (michael) didnt check them, these are just the alignents which i think could be reached easily ... | |
89 | |
90 !future video codecs might need functions with less strict alignment | |
91 */ | |
0 | 92 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
93 /* |
1064 | 94 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size); |
95 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride); | |
96 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
97 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
296 | 98 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
|
99 */ |
0 | 100 |
101 /* add and put pixel (decoding) */ | |
675 | 102 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 |
1709 | 103 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4 |
1064 | 104 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
|
105 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 | 106 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); |
1168 | 107 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 | 108 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset); |
109 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offsetd, int offsets); | |
0 | 110 |
984 | 111 #define DEF_OLD_QPEL(name)\ |
1064 | 112 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ |
113 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ | |
114 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
|
115 |
984 | 116 DEF_OLD_QPEL(qpel16_mc11_old_c) |
117 DEF_OLD_QPEL(qpel16_mc31_old_c) | |
118 DEF_OLD_QPEL(qpel16_mc12_old_c) | |
119 DEF_OLD_QPEL(qpel16_mc32_old_c) | |
120 DEF_OLD_QPEL(qpel16_mc13_old_c) | |
121 DEF_OLD_QPEL(qpel16_mc33_old_c) | |
122 DEF_OLD_QPEL(qpel8_mc11_old_c) | |
123 DEF_OLD_QPEL(qpel8_mc31_old_c) | |
124 DEF_OLD_QPEL(qpel8_mc12_old_c) | |
125 DEF_OLD_QPEL(qpel8_mc32_old_c) | |
126 DEF_OLD_QPEL(qpel8_mc13_old_c) | |
127 DEF_OLD_QPEL(qpel8_mc33_old_c) | |
651 | 128 |
129 #define CALL_2X_PIXELS(a, b, n)\ | |
130 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ | |
131 b(block , pixels , line_size, h);\ | |
132 b(block+n, pixels+n, line_size, h);\ | |
133 } | |
255 | 134 |
0 | 135 /* motion estimation */ |
1709 | 136 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2 |
137 // allthough currently h<4 is not used as functions with width <8 are not used and neither implemented | |
1708 | 138 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 | 139 |
1168 | 140 |
1102 | 141 /** |
142 * DSPContext. | |
143 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
144 typedef struct DSPContext { |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
145 /* pixel ops : interface with DCT */ |
1064 | 146 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); |
147 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); | |
148 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
|
149 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
1064 | 150 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
1136 | 151 /** |
152 * translational global motion compensation. | |
153 */ | |
1064 | 154 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); |
1136 | 155 /** |
156 * global motion compensation. | |
157 */ | |
1064 | 158 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
|
159 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
|
160 void (*clear_blocks)(DCTELEM *blocks/*align 16*/); |
1064 | 161 int (*pix_sum)(uint8_t * pix, int line_size); |
162 int (*pix_norm1)(uint8_t * pix, int line_size); | |
1708 | 163 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4 |
164 | |
1729 | 165 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */ |
166 me_cmp_func sse[5]; | |
167 me_cmp_func hadamard8_diff[5]; | |
168 me_cmp_func dct_sad[5]; | |
169 me_cmp_func quant_psnr[5]; | |
170 me_cmp_func bit[5]; | |
171 me_cmp_func rd[5]; | |
172 me_cmp_func vsad[5]; | |
173 me_cmp_func vsse[5]; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
174 me_cmp_func nsse[5]; |
2184 | 175 me_cmp_func w53[5]; |
176 me_cmp_func w97[5]; | |
2382 | 177 me_cmp_func dct_max[5]; |
936 | 178 |
1708 | 179 me_cmp_func me_pre_cmp[5]; |
180 me_cmp_func me_cmp[5]; | |
181 me_cmp_func me_sub_cmp[5]; | |
182 me_cmp_func mb_cmp[5]; | |
1729 | 183 me_cmp_func ildct_cmp[5]; //only width 16 used |
2382 | 184 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
|
185 |
1136 | 186 /** |
187 * 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
|
188 * this is an array[4][4] of motion compensation funcions for 4 |
1713 | 189 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 190 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 191 * @param block destination where the result is stored |
192 * @param pixels source | |
193 * @param line_size number of bytes in a horizontal line of block | |
194 * @param h height | |
195 */ | |
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
|
196 op_pixels_func put_pixels_tab[4][4]; |
1136 | 197 |
198 /** | |
199 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
1320 | 200 * This is an array[4][4] of motion compensation functions for 4 |
1713 | 201 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 202 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 203 * @param block destination into which the result is averaged (a+b+1)>>1 |
204 * @param pixels source | |
205 * @param line_size number of bytes in a horizontal line of block | |
206 * @param h height | |
207 */ | |
1319 | 208 op_pixels_func avg_pixels_tab[4][4]; |
1136 | 209 |
210 /** | |
211 * Halfpel motion compensation with no rounding (a+b)>>1. | |
1225 | 212 * this is an array[2][4] of motion compensation funcions for 2 |
213 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> | |
1213 | 214 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 215 * @param block destination where the result is stored |
216 * @param pixels source | |
217 * @param line_size number of bytes in a horizontal line of block | |
218 * @param h height | |
219 */ | |
2075 | 220 op_pixels_func put_no_rnd_pixels_tab[4][4]; |
1136 | 221 |
222 /** | |
223 * Halfpel motion compensation with no rounding (a+b)>>1. | |
1225 | 224 * this is an array[2][4] of motion compensation funcions for 2 |
225 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> | |
1213 | 226 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 227 * @param block destination into which the result is averaged (a+b)>>1 |
228 * @param pixels source | |
229 * @param line_size number of bytes in a horizontal line of block | |
230 * @param h height | |
231 */ | |
2075 | 232 op_pixels_func avg_no_rnd_pixels_tab[4][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
|
233 |
1864 | 234 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); |
235 | |
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
|
236 /** |
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
|
237 * 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
|
238 * 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
|
239 * *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
|
240 * @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
|
241 * @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
|
242 * @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
|
243 * @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
|
244 */ |
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
|
245 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
1319 | 246 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
247 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
248 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
|
249 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
|
250 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
|
251 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; |
936 | 252 qpel_mc_func put_mspel_pixels_tab[8]; |
1168 | 253 |
254 /** | |
255 * h264 Chram MC | |
256 */ | |
257 h264_chroma_mc_func put_h264_chroma_pixels_tab[3]; | |
258 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
|
259 |
1168 | 260 qpel_mc_func put_h264_qpel_pixels_tab[3][16]; |
261 qpel_mc_func avg_h264_qpel_pixels_tab[3][16]; | |
262 | |
2415 | 263 h264_weight_func weight_h264_pixels_tab[10]; |
264 h264_biweight_func biweight_h264_pixels_tab[10]; | |
265 | |
1708 | 266 me_cmp_func pix_abs[2][4]; |
866 | 267 |
268 /* huffyuv specific */ | |
269 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); | |
936 | 270 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); |
1527 | 271 /** |
272 * subtract huffyuv's variant of median prediction | |
273 * note, this might read from src1[-1], src2[-1] | |
274 */ | |
275 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top); | |
1273 | 276 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w); |
2633 | 277 |
278 void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int *tc0); | |
279 void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int *tc0); | |
280 void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int *tc0); | |
281 void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int *tc0); | |
1092 | 282 |
1644 | 283 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale); |
284 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale); | |
285 | |
2045 | 286 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
|
287 |
1092 | 288 /* (I)DCT */ |
289 void (*fdct)(DCTELEM *block/* align 16*/); | |
1567 | 290 void (*fdct248)(DCTELEM *block/* align 16*/); |
1102 | 291 |
1324
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
292 /* IDCT really*/ |
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
293 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
|
294 |
1102 | 295 /** |
1104 | 296 * block -> idct -> clip to unsigned 8 bit -> dest. |
1102 | 297 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) |
1111 | 298 * @param line_size size in bytes of a horizotal line of dest |
1102 | 299 */ |
1092 | 300 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
1102 | 301 |
302 /** | |
303 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. | |
1111 | 304 * @param line_size size in bytes of a horizotal line of dest |
1102 | 305 */ |
1092 | 306 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
1102 | 307 |
308 /** | |
1104 | 309 * idct input permutation. |
1241 | 310 * several optimized IDCTs need a permutated input (relative to the normal order of the reference |
311 * IDCT) | |
312 * this permutation must be performed before the idct_put/add, note, normally this can be merged | |
313 * with the zigzag/alternate scan<br> | |
1102 | 314 * an example to avoid confusion: |
315 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...) | |
316 * - (x -> referece dct -> reference idct -> x) | |
317 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x) | |
318 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...) | |
319 */ | |
1092 | 320 uint8_t idct_permutation[64]; |
321 int idct_permutation_type; | |
322 #define FF_NO_IDCT_PERM 1 | |
323 #define FF_LIBMPEG2_IDCT_PERM 2 | |
324 #define FF_SIMPLE_IDCT_PERM 3 | |
325 #define FF_TRANSPOSE_IDCT_PERM 4 | |
326 | |
1784 | 327 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale); |
328 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); | |
329 #define BASIS_SHIFT 16 | |
330 #define RECON_SHIFT 6 | |
331 | |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
332 /** |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
333 * This function handles any initialization for the VP3 DSP functions. |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
334 */ |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
335 void (*vp3_dsp_init)(void); |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
336 |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
337 /** |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
338 * This function is responsible for taking a block of zigzag'd, |
1977 | 339 * quantized DCT coefficients and reconstructing the original block of |
340 * samples. | |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
341 * @param input_data 64 zigzag'd, quantized DCT coefficients |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
342 * @param dequant_matrix 64 zigzag'd quantizer coefficients |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
343 * @param coeff_count index of the last coefficient |
1977 | 344 * @param output_samples space for 64 DCTELEMs where the transformed |
345 * samples will be stored | |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
346 */ |
1977 | 347 void (*vp3_idct)(int16_t *input_data, int16_t *dequant_matrix, |
348 int coeff_count, DCTELEM *output_samples); | |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
349 |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
350 void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
351 } DSPContext; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
352 |
1201 | 353 void dsputil_static_init(void); |
1092 | 354 void dsputil_init(DSPContext* p, AVCodecContext *avctx); |
0 | 355 |
764 | 356 /** |
357 * permute block according to permuatation. | |
358 * @param last last non zero element in scantable order | |
359 */ | |
1064 | 360 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); |
34 | 361 |
1729 | 362 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type); |
363 | |
1264 | 364 #define BYTE_VEC32(c) ((c)*0x01010101UL) |
365 | |
366 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b) | |
367 { | |
368 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
369 } | |
370 | |
371 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b) | |
372 { | |
373 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
374 } | |
375 | |
2184 | 376 static inline int get_penalty_factor(int lambda, int lambda2, int type){ |
377 switch(type&0xFF){ | |
378 default: | |
379 case FF_CMP_SAD: | |
380 return lambda>>FF_LAMBDA_SHIFT; | |
381 case FF_CMP_DCT: | |
382 return (3*lambda)>>(FF_LAMBDA_SHIFT+1); | |
383 case FF_CMP_W53: | |
384 return (4*lambda)>>(FF_LAMBDA_SHIFT); | |
385 case FF_CMP_W97: | |
386 return (2*lambda)>>(FF_LAMBDA_SHIFT); | |
387 case FF_CMP_SATD: | |
388 return (2*lambda)>>FF_LAMBDA_SHIFT; | |
389 case FF_CMP_RD: | |
390 case FF_CMP_PSNR: | |
391 case FF_CMP_SSE: | |
392 case FF_CMP_NSSE: | |
393 return lambda2>>FF_LAMBDA_SHIFT; | |
394 case FF_CMP_BIT: | |
395 return 1; | |
396 } | |
397 } | |
398 | |
1102 | 399 /** |
1104 | 400 * Empty mmx state. |
1102 | 401 * this must be called between any dsp function and float/double code. |
402 * for example sin(); dsp->idct_put(); emms_c(); cos() | |
403 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
404 #define emms_c() |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
405 |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
406 /* should be defined by architectures supporting |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
407 one or more MultiMedia extension */ |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
408 int mm_support(void); |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
409 |
1974
8c5489b2cf3e
move __align16 some place where non-MMX machines can see it
melanson
parents:
1972
diff
changeset
|
410 #define __align16 __attribute__ ((aligned (16))) |
8c5489b2cf3e
move __align16 some place where non-MMX machines can see it
melanson
parents:
1972
diff
changeset
|
411 |
62 | 412 #if defined(HAVE_MMX) |
0 | 413 |
862 | 414 #undef emms_c |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
415 |
0 | 416 #define MM_MMX 0x0001 /* standard MMX */ |
417 #define MM_3DNOW 0x0004 /* AMD 3DNOW */ | |
418 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */ | |
419 #define MM_SSE 0x0008 /* SSE functions */ | |
420 #define MM_SSE2 0x0010 /* PIV SSE2 functions */ | |
2388 | 421 #define MM_3DNOWEXT 0x0020 /* AMD 3DNowExt */ |
0 | 422 |
423 extern int mm_flags; | |
424 | |
1064 | 425 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
426 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
|
427 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
0 | 428 |
429 static inline void emms(void) | |
430 { | |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
431 __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
|
432 } |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
433 |
936 | 434 |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
435 #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
|
436 {\ |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
437 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
|
438 emms();\ |
0 | 439 } |
440 | |
441 #define __align8 __attribute__ ((aligned (8))) | |
2324 | 442 #define STRIDE_ALIGN 8 |
0 | 443 |
1092 | 444 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); |
445 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx); | |
1065 | 446 |
62 | 447 #elif defined(ARCH_ARMV4L) |
448 | |
449 /* 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
|
450 line optimizations */ |
62 | 451 #define __align8 __attribute__ ((aligned (4))) |
2324 | 452 #define STRIDE_ALIGN 4 |
62 | 453 |
1092 | 454 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx); |
62 | 455 |
88 | 456 #elif defined(HAVE_MLIB) |
457 | |
458 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ | |
459 #define __align8 __attribute__ ((aligned (8))) | |
2324 | 460 #define STRIDE_ALIGN 8 |
88 | 461 |
1092 | 462 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx); |
88 | 463 |
1959
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
464 #elif defined(ARCH_SPARC) |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
465 |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
466 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
467 #define __align8 __attribute__ ((aligned (8))) |
2324 | 468 #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
|
469 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
|
470 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
471 #elif defined(ARCH_ALPHA) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
472 |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
473 #define __align8 __attribute__ ((aligned (8))) |
2324 | 474 #define STRIDE_ALIGN 8 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
475 |
1092 | 476 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
|
477 |
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
|
478 #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
|
479 |
894
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
480 #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
|
481 |
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
482 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
|
483 |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
484 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN) |
1653 | 485 #define pixel altivec_pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
486 #include <altivec.h> |
1653 | 487 #undef pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
488 #endif |
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
489 |
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
|
490 #define __align8 __attribute__ ((aligned (16))) |
2324 | 491 #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
|
492 |
1092 | 493 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
|
494 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
495 #elif defined(HAVE_MMI) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
496 |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
497 #define __align8 __attribute__ ((aligned (16))) |
2324 | 498 #define STRIDE_ALIGN 16 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
499 |
1092 | 500 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
|
501 |
1259
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
502 #elif defined(ARCH_SH4) |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
503 |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
504 #define __align8 __attribute__ ((aligned (8))) |
2324 | 505 #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
|
506 |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
507 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
|
508 |
0 | 509 #else |
510 | |
2324 | 511 #define __align8 __attribute__ ((aligned (8))) |
512 #define STRIDE_ALIGN 8 | |
0 | 513 |
514 #endif | |
515 | |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
516 #ifdef __GNUC__ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
517 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
518 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
|
519 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
|
520 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
|
521 |
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
|
522 #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
|
523 #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
|
524 #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
|
525 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
526 #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
|
527 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
528 #else /* __GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
529 |
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
|
530 #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
|
531 #define LD32(a) (*((uint32_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
532 #define LD64(a) (*((uint64_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
533 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
534 #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
|
535 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
536 #endif /* !__GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
537 |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
538 /* PSNR */ |
1064 | 539 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
|
540 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
|
541 AVCodecContext *avctx); |
781 | 542 |
543 /* FFT computation */ | |
544 | |
545 /* NOTE: soon integer code will be added, so you must use the | |
546 FFTSample type */ | |
547 typedef float FFTSample; | |
548 | |
549 typedef struct FFTComplex { | |
550 FFTSample re, im; | |
551 } FFTComplex; | |
552 | |
553 typedef struct FFTContext { | |
554 int nbits; | |
555 int inverse; | |
556 uint16_t *revtab; | |
557 FFTComplex *exptab; | |
558 FFTComplex *exptab1; /* only used by SSE code */ | |
559 void (*fft_calc)(struct FFTContext *s, FFTComplex *z); | |
560 } FFTContext; | |
561 | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
562 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
|
563 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
|
564 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
|
565 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z); |
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
566 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
|
567 |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
568 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z) |
781 | 569 { |
570 s->fft_calc(s, z); | |
571 } | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
572 void ff_fft_end(FFTContext *s); |
781 | 573 |
574 /* MDCT computation */ | |
575 | |
576 typedef struct MDCTContext { | |
577 int n; /* size of MDCT (i.e. number of input data * 2) */ | |
578 int nbits; /* n = 2^nbits */ | |
579 /* pre/post rotation tables */ | |
580 FFTSample *tcos; | |
581 FFTSample *tsin; | |
582 FFTContext fft; | |
583 } MDCTContext; | |
584 | |
794 | 585 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
|
586 void ff_imdct_calc(MDCTContext *s, FFTSample *output, |
781 | 587 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
|
588 void ff_mdct_calc(MDCTContext *s, FFTSample *out, |
781 | 589 const FFTSample *input, FFTSample *tmp); |
794 | 590 void ff_mdct_end(MDCTContext *s); |
781 | 591 |
1708 | 592 #define WARPER8_16(name8, name16)\ |
593 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
594 return name8(s, dst , src , stride, h)\ | |
595 +name8(s, dst+8 , src+8 , stride, h);\ | |
596 } | |
597 | |
598 #define WARPER8_16_SQ(name8, name16)\ | |
599 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
600 int score=0;\ | |
601 score +=name8(s, dst , src , stride, 8);\ | |
602 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
603 if(h==16){\ | |
604 dst += 8*stride;\ | |
605 src += 8*stride;\ | |
606 score +=name8(s, dst , src , stride, 8);\ | |
607 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
608 }\ | |
609 return score;\ | |
936 | 610 } |
611 | |
838
b78812db886f
lrintf detection (based upon a patch by Franois Revol <revol at free dot fr>)
michaelni
parents:
802
diff
changeset
|
612 #ifndef HAVE_LRINTF |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
613 /* 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
|
614 /* better than nothing implementation. */ |
802 | 615 /* btw, rintf() is existing on fbsd too -- alex */ |
2185 | 616 static always_inline long int lrintf(float x) |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
617 { |
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
|
618 #ifdef CONFIG_WIN32 |
2185 | 619 # ifdef ARCH_X86 |
620 int32_t i; | |
621 asm volatile( | |
622 "fistpl %0\n\t" | |
623 : "=m" (i) : "t" (x) : "st" | |
624 ); | |
625 return i; | |
626 # else | |
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
|
627 /* XXX: incorrect, but make it compile */ |
2185 | 628 return (int)(x + (x < 0 ? -0.5 : 0.5)); |
629 # endif | |
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
|
630 #else |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
631 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
|
632 #endif |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
633 } |
2024
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
1984
diff
changeset
|
634 #else |
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
1984
diff
changeset
|
635 #ifndef _ISOC9X_SOURCE |
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
1984
diff
changeset
|
636 #define _ISOC9X_SOURCE |
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
1984
diff
changeset
|
637 #endif |
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
1984
diff
changeset
|
638 #include <math.h> |
0 | 639 #endif |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
640 |
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
641 #endif |