Mercurial > libavcodec.hg
annotate dsputil.h @ 1893:779bdf5063ce libavcodec
1000000l (using uninitalized variable for initalizing bits per MV table)
author | michael |
---|---|
date | Tue, 16 Mar 2004 16:57:27 +0000 |
parents | dd63cb7e5080 |
children | 55b7435c59b8 |
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 |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
63 /* VP3 DSP functions */ |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
64 void vp3_dsp_init_c(void); |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
65 void vp3_idct_put_c(int16_t *input_data, int16_t *dequant_matrix, |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
66 int coeff_count, uint8_t *dest, int stride); |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
67 void vp3_idct_add_c(int16_t *input_data, int16_t *dequant_matrix, |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
68 int coeff_count, uint8_t *dest, int stride); |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
69 |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
70 void vp3_dsp_init_mmx(void); |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
71 void vp3_idct_put_mmx(int16_t *input_data, int16_t *dequant_matrix, |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
72 int coeff_count, uint8_t *dest, int stride); |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
73 void vp3_idct_add_mmx(int16_t *input_data, int16_t *dequant_matrix, |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
74 int coeff_count, uint8_t *dest, int stride); |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
75 |
0 | 76 |
675 | 77 /* minimum alignment rules ;) |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
78 if u notice errors in the align stuff, need more alignment for some asm code for some cpu |
675 | 79 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ... |
80 | |
81 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible) | |
82 i (michael) didnt check them, these are just the alignents which i think could be reached easily ... | |
83 | |
84 !future video codecs might need functions with less strict alignment | |
85 */ | |
0 | 86 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
87 /* |
1064 | 88 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size); |
89 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride); | |
90 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
91 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
296 | 92 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
|
93 */ |
0 | 94 |
95 /* add and put pixel (decoding) */ | |
675 | 96 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 |
1709 | 97 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4 |
1064 | 98 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
|
99 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 | 100 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); |
1168 | 101 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 | 102 |
984 | 103 #define DEF_OLD_QPEL(name)\ |
1064 | 104 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ |
105 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ | |
106 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
|
107 |
984 | 108 DEF_OLD_QPEL(qpel16_mc11_old_c) |
109 DEF_OLD_QPEL(qpel16_mc31_old_c) | |
110 DEF_OLD_QPEL(qpel16_mc12_old_c) | |
111 DEF_OLD_QPEL(qpel16_mc32_old_c) | |
112 DEF_OLD_QPEL(qpel16_mc13_old_c) | |
113 DEF_OLD_QPEL(qpel16_mc33_old_c) | |
114 DEF_OLD_QPEL(qpel8_mc11_old_c) | |
115 DEF_OLD_QPEL(qpel8_mc31_old_c) | |
116 DEF_OLD_QPEL(qpel8_mc12_old_c) | |
117 DEF_OLD_QPEL(qpel8_mc32_old_c) | |
118 DEF_OLD_QPEL(qpel8_mc13_old_c) | |
119 DEF_OLD_QPEL(qpel8_mc33_old_c) | |
651 | 120 |
121 #define CALL_2X_PIXELS(a, b, n)\ | |
122 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ | |
123 b(block , pixels , line_size, h);\ | |
124 b(block+n, pixels+n, line_size, h);\ | |
125 } | |
255 | 126 |
0 | 127 /* motion estimation */ |
1709 | 128 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2 |
129 // allthough currently h<4 is not used as functions with width <8 are not used and neither implemented | |
1708 | 130 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 | 131 |
1168 | 132 |
1102 | 133 /** |
134 * DSPContext. | |
135 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
136 typedef struct DSPContext { |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
137 /* pixel ops : interface with DCT */ |
1064 | 138 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); |
139 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); | |
140 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); | |
141 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); | |
1136 | 142 /** |
143 * translational global motion compensation. | |
144 */ | |
1064 | 145 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); |
1136 | 146 /** |
147 * global motion compensation. | |
148 */ | |
1064 | 149 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
|
150 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
|
151 void (*clear_blocks)(DCTELEM *blocks/*align 16*/); |
1064 | 152 int (*pix_sum)(uint8_t * pix, int line_size); |
153 int (*pix_norm1)(uint8_t * pix, int line_size); | |
1708 | 154 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4 |
155 | |
1729 | 156 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */ |
157 me_cmp_func sse[5]; | |
158 me_cmp_func hadamard8_diff[5]; | |
159 me_cmp_func dct_sad[5]; | |
160 me_cmp_func quant_psnr[5]; | |
161 me_cmp_func bit[5]; | |
162 me_cmp_func rd[5]; | |
163 me_cmp_func vsad[5]; | |
164 me_cmp_func vsse[5]; | |
936 | 165 |
1708 | 166 me_cmp_func me_pre_cmp[5]; |
167 me_cmp_func me_cmp[5]; | |
168 me_cmp_func me_sub_cmp[5]; | |
169 me_cmp_func mb_cmp[5]; | |
1729 | 170 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
|
171 |
1136 | 172 /** |
173 * 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
|
174 * this is an array[4][4] of motion compensation funcions for 4 |
1713 | 175 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 176 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 177 * @param block destination where the result is stored |
178 * @param pixels source | |
179 * @param line_size number of bytes in a horizontal line of block | |
180 * @param h height | |
181 */ | |
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
|
182 op_pixels_func put_pixels_tab[4][4]; |
1136 | 183 |
184 /** | |
185 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
1320 | 186 * This is an array[4][4] of motion compensation functions for 4 |
1713 | 187 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 188 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 189 * @param block destination into which the result is averaged (a+b+1)>>1 |
190 * @param pixels source | |
191 * @param line_size number of bytes in a horizontal line of block | |
192 * @param h height | |
193 */ | |
1319 | 194 op_pixels_func avg_pixels_tab[4][4]; |
1136 | 195 |
196 /** | |
197 * Halfpel motion compensation with no rounding (a+b)>>1. | |
1225 | 198 * this is an array[2][4] of motion compensation funcions for 2 |
199 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> | |
1213 | 200 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 201 * @param block destination where the result is stored |
202 * @param pixels source | |
203 * @param line_size number of bytes in a horizontal line of block | |
204 * @param h height | |
205 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
206 op_pixels_func put_no_rnd_pixels_tab[2][4]; |
1136 | 207 |
208 /** | |
209 * Halfpel motion compensation with no rounding (a+b)>>1. | |
1225 | 210 * this is an array[2][4] of motion compensation funcions for 2 |
211 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> | |
1213 | 212 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 213 * @param block destination into which the result is averaged (a+b)>>1 |
214 * @param pixels source | |
215 * @param line_size number of bytes in a horizontal line of block | |
216 * @param h height | |
217 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
218 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
|
219 |
1864 | 220 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); |
221 | |
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
|
222 /** |
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
|
223 * 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
|
224 * 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
|
225 * *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
|
226 * @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
|
227 * @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
|
228 * @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
|
229 * @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
|
230 */ |
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
|
231 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
1319 | 232 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
233 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
234 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
|
235 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
|
236 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
|
237 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; |
936 | 238 qpel_mc_func put_mspel_pixels_tab[8]; |
1168 | 239 |
240 /** | |
241 * h264 Chram MC | |
242 */ | |
243 h264_chroma_mc_func put_h264_chroma_pixels_tab[3]; | |
244 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
|
245 |
1168 | 246 qpel_mc_func put_h264_qpel_pixels_tab[3][16]; |
247 qpel_mc_func avg_h264_qpel_pixels_tab[3][16]; | |
248 | |
1708 | 249 me_cmp_func pix_abs[2][4]; |
866 | 250 |
251 /* huffyuv specific */ | |
252 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); | |
936 | 253 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); |
1527 | 254 /** |
255 * subtract huffyuv's variant of median prediction | |
256 * note, this might read from src1[-1], src2[-1] | |
257 */ | |
258 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top); | |
1273 | 259 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w); |
1092 | 260 |
1644 | 261 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale); |
262 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale); | |
263 | |
1092 | 264 /* (I)DCT */ |
265 void (*fdct)(DCTELEM *block/* align 16*/); | |
1567 | 266 void (*fdct248)(DCTELEM *block/* align 16*/); |
1102 | 267 |
1324
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
268 /* IDCT really*/ |
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
269 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
|
270 |
1102 | 271 /** |
1104 | 272 * block -> idct -> clip to unsigned 8 bit -> dest. |
1102 | 273 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) |
1111 | 274 * @param line_size size in bytes of a horizotal line of dest |
1102 | 275 */ |
1092 | 276 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
1102 | 277 |
278 /** | |
279 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. | |
1111 | 280 * @param line_size size in bytes of a horizotal line of dest |
1102 | 281 */ |
1092 | 282 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
1102 | 283 |
284 /** | |
1104 | 285 * idct input permutation. |
1241 | 286 * several optimized IDCTs need a permutated input (relative to the normal order of the reference |
287 * IDCT) | |
288 * this permutation must be performed before the idct_put/add, note, normally this can be merged | |
289 * with the zigzag/alternate scan<br> | |
1102 | 290 * an example to avoid confusion: |
291 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...) | |
292 * - (x -> referece dct -> reference idct -> x) | |
293 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x) | |
294 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...) | |
295 */ | |
1092 | 296 uint8_t idct_permutation[64]; |
297 int idct_permutation_type; | |
298 #define FF_NO_IDCT_PERM 1 | |
299 #define FF_LIBMPEG2_IDCT_PERM 2 | |
300 #define FF_SIMPLE_IDCT_PERM 3 | |
301 #define FF_TRANSPOSE_IDCT_PERM 4 | |
302 | |
1784 | 303 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale); |
304 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); | |
305 #define BASIS_SHIFT 16 | |
306 #define RECON_SHIFT 6 | |
307 | |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
308 /** |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
309 * 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
|
310 */ |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
311 void (*vp3_dsp_init)(void); |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
312 |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
313 /** |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
314 * This function is responsible for taking a block of zigzag'd, |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
315 * quantized DCT coefficients, reconstructing the original block of |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
316 * samples, and placing it into the output. |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
317 * @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
|
318 * @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
|
319 * @param coeff_count index of the last coefficient |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
320 * @param dest the final output location where the transformed samples |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
321 * are to be placed |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
322 * @param stride the width in 8-bit samples of a line on this plane |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
323 */ |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
324 void (*vp3_idct_put)(int16_t *input_data, int16_t *dequant_matrix, |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
325 int coeff_count, uint8_t *dest, int stride); |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
326 |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
327 /** |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
328 * This function is responsible for taking a block of zigzag'd, |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
329 * quantized DCT coefficients, reconstructing the original block of |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
330 * samples, and adding the transformed samples to an existing block of |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
331 * samples in the output. |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
332 * @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
|
333 * @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
|
334 * @param coeff_count index of the last coefficient |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
335 * @param dest the final output location where the transformed samples |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
336 * are to be placed |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
337 * @param stride the width in 8-bit samples of a line on this plane |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
338 */ |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
339 void (*vp3_idct_add)(int16_t *input_data, int16_t *dequant_matrix, |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
340 int coeff_count, uint8_t *dest, int stride); |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
341 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
342 } DSPContext; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
343 |
1201 | 344 void dsputil_static_init(void); |
1092 | 345 void dsputil_init(DSPContext* p, AVCodecContext *avctx); |
0 | 346 |
764 | 347 /** |
348 * permute block according to permuatation. | |
349 * @param last last non zero element in scantable order | |
350 */ | |
1064 | 351 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); |
34 | 352 |
1729 | 353 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type); |
354 | |
1264 | 355 #define BYTE_VEC32(c) ((c)*0x01010101UL) |
356 | |
357 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b) | |
358 { | |
359 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
360 } | |
361 | |
362 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b) | |
363 { | |
364 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
365 } | |
366 | |
1102 | 367 /** |
1104 | 368 * Empty mmx state. |
1102 | 369 * this must be called between any dsp function and float/double code. |
370 * for example sin(); dsp->idct_put(); emms_c(); cos() | |
371 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
372 #define emms_c() |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
373 |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
374 /* should be defined by architectures supporting |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
375 one or more MultiMedia extension */ |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
376 int mm_support(void); |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
377 |
62 | 378 #if defined(HAVE_MMX) |
0 | 379 |
862 | 380 #undef emms_c |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
381 |
0 | 382 #define MM_MMX 0x0001 /* standard MMX */ |
383 #define MM_3DNOW 0x0004 /* AMD 3DNOW */ | |
384 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */ | |
385 #define MM_SSE 0x0008 /* SSE functions */ | |
386 #define MM_SSE2 0x0010 /* PIV SSE2 functions */ | |
387 | |
388 extern int mm_flags; | |
389 | |
1064 | 390 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
391 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); | |
0 | 392 |
393 static inline void emms(void) | |
394 { | |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
395 __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
|
396 } |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
397 |
936 | 398 |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
399 #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
|
400 {\ |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
401 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
|
402 emms();\ |
0 | 403 } |
404 | |
405 #define __align8 __attribute__ ((aligned (8))) | |
406 | |
1092 | 407 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); |
408 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx); | |
1065 | 409 |
62 | 410 #elif defined(ARCH_ARMV4L) |
411 | |
412 /* This is to use 4 bytes read to the IDCT pointers for some 'zero' | |
413 line ptimizations */ | |
414 #define __align8 __attribute__ ((aligned (4))) | |
415 | |
1092 | 416 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx); |
62 | 417 |
88 | 418 #elif defined(HAVE_MLIB) |
419 | |
420 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ | |
421 #define __align8 __attribute__ ((aligned (8))) | |
422 | |
1092 | 423 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx); |
88 | 424 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
425 #elif defined(ARCH_ALPHA) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
426 |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
427 #define __align8 __attribute__ ((aligned (8))) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
428 |
1092 | 429 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
|
430 |
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
|
431 #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
|
432 |
894
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
433 #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
|
434 |
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
435 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
|
436 |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
437 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN) |
1653 | 438 #define pixel altivec_pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
439 #include <altivec.h> |
1653 | 440 #undef pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
441 #endif |
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
442 |
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
|
443 #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
|
444 |
1092 | 445 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
|
446 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
447 #elif defined(HAVE_MMI) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
448 |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
449 #define __align8 __attribute__ ((aligned (16))) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
450 |
1092 | 451 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
|
452 |
1259
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
453 #elif defined(ARCH_SH4) |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
454 |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
455 #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
|
456 |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
457 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
|
458 |
0 | 459 #else |
460 | |
461 #define __align8 | |
462 | |
463 #endif | |
464 | |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
465 #ifdef __GNUC__ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
466 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
467 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
|
468 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
|
469 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
|
470 |
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
|
471 #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
|
472 #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
|
473 #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
|
474 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
475 #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
|
476 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
477 #else /* __GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
478 |
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
|
479 #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
|
480 #define LD32(a) (*((uint32_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
481 #define LD64(a) (*((uint64_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
482 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
483 #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
|
484 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
485 #endif /* !__GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
486 |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
487 /* PSNR */ |
1064 | 488 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
|
489 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
|
490 AVCodecContext *avctx); |
781 | 491 |
492 /* FFT computation */ | |
493 | |
494 /* NOTE: soon integer code will be added, so you must use the | |
495 FFTSample type */ | |
496 typedef float FFTSample; | |
497 | |
498 typedef struct FFTComplex { | |
499 FFTSample re, im; | |
500 } FFTComplex; | |
501 | |
502 typedef struct FFTContext { | |
503 int nbits; | |
504 int inverse; | |
505 uint16_t *revtab; | |
506 FFTComplex *exptab; | |
507 FFTComplex *exptab1; /* only used by SSE code */ | |
508 void (*fft_calc)(struct FFTContext *s, FFTComplex *z); | |
509 } FFTContext; | |
510 | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
511 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
|
512 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
|
513 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
|
514 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
|
515 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
|
516 |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
517 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z) |
781 | 518 { |
519 s->fft_calc(s, z); | |
520 } | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
521 void ff_fft_end(FFTContext *s); |
781 | 522 |
523 /* MDCT computation */ | |
524 | |
525 typedef struct MDCTContext { | |
526 int n; /* size of MDCT (i.e. number of input data * 2) */ | |
527 int nbits; /* n = 2^nbits */ | |
528 /* pre/post rotation tables */ | |
529 FFTSample *tcos; | |
530 FFTSample *tsin; | |
531 FFTContext fft; | |
532 } MDCTContext; | |
533 | |
794 | 534 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
|
535 void ff_imdct_calc(MDCTContext *s, FFTSample *output, |
781 | 536 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
|
537 void ff_mdct_calc(MDCTContext *s, FFTSample *out, |
781 | 538 const FFTSample *input, FFTSample *tmp); |
794 | 539 void ff_mdct_end(MDCTContext *s); |
781 | 540 |
1708 | 541 #define WARPER8_16(name8, name16)\ |
542 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
543 return name8(s, dst , src , stride, h)\ | |
544 +name8(s, dst+8 , src+8 , stride, h);\ | |
545 } | |
546 | |
547 #define WARPER8_16_SQ(name8, name16)\ | |
548 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
549 int score=0;\ | |
550 score +=name8(s, dst , src , stride, 8);\ | |
551 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
552 if(h==16){\ | |
553 dst += 8*stride;\ | |
554 src += 8*stride;\ | |
555 score +=name8(s, dst , src , stride, 8);\ | |
556 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
557 }\ | |
558 return score;\ | |
936 | 559 } |
560 | |
838
b78812db886f
lrintf detection (based upon a patch by Franois Revol <revol at free dot fr>)
michaelni
parents:
802
diff
changeset
|
561 #ifndef HAVE_LRINTF |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
562 /* 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
|
563 /* better than nothing implementation. */ |
802 | 564 /* 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
|
565 static inline long int lrintf(float x) |
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
566 { |
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
|
567 #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
|
568 /* 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
|
569 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
|
570 #else |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
571 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
|
572 #endif |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
573 } |
0 | 574 #endif |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
575 |
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
576 #endif |