Mercurial > libavcodec.hg
annotate dsputil.h @ 1972:8a556283601d libavcodec
hook up support for SSE2-optimized VP3 IDCT
author | melanson |
---|---|
date | Sun, 25 Apr 2004 03:33:30 +0000 |
parents | 55b7435c59b8 |
children | 8c5489b2cf3e |
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 |
1972 | 76 void vp3_dsp_init_sse2(void); |
77 void vp3_idct_put_sse2(int16_t *input_data, int16_t *dequant_matrix, | |
78 int coeff_count, uint8_t *dest, int stride); | |
79 void vp3_idct_add_sse2(int16_t *input_data, int16_t *dequant_matrix, | |
80 int coeff_count, uint8_t *dest, int stride); | |
81 | |
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); |
0 | 108 |
984 | 109 #define DEF_OLD_QPEL(name)\ |
1064 | 110 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ |
111 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ | |
112 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
|
113 |
984 | 114 DEF_OLD_QPEL(qpel16_mc11_old_c) |
115 DEF_OLD_QPEL(qpel16_mc31_old_c) | |
116 DEF_OLD_QPEL(qpel16_mc12_old_c) | |
117 DEF_OLD_QPEL(qpel16_mc32_old_c) | |
118 DEF_OLD_QPEL(qpel16_mc13_old_c) | |
119 DEF_OLD_QPEL(qpel16_mc33_old_c) | |
120 DEF_OLD_QPEL(qpel8_mc11_old_c) | |
121 DEF_OLD_QPEL(qpel8_mc31_old_c) | |
122 DEF_OLD_QPEL(qpel8_mc12_old_c) | |
123 DEF_OLD_QPEL(qpel8_mc32_old_c) | |
124 DEF_OLD_QPEL(qpel8_mc13_old_c) | |
125 DEF_OLD_QPEL(qpel8_mc33_old_c) | |
651 | 126 |
127 #define CALL_2X_PIXELS(a, b, n)\ | |
128 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ | |
129 b(block , pixels , line_size, h);\ | |
130 b(block+n, pixels+n, line_size, h);\ | |
131 } | |
255 | 132 |
0 | 133 /* motion estimation */ |
1709 | 134 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2 |
135 // allthough currently h<4 is not used as functions with width <8 are not used and neither implemented | |
1708 | 136 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 | 137 |
1168 | 138 |
1102 | 139 /** |
140 * DSPContext. | |
141 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
142 typedef struct DSPContext { |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
143 /* pixel ops : interface with DCT */ |
1064 | 144 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); |
145 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); | |
146 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); | |
147 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); | |
1136 | 148 /** |
149 * translational global motion compensation. | |
150 */ | |
1064 | 151 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); |
1136 | 152 /** |
153 * global motion compensation. | |
154 */ | |
1064 | 155 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
|
156 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
|
157 void (*clear_blocks)(DCTELEM *blocks/*align 16*/); |
1064 | 158 int (*pix_sum)(uint8_t * pix, int line_size); |
159 int (*pix_norm1)(uint8_t * pix, int line_size); | |
1708 | 160 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4 |
161 | |
1729 | 162 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */ |
163 me_cmp_func sse[5]; | |
164 me_cmp_func hadamard8_diff[5]; | |
165 me_cmp_func dct_sad[5]; | |
166 me_cmp_func quant_psnr[5]; | |
167 me_cmp_func bit[5]; | |
168 me_cmp_func rd[5]; | |
169 me_cmp_func vsad[5]; | |
170 me_cmp_func vsse[5]; | |
936 | 171 |
1708 | 172 me_cmp_func me_pre_cmp[5]; |
173 me_cmp_func me_cmp[5]; | |
174 me_cmp_func me_sub_cmp[5]; | |
175 me_cmp_func mb_cmp[5]; | |
1729 | 176 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
|
177 |
1136 | 178 /** |
179 * 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
|
180 * this is an array[4][4] of motion compensation funcions for 4 |
1713 | 181 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 182 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 183 * @param block destination where the result is stored |
184 * @param pixels source | |
185 * @param line_size number of bytes in a horizontal line of block | |
186 * @param h height | |
187 */ | |
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 op_pixels_func put_pixels_tab[4][4]; |
1136 | 189 |
190 /** | |
191 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
1320 | 192 * This is an array[4][4] of motion compensation functions for 4 |
1713 | 193 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 194 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 195 * @param block destination into which the result is averaged (a+b+1)>>1 |
196 * @param pixels source | |
197 * @param line_size number of bytes in a horizontal line of block | |
198 * @param h height | |
199 */ | |
1319 | 200 op_pixels_func avg_pixels_tab[4][4]; |
1136 | 201 |
202 /** | |
203 * Halfpel motion compensation with no rounding (a+b)>>1. | |
1225 | 204 * this is an array[2][4] of motion compensation funcions for 2 |
205 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> | |
1213 | 206 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 207 * @param block destination where the result is stored |
208 * @param pixels source | |
209 * @param line_size number of bytes in a horizontal line of block | |
210 * @param h height | |
211 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
212 op_pixels_func put_no_rnd_pixels_tab[2][4]; |
1136 | 213 |
214 /** | |
215 * Halfpel motion compensation with no rounding (a+b)>>1. | |
1225 | 216 * this is an array[2][4] of motion compensation funcions for 2 |
217 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> | |
1213 | 218 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 219 * @param block destination into which the result is averaged (a+b)>>1 |
220 * @param pixels source | |
221 * @param line_size number of bytes in a horizontal line of block | |
222 * @param h height | |
223 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
224 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
|
225 |
1864 | 226 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); |
227 | |
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
|
228 /** |
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 * 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
|
230 * 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
|
231 * *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
|
232 * @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
|
233 * @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
|
234 * @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
|
235 * @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
|
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 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
1319 | 238 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
239 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
240 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
|
241 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
|
242 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
|
243 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; |
936 | 244 qpel_mc_func put_mspel_pixels_tab[8]; |
1168 | 245 |
246 /** | |
247 * h264 Chram MC | |
248 */ | |
249 h264_chroma_mc_func put_h264_chroma_pixels_tab[3]; | |
250 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
|
251 |
1168 | 252 qpel_mc_func put_h264_qpel_pixels_tab[3][16]; |
253 qpel_mc_func avg_h264_qpel_pixels_tab[3][16]; | |
254 | |
1708 | 255 me_cmp_func pix_abs[2][4]; |
866 | 256 |
257 /* huffyuv specific */ | |
258 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); | |
936 | 259 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); |
1527 | 260 /** |
261 * subtract huffyuv's variant of median prediction | |
262 * note, this might read from src1[-1], src2[-1] | |
263 */ | |
264 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top); | |
1273 | 265 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w); |
1092 | 266 |
1644 | 267 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale); |
268 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale); | |
269 | |
1092 | 270 /* (I)DCT */ |
271 void (*fdct)(DCTELEM *block/* align 16*/); | |
1567 | 272 void (*fdct248)(DCTELEM *block/* align 16*/); |
1102 | 273 |
1324
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
274 /* IDCT really*/ |
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
275 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
|
276 |
1102 | 277 /** |
1104 | 278 * block -> idct -> clip to unsigned 8 bit -> dest. |
1102 | 279 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) |
1111 | 280 * @param line_size size in bytes of a horizotal line of dest |
1102 | 281 */ |
1092 | 282 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
1102 | 283 |
284 /** | |
285 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. | |
1111 | 286 * @param line_size size in bytes of a horizotal line of dest |
1102 | 287 */ |
1092 | 288 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
1102 | 289 |
290 /** | |
1104 | 291 * idct input permutation. |
1241 | 292 * several optimized IDCTs need a permutated input (relative to the normal order of the reference |
293 * IDCT) | |
294 * this permutation must be performed before the idct_put/add, note, normally this can be merged | |
295 * with the zigzag/alternate scan<br> | |
1102 | 296 * an example to avoid confusion: |
297 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...) | |
298 * - (x -> referece dct -> reference idct -> x) | |
299 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x) | |
300 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...) | |
301 */ | |
1092 | 302 uint8_t idct_permutation[64]; |
303 int idct_permutation_type; | |
304 #define FF_NO_IDCT_PERM 1 | |
305 #define FF_LIBMPEG2_IDCT_PERM 2 | |
306 #define FF_SIMPLE_IDCT_PERM 3 | |
307 #define FF_TRANSPOSE_IDCT_PERM 4 | |
308 | |
1784 | 309 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale); |
310 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); | |
311 #define BASIS_SHIFT 16 | |
312 #define RECON_SHIFT 6 | |
313 | |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
314 /** |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
315 * 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
|
316 */ |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
317 void (*vp3_dsp_init)(void); |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
318 |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
319 /** |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
320 * 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
|
321 * 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
|
322 * 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
|
323 * @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
|
324 * @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
|
325 * @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
|
326 * @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
|
327 * are to be placed |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
328 * @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
|
329 */ |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
330 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
|
331 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
|
332 |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
333 /** |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
334 * 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
|
335 * 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
|
336 * 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
|
337 * samples in the output. |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
338 * @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
|
339 * @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
|
340 * @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
|
341 * @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
|
342 * are to be placed |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
343 * @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
|
344 */ |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
345 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
|
346 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
|
347 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
348 } DSPContext; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
349 |
1201 | 350 void dsputil_static_init(void); |
1092 | 351 void dsputil_init(DSPContext* p, AVCodecContext *avctx); |
0 | 352 |
764 | 353 /** |
354 * permute block according to permuatation. | |
355 * @param last last non zero element in scantable order | |
356 */ | |
1064 | 357 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); |
34 | 358 |
1729 | 359 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type); |
360 | |
1264 | 361 #define BYTE_VEC32(c) ((c)*0x01010101UL) |
362 | |
363 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b) | |
364 { | |
365 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
366 } | |
367 | |
368 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b) | |
369 { | |
370 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
371 } | |
372 | |
1102 | 373 /** |
1104 | 374 * Empty mmx state. |
1102 | 375 * this must be called between any dsp function and float/double code. |
376 * for example sin(); dsp->idct_put(); emms_c(); cos() | |
377 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
378 #define emms_c() |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
379 |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
380 /* should be defined by architectures supporting |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
381 one or more MultiMedia extension */ |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
382 int mm_support(void); |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
383 |
62 | 384 #if defined(HAVE_MMX) |
0 | 385 |
862 | 386 #undef emms_c |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
387 |
0 | 388 #define MM_MMX 0x0001 /* standard MMX */ |
389 #define MM_3DNOW 0x0004 /* AMD 3DNOW */ | |
390 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */ | |
391 #define MM_SSE 0x0008 /* SSE functions */ | |
392 #define MM_SSE2 0x0010 /* PIV SSE2 functions */ | |
393 | |
394 extern int mm_flags; | |
395 | |
1064 | 396 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
397 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); | |
0 | 398 |
399 static inline void emms(void) | |
400 { | |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
401 __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
|
402 } |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
403 |
936 | 404 |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
405 #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
|
406 {\ |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
407 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
|
408 emms();\ |
0 | 409 } |
410 | |
411 #define __align8 __attribute__ ((aligned (8))) | |
1972 | 412 #define __align16 __attribute__ ((aligned (16))) |
0 | 413 |
1092 | 414 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); |
415 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx); | |
1065 | 416 |
62 | 417 #elif defined(ARCH_ARMV4L) |
418 | |
419 /* This is to use 4 bytes read to the IDCT pointers for some 'zero' | |
420 line ptimizations */ | |
421 #define __align8 __attribute__ ((aligned (4))) | |
422 | |
1092 | 423 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx); |
62 | 424 |
88 | 425 #elif defined(HAVE_MLIB) |
426 | |
427 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ | |
428 #define __align8 __attribute__ ((aligned (8))) | |
429 | |
1092 | 430 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx); |
88 | 431 |
1959
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
432 #elif defined(ARCH_SPARC) |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
433 |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
434 /* 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
|
435 #define __align8 __attribute__ ((aligned (8))) |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
436 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
|
437 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
438 #elif defined(ARCH_ALPHA) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
439 |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
440 #define __align8 __attribute__ ((aligned (8))) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
441 |
1092 | 442 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
|
443 |
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
|
444 #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
|
445 |
894
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
446 #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
|
447 |
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
448 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
|
449 |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
450 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN) |
1653 | 451 #define pixel altivec_pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
452 #include <altivec.h> |
1653 | 453 #undef pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
454 #endif |
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
455 |
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
|
456 #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
|
457 |
1092 | 458 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
|
459 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
460 #elif defined(HAVE_MMI) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
461 |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
462 #define __align8 __attribute__ ((aligned (16))) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
463 |
1092 | 464 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
|
465 |
1259
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
466 #elif defined(ARCH_SH4) |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
467 |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
468 #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
|
469 |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
470 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
|
471 |
0 | 472 #else |
473 | |
474 #define __align8 | |
475 | |
476 #endif | |
477 | |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
478 #ifdef __GNUC__ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
479 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
480 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
|
481 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
|
482 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
|
483 |
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
|
484 #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
|
485 #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
|
486 #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
|
487 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
488 #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
|
489 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
490 #else /* __GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
491 |
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
|
492 #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
|
493 #define LD32(a) (*((uint32_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
494 #define LD64(a) (*((uint64_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
495 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
496 #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
|
497 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
498 #endif /* !__GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
499 |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
500 /* PSNR */ |
1064 | 501 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
|
502 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
|
503 AVCodecContext *avctx); |
781 | 504 |
505 /* FFT computation */ | |
506 | |
507 /* NOTE: soon integer code will be added, so you must use the | |
508 FFTSample type */ | |
509 typedef float FFTSample; | |
510 | |
511 typedef struct FFTComplex { | |
512 FFTSample re, im; | |
513 } FFTComplex; | |
514 | |
515 typedef struct FFTContext { | |
516 int nbits; | |
517 int inverse; | |
518 uint16_t *revtab; | |
519 FFTComplex *exptab; | |
520 FFTComplex *exptab1; /* only used by SSE code */ | |
521 void (*fft_calc)(struct FFTContext *s, FFTComplex *z); | |
522 } FFTContext; | |
523 | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
524 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
|
525 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
|
526 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
|
527 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
|
528 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
|
529 |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
530 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z) |
781 | 531 { |
532 s->fft_calc(s, z); | |
533 } | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
534 void ff_fft_end(FFTContext *s); |
781 | 535 |
536 /* MDCT computation */ | |
537 | |
538 typedef struct MDCTContext { | |
539 int n; /* size of MDCT (i.e. number of input data * 2) */ | |
540 int nbits; /* n = 2^nbits */ | |
541 /* pre/post rotation tables */ | |
542 FFTSample *tcos; | |
543 FFTSample *tsin; | |
544 FFTContext fft; | |
545 } MDCTContext; | |
546 | |
794 | 547 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
|
548 void ff_imdct_calc(MDCTContext *s, FFTSample *output, |
781 | 549 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
|
550 void ff_mdct_calc(MDCTContext *s, FFTSample *out, |
781 | 551 const FFTSample *input, FFTSample *tmp); |
794 | 552 void ff_mdct_end(MDCTContext *s); |
781 | 553 |
1708 | 554 #define WARPER8_16(name8, name16)\ |
555 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
556 return name8(s, dst , src , stride, h)\ | |
557 +name8(s, dst+8 , src+8 , stride, h);\ | |
558 } | |
559 | |
560 #define WARPER8_16_SQ(name8, name16)\ | |
561 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
562 int score=0;\ | |
563 score +=name8(s, dst , src , stride, 8);\ | |
564 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
565 if(h==16){\ | |
566 dst += 8*stride;\ | |
567 src += 8*stride;\ | |
568 score +=name8(s, dst , src , stride, 8);\ | |
569 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
570 }\ | |
571 return score;\ | |
936 | 572 } |
573 | |
838
b78812db886f
lrintf detection (based upon a patch by Franois Revol <revol at free dot fr>)
michaelni
parents:
802
diff
changeset
|
574 #ifndef HAVE_LRINTF |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
575 /* 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
|
576 /* better than nothing implementation. */ |
802 | 577 /* 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
|
578 static inline long int lrintf(float x) |
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
579 { |
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
|
580 #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
|
581 /* 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
|
582 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
|
583 #else |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
584 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
|
585 #endif |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
586 } |
0 | 587 #endif |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
588 |
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
589 #endif |