Mercurial > libavcodec.hg
annotate dsputil.h @ 2973:bb0222145083 libavcodec
forgot
author | michael |
---|---|
date | Mon, 19 Dec 2005 03:08:08 +0000 |
parents | ef2149182f1c |
children | bfabfdf9ce55 |
rev | line source |
---|---|
429 | 1 /* |
2 * DSP utils | |
3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard. | |
1739
07a484280a82
copyright year update of the files i touched and remembered, things look annoyingly unmaintained otherwise
michael
parents:
1729
diff
changeset
|
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> |
429 | 5 * |
6 * This library is free software; you can redistribute it and/or | |
7 * modify it under the terms of the GNU Lesser General Public | |
8 * License as published by the Free Software Foundation; either | |
9 * version 2 of the License, or (at your option) any later version. | |
10 * | |
11 * This library is distributed in the hope that it will be useful, | |
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
14 * Lesser General Public License for more details. | |
15 * | |
16 * You should have received a copy of the GNU Lesser General Public | |
17 * License along with this library; if not, write to the Free Software | |
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
19 */ | |
1102 | 20 |
21 /** | |
22 * @file dsputil.h | |
1106 | 23 * DSP utils. |
1213 | 24 * note, many functions in here may use MMX which trashes the FPU state, it is |
25 * absolutely necessary to call emms_c() between dsp & float/double code | |
1102 | 26 */ |
27 | |
0 | 28 #ifndef DSPUTIL_H |
29 #define DSPUTIL_H | |
30 | |
31 #include "common.h" | |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
32 #include "avcodec.h" |
0 | 33 |
1102 | 34 |
255 | 35 //#define DEBUG |
0 | 36 /* dct code */ |
37 typedef short DCTELEM; | |
38 | |
474
11dbd00682fc
avoid name clash with libjpeg - added missing externs
bellard
parents:
429
diff
changeset
|
39 void fdct_ifast (DCTELEM *data); |
1571
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1567
diff
changeset
|
40 void fdct_ifast248 (DCTELEM *data); |
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
623
diff
changeset
|
41 void ff_jpeg_fdct_islow (DCTELEM *data); |
1567 | 42 void ff_fdct248_islow (DCTELEM *data); |
0 | 43 |
44 void j_rev_dct (DCTELEM *data); | |
2256 | 45 void j_rev_dct4 (DCTELEM *data); |
2257 | 46 void j_rev_dct2 (DCTELEM *data); |
2259 | 47 void j_rev_dct1 (DCTELEM *data); |
0 | 48 |
687
9abb13c21fbe
fdct_mmx -> ff_fdct_mmx (renamed to avoid namespace conflict with xvid)
arpi_esp
parents:
675
diff
changeset
|
49 void ff_fdct_mmx(DCTELEM *block); |
1565 | 50 void ff_fdct_mmx2(DCTELEM *block); |
1765
e31754bc5b65
SSE2 fdct by (Balatoni Denes <pnis at coder dot hu>)
michael
parents:
1739
diff
changeset
|
51 void ff_fdct_sse2(DCTELEM *block); |
0 | 52 |
2755 | 53 void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride); |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
54 void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride); |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
55 void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block); |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
56 void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block); |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
57 |
34 | 58 /* encoding scans */ |
1064 | 59 extern const uint8_t ff_alternate_horizontal_scan[64]; |
60 extern const uint8_t ff_alternate_vertical_scan[64]; | |
61 extern const uint8_t ff_zigzag_direct[64]; | |
1567 | 62 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
|
63 |
0 | 64 /* pixel operations */ |
2090 | 65 #define MAX_NEG_CROP 1024 |
0 | 66 |
67 /* temporary */ | |
1064 | 68 extern uint32_t squareTbl[512]; |
69 extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP]; | |
0 | 70 |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
71 /* VP3 DSP functions */ |
2693 | 72 void ff_vp3_idct_c(DCTELEM *block/* align 16*/); |
73 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); | |
74 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); | |
0 | 75 |
675 | 76 /* minimum alignment rules ;) |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
77 if u notice errors in the align stuff, need more alignment for some asm code for some cpu |
675 | 78 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ... |
79 | |
80 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible) | |
81 i (michael) didnt check them, these are just the alignents which i think could be reached easily ... | |
82 | |
83 !future video codecs might need functions with less strict alignment | |
84 */ | |
0 | 85 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
86 /* |
1064 | 87 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size); |
88 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride); | |
89 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
90 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
296 | 91 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
|
92 */ |
0 | 93 |
94 /* add and put pixel (decoding) */ | |
675 | 95 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 |
1709 | 96 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4 |
1064 | 97 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
|
98 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 | 99 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); |
1168 | 100 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y); |
2415 | 101 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset); |
102 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offsetd, int offsets); | |
0 | 103 |
984 | 104 #define DEF_OLD_QPEL(name)\ |
1064 | 105 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ |
106 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ | |
107 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
|
108 |
984 | 109 DEF_OLD_QPEL(qpel16_mc11_old_c) |
110 DEF_OLD_QPEL(qpel16_mc31_old_c) | |
111 DEF_OLD_QPEL(qpel16_mc12_old_c) | |
112 DEF_OLD_QPEL(qpel16_mc32_old_c) | |
113 DEF_OLD_QPEL(qpel16_mc13_old_c) | |
114 DEF_OLD_QPEL(qpel16_mc33_old_c) | |
115 DEF_OLD_QPEL(qpel8_mc11_old_c) | |
116 DEF_OLD_QPEL(qpel8_mc31_old_c) | |
117 DEF_OLD_QPEL(qpel8_mc12_old_c) | |
118 DEF_OLD_QPEL(qpel8_mc32_old_c) | |
119 DEF_OLD_QPEL(qpel8_mc13_old_c) | |
120 DEF_OLD_QPEL(qpel8_mc33_old_c) | |
651 | 121 |
122 #define CALL_2X_PIXELS(a, b, n)\ | |
123 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ | |
124 b(block , pixels , line_size, h);\ | |
125 b(block+n, pixels+n, line_size, h);\ | |
126 } | |
255 | 127 |
0 | 128 /* motion estimation */ |
1709 | 129 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2 |
130 // allthough currently h<4 is not used as functions with width <8 are not used and neither implemented | |
1708 | 131 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 | 132 |
1168 | 133 |
1102 | 134 /** |
135 * DSPContext. | |
136 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
137 typedef struct DSPContext { |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
138 /* pixel ops : interface with DCT */ |
1064 | 139 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); |
140 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); | |
141 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); | |
1984
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
142 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
1064 | 143 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
2763 | 144 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size); |
145 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size); | |
1136 | 146 /** |
147 * translational global motion compensation. | |
148 */ | |
1064 | 149 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); |
1136 | 150 /** |
151 * global motion compensation. | |
152 */ | |
1064 | 153 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
|
154 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
|
155 void (*clear_blocks)(DCTELEM *blocks/*align 16*/); |
1064 | 156 int (*pix_sum)(uint8_t * pix, int line_size); |
157 int (*pix_norm1)(uint8_t * pix, int line_size); | |
1708 | 158 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4 |
2967 | 159 |
1729 | 160 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */ |
161 me_cmp_func sse[5]; | |
162 me_cmp_func hadamard8_diff[5]; | |
163 me_cmp_func dct_sad[5]; | |
164 me_cmp_func quant_psnr[5]; | |
165 me_cmp_func bit[5]; | |
166 me_cmp_func rd[5]; | |
167 me_cmp_func vsad[5]; | |
168 me_cmp_func vsse[5]; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
169 me_cmp_func nsse[5]; |
2184 | 170 me_cmp_func w53[5]; |
171 me_cmp_func w97[5]; | |
2382 | 172 me_cmp_func dct_max[5]; |
936 | 173 |
1708 | 174 me_cmp_func me_pre_cmp[5]; |
175 me_cmp_func me_cmp[5]; | |
176 me_cmp_func me_sub_cmp[5]; | |
177 me_cmp_func mb_cmp[5]; | |
1729 | 178 me_cmp_func ildct_cmp[5]; //only width 16 used |
2382 | 179 me_cmp_func frame_skip_cmp[5]; //only width 8 used |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
180 |
1136 | 181 /** |
182 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
2967 | 183 * this is an array[4][4] of motion compensation funcions for 4 |
1713 | 184 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 185 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 186 * @param block destination where the result is stored |
187 * @param pixels source | |
188 * @param line_size number of bytes in a horizontal line of block | |
189 * @param h height | |
190 */ | |
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
|
191 op_pixels_func put_pixels_tab[4][4]; |
1136 | 192 |
193 /** | |
194 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
2967 | 195 * This is an array[4][4] of motion compensation functions for 4 |
1713 | 196 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 197 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 198 * @param block destination into which the result is averaged (a+b+1)>>1 |
199 * @param pixels source | |
200 * @param line_size number of bytes in a horizontal line of block | |
201 * @param h height | |
202 */ | |
1319 | 203 op_pixels_func avg_pixels_tab[4][4]; |
1136 | 204 |
205 /** | |
206 * Halfpel motion compensation with no rounding (a+b)>>1. | |
2967 | 207 * this is an array[2][4] of motion compensation funcions for 2 |
1225 | 208 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 209 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 210 * @param block destination where the result is stored |
211 * @param pixels source | |
212 * @param line_size number of bytes in a horizontal line of block | |
213 * @param h height | |
214 */ | |
2075 | 215 op_pixels_func put_no_rnd_pixels_tab[4][4]; |
1136 | 216 |
217 /** | |
218 * Halfpel motion compensation with no rounding (a+b)>>1. | |
2967 | 219 * this is an array[2][4] of motion compensation funcions for 2 |
1225 | 220 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 221 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 222 * @param block destination into which the result is averaged (a+b)>>1 |
223 * @param pixels source | |
224 * @param line_size number of bytes in a horizontal line of block | |
225 * @param h height | |
226 */ | |
2075 | 227 op_pixels_func avg_no_rnd_pixels_tab[4][4]; |
2967 | 228 |
1864 | 229 void (*put_no_rnd_pixels_l2[2])(uint8_t *block/*align width (8 or 16)*/, const uint8_t *a/*align 1*/, const uint8_t *b/*align 1*/, int line_size, int h); |
2967 | 230 |
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
|
231 /** |
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 * 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
|
233 * 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
|
234 * *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
|
235 * @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
|
236 * @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
|
237 * @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
|
238 * @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
|
239 */ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
240 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
1319 | 241 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
242 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
243 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
|
244 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
|
245 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
|
246 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; |
936 | 247 qpel_mc_func put_mspel_pixels_tab[8]; |
2967 | 248 |
1168 | 249 /** |
250 * h264 Chram MC | |
251 */ | |
252 h264_chroma_mc_func put_h264_chroma_pixels_tab[3]; | |
253 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
|
254 |
1168 | 255 qpel_mc_func put_h264_qpel_pixels_tab[3][16]; |
256 qpel_mc_func avg_h264_qpel_pixels_tab[3][16]; | |
2967 | 257 |
2415 | 258 h264_weight_func weight_h264_pixels_tab[10]; |
259 h264_biweight_func biweight_h264_pixels_tab[10]; | |
2967 | 260 |
1708 | 261 me_cmp_func pix_abs[2][4]; |
2967 | 262 |
866 | 263 /* huffyuv specific */ |
264 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); | |
936 | 265 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); |
1527 | 266 /** |
267 * subtract huffyuv's variant of median prediction | |
268 * note, this might read from src1[-1], src2[-1] | |
269 */ | |
270 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top); | |
1273 | 271 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w); |
2633 | 272 |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
273 void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
274 void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
275 void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
276 void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
277 void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta); |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
278 void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta); |
2967 | 279 |
1644 | 280 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale); |
281 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale); | |
282 | |
2045 | 283 void (*h261_loop_filter)(uint8_t *src, int stride); |
2044
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
2024
diff
changeset
|
284 |
1092 | 285 /* (I)DCT */ |
286 void (*fdct)(DCTELEM *block/* align 16*/); | |
1567 | 287 void (*fdct248)(DCTELEM *block/* align 16*/); |
2967 | 288 |
1324
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
289 /* IDCT really*/ |
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
290 void (*idct)(DCTELEM *block/* align 16*/); |
2967 | 291 |
1102 | 292 /** |
1104 | 293 * block -> idct -> clip to unsigned 8 bit -> dest. |
1102 | 294 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) |
1111 | 295 * @param line_size size in bytes of a horizotal line of dest |
1102 | 296 */ |
1092 | 297 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
2967 | 298 |
1102 | 299 /** |
300 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. | |
1111 | 301 * @param line_size size in bytes of a horizotal line of dest |
1102 | 302 */ |
1092 | 303 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
2967 | 304 |
1102 | 305 /** |
1104 | 306 * idct input permutation. |
1241 | 307 * several optimized IDCTs need a permutated input (relative to the normal order of the reference |
308 * IDCT) | |
309 * this permutation must be performed before the idct_put/add, note, normally this can be merged | |
310 * with the zigzag/alternate scan<br> | |
1102 | 311 * an example to avoid confusion: |
312 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...) | |
313 * - (x -> referece dct -> reference idct -> x) | |
314 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x) | |
315 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...) | |
316 */ | |
1092 | 317 uint8_t idct_permutation[64]; |
318 int idct_permutation_type; | |
319 #define FF_NO_IDCT_PERM 1 | |
320 #define FF_LIBMPEG2_IDCT_PERM 2 | |
321 #define FF_SIMPLE_IDCT_PERM 3 | |
322 #define FF_TRANSPOSE_IDCT_PERM 4 | |
2696
9699d325049d
porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents:
2693
diff
changeset
|
323 #define FF_PARTTRANS_IDCT_PERM 5 |
1092 | 324 |
1784 | 325 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale); |
326 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); | |
327 #define BASIS_SHIFT 16 | |
328 #define RECON_SHIFT 6 | |
2967 | 329 |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
330 void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride); |
2755 | 331 void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
332 } DSPContext; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
333 |
1201 | 334 void dsputil_static_init(void); |
1092 | 335 void dsputil_init(DSPContext* p, AVCodecContext *avctx); |
0 | 336 |
764 | 337 /** |
338 * permute block according to permuatation. | |
339 * @param last last non zero element in scantable order | |
340 */ | |
1064 | 341 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); |
34 | 342 |
1729 | 343 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type); |
344 | |
1264 | 345 #define BYTE_VEC32(c) ((c)*0x01010101UL) |
346 | |
347 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b) | |
348 { | |
349 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
350 } | |
351 | |
352 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b) | |
353 { | |
354 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
355 } | |
356 | |
2184 | 357 static inline int get_penalty_factor(int lambda, int lambda2, int type){ |
358 switch(type&0xFF){ | |
359 default: | |
360 case FF_CMP_SAD: | |
361 return lambda>>FF_LAMBDA_SHIFT; | |
362 case FF_CMP_DCT: | |
363 return (3*lambda)>>(FF_LAMBDA_SHIFT+1); | |
364 case FF_CMP_W53: | |
365 return (4*lambda)>>(FF_LAMBDA_SHIFT); | |
366 case FF_CMP_W97: | |
367 return (2*lambda)>>(FF_LAMBDA_SHIFT); | |
368 case FF_CMP_SATD: | |
369 return (2*lambda)>>FF_LAMBDA_SHIFT; | |
370 case FF_CMP_RD: | |
371 case FF_CMP_PSNR: | |
372 case FF_CMP_SSE: | |
373 case FF_CMP_NSSE: | |
374 return lambda2>>FF_LAMBDA_SHIFT; | |
375 case FF_CMP_BIT: | |
376 return 1; | |
377 } | |
378 } | |
379 | |
1102 | 380 /** |
1104 | 381 * Empty mmx state. |
1102 | 382 * this must be called between any dsp function and float/double code. |
383 * for example sin(); dsp->idct_put(); emms_c(); cos() | |
384 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
385 #define emms_c() |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
386 |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
387 /* should be defined by architectures supporting |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
388 one or more MultiMedia extension */ |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
389 int mm_support(void); |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
390 |
1974
8c5489b2cf3e
move __align16 some place where non-MMX machines can see it
melanson
parents:
1972
diff
changeset
|
391 #define __align16 __attribute__ ((aligned (16))) |
8c5489b2cf3e
move __align16 some place where non-MMX machines can see it
melanson
parents:
1972
diff
changeset
|
392 |
62 | 393 #if defined(HAVE_MMX) |
0 | 394 |
862 | 395 #undef emms_c |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
396 |
0 | 397 #define MM_MMX 0x0001 /* standard MMX */ |
398 #define MM_3DNOW 0x0004 /* AMD 3DNOW */ | |
399 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */ | |
400 #define MM_SSE 0x0008 /* SSE functions */ | |
401 #define MM_SSE2 0x0010 /* PIV SSE2 functions */ | |
2388 | 402 #define MM_3DNOWEXT 0x0020 /* AMD 3DNowExt */ |
0 | 403 |
404 extern int mm_flags; | |
405 | |
1064 | 406 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
407 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); | |
1984
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
408 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
0 | 409 |
410 static inline void emms(void) | |
411 { | |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
412 __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
|
413 } |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
414 |
936 | 415 |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
416 #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
|
417 {\ |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
418 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
|
419 emms();\ |
0 | 420 } |
421 | |
422 #define __align8 __attribute__ ((aligned (8))) | |
2324 | 423 #define STRIDE_ALIGN 8 |
0 | 424 |
1092 | 425 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); |
426 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx); | |
1065 | 427 |
62 | 428 #elif defined(ARCH_ARMV4L) |
429 | |
430 /* This is to use 4 bytes read to the IDCT pointers for some 'zero' | |
1974
8c5489b2cf3e
move __align16 some place where non-MMX machines can see it
melanson
parents:
1972
diff
changeset
|
431 line optimizations */ |
62 | 432 #define __align8 __attribute__ ((aligned (4))) |
2324 | 433 #define STRIDE_ALIGN 4 |
62 | 434 |
2776
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
435 #define MM_IWMMXT 0x0100 /* XScale IWMMXT */ |
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
436 |
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
437 extern int mm_flags; |
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
438 |
1092 | 439 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx); |
62 | 440 |
88 | 441 #elif defined(HAVE_MLIB) |
442 | |
443 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ | |
444 #define __align8 __attribute__ ((aligned (8))) | |
2324 | 445 #define STRIDE_ALIGN 8 |
88 | 446 |
1092 | 447 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx); |
88 | 448 |
1959
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
449 #elif defined(ARCH_SPARC) |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
450 |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
451 /* 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
|
452 #define __align8 __attribute__ ((aligned (8))) |
2324 | 453 #define STRIDE_ALIGN 8 |
1959
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
454 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
|
455 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
456 #elif defined(ARCH_ALPHA) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
457 |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
458 #define __align8 __attribute__ ((aligned (8))) |
2324 | 459 #define STRIDE_ALIGN 8 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
460 |
1092 | 461 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
|
462 |
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
|
463 #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
|
464 |
894
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
465 #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
|
466 |
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
467 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
|
468 |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
469 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN) |
1653 | 470 #define pixel altivec_pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
471 #include <altivec.h> |
1653 | 472 #undef pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
473 #endif |
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
474 |
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
|
475 #define __align8 __attribute__ ((aligned (16))) |
2324 | 476 #define STRIDE_ALIGN 16 |
623
92e99e506920
first cut at altivec support on darwin patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
612
diff
changeset
|
477 |
1092 | 478 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
|
479 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
480 #elif defined(HAVE_MMI) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
481 |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
482 #define __align8 __attribute__ ((aligned (16))) |
2324 | 483 #define STRIDE_ALIGN 16 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
484 |
1092 | 485 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
|
486 |
1259
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
487 #elif defined(ARCH_SH4) |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
488 |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
489 #define __align8 __attribute__ ((aligned (8))) |
2324 | 490 #define STRIDE_ALIGN 8 |
1259
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
491 |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
492 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
|
493 |
0 | 494 #else |
495 | |
2324 | 496 #define __align8 __attribute__ ((aligned (8))) |
497 #define STRIDE_ALIGN 8 | |
0 | 498 |
499 #endif | |
500 | |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
501 #ifdef __GNUC__ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
502 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
503 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
|
504 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
|
505 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
|
506 |
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
|
507 #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
|
508 #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
|
509 #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
|
510 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
511 #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
|
512 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
513 #else /* __GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
514 |
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
|
515 #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
|
516 #define LD32(a) (*((uint32_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
517 #define LD64(a) (*((uint64_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
518 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
519 #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
|
520 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
521 #endif /* !__GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
522 |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
523 /* PSNR */ |
1064 | 524 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
|
525 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
|
526 AVCodecContext *avctx); |
781 | 527 |
528 /* FFT computation */ | |
529 | |
530 /* NOTE: soon integer code will be added, so you must use the | |
531 FFTSample type */ | |
532 typedef float FFTSample; | |
533 | |
534 typedef struct FFTComplex { | |
535 FFTSample re, im; | |
536 } FFTComplex; | |
537 | |
538 typedef struct FFTContext { | |
539 int nbits; | |
540 int inverse; | |
541 uint16_t *revtab; | |
542 FFTComplex *exptab; | |
543 FFTComplex *exptab1; /* only used by SSE code */ | |
544 void (*fft_calc)(struct FFTContext *s, FFTComplex *z); | |
545 } FFTContext; | |
546 | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
547 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
|
548 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
|
549 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
|
550 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
|
551 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
|
552 |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
553 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z) |
781 | 554 { |
555 s->fft_calc(s, z); | |
556 } | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
557 void ff_fft_end(FFTContext *s); |
781 | 558 |
559 /* MDCT computation */ | |
560 | |
561 typedef struct MDCTContext { | |
562 int n; /* size of MDCT (i.e. number of input data * 2) */ | |
563 int nbits; /* n = 2^nbits */ | |
564 /* pre/post rotation tables */ | |
565 FFTSample *tcos; | |
566 FFTSample *tsin; | |
567 FFTContext fft; | |
568 } MDCTContext; | |
569 | |
794 | 570 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
|
571 void ff_imdct_calc(MDCTContext *s, FFTSample *output, |
781 | 572 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
|
573 void ff_mdct_calc(MDCTContext *s, FFTSample *out, |
781 | 574 const FFTSample *input, FFTSample *tmp); |
794 | 575 void ff_mdct_end(MDCTContext *s); |
781 | 576 |
1708 | 577 #define WARPER8_16(name8, name16)\ |
578 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
579 return name8(s, dst , src , stride, h)\ | |
580 +name8(s, dst+8 , src+8 , stride, h);\ | |
581 } | |
582 | |
583 #define WARPER8_16_SQ(name8, name16)\ | |
584 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
585 int score=0;\ | |
586 score +=name8(s, dst , src , stride, 8);\ | |
587 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
588 if(h==16){\ | |
589 dst += 8*stride;\ | |
590 src += 8*stride;\ | |
591 score +=name8(s, dst , src , stride, 8);\ | |
592 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
593 }\ | |
594 return score;\ | |
936 | 595 } |
596 | |
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
|
597 #endif |