Mercurial > libavcodec.hg
annotate dsputil.h @ 2693:02925a3903b6 libavcodec
porting vp3 idct over to lavc idct api
author | michael |
---|---|
date | Tue, 17 May 2005 11:12:04 +0000 |
parents | 72e6ffa1f3a5 |
children | 9699d325049d |
rev | line source |
---|---|
429 | 1 /* |
2 * DSP utils | |
3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard. | |
1739
07a484280a82
copyright year update of the files i touched and remembered, things look annoyingly unmaintained otherwise
michael
parents:
1729
diff
changeset
|
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> |
429 | 5 * |
6 * This library is free software; you can redistribute it and/or | |
7 * modify it under the terms of the GNU Lesser General Public | |
8 * License as published by the Free Software Foundation; either | |
9 * version 2 of the License, or (at your option) any later version. | |
10 * | |
11 * This library is distributed in the hope that it will be useful, | |
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
14 * Lesser General Public License for more details. | |
15 * | |
16 * You should have received a copy of the GNU Lesser General Public | |
17 * License along with this library; if not, write to the Free Software | |
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
19 */ | |
1102 | 20 |
21 /** | |
22 * @file dsputil.h | |
1106 | 23 * DSP utils. |
1213 | 24 * note, many functions in here may use MMX which trashes the FPU state, it is |
25 * absolutely necessary to call emms_c() between dsp & float/double code | |
1102 | 26 */ |
27 | |
0 | 28 #ifndef DSPUTIL_H |
29 #define DSPUTIL_H | |
30 | |
31 #include "common.h" | |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
32 #include "avcodec.h" |
0 | 33 |
1102 | 34 |
255 | 35 //#define DEBUG |
0 | 36 /* dct code */ |
37 typedef short DCTELEM; | |
38 | |
474
11dbd00682fc
avoid name clash with libjpeg - added missing externs
bellard
parents:
429
diff
changeset
|
39 void fdct_ifast (DCTELEM *data); |
1571
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1567
diff
changeset
|
40 void fdct_ifast248 (DCTELEM *data); |
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
623
diff
changeset
|
41 void ff_jpeg_fdct_islow (DCTELEM *data); |
1567 | 42 void ff_fdct248_islow (DCTELEM *data); |
0 | 43 |
44 void j_rev_dct (DCTELEM *data); | |
2256 | 45 void j_rev_dct4 (DCTELEM *data); |
2257 | 46 void j_rev_dct2 (DCTELEM *data); |
2259 | 47 void j_rev_dct1 (DCTELEM *data); |
0 | 48 |
687
9abb13c21fbe
fdct_mmx -> ff_fdct_mmx (renamed to avoid namespace conflict with xvid)
arpi_esp
parents:
675
diff
changeset
|
49 void ff_fdct_mmx(DCTELEM *block); |
1565 | 50 void ff_fdct_mmx2(DCTELEM *block); |
1765
e31754bc5b65
SSE2 fdct by (Balatoni Denes <pnis at coder dot hu>)
michael
parents:
1739
diff
changeset
|
51 void ff_fdct_sse2(DCTELEM *block); |
0 | 52 |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
53 void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride); |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
54 void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block); |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
55 void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block); |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
56 |
34 | 57 /* encoding scans */ |
1064 | 58 extern const uint8_t ff_alternate_horizontal_scan[64]; |
59 extern const uint8_t ff_alternate_vertical_scan[64]; | |
60 extern const uint8_t ff_zigzag_direct[64]; | |
1567 | 61 extern const uint8_t ff_zigzag248_direct[64]; |
190
9e0e56869d05
fix for non-mmx runtimedetect encoding bugs - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
174
diff
changeset
|
62 |
0 | 63 /* pixel operations */ |
2090 | 64 #define MAX_NEG_CROP 1024 |
0 | 65 |
66 /* temporary */ | |
1064 | 67 extern uint32_t squareTbl[512]; |
68 extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP]; | |
0 | 69 |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
70 /* VP3 DSP functions */ |
2693 | 71 void ff_vp3_idct_c(DCTELEM *block/* align 16*/); |
72 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); | |
73 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); | |
0 | 74 |
675 | 75 /* minimum alignment rules ;) |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
76 if u notice errors in the align stuff, need more alignment for some asm code for some cpu |
675 | 77 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ... |
78 | |
79 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible) | |
80 i (michael) didnt check them, these are just the alignents which i think could be reached easily ... | |
81 | |
82 !future video codecs might need functions with less strict alignment | |
83 */ | |
0 | 84 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
85 /* |
1064 | 86 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size); |
87 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride); | |
88 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
89 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
296 | 90 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
|
91 */ |
0 | 92 |
93 /* add and put pixel (decoding) */ | |
675 | 94 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 |
1709 | 95 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4 |
1064 | 96 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
|
97 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 | 98 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); |
1168 | 99 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 | 100 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset); |
101 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 | 102 |
984 | 103 #define DEF_OLD_QPEL(name)\ |
1064 | 104 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ |
105 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ | |
106 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
107 |
984 | 108 DEF_OLD_QPEL(qpel16_mc11_old_c) |
109 DEF_OLD_QPEL(qpel16_mc31_old_c) | |
110 DEF_OLD_QPEL(qpel16_mc12_old_c) | |
111 DEF_OLD_QPEL(qpel16_mc32_old_c) | |
112 DEF_OLD_QPEL(qpel16_mc13_old_c) | |
113 DEF_OLD_QPEL(qpel16_mc33_old_c) | |
114 DEF_OLD_QPEL(qpel8_mc11_old_c) | |
115 DEF_OLD_QPEL(qpel8_mc31_old_c) | |
116 DEF_OLD_QPEL(qpel8_mc12_old_c) | |
117 DEF_OLD_QPEL(qpel8_mc32_old_c) | |
118 DEF_OLD_QPEL(qpel8_mc13_old_c) | |
119 DEF_OLD_QPEL(qpel8_mc33_old_c) | |
651 | 120 |
121 #define CALL_2X_PIXELS(a, b, n)\ | |
122 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ | |
123 b(block , pixels , line_size, h);\ | |
124 b(block+n, pixels+n, line_size, h);\ | |
125 } | |
255 | 126 |
0 | 127 /* motion estimation */ |
1709 | 128 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2 |
129 // allthough currently h<4 is not used as functions with width <8 are not used and neither implemented | |
1708 | 130 typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size, int h)/* __attribute__ ((const))*/; |
936 | 131 |
1168 | 132 |
1102 | 133 /** |
134 * DSPContext. | |
135 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
136 typedef struct DSPContext { |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
137 /* pixel ops : interface with DCT */ |
1064 | 138 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); |
139 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); | |
140 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); | |
1984
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
141 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
1064 | 142 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
1136 | 143 /** |
144 * translational global motion compensation. | |
145 */ | |
1064 | 146 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); |
1136 | 147 /** |
148 * global motion compensation. | |
149 */ | |
1064 | 150 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
|
151 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
|
152 void (*clear_blocks)(DCTELEM *blocks/*align 16*/); |
1064 | 153 int (*pix_sum)(uint8_t * pix, int line_size); |
154 int (*pix_norm1)(uint8_t * pix, int line_size); | |
1708 | 155 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4 |
156 | |
1729 | 157 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */ |
158 me_cmp_func sse[5]; | |
159 me_cmp_func hadamard8_diff[5]; | |
160 me_cmp_func dct_sad[5]; | |
161 me_cmp_func quant_psnr[5]; | |
162 me_cmp_func bit[5]; | |
163 me_cmp_func rd[5]; | |
164 me_cmp_func vsad[5]; | |
165 me_cmp_func vsse[5]; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
166 me_cmp_func nsse[5]; |
2184 | 167 me_cmp_func w53[5]; |
168 me_cmp_func w97[5]; | |
2382 | 169 me_cmp_func dct_max[5]; |
936 | 170 |
1708 | 171 me_cmp_func me_pre_cmp[5]; |
172 me_cmp_func me_cmp[5]; | |
173 me_cmp_func me_sub_cmp[5]; | |
174 me_cmp_func mb_cmp[5]; | |
1729 | 175 me_cmp_func ildct_cmp[5]; //only width 16 used |
2382 | 176 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
|
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 */ | |
2075 | 212 op_pixels_func put_no_rnd_pixels_tab[4][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 */ | |
2075 | 224 op_pixels_func avg_no_rnd_pixels_tab[4][4]; |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
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 | |
2415 | 255 h264_weight_func weight_h264_pixels_tab[10]; |
256 h264_biweight_func biweight_h264_pixels_tab[10]; | |
257 | |
1708 | 258 me_cmp_func pix_abs[2][4]; |
866 | 259 |
260 /* huffyuv specific */ | |
261 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); | |
936 | 262 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); |
1527 | 263 /** |
264 * subtract huffyuv's variant of median prediction | |
265 * note, this might read from src1[-1], src2[-1] | |
266 */ | |
267 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top); | |
1273 | 268 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w); |
2633 | 269 |
270 void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int *tc0); | |
271 void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int *tc0); | |
272 void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int *tc0); | |
273 void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int *tc0); | |
1092 | 274 |
1644 | 275 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale); |
276 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale); | |
277 | |
2045 | 278 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
|
279 |
1092 | 280 /* (I)DCT */ |
281 void (*fdct)(DCTELEM *block/* align 16*/); | |
1567 | 282 void (*fdct248)(DCTELEM *block/* align 16*/); |
1102 | 283 |
1324
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
284 /* IDCT really*/ |
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
285 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
|
286 |
1102 | 287 /** |
1104 | 288 * block -> idct -> clip to unsigned 8 bit -> dest. |
1102 | 289 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) |
1111 | 290 * @param line_size size in bytes of a horizotal line of dest |
1102 | 291 */ |
1092 | 292 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
1102 | 293 |
294 /** | |
295 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. | |
1111 | 296 * @param line_size size in bytes of a horizotal line of dest |
1102 | 297 */ |
1092 | 298 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
1102 | 299 |
300 /** | |
1104 | 301 * idct input permutation. |
1241 | 302 * several optimized IDCTs need a permutated input (relative to the normal order of the reference |
303 * IDCT) | |
304 * this permutation must be performed before the idct_put/add, note, normally this can be merged | |
305 * with the zigzag/alternate scan<br> | |
1102 | 306 * an example to avoid confusion: |
307 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...) | |
308 * - (x -> referece dct -> reference idct -> x) | |
309 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x) | |
310 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...) | |
311 */ | |
1092 | 312 uint8_t idct_permutation[64]; |
313 int idct_permutation_type; | |
314 #define FF_NO_IDCT_PERM 1 | |
315 #define FF_LIBMPEG2_IDCT_PERM 2 | |
316 #define FF_SIMPLE_IDCT_PERM 3 | |
317 #define FF_TRANSPOSE_IDCT_PERM 4 | |
318 | |
1784 | 319 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale); |
320 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); | |
321 #define BASIS_SHIFT 16 | |
322 #define RECON_SHIFT 6 | |
323 | |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
324 /** |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
325 * 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
|
326 */ |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
327 void (*vp3_dsp_init)(void); |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
328 |
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 * This function is responsible for taking a block of zigzag'd, |
1977 | 331 * quantized DCT coefficients and reconstructing the original block of |
332 * samples. | |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
333 * @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
|
334 * @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
|
335 * @param coeff_count index of the last coefficient |
1977 | 336 * @param output_samples space for 64 DCTELEMs where the transformed |
337 * samples will be stored | |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
338 */ |
1977 | 339 void (*vp3_idct)(int16_t *input_data, int16_t *dequant_matrix, |
340 int coeff_count, DCTELEM *output_samples); | |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
341 |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
342 void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
343 } DSPContext; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
344 |
1201 | 345 void dsputil_static_init(void); |
1092 | 346 void dsputil_init(DSPContext* p, AVCodecContext *avctx); |
0 | 347 |
764 | 348 /** |
349 * permute block according to permuatation. | |
350 * @param last last non zero element in scantable order | |
351 */ | |
1064 | 352 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); |
34 | 353 |
1729 | 354 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type); |
355 | |
1264 | 356 #define BYTE_VEC32(c) ((c)*0x01010101UL) |
357 | |
358 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b) | |
359 { | |
360 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
361 } | |
362 | |
363 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b) | |
364 { | |
365 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
366 } | |
367 | |
2184 | 368 static inline int get_penalty_factor(int lambda, int lambda2, int type){ |
369 switch(type&0xFF){ | |
370 default: | |
371 case FF_CMP_SAD: | |
372 return lambda>>FF_LAMBDA_SHIFT; | |
373 case FF_CMP_DCT: | |
374 return (3*lambda)>>(FF_LAMBDA_SHIFT+1); | |
375 case FF_CMP_W53: | |
376 return (4*lambda)>>(FF_LAMBDA_SHIFT); | |
377 case FF_CMP_W97: | |
378 return (2*lambda)>>(FF_LAMBDA_SHIFT); | |
379 case FF_CMP_SATD: | |
380 return (2*lambda)>>FF_LAMBDA_SHIFT; | |
381 case FF_CMP_RD: | |
382 case FF_CMP_PSNR: | |
383 case FF_CMP_SSE: | |
384 case FF_CMP_NSSE: | |
385 return lambda2>>FF_LAMBDA_SHIFT; | |
386 case FF_CMP_BIT: | |
387 return 1; | |
388 } | |
389 } | |
390 | |
1102 | 391 /** |
1104 | 392 * Empty mmx state. |
1102 | 393 * this must be called between any dsp function and float/double code. |
394 * for example sin(); dsp->idct_put(); emms_c(); cos() | |
395 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
396 #define emms_c() |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
397 |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
398 /* should be defined by architectures supporting |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
399 one or more MultiMedia extension */ |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
400 int mm_support(void); |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
401 |
1974
8c5489b2cf3e
move __align16 some place where non-MMX machines can see it
melanson
parents:
1972
diff
changeset
|
402 #define __align16 __attribute__ ((aligned (16))) |
8c5489b2cf3e
move __align16 some place where non-MMX machines can see it
melanson
parents:
1972
diff
changeset
|
403 |
62 | 404 #if defined(HAVE_MMX) |
0 | 405 |
862 | 406 #undef emms_c |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
407 |
0 | 408 #define MM_MMX 0x0001 /* standard MMX */ |
409 #define MM_3DNOW 0x0004 /* AMD 3DNOW */ | |
410 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */ | |
411 #define MM_SSE 0x0008 /* SSE functions */ | |
412 #define MM_SSE2 0x0010 /* PIV SSE2 functions */ | |
2388 | 413 #define MM_3DNOWEXT 0x0020 /* AMD 3DNowExt */ |
0 | 414 |
415 extern int mm_flags; | |
416 | |
1064 | 417 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
418 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
|
419 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
0 | 420 |
421 static inline void emms(void) | |
422 { | |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
423 __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
|
424 } |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
425 |
936 | 426 |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
427 #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
|
428 {\ |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
429 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
|
430 emms();\ |
0 | 431 } |
432 | |
433 #define __align8 __attribute__ ((aligned (8))) | |
2324 | 434 #define STRIDE_ALIGN 8 |
0 | 435 |
1092 | 436 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); |
437 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx); | |
1065 | 438 |
62 | 439 #elif defined(ARCH_ARMV4L) |
440 | |
441 /* 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
|
442 line optimizations */ |
62 | 443 #define __align8 __attribute__ ((aligned (4))) |
2324 | 444 #define STRIDE_ALIGN 4 |
62 | 445 |
1092 | 446 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx); |
62 | 447 |
88 | 448 #elif defined(HAVE_MLIB) |
449 | |
450 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ | |
451 #define __align8 __attribute__ ((aligned (8))) | |
2324 | 452 #define STRIDE_ALIGN 8 |
88 | 453 |
1092 | 454 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx); |
88 | 455 |
1959
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
456 #elif defined(ARCH_SPARC) |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
457 |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
458 /* 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
|
459 #define __align8 __attribute__ ((aligned (8))) |
2324 | 460 #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
|
461 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
|
462 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
463 #elif defined(ARCH_ALPHA) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
464 |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
465 #define __align8 __attribute__ ((aligned (8))) |
2324 | 466 #define STRIDE_ALIGN 8 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
467 |
1092 | 468 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
|
469 |
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
|
470 #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
|
471 |
894
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
472 #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
|
473 |
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
474 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
|
475 |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
476 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN) |
1653 | 477 #define pixel altivec_pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
478 #include <altivec.h> |
1653 | 479 #undef pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
480 #endif |
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
481 |
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
|
482 #define __align8 __attribute__ ((aligned (16))) |
2324 | 483 #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
|
484 |
1092 | 485 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
|
486 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
487 #elif defined(HAVE_MMI) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
488 |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
489 #define __align8 __attribute__ ((aligned (16))) |
2324 | 490 #define STRIDE_ALIGN 16 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
491 |
1092 | 492 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
|
493 |
1259
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
494 #elif defined(ARCH_SH4) |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
495 |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
496 #define __align8 __attribute__ ((aligned (8))) |
2324 | 497 #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
|
498 |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
499 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
|
500 |
0 | 501 #else |
502 | |
2324 | 503 #define __align8 __attribute__ ((aligned (8))) |
504 #define STRIDE_ALIGN 8 | |
0 | 505 |
506 #endif | |
507 | |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
508 #ifdef __GNUC__ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
509 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
510 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
|
511 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
|
512 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
|
513 |
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
|
514 #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
|
515 #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
|
516 #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
|
517 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
518 #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
|
519 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
520 #else /* __GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
521 |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
522 #define LD16(a) (*((uint16_t*)(a))) |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
523 #define LD32(a) (*((uint32_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
524 #define LD64(a) (*((uint64_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
525 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
526 #define ST32(a, b) *((uint32_t*)(a)) = (b) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
527 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
528 #endif /* !__GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
529 |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
530 /* PSNR */ |
1064 | 531 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
|
532 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
|
533 AVCodecContext *avctx); |
781 | 534 |
535 /* FFT computation */ | |
536 | |
537 /* NOTE: soon integer code will be added, so you must use the | |
538 FFTSample type */ | |
539 typedef float FFTSample; | |
540 | |
541 typedef struct FFTComplex { | |
542 FFTSample re, im; | |
543 } FFTComplex; | |
544 | |
545 typedef struct FFTContext { | |
546 int nbits; | |
547 int inverse; | |
548 uint16_t *revtab; | |
549 FFTComplex *exptab; | |
550 FFTComplex *exptab1; /* only used by SSE code */ | |
551 void (*fft_calc)(struct FFTContext *s, FFTComplex *z); | |
552 } FFTContext; | |
553 | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
554 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
|
555 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
|
556 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
|
557 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
|
558 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
|
559 |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
560 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z) |
781 | 561 { |
562 s->fft_calc(s, z); | |
563 } | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
564 void ff_fft_end(FFTContext *s); |
781 | 565 |
566 /* MDCT computation */ | |
567 | |
568 typedef struct MDCTContext { | |
569 int n; /* size of MDCT (i.e. number of input data * 2) */ | |
570 int nbits; /* n = 2^nbits */ | |
571 /* pre/post rotation tables */ | |
572 FFTSample *tcos; | |
573 FFTSample *tsin; | |
574 FFTContext fft; | |
575 } MDCTContext; | |
576 | |
794 | 577 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
|
578 void ff_imdct_calc(MDCTContext *s, FFTSample *output, |
781 | 579 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
|
580 void ff_mdct_calc(MDCTContext *s, FFTSample *out, |
781 | 581 const FFTSample *input, FFTSample *tmp); |
794 | 582 void ff_mdct_end(MDCTContext *s); |
781 | 583 |
1708 | 584 #define WARPER8_16(name8, name16)\ |
585 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
586 return name8(s, dst , src , stride, h)\ | |
587 +name8(s, dst+8 , src+8 , stride, h);\ | |
588 } | |
589 | |
590 #define WARPER8_16_SQ(name8, name16)\ | |
591 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
592 int score=0;\ | |
593 score +=name8(s, dst , src , stride, 8);\ | |
594 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
595 if(h==16){\ | |
596 dst += 8*stride;\ | |
597 src += 8*stride;\ | |
598 score +=name8(s, dst , src , stride, 8);\ | |
599 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
600 }\ | |
601 return score;\ | |
936 | 602 } |
603 | |
838
b78812db886f
lrintf detection (based upon a patch by Franois Revol <revol at free dot fr>)
michaelni
parents:
802
diff
changeset
|
604 #ifndef HAVE_LRINTF |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
605 /* 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
|
606 /* better than nothing implementation. */ |
802 | 607 /* btw, rintf() is existing on fbsd too -- alex */ |
2185 | 608 static always_inline long int lrintf(float x) |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
609 { |
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
|
610 #ifdef CONFIG_WIN32 |
2185 | 611 # ifdef ARCH_X86 |
612 int32_t i; | |
613 asm volatile( | |
614 "fistpl %0\n\t" | |
615 : "=m" (i) : "t" (x) : "st" | |
616 ); | |
617 return i; | |
618 # else | |
1040
998d5035b15b
win32: rint() does not seem to be defined with mingw32-gcc 2.95 - do you have a better solution ?
bellard
parents:
1033
diff
changeset
|
619 /* XXX: incorrect, but make it compile */ |
2185 | 620 return (int)(x + (x < 0 ? -0.5 : 0.5)); |
621 # endif | |
1040
998d5035b15b
win32: rint() does not seem to be defined with mingw32-gcc 2.95 - do you have a better solution ?
bellard
parents:
1033
diff
changeset
|
622 #else |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
623 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
|
624 #endif |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
625 } |
2024
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
1984
diff
changeset
|
626 #else |
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
1984
diff
changeset
|
627 #ifndef _ISOC9X_SOURCE |
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
1984
diff
changeset
|
628 #define _ISOC9X_SOURCE |
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
1984
diff
changeset
|
629 #endif |
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
1984
diff
changeset
|
630 #include <math.h> |
0 | 631 #endif |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
632 |
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
633 #endif |