Mercurial > libavcodec.hg
annotate dsputil.h @ 6441:0cc3651e0bb2 libavcodec
fix wmv3 dependencies
author | aurel |
---|---|
date | Tue, 04 Mar 2008 18:38:32 +0000 |
parents | b6d05f8827cd |
children | 2b553c57ec51 |
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 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
6 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
7 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
8 * FFmpeg is free software; you can redistribute it and/or |
429 | 9 * modify it under the terms of the GNU Lesser General Public |
10 * License as published by the Free Software Foundation; either | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
11 * version 2.1 of the License, or (at your option) any later version. |
429 | 12 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
13 * FFmpeg is distributed in the hope that it will be useful, |
429 | 14 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
16 * Lesser General Public License for more details. | |
17 * | |
18 * You should have received a copy of the GNU Lesser General Public | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
19 * License along with FFmpeg; if not, write to the Free Software |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
3029
diff
changeset
|
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
429 | 21 */ |
1102 | 22 |
23 /** | |
24 * @file dsputil.h | |
1106 | 25 * DSP utils. |
1213 | 26 * note, many functions in here may use MMX which trashes the FPU state, it is |
27 * absolutely necessary to call emms_c() between dsp & float/double code | |
1102 | 28 */ |
29 | |
5830
1d83e9c34641
Add FFMPEG_ prefix to all multiple inclusion guards.
diego
parents:
5750
diff
changeset
|
30 #ifndef FFMPEG_DSPUTIL_H |
1d83e9c34641
Add FFMPEG_ prefix to all multiple inclusion guards.
diego
parents:
5750
diff
changeset
|
31 #define FFMPEG_DSPUTIL_H |
0 | 32 |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
33 #include "avcodec.h" |
0 | 34 |
1102 | 35 |
255 | 36 //#define DEBUG |
0 | 37 /* dct code */ |
38 typedef short DCTELEM; | |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
39 typedef int DWTELEM; |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5573
diff
changeset
|
40 typedef short IDWTELEM; |
0 | 41 |
474
11dbd00682fc
avoid name clash with libjpeg - added missing externs
bellard
parents:
429
diff
changeset
|
42 void fdct_ifast (DCTELEM *data); |
1571
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1567
diff
changeset
|
43 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
|
44 void ff_jpeg_fdct_islow (DCTELEM *data); |
1567 | 45 void ff_fdct248_islow (DCTELEM *data); |
0 | 46 |
47 void j_rev_dct (DCTELEM *data); | |
2256 | 48 void j_rev_dct4 (DCTELEM *data); |
2257 | 49 void j_rev_dct2 (DCTELEM *data); |
2259 | 50 void j_rev_dct1 (DCTELEM *data); |
5887 | 51 void ff_wmv2_idct_c(DCTELEM *data); |
0 | 52 |
687
9abb13c21fbe
fdct_mmx -> ff_fdct_mmx (renamed to avoid namespace conflict with xvid)
arpi_esp
parents:
675
diff
changeset
|
53 void ff_fdct_mmx(DCTELEM *block); |
1565 | 54 void ff_fdct_mmx2(DCTELEM *block); |
1765
e31754bc5b65
SSE2 fdct by (Balatoni Denes <pnis at coder dot hu>)
michael
parents:
1739
diff
changeset
|
55 void ff_fdct_sse2(DCTELEM *block); |
0 | 56 |
2755 | 57 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
|
58 void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride); |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3089
diff
changeset
|
59 void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3089
diff
changeset
|
60 void ff_h264_idct_dc_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
|
61 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
|
62 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
|
63 |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
64 void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1, |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
65 const float *src2, int src3, int blocksize, int step); |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
66 void ff_float_to_int16_c(int16_t *dst, const float *src, int len); |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
67 |
34 | 68 /* encoding scans */ |
1064 | 69 extern const uint8_t ff_alternate_horizontal_scan[64]; |
70 extern const uint8_t ff_alternate_vertical_scan[64]; | |
71 extern const uint8_t ff_zigzag_direct[64]; | |
1567 | 72 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
|
73 |
0 | 74 /* pixel operations */ |
2090 | 75 #define MAX_NEG_CROP 1024 |
0 | 76 |
77 /* temporary */ | |
4179 | 78 extern uint32_t ff_squareTbl[512]; |
4176 | 79 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP]; |
0 | 80 |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
81 /* VP3 DSP functions */ |
2693 | 82 void ff_vp3_idct_c(DCTELEM *block/* align 16*/); |
83 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); | |
84 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); | |
0 | 85 |
3245 | 86 /* 1/2^n downscaling functions from imgconvert.c */ |
87 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
88 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
89 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
90 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
91 | |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
92 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy, |
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
93 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); |
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
94 |
675 | 95 /* minimum alignment rules ;) |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
96 if u notice errors in the align stuff, need more alignment for some asm code for some cpu |
675 | 97 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ... |
98 | |
99 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible) | |
100 i (michael) didnt check them, these are just the alignents which i think could be reached easily ... | |
101 | |
102 !future video codecs might need functions with less strict alignment | |
103 */ | |
0 | 104 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
105 /* |
1064 | 106 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size); |
107 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride); | |
108 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
109 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
296 | 110 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
|
111 */ |
0 | 112 |
113 /* add and put pixel (decoding) */ | |
675 | 114 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 |
1709 | 115 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4 |
1064 | 116 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
|
117 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 | 118 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); |
1168 | 119 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 | 120 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset); |
3029 | 121 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset); |
0 | 122 |
984 | 123 #define DEF_OLD_QPEL(name)\ |
1064 | 124 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ |
125 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ | |
126 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
|
127 |
984 | 128 DEF_OLD_QPEL(qpel16_mc11_old_c) |
129 DEF_OLD_QPEL(qpel16_mc31_old_c) | |
130 DEF_OLD_QPEL(qpel16_mc12_old_c) | |
131 DEF_OLD_QPEL(qpel16_mc32_old_c) | |
132 DEF_OLD_QPEL(qpel16_mc13_old_c) | |
133 DEF_OLD_QPEL(qpel16_mc33_old_c) | |
134 DEF_OLD_QPEL(qpel8_mc11_old_c) | |
135 DEF_OLD_QPEL(qpel8_mc31_old_c) | |
136 DEF_OLD_QPEL(qpel8_mc12_old_c) | |
137 DEF_OLD_QPEL(qpel8_mc32_old_c) | |
138 DEF_OLD_QPEL(qpel8_mc13_old_c) | |
139 DEF_OLD_QPEL(qpel8_mc33_old_c) | |
651 | 140 |
141 #define CALL_2X_PIXELS(a, b, n)\ | |
142 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ | |
143 b(block , pixels , line_size, h);\ | |
144 b(block+n, pixels+n, line_size, h);\ | |
145 } | |
255 | 146 |
0 | 147 /* motion estimation */ |
1709 | 148 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2 |
5127 | 149 // although currently h<4 is not used as functions with width <8 are neither used nor implemented |
1708 | 150 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 | 151 |
1168 | 152 |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
153 // for snow slices |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
154 typedef struct slice_buffer_s slice_buffer; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
155 |
1102 | 156 /** |
6438 | 157 * Scantable. |
158 */ | |
159 typedef struct ScanTable{ | |
160 const uint8_t *scantable; | |
161 uint8_t permutated[64]; | |
162 uint8_t raster_end[64]; | |
163 #ifdef ARCH_POWERPC | |
164 /** Used by dct_quantize_altivec to find last-non-zero */ | |
6440 | 165 DECLARE_ALIGNED(16, uint8_t, inverse[64]); |
6438 | 166 #endif |
167 } ScanTable; | |
168 | |
169 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable); | |
170 | |
171 /** | |
1102 | 172 * DSPContext. |
173 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
174 typedef struct DSPContext { |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
175 /* pixel ops : interface with DCT */ |
1064 | 176 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); |
177 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); | |
178 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
|
179 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
1064 | 180 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
2763 | 181 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size); |
182 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size); | |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4962
diff
changeset
|
183 int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/); |
1136 | 184 /** |
185 * translational global motion compensation. | |
186 */ | |
1064 | 187 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); |
1136 | 188 /** |
189 * global motion compensation. | |
190 */ | |
1064 | 191 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy, |
2979 | 192 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
193 void (*clear_blocks)(DCTELEM *blocks/*align 16*/); |
1064 | 194 int (*pix_sum)(uint8_t * pix, int line_size); |
195 int (*pix_norm1)(uint8_t * pix, int line_size); | |
1708 | 196 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4 |
2967 | 197 |
1729 | 198 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */ |
199 me_cmp_func sse[5]; | |
200 me_cmp_func hadamard8_diff[5]; | |
201 me_cmp_func dct_sad[5]; | |
202 me_cmp_func quant_psnr[5]; | |
203 me_cmp_func bit[5]; | |
204 me_cmp_func rd[5]; | |
205 me_cmp_func vsad[5]; | |
206 me_cmp_func vsse[5]; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
207 me_cmp_func nsse[5]; |
2184 | 208 me_cmp_func w53[5]; |
209 me_cmp_func w97[5]; | |
2382 | 210 me_cmp_func dct_max[5]; |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
211 me_cmp_func dct264_sad[5]; |
936 | 212 |
1708 | 213 me_cmp_func me_pre_cmp[5]; |
214 me_cmp_func me_cmp[5]; | |
215 me_cmp_func me_sub_cmp[5]; | |
216 me_cmp_func mb_cmp[5]; | |
1729 | 217 me_cmp_func ildct_cmp[5]; //only width 16 used |
2382 | 218 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
|
219 |
5255 | 220 int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2, |
221 int size); | |
4749 | 222 |
1136 | 223 /** |
224 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
4751 | 225 * this is an array[4][4] of motion compensation functions for 4 |
1713 | 226 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 227 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 228 * @param block destination where the result is stored |
229 * @param pixels source | |
230 * @param line_size number of bytes in a horizontal line of block | |
231 * @param h height | |
232 */ | |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
233 op_pixels_func put_pixels_tab[4][4]; |
1136 | 234 |
235 /** | |
236 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
2967 | 237 * This is an array[4][4] of motion compensation functions for 4 |
1713 | 238 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 239 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 240 * @param block destination into which the result is averaged (a+b+1)>>1 |
241 * @param pixels source | |
242 * @param line_size number of bytes in a horizontal line of block | |
243 * @param h height | |
244 */ | |
1319 | 245 op_pixels_func avg_pixels_tab[4][4]; |
1136 | 246 |
247 /** | |
248 * Halfpel motion compensation with no rounding (a+b)>>1. | |
4751 | 249 * this is an array[2][4] of motion compensation functions for 2 |
1225 | 250 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 251 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 252 * @param block destination where the result is stored |
253 * @param pixels source | |
254 * @param line_size number of bytes in a horizontal line of block | |
255 * @param h height | |
256 */ | |
2075 | 257 op_pixels_func put_no_rnd_pixels_tab[4][4]; |
1136 | 258 |
259 /** | |
260 * Halfpel motion compensation with no rounding (a+b)>>1. | |
4751 | 261 * this is an array[2][4] of motion compensation functions for 2 |
1225 | 262 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 263 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 264 * @param block destination into which the result is averaged (a+b)>>1 |
265 * @param pixels source | |
266 * @param line_size number of bytes in a horizontal line of block | |
267 * @param h height | |
268 */ | |
2075 | 269 op_pixels_func avg_no_rnd_pixels_tab[4][4]; |
2967 | 270 |
1864 | 271 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 | 272 |
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
|
273 /** |
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
|
274 * Thirdpel motion compensation with rounding (a+b+1)>>1. |
4751 | 275 * this is an array[12] of motion compensation functions for the 9 thirdpe |
276 * positions<br> | |
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
|
277 * *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
|
278 * @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
|
279 * @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
|
280 * @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
|
281 * @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
|
282 */ |
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
|
283 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
1319 | 284 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
285 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
286 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
|
287 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
|
288 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
|
289 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; |
936 | 290 qpel_mc_func put_mspel_pixels_tab[8]; |
2967 | 291 |
1168 | 292 /** |
4751 | 293 * h264 Chroma MC |
1168 | 294 */ |
295 h264_chroma_mc_func put_h264_chroma_pixels_tab[3]; | |
3663 | 296 /* This is really one func used in VC-1 decoding */ |
297 h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3]; | |
1168 | 298 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
|
299 |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3010
diff
changeset
|
300 qpel_mc_func put_h264_qpel_pixels_tab[4][16]; |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3010
diff
changeset
|
301 qpel_mc_func avg_h264_qpel_pixels_tab[4][16]; |
2967 | 302 |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3746
diff
changeset
|
303 qpel_mc_func put_2tap_qpel_pixels_tab[4][16]; |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3746
diff
changeset
|
304 qpel_mc_func avg_2tap_qpel_pixels_tab[4][16]; |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3746
diff
changeset
|
305 |
2415 | 306 h264_weight_func weight_h264_pixels_tab[10]; |
307 h264_biweight_func biweight_h264_pixels_tab[10]; | |
2967 | 308 |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
309 /* AVS specific */ |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
310 qpel_mc_func put_cavs_qpel_pixels_tab[2][16]; |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
311 qpel_mc_func avg_cavs_qpel_pixels_tab[2][16]; |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
312 void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
313 void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
314 void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
315 void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
316 void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride); |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
317 |
1708 | 318 me_cmp_func pix_abs[2][4]; |
2967 | 319 |
866 | 320 /* huffyuv specific */ |
321 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); | |
6384 | 322 void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w); |
936 | 323 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); |
1527 | 324 /** |
325 * subtract huffyuv's variant of median prediction | |
326 * note, this might read from src1[-1], src2[-1] | |
327 */ | |
328 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top); | |
6384 | 329 /* this might write to dst[w] */ |
330 void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp); | |
6241 | 331 void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w); |
2633 | 332 |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
333 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
|
334 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
|
335 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
|
336 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
|
337 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
|
338 void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta); |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3574
diff
changeset
|
339 // h264_loop_filter_strength: simd only. the C version is inlined in h264.c |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3574
diff
changeset
|
340 void (*h264_loop_filter_strength)(int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2], |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3574
diff
changeset
|
341 int bidir, int edges, int step, int mask_mv0, int mask_mv1); |
2967 | 342 |
1644 | 343 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale); |
344 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale); | |
345 | |
2045 | 346 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
|
347 |
5887 | 348 void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale); |
349 void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale); | |
350 | |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
351 /* assume len is a multiple of 4, and arrays are 16-byte aligned */ |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
352 void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize); |
5737 | 353 /* no alignment needed */ |
354 void (*flac_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc); | |
3574 | 355 /* assume len is a multiple of 8, and arrays are 16-byte aligned */ |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
356 void (*vector_fmul)(float *dst, const float *src, int len); |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
357 void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len); |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
358 /* assume len is a multiple of 8, and src arrays are 16-byte aligned */ |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
359 void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step); |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
360 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
361 /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767] |
3660
7e1ee254a3ee
Align the input buffer in ffplay, introduce a public macro for aligned declarations
lu_zero
parents:
3656
diff
changeset
|
362 * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */ |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
363 void (*float_to_int16)(int16_t *dst, const float *src, int len); |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
364 |
1092 | 365 /* (I)DCT */ |
366 void (*fdct)(DCTELEM *block/* align 16*/); | |
1567 | 367 void (*fdct248)(DCTELEM *block/* align 16*/); |
2967 | 368 |
1324
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
369 /* IDCT really*/ |
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
370 void (*idct)(DCTELEM *block/* align 16*/); |
2967 | 371 |
1102 | 372 /** |
1104 | 373 * block -> idct -> clip to unsigned 8 bit -> dest. |
1102 | 374 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) |
5728 | 375 * @param line_size size in bytes of a horizontal line of dest |
1102 | 376 */ |
1092 | 377 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
2967 | 378 |
1102 | 379 /** |
380 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. | |
5728 | 381 * @param line_size size in bytes of a horizontal line of dest |
1102 | 382 */ |
1092 | 383 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
2967 | 384 |
1102 | 385 /** |
1104 | 386 * idct input permutation. |
1241 | 387 * several optimized IDCTs need a permutated input (relative to the normal order of the reference |
388 * IDCT) | |
389 * this permutation must be performed before the idct_put/add, note, normally this can be merged | |
390 * with the zigzag/alternate scan<br> | |
1102 | 391 * an example to avoid confusion: |
392 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...) | |
393 * - (x -> referece dct -> reference idct -> x) | |
394 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x) | |
395 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...) | |
396 */ | |
1092 | 397 uint8_t idct_permutation[64]; |
398 int idct_permutation_type; | |
399 #define FF_NO_IDCT_PERM 1 | |
400 #define FF_LIBMPEG2_IDCT_PERM 2 | |
401 #define FF_SIMPLE_IDCT_PERM 3 | |
402 #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
|
403 #define FF_PARTTRANS_IDCT_PERM 5 |
1092 | 404 |
1784 | 405 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale); |
406 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); | |
407 #define BASIS_SHIFT 16 | |
408 #define RECON_SHIFT 6 | |
2967 | 409 |
6437 | 410 void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w); |
411 | |
4268 | 412 /* h264 functions */ |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
413 void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride); |
2755 | 414 void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride); |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3089
diff
changeset
|
415 void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3089
diff
changeset
|
416 void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride); |
4279 | 417 void (*h264_dct)(DCTELEM block[4][4]); |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
418 |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
419 /* snow wavelet */ |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5573
diff
changeset
|
420 void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width); |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5573
diff
changeset
|
421 void (*horizontal_compose97i)(IDWTELEM *b, int width); |
4436
d3e389536b0a
Add the const specifier as needed to reduce the number of warnings.
takis
parents:
4311
diff
changeset
|
422 void (*inner_add_yblock)(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8); |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3198
diff
changeset
|
423 |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3198
diff
changeset
|
424 void (*prefetch)(void *mem, int stride, int h); |
3245 | 425 |
426 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
3526 | 427 |
428 /* vc1 functions */ | |
429 void (*vc1_inv_trans_8x8)(DCTELEM *b); | |
5997
90de28dfd8d6
Switch VC-1 decoder to output decoded residual immediately.
kostya
parents:
5907
diff
changeset
|
430 void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block); |
90de28dfd8d6
Switch VC-1 decoder to output decoded residual immediately.
kostya
parents:
5907
diff
changeset
|
431 void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block); |
90de28dfd8d6
Switch VC-1 decoder to output decoded residual immediately.
kostya
parents:
5907
diff
changeset
|
432 void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block); |
4239 | 433 void (*vc1_v_overlap)(uint8_t* src, int stride); |
434 void (*vc1_h_overlap)(uint8_t* src, int stride); | |
3526 | 435 /* put 8x8 block with bicubic interpolation and quarterpel precision |
436 * last argument is actually round value instead of height | |
437 */ | |
438 op_pixels_func put_vc1_mspel_pixels_tab[16]; | |
5887 | 439 |
440 /* intrax8 functions */ | |
5907
fbd10e6dfbe1
cosmetics: Fix spacial --> spatial typo in function names.
diego
parents:
5887
diff
changeset
|
441 void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize); |
fbd10e6dfbe1
cosmetics: Fix spacial --> spatial typo in function names.
diego
parents:
5887
diff
changeset
|
442 void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize, |
5887 | 443 int * range, int * sum, int edges); |
444 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
445 } DSPContext; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
446 |
4197 | 447 void dsputil_static_init(void); |
1092 | 448 void dsputil_init(DSPContext* p, AVCodecContext *avctx); |
0 | 449 |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4279
diff
changeset
|
450 int ff_check_alignment(void); |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4279
diff
changeset
|
451 |
764 | 452 /** |
453 * permute block according to permuatation. | |
454 * @param last last non zero element in scantable order | |
455 */ | |
1064 | 456 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); |
34 | 457 |
1729 | 458 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type); |
459 | |
2979 | 460 #define BYTE_VEC32(c) ((c)*0x01010101UL) |
1264 | 461 |
462 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b) | |
463 { | |
464 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
465 } | |
466 | |
467 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b) | |
468 { | |
469 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
470 } | |
471 | |
2184 | 472 static inline int get_penalty_factor(int lambda, int lambda2, int type){ |
473 switch(type&0xFF){ | |
474 default: | |
475 case FF_CMP_SAD: | |
476 return lambda>>FF_LAMBDA_SHIFT; | |
477 case FF_CMP_DCT: | |
478 return (3*lambda)>>(FF_LAMBDA_SHIFT+1); | |
479 case FF_CMP_W53: | |
480 return (4*lambda)>>(FF_LAMBDA_SHIFT); | |
481 case FF_CMP_W97: | |
482 return (2*lambda)>>(FF_LAMBDA_SHIFT); | |
483 case FF_CMP_SATD: | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
484 case FF_CMP_DCT264: |
2184 | 485 return (2*lambda)>>FF_LAMBDA_SHIFT; |
486 case FF_CMP_RD: | |
487 case FF_CMP_PSNR: | |
488 case FF_CMP_SSE: | |
489 case FF_CMP_NSSE: | |
490 return lambda2>>FF_LAMBDA_SHIFT; | |
491 case FF_CMP_BIT: | |
492 return 1; | |
493 } | |
494 } | |
495 | |
1102 | 496 /** |
1104 | 497 * Empty mmx state. |
1102 | 498 * this must be called between any dsp function and float/double code. |
499 * for example sin(); dsp->idct_put(); emms_c(); cos() | |
500 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
501 #define emms_c() |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
502 |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
503 /* should be defined by architectures supporting |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
504 one or more MultiMedia extension */ |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
505 int mm_support(void); |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
506 |
5149 | 507 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx); |
508 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx); | |
509 void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx); | |
510 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx); | |
511 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx); | |
512 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); | |
513 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx); | |
514 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx); | |
515 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx); | |
516 | |
4999
0c4bf6b7d1c6
use DECLARE_ALIGNED from mem.h (which is included through common.h)
alex
parents:
4988
diff
changeset
|
517 #define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v) |
1974
8c5489b2cf3e
move __align16 some place where non-MMX machines can see it
melanson
parents:
1972
diff
changeset
|
518 |
62 | 519 #if defined(HAVE_MMX) |
0 | 520 |
862 | 521 #undef emms_c |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
522 |
0 | 523 #define MM_MMX 0x0001 /* standard MMX */ |
524 #define MM_3DNOW 0x0004 /* AMD 3DNOW */ | |
525 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */ | |
526 #define MM_SSE 0x0008 /* SSE functions */ | |
527 #define MM_SSE2 0x0010 /* PIV SSE2 functions */ | |
2388 | 528 #define MM_3DNOWEXT 0x0020 /* AMD 3DNowExt */ |
3279
647a677c00a4
Remove unused and unsupported Cyrix's "Extended MMX",
gpoirier
parents:
3248
diff
changeset
|
529 #define MM_SSE3 0x0040 /* Prescott SSE3 functions */ |
4311
f9325f2bc762
Add SSSE3 (Core2 aka Conroe/Merom/Woodcrester new instructions) detection
gpoirier
parents:
4281
diff
changeset
|
530 #define MM_SSSE3 0x0080 /* Conroe SSSE3 functions */ |
0 | 531 |
4197 | 532 extern int mm_flags; |
533 | |
1064 | 534 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
535 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
|
536 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
0 | 537 |
538 static inline void emms(void) | |
539 { | |
6391
3dc36ec2dcad
__asm __volatile -> asm volatile, improves code consistency and works
reimar
parents:
6384
diff
changeset
|
540 asm volatile ("emms;":::"memory"); |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
541 } |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
542 |
936 | 543 |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
544 #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
|
545 {\ |
4197 | 546 if (mm_flags & MM_MMX)\ |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
547 emms();\ |
0 | 548 } |
549 | |
1092 | 550 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx); |
1065 | 551 |
62 | 552 #elif defined(ARCH_ARMV4L) |
553 | |
2776
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
554 #define MM_IWMMXT 0x0100 /* XScale IWMMXT */ |
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
555 |
4197 | 556 extern int mm_flags; |
557 | |
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
|
558 #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
|
559 |
894
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
560 #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
|
561 |
4197 | 562 extern int mm_flags; |
563 | |
4999
0c4bf6b7d1c6
use DECLARE_ALIGNED from mem.h (which is included through common.h)
alex
parents:
4988
diff
changeset
|
564 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v) |
2324 | 565 #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
|
566 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
567 #elif defined(HAVE_MMI) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
568 |
4999
0c4bf6b7d1c6
use DECLARE_ALIGNED from mem.h (which is included through common.h)
alex
parents:
4988
diff
changeset
|
569 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v) |
2324 | 570 #define STRIDE_ALIGN 16 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
571 |
6363
7ebd1cdb2142
clean up definition of DECLARE_ALIGNED_8 and STRIDE_ALIGN
mru
parents:
6241
diff
changeset
|
572 #endif |
3728 | 573 |
6363
7ebd1cdb2142
clean up definition of DECLARE_ALIGNED_8 and STRIDE_ALIGN
mru
parents:
6241
diff
changeset
|
574 #ifndef DECLARE_ALIGNED_8 |
7ebd1cdb2142
clean up definition of DECLARE_ALIGNED_8 and STRIDE_ALIGN
mru
parents:
6241
diff
changeset
|
575 # define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v) |
7ebd1cdb2142
clean up definition of DECLARE_ALIGNED_8 and STRIDE_ALIGN
mru
parents:
6241
diff
changeset
|
576 #endif |
3728 | 577 |
6363
7ebd1cdb2142
clean up definition of DECLARE_ALIGNED_8 and STRIDE_ALIGN
mru
parents:
6241
diff
changeset
|
578 #ifndef STRIDE_ALIGN |
7ebd1cdb2142
clean up definition of DECLARE_ALIGNED_8 and STRIDE_ALIGN
mru
parents:
6241
diff
changeset
|
579 # define STRIDE_ALIGN 8 |
0 | 580 #endif |
581 | |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
582 /* PSNR */ |
1064 | 583 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
|
584 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
|
585 AVCodecContext *avctx); |
781 | 586 |
587 /* FFT computation */ | |
588 | |
589 /* NOTE: soon integer code will be added, so you must use the | |
590 FFTSample type */ | |
591 typedef float FFTSample; | |
592 | |
3555 | 593 struct MDCTContext; |
594 | |
781 | 595 typedef struct FFTComplex { |
596 FFTSample re, im; | |
597 } FFTComplex; | |
598 | |
599 typedef struct FFTContext { | |
600 int nbits; | |
601 int inverse; | |
602 uint16_t *revtab; | |
603 FFTComplex *exptab; | |
604 FFTComplex *exptab1; /* only used by SSE code */ | |
605 void (*fft_calc)(struct FFTContext *s, FFTComplex *z); | |
3555 | 606 void (*imdct_calc)(struct MDCTContext *s, FFTSample *output, |
607 const FFTSample *input, FFTSample *tmp); | |
781 | 608 } FFTContext; |
609 | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
610 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
|
611 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
|
612 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
|
613 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z); |
3175 | 614 void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z); |
615 void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z); | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
616 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
|
617 |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
618 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z) |
781 | 619 { |
620 s->fft_calc(s, z); | |
621 } | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
622 void ff_fft_end(FFTContext *s); |
781 | 623 |
624 /* MDCT computation */ | |
625 | |
626 typedef struct MDCTContext { | |
627 int n; /* size of MDCT (i.e. number of input data * 2) */ | |
628 int nbits; /* n = 2^nbits */ | |
629 /* pre/post rotation tables */ | |
630 FFTSample *tcos; | |
631 FFTSample *tsin; | |
632 FFTContext fft; | |
633 } MDCTContext; | |
634 | |
6139
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
6056
diff
changeset
|
635 /** |
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
6056
diff
changeset
|
636 * Generate a Kaiser-Bessel Derived Window. |
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
6056
diff
changeset
|
637 * @param window pointer to half window |
6142
a35b838ab955
Add variable alpha and size of half window for Kaiser-Bessel Derived window
superdump
parents:
6139
diff
changeset
|
638 * @param alpha determines window shape |
a35b838ab955
Add variable alpha and size of half window for Kaiser-Bessel Derived window
superdump
parents:
6139
diff
changeset
|
639 * @param n size of half window |
6139
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
6056
diff
changeset
|
640 */ |
6142
a35b838ab955
Add variable alpha and size of half window for Kaiser-Bessel Derived window
superdump
parents:
6139
diff
changeset
|
641 void ff_kbd_window_init(float *window, float alpha, int n); |
6139
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
6056
diff
changeset
|
642 |
794 | 643 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
|
644 void ff_imdct_calc(MDCTContext *s, FFTSample *output, |
781 | 645 const FFTSample *input, FFTSample *tmp); |
3555 | 646 void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output, |
647 const FFTSample *input, FFTSample *tmp); | |
3746 | 648 void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output, |
649 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
|
650 void ff_mdct_calc(MDCTContext *s, FFTSample *out, |
781 | 651 const FFTSample *input, FFTSample *tmp); |
794 | 652 void ff_mdct_end(MDCTContext *s); |
781 | 653 |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
5997
diff
changeset
|
654 #define WRAPPER8_16(name8, name16)\ |
1708 | 655 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ |
656 return name8(s, dst , src , stride, h)\ | |
657 +name8(s, dst+8 , src+8 , stride, h);\ | |
658 } | |
659 | |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
5997
diff
changeset
|
660 #define WRAPPER8_16_SQ(name8, name16)\ |
1708 | 661 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ |
662 int score=0;\ | |
663 score +=name8(s, dst , src , stride, 8);\ | |
664 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
665 if(h==16){\ | |
666 dst += 8*stride;\ | |
667 src += 8*stride;\ | |
668 score +=name8(s, dst , src , stride, 8);\ | |
669 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
670 }\ | |
671 return score;\ | |
936 | 672 } |
673 | |
4240 | 674 |
675 static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
676 { | |
677 int i; | |
678 for(i=0; i<h; i++) | |
679 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
680 AV_WN16(dst , AV_RN16(src )); |
4240 | 681 dst+=dstStride; |
682 src+=srcStride; | |
683 } | |
684 } | |
685 | |
686 static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
687 { | |
688 int i; | |
689 for(i=0; i<h; i++) | |
690 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
691 AV_WN32(dst , AV_RN32(src )); |
4240 | 692 dst+=dstStride; |
693 src+=srcStride; | |
694 } | |
695 } | |
696 | |
697 static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
698 { | |
699 int i; | |
700 for(i=0; i<h; i++) | |
701 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
702 AV_WN32(dst , AV_RN32(src )); |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
703 AV_WN32(dst+4 , AV_RN32(src+4 )); |
4240 | 704 dst+=dstStride; |
705 src+=srcStride; | |
706 } | |
707 } | |
708 | |
709 static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
710 { | |
711 int i; | |
712 for(i=0; i<h; i++) | |
713 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
714 AV_WN32(dst , AV_RN32(src )); |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
715 AV_WN32(dst+4 , AV_RN32(src+4 )); |
4240 | 716 dst[8]= src[8]; |
717 dst+=dstStride; | |
718 src+=srcStride; | |
719 } | |
720 } | |
721 | |
722 static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
723 { | |
724 int i; | |
725 for(i=0; i<h; i++) | |
726 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
727 AV_WN32(dst , AV_RN32(src )); |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
728 AV_WN32(dst+4 , AV_RN32(src+4 )); |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
729 AV_WN32(dst+8 , AV_RN32(src+8 )); |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
730 AV_WN32(dst+12, AV_RN32(src+12)); |
4240 | 731 dst+=dstStride; |
732 src+=srcStride; | |
733 } | |
734 } | |
735 | |
736 static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
737 { | |
738 int i; | |
739 for(i=0; i<h; i++) | |
740 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
741 AV_WN32(dst , AV_RN32(src )); |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
742 AV_WN32(dst+4 , AV_RN32(src+4 )); |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
743 AV_WN32(dst+8 , AV_RN32(src+8 )); |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
744 AV_WN32(dst+12, AV_RN32(src+12)); |
4240 | 745 dst[16]= src[16]; |
746 dst+=dstStride; | |
747 src+=srcStride; | |
748 } | |
749 } | |
750 | |
5830
1d83e9c34641
Add FFMPEG_ prefix to all multiple inclusion guards.
diego
parents:
5750
diff
changeset
|
751 #endif /* FFMPEG_DSPUTIL_H */ |