Mercurial > libavcodec.hg
annotate dsputil.h @ 3715:cdaee53c5da1 libavcodec
replaces back some C code by dsp.put_pixels_tab[0][0] (revert r6231)
but this time ensure proper alignment using av_malloc instead of av_realloc
author | aurel |
---|---|
date | Wed, 13 Sep 2006 20:16:33 +0000 |
parents | 9d942506b1f2 |
children | 4d346abd723c |
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 | |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
3029
diff
changeset
|
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
429 | 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; | |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
38 typedef int DWTELEM; |
0 | 39 |
474
11dbd00682fc
avoid name clash with libjpeg - added missing externs
bellard
parents:
429
diff
changeset
|
40 void fdct_ifast (DCTELEM *data); |
1571
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1567
diff
changeset
|
41 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
|
42 void ff_jpeg_fdct_islow (DCTELEM *data); |
1567 | 43 void ff_fdct248_islow (DCTELEM *data); |
0 | 44 |
45 void j_rev_dct (DCTELEM *data); | |
2256 | 46 void j_rev_dct4 (DCTELEM *data); |
2257 | 47 void j_rev_dct2 (DCTELEM *data); |
2259 | 48 void j_rev_dct1 (DCTELEM *data); |
0 | 49 |
687
9abb13c21fbe
fdct_mmx -> ff_fdct_mmx (renamed to avoid namespace conflict with xvid)
arpi_esp
parents:
675
diff
changeset
|
50 void ff_fdct_mmx(DCTELEM *block); |
1565 | 51 void ff_fdct_mmx2(DCTELEM *block); |
1765
e31754bc5b65
SSE2 fdct by (Balatoni Denes <pnis at coder dot hu>)
michael
parents:
1739
diff
changeset
|
52 void ff_fdct_sse2(DCTELEM *block); |
0 | 53 |
2755 | 54 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
|
55 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
|
56 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
|
57 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
|
58 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
|
59 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
|
60 |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
61 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
|
62 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
|
63 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
|
64 |
34 | 65 /* encoding scans */ |
1064 | 66 extern const uint8_t ff_alternate_horizontal_scan[64]; |
67 extern const uint8_t ff_alternate_vertical_scan[64]; | |
68 extern const uint8_t ff_zigzag_direct[64]; | |
1567 | 69 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
|
70 |
0 | 71 /* pixel operations */ |
2090 | 72 #define MAX_NEG_CROP 1024 |
0 | 73 |
74 /* temporary */ | |
1064 | 75 extern uint32_t squareTbl[512]; |
76 extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP]; | |
0 | 77 |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
78 /* VP3 DSP functions */ |
2693 | 79 void ff_vp3_idct_c(DCTELEM *block/* align 16*/); |
80 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); | |
81 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); | |
0 | 82 |
3245 | 83 /* 1/2^n downscaling functions from imgconvert.c */ |
84 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
85 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
86 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
87 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
88 | |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
89 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
|
90 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
|
91 |
675 | 92 /* minimum alignment rules ;) |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
93 if u notice errors in the align stuff, need more alignment for some asm code for some cpu |
675 | 94 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ... |
95 | |
96 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible) | |
97 i (michael) didnt check them, these are just the alignents which i think could be reached easily ... | |
98 | |
99 !future video codecs might need functions with less strict alignment | |
100 */ | |
0 | 101 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
102 /* |
1064 | 103 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size); |
104 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride); | |
105 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
106 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
296 | 107 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
|
108 */ |
0 | 109 |
110 /* add and put pixel (decoding) */ | |
675 | 111 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 |
1709 | 112 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4 |
1064 | 113 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
|
114 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 | 115 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); |
1168 | 116 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 | 117 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset); |
3029 | 118 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset); |
0 | 119 |
984 | 120 #define DEF_OLD_QPEL(name)\ |
1064 | 121 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ |
122 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ | |
123 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
|
124 |
984 | 125 DEF_OLD_QPEL(qpel16_mc11_old_c) |
126 DEF_OLD_QPEL(qpel16_mc31_old_c) | |
127 DEF_OLD_QPEL(qpel16_mc12_old_c) | |
128 DEF_OLD_QPEL(qpel16_mc32_old_c) | |
129 DEF_OLD_QPEL(qpel16_mc13_old_c) | |
130 DEF_OLD_QPEL(qpel16_mc33_old_c) | |
131 DEF_OLD_QPEL(qpel8_mc11_old_c) | |
132 DEF_OLD_QPEL(qpel8_mc31_old_c) | |
133 DEF_OLD_QPEL(qpel8_mc12_old_c) | |
134 DEF_OLD_QPEL(qpel8_mc32_old_c) | |
135 DEF_OLD_QPEL(qpel8_mc13_old_c) | |
136 DEF_OLD_QPEL(qpel8_mc33_old_c) | |
651 | 137 |
138 #define CALL_2X_PIXELS(a, b, n)\ | |
139 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ | |
140 b(block , pixels , line_size, h);\ | |
141 b(block+n, pixels+n, line_size, h);\ | |
142 } | |
255 | 143 |
0 | 144 /* motion estimation */ |
1709 | 145 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2 |
146 // allthough currently h<4 is not used as functions with width <8 are not used and neither implemented | |
1708 | 147 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 | 148 |
1168 | 149 |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
150 // for snow slices |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
151 typedef struct slice_buffer_s slice_buffer; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
152 |
1102 | 153 /** |
154 * DSPContext. | |
155 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
156 typedef struct DSPContext { |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
157 /* pixel ops : interface with DCT */ |
1064 | 158 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); |
159 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); | |
160 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
|
161 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
1064 | 162 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
2763 | 163 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size); |
164 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size); | |
1136 | 165 /** |
166 * translational global motion compensation. | |
167 */ | |
1064 | 168 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); |
1136 | 169 /** |
170 * global motion compensation. | |
171 */ | |
1064 | 172 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy, |
2979 | 173 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
|
174 void (*clear_blocks)(DCTELEM *blocks/*align 16*/); |
1064 | 175 int (*pix_sum)(uint8_t * pix, int line_size); |
176 int (*pix_norm1)(uint8_t * pix, int line_size); | |
1708 | 177 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4 |
2967 | 178 |
1729 | 179 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */ |
180 me_cmp_func sse[5]; | |
181 me_cmp_func hadamard8_diff[5]; | |
182 me_cmp_func dct_sad[5]; | |
183 me_cmp_func quant_psnr[5]; | |
184 me_cmp_func bit[5]; | |
185 me_cmp_func rd[5]; | |
186 me_cmp_func vsad[5]; | |
187 me_cmp_func vsse[5]; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
188 me_cmp_func nsse[5]; |
2184 | 189 me_cmp_func w53[5]; |
190 me_cmp_func w97[5]; | |
2382 | 191 me_cmp_func dct_max[5]; |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
192 me_cmp_func dct264_sad[5]; |
936 | 193 |
1708 | 194 me_cmp_func me_pre_cmp[5]; |
195 me_cmp_func me_cmp[5]; | |
196 me_cmp_func me_sub_cmp[5]; | |
197 me_cmp_func mb_cmp[5]; | |
1729 | 198 me_cmp_func ildct_cmp[5]; //only width 16 used |
2382 | 199 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
|
200 |
1136 | 201 /** |
202 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
2967 | 203 * this is an array[4][4] of motion compensation funcions for 4 |
1713 | 204 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 205 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 206 * @param block destination where the result is stored |
207 * @param pixels source | |
208 * @param line_size number of bytes in a horizontal line of block | |
209 * @param h height | |
210 */ | |
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
|
211 op_pixels_func put_pixels_tab[4][4]; |
1136 | 212 |
213 /** | |
214 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
2967 | 215 * This is an array[4][4] of motion compensation functions for 4 |
1713 | 216 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 217 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 218 * @param block destination into which the result is averaged (a+b+1)>>1 |
219 * @param pixels source | |
220 * @param line_size number of bytes in a horizontal line of block | |
221 * @param h height | |
222 */ | |
1319 | 223 op_pixels_func avg_pixels_tab[4][4]; |
1136 | 224 |
225 /** | |
226 * Halfpel motion compensation with no rounding (a+b)>>1. | |
2967 | 227 * this is an array[2][4] of motion compensation funcions for 2 |
1225 | 228 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 229 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 230 * @param block destination where the result is stored |
231 * @param pixels source | |
232 * @param line_size number of bytes in a horizontal line of block | |
233 * @param h height | |
234 */ | |
2075 | 235 op_pixels_func put_no_rnd_pixels_tab[4][4]; |
1136 | 236 |
237 /** | |
238 * Halfpel motion compensation with no rounding (a+b)>>1. | |
2967 | 239 * this is an array[2][4] of motion compensation funcions for 2 |
1225 | 240 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 241 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 242 * @param block destination into which the result is averaged (a+b)>>1 |
243 * @param pixels source | |
244 * @param line_size number of bytes in a horizontal line of block | |
245 * @param h height | |
246 */ | |
2075 | 247 op_pixels_func avg_no_rnd_pixels_tab[4][4]; |
2967 | 248 |
1864 | 249 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 | 250 |
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
|
251 /** |
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
|
252 * 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
|
253 * 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
|
254 * *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
|
255 * @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
|
256 * @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
|
257 * @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
|
258 * @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
|
259 */ |
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
|
260 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
1319 | 261 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
262 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
263 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
|
264 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
|
265 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
|
266 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; |
936 | 267 qpel_mc_func put_mspel_pixels_tab[8]; |
2967 | 268 |
1168 | 269 /** |
270 * h264 Chram MC | |
271 */ | |
272 h264_chroma_mc_func put_h264_chroma_pixels_tab[3]; | |
3663 | 273 /* This is really one func used in VC-1 decoding */ |
274 h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3]; | |
1168 | 275 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
|
276 |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3010
diff
changeset
|
277 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
|
278 qpel_mc_func avg_h264_qpel_pixels_tab[4][16]; |
2967 | 279 |
2415 | 280 h264_weight_func weight_h264_pixels_tab[10]; |
281 h264_biweight_func biweight_h264_pixels_tab[10]; | |
2967 | 282 |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
283 /* AVS specific */ |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
284 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
|
285 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
|
286 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
|
287 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
|
288 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
|
289 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
|
290 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
|
291 |
1708 | 292 me_cmp_func pix_abs[2][4]; |
2967 | 293 |
866 | 294 /* huffyuv specific */ |
295 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); | |
936 | 296 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); |
1527 | 297 /** |
298 * subtract huffyuv's variant of median prediction | |
299 * note, this might read from src1[-1], src2[-1] | |
300 */ | |
301 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top); | |
1273 | 302 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w); |
2633 | 303 |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
304 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
|
305 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
|
306 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
|
307 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
|
308 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
|
309 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
|
310 // 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
|
311 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
|
312 int bidir, int edges, int step, int mask_mv0, int mask_mv1); |
2967 | 313 |
1644 | 314 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale); |
315 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale); | |
316 | |
2045 | 317 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
|
318 |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
319 /* 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
|
320 void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize); |
3574 | 321 /* 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
|
322 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
|
323 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
|
324 /* 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
|
325 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
|
326 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
327 /* 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
|
328 * 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
|
329 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
|
330 |
1092 | 331 /* (I)DCT */ |
332 void (*fdct)(DCTELEM *block/* align 16*/); | |
1567 | 333 void (*fdct248)(DCTELEM *block/* align 16*/); |
2967 | 334 |
1324
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
335 /* IDCT really*/ |
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
336 void (*idct)(DCTELEM *block/* align 16*/); |
2967 | 337 |
1102 | 338 /** |
1104 | 339 * block -> idct -> clip to unsigned 8 bit -> dest. |
1102 | 340 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) |
1111 | 341 * @param line_size size in bytes of a horizotal line of dest |
1102 | 342 */ |
1092 | 343 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
2967 | 344 |
1102 | 345 /** |
346 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. | |
1111 | 347 * @param line_size size in bytes of a horizotal line of dest |
1102 | 348 */ |
1092 | 349 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
2967 | 350 |
1102 | 351 /** |
1104 | 352 * idct input permutation. |
1241 | 353 * several optimized IDCTs need a permutated input (relative to the normal order of the reference |
354 * IDCT) | |
355 * this permutation must be performed before the idct_put/add, note, normally this can be merged | |
356 * with the zigzag/alternate scan<br> | |
1102 | 357 * an example to avoid confusion: |
358 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...) | |
359 * - (x -> referece dct -> reference idct -> x) | |
360 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x) | |
361 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...) | |
362 */ | |
1092 | 363 uint8_t idct_permutation[64]; |
364 int idct_permutation_type; | |
365 #define FF_NO_IDCT_PERM 1 | |
366 #define FF_LIBMPEG2_IDCT_PERM 2 | |
367 #define FF_SIMPLE_IDCT_PERM 3 | |
368 #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
|
369 #define FF_PARTTRANS_IDCT_PERM 5 |
1092 | 370 |
1784 | 371 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale); |
372 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); | |
373 #define BASIS_SHIFT 16 | |
374 #define RECON_SHIFT 6 | |
2967 | 375 |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
376 void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride); |
2755 | 377 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
|
378 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
|
379 void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride); |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
380 |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
381 /* snow wavelet */ |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
382 void (*vertical_compose97i)(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width); |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
383 void (*horizontal_compose97i)(DWTELEM *b, int width); |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
384 void (*inner_add_yblock)(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
|
385 |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3198
diff
changeset
|
386 void (*prefetch)(void *mem, int stride, int h); |
3245 | 387 |
388 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
3526 | 389 |
390 /* vc1 functions */ | |
391 void (*vc1_inv_trans_8x8)(DCTELEM *b); | |
392 void (*vc1_inv_trans_8x4)(DCTELEM *b, int n); | |
393 void (*vc1_inv_trans_4x8)(DCTELEM *b, int n); | |
394 void (*vc1_inv_trans_4x4)(DCTELEM *b, int n); | |
395 void (*vc1_v_overlap)(uint8_t* src, int stride, int rnd); | |
396 void (*vc1_h_overlap)(uint8_t* src, int stride, int rnd); | |
397 /* put 8x8 block with bicubic interpolation and quarterpel precision | |
398 * last argument is actually round value instead of height | |
399 */ | |
400 op_pixels_func put_vc1_mspel_pixels_tab[16]; | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
401 } DSPContext; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
402 |
1201 | 403 void dsputil_static_init(void); |
1092 | 404 void dsputil_init(DSPContext* p, AVCodecContext *avctx); |
0 | 405 |
764 | 406 /** |
407 * permute block according to permuatation. | |
408 * @param last last non zero element in scantable order | |
409 */ | |
1064 | 410 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); |
34 | 411 |
1729 | 412 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type); |
413 | |
2979 | 414 #define BYTE_VEC32(c) ((c)*0x01010101UL) |
1264 | 415 |
416 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b) | |
417 { | |
418 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
419 } | |
420 | |
421 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b) | |
422 { | |
423 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
424 } | |
425 | |
2184 | 426 static inline int get_penalty_factor(int lambda, int lambda2, int type){ |
427 switch(type&0xFF){ | |
428 default: | |
429 case FF_CMP_SAD: | |
430 return lambda>>FF_LAMBDA_SHIFT; | |
431 case FF_CMP_DCT: | |
432 return (3*lambda)>>(FF_LAMBDA_SHIFT+1); | |
433 case FF_CMP_W53: | |
434 return (4*lambda)>>(FF_LAMBDA_SHIFT); | |
435 case FF_CMP_W97: | |
436 return (2*lambda)>>(FF_LAMBDA_SHIFT); | |
437 case FF_CMP_SATD: | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
438 case FF_CMP_DCT264: |
2184 | 439 return (2*lambda)>>FF_LAMBDA_SHIFT; |
440 case FF_CMP_RD: | |
441 case FF_CMP_PSNR: | |
442 case FF_CMP_SSE: | |
443 case FF_CMP_NSSE: | |
444 return lambda2>>FF_LAMBDA_SHIFT; | |
445 case FF_CMP_BIT: | |
446 return 1; | |
447 } | |
448 } | |
449 | |
1102 | 450 /** |
1104 | 451 * Empty mmx state. |
1102 | 452 * this must be called between any dsp function and float/double code. |
453 * for example sin(); dsp->idct_put(); emms_c(); cos() | |
454 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
455 #define emms_c() |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
456 |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
457 /* should be defined by architectures supporting |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
458 one or more MultiMedia extension */ |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
459 int mm_support(void); |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
460 |
3089 | 461 #ifdef __GNUC__ |
462 #define DECLARE_ALIGNED_16(t,v) t v __attribute__ ((aligned (16))) | |
463 #else | |
464 #define DECLARE_ALIGNED_16(t,v) __declspec(align(16)) t v | |
465 #endif | |
1974
8c5489b2cf3e
move __align16 some place where non-MMX machines can see it
melanson
parents:
1972
diff
changeset
|
466 |
62 | 467 #if defined(HAVE_MMX) |
0 | 468 |
862 | 469 #undef emms_c |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
470 |
0 | 471 #define MM_MMX 0x0001 /* standard MMX */ |
472 #define MM_3DNOW 0x0004 /* AMD 3DNOW */ | |
473 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */ | |
474 #define MM_SSE 0x0008 /* SSE functions */ | |
475 #define MM_SSE2 0x0010 /* PIV SSE2 functions */ | |
2388 | 476 #define MM_3DNOWEXT 0x0020 /* AMD 3DNowExt */ |
3279
647a677c00a4
Remove unused and unsupported Cyrix's "Extended MMX",
gpoirier
parents:
3248
diff
changeset
|
477 #define MM_SSE3 0x0040 /* Prescott SSE3 functions */ |
0 | 478 |
479 extern int mm_flags; | |
480 | |
1064 | 481 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
482 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
|
483 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
0 | 484 |
485 static inline void emms(void) | |
486 { | |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
487 __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
|
488 } |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
489 |
936 | 490 |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
491 #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
|
492 {\ |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
493 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
|
494 emms();\ |
0 | 495 } |
496 | |
3089 | 497 #ifdef __GNUC__ |
498 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) | |
499 #else | |
500 #define DECLARE_ALIGNED_8(t,v) __declspec(align(8)) t v | |
501 #endif | |
502 | |
2324 | 503 #define STRIDE_ALIGN 8 |
0 | 504 |
1092 | 505 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); |
506 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx); | |
1065 | 507 |
62 | 508 #elif defined(ARCH_ARMV4L) |
509 | |
510 /* 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
|
511 line optimizations */ |
3089 | 512 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (4))) |
2324 | 513 #define STRIDE_ALIGN 4 |
62 | 514 |
2776
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
515 #define MM_IWMMXT 0x0100 /* XScale IWMMXT */ |
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
516 |
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
517 extern int mm_flags; |
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
518 |
1092 | 519 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx); |
62 | 520 |
88 | 521 #elif defined(HAVE_MLIB) |
522 | |
523 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ | |
3089 | 524 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 525 #define STRIDE_ALIGN 8 |
88 | 526 |
1092 | 527 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx); |
88 | 528 |
1959
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
529 #elif defined(ARCH_SPARC) |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
530 |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
531 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ |
3089 | 532 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 533 #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
|
534 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
|
535 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
536 #elif defined(ARCH_ALPHA) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
537 |
3089 | 538 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 539 #define STRIDE_ALIGN 8 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
540 |
1092 | 541 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
|
542 |
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
|
543 #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
|
544 |
894
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
545 #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
|
546 |
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
547 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
|
548 |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
549 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN) |
1653 | 550 #define pixel altivec_pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
551 #include <altivec.h> |
1653 | 552 #undef pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
553 #endif |
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
554 |
3089 | 555 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16))) |
2324 | 556 #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
|
557 |
1092 | 558 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
|
559 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
560 #elif defined(HAVE_MMI) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
561 |
3089 | 562 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16))) |
2324 | 563 #define STRIDE_ALIGN 16 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
564 |
1092 | 565 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
|
566 |
1259
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
567 #elif defined(ARCH_SH4) |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
568 |
3089 | 569 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 570 #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
|
571 |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
572 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
|
573 |
0 | 574 #else |
575 | |
3089 | 576 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 577 #define STRIDE_ALIGN 8 |
0 | 578 |
579 #endif | |
580 | |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
581 #ifdef __GNUC__ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
582 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
583 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
|
584 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
|
585 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
|
586 |
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
|
587 #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
|
588 #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
|
589 #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
|
590 |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3010
diff
changeset
|
591 #define ST16(a, b) (((struct unaligned_16 *) (a))->l) = (b) |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
592 #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
|
593 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
594 #else /* __GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
595 |
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
|
596 #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
|
597 #define LD32(a) (*((uint32_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
598 #define LD64(a) (*((uint64_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
599 |
3217 | 600 #define ST16(a, b) *((uint16_t*)(a)) = (b) |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
601 #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
|
602 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
603 #endif /* !__GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
604 |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
605 /* PSNR */ |
1064 | 606 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
|
607 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
|
608 AVCodecContext *avctx); |
781 | 609 |
610 /* FFT computation */ | |
611 | |
612 /* NOTE: soon integer code will be added, so you must use the | |
613 FFTSample type */ | |
614 typedef float FFTSample; | |
615 | |
3555 | 616 struct MDCTContext; |
617 | |
781 | 618 typedef struct FFTComplex { |
619 FFTSample re, im; | |
620 } FFTComplex; | |
621 | |
622 typedef struct FFTContext { | |
623 int nbits; | |
624 int inverse; | |
625 uint16_t *revtab; | |
626 FFTComplex *exptab; | |
627 FFTComplex *exptab1; /* only used by SSE code */ | |
628 void (*fft_calc)(struct FFTContext *s, FFTComplex *z); | |
3555 | 629 void (*imdct_calc)(struct MDCTContext *s, FFTSample *output, |
630 const FFTSample *input, FFTSample *tmp); | |
781 | 631 } FFTContext; |
632 | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
633 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
|
634 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
|
635 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
|
636 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z); |
3175 | 637 void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z); |
638 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
|
639 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
|
640 |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
641 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z) |
781 | 642 { |
643 s->fft_calc(s, z); | |
644 } | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
645 void ff_fft_end(FFTContext *s); |
781 | 646 |
647 /* MDCT computation */ | |
648 | |
649 typedef struct MDCTContext { | |
650 int n; /* size of MDCT (i.e. number of input data * 2) */ | |
651 int nbits; /* n = 2^nbits */ | |
652 /* pre/post rotation tables */ | |
653 FFTSample *tcos; | |
654 FFTSample *tsin; | |
655 FFTContext fft; | |
656 } MDCTContext; | |
657 | |
794 | 658 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
|
659 void ff_imdct_calc(MDCTContext *s, FFTSample *output, |
781 | 660 const FFTSample *input, FFTSample *tmp); |
3555 | 661 void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output, |
662 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
|
663 void ff_mdct_calc(MDCTContext *s, FFTSample *out, |
781 | 664 const FFTSample *input, FFTSample *tmp); |
794 | 665 void ff_mdct_end(MDCTContext *s); |
781 | 666 |
1708 | 667 #define WARPER8_16(name8, name16)\ |
668 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
669 return name8(s, dst , src , stride, h)\ | |
670 +name8(s, dst+8 , src+8 , stride, h);\ | |
671 } | |
672 | |
673 #define WARPER8_16_SQ(name8, name16)\ | |
674 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
675 int score=0;\ | |
676 score +=name8(s, dst , src , stride, 8);\ | |
677 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
678 if(h==16){\ | |
679 dst += 8*stride;\ | |
680 src += 8*stride;\ | |
681 score +=name8(s, dst , src , stride, 8);\ | |
682 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
683 }\ | |
684 return score;\ | |
936 | 685 } |
686 | |
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
|
687 #endif |