Mercurial > libavcodec.hg
annotate dsputil.h @ 3247:ac7bfc9bf1a4 libavcodec
2x faster ff_mpeg4_set_direct_mv
author | lorenm |
---|---|
date | Mon, 03 Apr 2006 07:52:24 +0000 |
parents | c2c29be6282e |
children | 7aa9f80e7954 |
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 |
34 | 61 /* encoding scans */ |
1064 | 62 extern const uint8_t ff_alternate_horizontal_scan[64]; |
63 extern const uint8_t ff_alternate_vertical_scan[64]; | |
64 extern const uint8_t ff_zigzag_direct[64]; | |
1567 | 65 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
|
66 |
0 | 67 /* pixel operations */ |
2090 | 68 #define MAX_NEG_CROP 1024 |
0 | 69 |
70 /* temporary */ | |
1064 | 71 extern uint32_t squareTbl[512]; |
72 extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP]; | |
0 | 73 |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
74 /* VP3 DSP functions */ |
2693 | 75 void ff_vp3_idct_c(DCTELEM *block/* align 16*/); |
76 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); | |
77 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); | |
0 | 78 |
3245 | 79 /* 1/2^n downscaling functions from imgconvert.c */ |
80 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
81 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
82 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
83 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
84 | |
675 | 85 /* minimum alignment rules ;) |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
86 if u notice errors in the align stuff, need more alignment for some asm code for some cpu |
675 | 87 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ... |
88 | |
89 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible) | |
90 i (michael) didnt check them, these are just the alignents which i think could be reached easily ... | |
91 | |
92 !future video codecs might need functions with less strict alignment | |
93 */ | |
0 | 94 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
95 /* |
1064 | 96 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size); |
97 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride); | |
98 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
99 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
296 | 100 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
|
101 */ |
0 | 102 |
103 /* add and put pixel (decoding) */ | |
675 | 104 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 |
1709 | 105 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4 |
1064 | 106 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
|
107 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 | 108 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); |
1168 | 109 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 | 110 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset); |
3029 | 111 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset); |
0 | 112 |
984 | 113 #define DEF_OLD_QPEL(name)\ |
1064 | 114 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ |
115 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ | |
116 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
|
117 |
984 | 118 DEF_OLD_QPEL(qpel16_mc11_old_c) |
119 DEF_OLD_QPEL(qpel16_mc31_old_c) | |
120 DEF_OLD_QPEL(qpel16_mc12_old_c) | |
121 DEF_OLD_QPEL(qpel16_mc32_old_c) | |
122 DEF_OLD_QPEL(qpel16_mc13_old_c) | |
123 DEF_OLD_QPEL(qpel16_mc33_old_c) | |
124 DEF_OLD_QPEL(qpel8_mc11_old_c) | |
125 DEF_OLD_QPEL(qpel8_mc31_old_c) | |
126 DEF_OLD_QPEL(qpel8_mc12_old_c) | |
127 DEF_OLD_QPEL(qpel8_mc32_old_c) | |
128 DEF_OLD_QPEL(qpel8_mc13_old_c) | |
129 DEF_OLD_QPEL(qpel8_mc33_old_c) | |
651 | 130 |
131 #define CALL_2X_PIXELS(a, b, n)\ | |
132 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ | |
133 b(block , pixels , line_size, h);\ | |
134 b(block+n, pixels+n, line_size, h);\ | |
135 } | |
255 | 136 |
0 | 137 /* motion estimation */ |
1709 | 138 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2 |
139 // allthough currently h<4 is not used as functions with width <8 are not used and neither implemented | |
1708 | 140 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 | 141 |
1168 | 142 |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
143 // for snow slices |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
144 typedef struct slice_buffer_s slice_buffer; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
145 |
1102 | 146 /** |
147 * DSPContext. | |
148 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
149 typedef struct DSPContext { |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
150 /* pixel ops : interface with DCT */ |
1064 | 151 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); |
152 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); | |
153 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
|
154 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
1064 | 155 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
2763 | 156 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size); |
157 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size); | |
1136 | 158 /** |
159 * translational global motion compensation. | |
160 */ | |
1064 | 161 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); |
1136 | 162 /** |
163 * global motion compensation. | |
164 */ | |
1064 | 165 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy, |
2979 | 166 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
|
167 void (*clear_blocks)(DCTELEM *blocks/*align 16*/); |
1064 | 168 int (*pix_sum)(uint8_t * pix, int line_size); |
169 int (*pix_norm1)(uint8_t * pix, int line_size); | |
1708 | 170 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4 |
2967 | 171 |
1729 | 172 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */ |
173 me_cmp_func sse[5]; | |
174 me_cmp_func hadamard8_diff[5]; | |
175 me_cmp_func dct_sad[5]; | |
176 me_cmp_func quant_psnr[5]; | |
177 me_cmp_func bit[5]; | |
178 me_cmp_func rd[5]; | |
179 me_cmp_func vsad[5]; | |
180 me_cmp_func vsse[5]; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
181 me_cmp_func nsse[5]; |
2184 | 182 me_cmp_func w53[5]; |
183 me_cmp_func w97[5]; | |
2382 | 184 me_cmp_func dct_max[5]; |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
185 me_cmp_func dct264_sad[5]; |
936 | 186 |
1708 | 187 me_cmp_func me_pre_cmp[5]; |
188 me_cmp_func me_cmp[5]; | |
189 me_cmp_func me_sub_cmp[5]; | |
190 me_cmp_func mb_cmp[5]; | |
1729 | 191 me_cmp_func ildct_cmp[5]; //only width 16 used |
2382 | 192 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
|
193 |
1136 | 194 /** |
195 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
2967 | 196 * this is an array[4][4] of motion compensation funcions for 4 |
1713 | 197 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 198 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 199 * @param block destination where the result is stored |
200 * @param pixels source | |
201 * @param line_size number of bytes in a horizontal line of block | |
202 * @param h height | |
203 */ | |
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
|
204 op_pixels_func put_pixels_tab[4][4]; |
1136 | 205 |
206 /** | |
207 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
2967 | 208 * This is an array[4][4] of motion compensation functions for 4 |
1713 | 209 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 210 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 211 * @param block destination into which the result is averaged (a+b+1)>>1 |
212 * @param pixels source | |
213 * @param line_size number of bytes in a horizontal line of block | |
214 * @param h height | |
215 */ | |
1319 | 216 op_pixels_func avg_pixels_tab[4][4]; |
1136 | 217 |
218 /** | |
219 * Halfpel motion compensation with no rounding (a+b)>>1. | |
2967 | 220 * this is an array[2][4] of motion compensation funcions for 2 |
1225 | 221 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 222 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 223 * @param block destination where the result is stored |
224 * @param pixels source | |
225 * @param line_size number of bytes in a horizontal line of block | |
226 * @param h height | |
227 */ | |
2075 | 228 op_pixels_func put_no_rnd_pixels_tab[4][4]; |
1136 | 229 |
230 /** | |
231 * Halfpel motion compensation with no rounding (a+b)>>1. | |
2967 | 232 * this is an array[2][4] of motion compensation funcions for 2 |
1225 | 233 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 234 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 235 * @param block destination into which the result is averaged (a+b)>>1 |
236 * @param pixels source | |
237 * @param line_size number of bytes in a horizontal line of block | |
238 * @param h height | |
239 */ | |
2075 | 240 op_pixels_func avg_no_rnd_pixels_tab[4][4]; |
2967 | 241 |
1864 | 242 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 | 243 |
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
|
244 /** |
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
|
245 * 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
|
246 * 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
|
247 * *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
|
248 * @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
|
249 * @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
|
250 * @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
|
251 * @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
|
252 */ |
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 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
1319 | 254 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
255 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
256 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
|
257 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
|
258 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
|
259 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; |
936 | 260 qpel_mc_func put_mspel_pixels_tab[8]; |
2967 | 261 |
1168 | 262 /** |
263 * h264 Chram MC | |
264 */ | |
265 h264_chroma_mc_func put_h264_chroma_pixels_tab[3]; | |
266 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
|
267 |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3010
diff
changeset
|
268 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
|
269 qpel_mc_func avg_h264_qpel_pixels_tab[4][16]; |
2967 | 270 |
2415 | 271 h264_weight_func weight_h264_pixels_tab[10]; |
272 h264_biweight_func biweight_h264_pixels_tab[10]; | |
2967 | 273 |
1708 | 274 me_cmp_func pix_abs[2][4]; |
2967 | 275 |
866 | 276 /* huffyuv specific */ |
277 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); | |
936 | 278 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); |
1527 | 279 /** |
280 * subtract huffyuv's variant of median prediction | |
281 * note, this might read from src1[-1], src2[-1] | |
282 */ | |
283 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top); | |
1273 | 284 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w); |
2633 | 285 |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
286 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
|
287 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
|
288 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
|
289 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
|
290 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
|
291 void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta); |
2967 | 292 |
1644 | 293 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale); |
294 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale); | |
295 | |
2045 | 296 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
|
297 |
1092 | 298 /* (I)DCT */ |
299 void (*fdct)(DCTELEM *block/* align 16*/); | |
1567 | 300 void (*fdct248)(DCTELEM *block/* align 16*/); |
2967 | 301 |
1324
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
302 /* IDCT really*/ |
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
303 void (*idct)(DCTELEM *block/* align 16*/); |
2967 | 304 |
1102 | 305 /** |
1104 | 306 * block -> idct -> clip to unsigned 8 bit -> dest. |
1102 | 307 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) |
1111 | 308 * @param line_size size in bytes of a horizotal line of dest |
1102 | 309 */ |
1092 | 310 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
2967 | 311 |
1102 | 312 /** |
313 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. | |
1111 | 314 * @param line_size size in bytes of a horizotal line of dest |
1102 | 315 */ |
1092 | 316 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
2967 | 317 |
1102 | 318 /** |
1104 | 319 * idct input permutation. |
1241 | 320 * several optimized IDCTs need a permutated input (relative to the normal order of the reference |
321 * IDCT) | |
322 * this permutation must be performed before the idct_put/add, note, normally this can be merged | |
323 * with the zigzag/alternate scan<br> | |
1102 | 324 * an example to avoid confusion: |
325 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...) | |
326 * - (x -> referece dct -> reference idct -> x) | |
327 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x) | |
328 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...) | |
329 */ | |
1092 | 330 uint8_t idct_permutation[64]; |
331 int idct_permutation_type; | |
332 #define FF_NO_IDCT_PERM 1 | |
333 #define FF_LIBMPEG2_IDCT_PERM 2 | |
334 #define FF_SIMPLE_IDCT_PERM 3 | |
335 #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
|
336 #define FF_PARTTRANS_IDCT_PERM 5 |
1092 | 337 |
1784 | 338 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale); |
339 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); | |
340 #define BASIS_SHIFT 16 | |
341 #define RECON_SHIFT 6 | |
2967 | 342 |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
343 void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride); |
2755 | 344 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
|
345 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
|
346 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
|
347 |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
348 /* snow wavelet */ |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
349 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
|
350 void (*horizontal_compose97i)(DWTELEM *b, int width); |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
351 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
|
352 |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3198
diff
changeset
|
353 void (*prefetch)(void *mem, int stride, int h); |
3245 | 354 |
355 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, 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
|
356 } DSPContext; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
357 |
1201 | 358 void dsputil_static_init(void); |
1092 | 359 void dsputil_init(DSPContext* p, AVCodecContext *avctx); |
0 | 360 |
764 | 361 /** |
362 * permute block according to permuatation. | |
363 * @param last last non zero element in scantable order | |
364 */ | |
1064 | 365 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); |
34 | 366 |
1729 | 367 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type); |
368 | |
2979 | 369 #define BYTE_VEC32(c) ((c)*0x01010101UL) |
1264 | 370 |
371 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b) | |
372 { | |
373 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
374 } | |
375 | |
376 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b) | |
377 { | |
378 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
379 } | |
380 | |
2184 | 381 static inline int get_penalty_factor(int lambda, int lambda2, int type){ |
382 switch(type&0xFF){ | |
383 default: | |
384 case FF_CMP_SAD: | |
385 return lambda>>FF_LAMBDA_SHIFT; | |
386 case FF_CMP_DCT: | |
387 return (3*lambda)>>(FF_LAMBDA_SHIFT+1); | |
388 case FF_CMP_W53: | |
389 return (4*lambda)>>(FF_LAMBDA_SHIFT); | |
390 case FF_CMP_W97: | |
391 return (2*lambda)>>(FF_LAMBDA_SHIFT); | |
392 case FF_CMP_SATD: | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
393 case FF_CMP_DCT264: |
2184 | 394 return (2*lambda)>>FF_LAMBDA_SHIFT; |
395 case FF_CMP_RD: | |
396 case FF_CMP_PSNR: | |
397 case FF_CMP_SSE: | |
398 case FF_CMP_NSSE: | |
399 return lambda2>>FF_LAMBDA_SHIFT; | |
400 case FF_CMP_BIT: | |
401 return 1; | |
402 } | |
403 } | |
404 | |
1102 | 405 /** |
1104 | 406 * Empty mmx state. |
1102 | 407 * this must be called between any dsp function and float/double code. |
408 * for example sin(); dsp->idct_put(); emms_c(); cos() | |
409 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
410 #define emms_c() |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
411 |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
412 /* should be defined by architectures supporting |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
413 one or more MultiMedia extension */ |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
414 int mm_support(void); |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
415 |
3089 | 416 #ifdef __GNUC__ |
417 #define DECLARE_ALIGNED_16(t,v) t v __attribute__ ((aligned (16))) | |
418 #else | |
419 #define DECLARE_ALIGNED_16(t,v) __declspec(align(16)) t v | |
420 #endif | |
1974
8c5489b2cf3e
move __align16 some place where non-MMX machines can see it
melanson
parents:
1972
diff
changeset
|
421 |
62 | 422 #if defined(HAVE_MMX) |
0 | 423 |
862 | 424 #undef emms_c |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
425 |
0 | 426 #define MM_MMX 0x0001 /* standard MMX */ |
427 #define MM_3DNOW 0x0004 /* AMD 3DNOW */ | |
428 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */ | |
429 #define MM_SSE 0x0008 /* SSE functions */ | |
430 #define MM_SSE2 0x0010 /* PIV SSE2 functions */ | |
2388 | 431 #define MM_3DNOWEXT 0x0020 /* AMD 3DNowExt */ |
0 | 432 |
433 extern int mm_flags; | |
434 | |
1064 | 435 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
436 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
|
437 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
0 | 438 |
439 static inline void emms(void) | |
440 { | |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
441 __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
|
442 } |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
443 |
936 | 444 |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
445 #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
|
446 {\ |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
447 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
|
448 emms();\ |
0 | 449 } |
450 | |
3089 | 451 #ifdef __GNUC__ |
452 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) | |
453 #else | |
454 #define DECLARE_ALIGNED_8(t,v) __declspec(align(8)) t v | |
455 #endif | |
456 | |
2324 | 457 #define STRIDE_ALIGN 8 |
0 | 458 |
1092 | 459 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); |
460 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx); | |
1065 | 461 |
62 | 462 #elif defined(ARCH_ARMV4L) |
463 | |
464 /* 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
|
465 line optimizations */ |
3089 | 466 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (4))) |
2324 | 467 #define STRIDE_ALIGN 4 |
62 | 468 |
2776
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
469 #define MM_IWMMXT 0x0100 /* XScale IWMMXT */ |
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
470 |
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
471 extern int mm_flags; |
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
472 |
1092 | 473 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx); |
62 | 474 |
88 | 475 #elif defined(HAVE_MLIB) |
476 | |
477 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ | |
3089 | 478 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 479 #define STRIDE_ALIGN 8 |
88 | 480 |
1092 | 481 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx); |
88 | 482 |
1959
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
483 #elif defined(ARCH_SPARC) |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
484 |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
485 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ |
3089 | 486 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 487 #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
|
488 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
|
489 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
490 #elif defined(ARCH_ALPHA) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
491 |
3089 | 492 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 493 #define STRIDE_ALIGN 8 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
494 |
1092 | 495 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
|
496 |
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
|
497 #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
|
498 |
894
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
499 #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
|
500 |
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
501 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
|
502 |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
503 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN) |
1653 | 504 #define pixel altivec_pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
505 #include <altivec.h> |
1653 | 506 #undef pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
507 #endif |
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
508 |
3089 | 509 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16))) |
2324 | 510 #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
|
511 |
1092 | 512 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
|
513 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
514 #elif defined(HAVE_MMI) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
515 |
3089 | 516 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16))) |
2324 | 517 #define STRIDE_ALIGN 16 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
518 |
1092 | 519 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
|
520 |
1259
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
521 #elif defined(ARCH_SH4) |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
522 |
3089 | 523 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 524 #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
|
525 |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
526 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
|
527 |
0 | 528 #else |
529 | |
3089 | 530 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 531 #define STRIDE_ALIGN 8 |
0 | 532 |
533 #endif | |
534 | |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
535 #ifdef __GNUC__ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
536 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
537 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
|
538 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
|
539 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
|
540 |
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
|
541 #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
|
542 #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
|
543 #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
|
544 |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3010
diff
changeset
|
545 #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
|
546 #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
|
547 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
548 #else /* __GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
549 |
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
|
550 #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
|
551 #define LD32(a) (*((uint32_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
552 #define LD64(a) (*((uint64_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
553 |
3217 | 554 #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
|
555 #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
|
556 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
557 #endif /* !__GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
558 |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
559 /* PSNR */ |
1064 | 560 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
|
561 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
|
562 AVCodecContext *avctx); |
781 | 563 |
564 /* FFT computation */ | |
565 | |
566 /* NOTE: soon integer code will be added, so you must use the | |
567 FFTSample type */ | |
568 typedef float FFTSample; | |
569 | |
570 typedef struct FFTComplex { | |
571 FFTSample re, im; | |
572 } FFTComplex; | |
573 | |
574 typedef struct FFTContext { | |
575 int nbits; | |
576 int inverse; | |
577 uint16_t *revtab; | |
578 FFTComplex *exptab; | |
579 FFTComplex *exptab1; /* only used by SSE code */ | |
580 void (*fft_calc)(struct FFTContext *s, FFTComplex *z); | |
581 } FFTContext; | |
582 | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
583 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
|
584 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
|
585 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
|
586 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z); |
3175 | 587 void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z); |
588 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
|
589 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
|
590 |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
591 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z) |
781 | 592 { |
593 s->fft_calc(s, z); | |
594 } | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
595 void ff_fft_end(FFTContext *s); |
781 | 596 |
597 /* MDCT computation */ | |
598 | |
599 typedef struct MDCTContext { | |
600 int n; /* size of MDCT (i.e. number of input data * 2) */ | |
601 int nbits; /* n = 2^nbits */ | |
602 /* pre/post rotation tables */ | |
603 FFTSample *tcos; | |
604 FFTSample *tsin; | |
605 FFTContext fft; | |
606 } MDCTContext; | |
607 | |
794 | 608 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
|
609 void ff_imdct_calc(MDCTContext *s, FFTSample *output, |
781 | 610 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
|
611 void ff_mdct_calc(MDCTContext *s, FFTSample *out, |
781 | 612 const FFTSample *input, FFTSample *tmp); |
794 | 613 void ff_mdct_end(MDCTContext *s); |
781 | 614 |
1708 | 615 #define WARPER8_16(name8, name16)\ |
616 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
617 return name8(s, dst , src , stride, h)\ | |
618 +name8(s, dst+8 , src+8 , stride, h);\ | |
619 } | |
620 | |
621 #define WARPER8_16_SQ(name8, name16)\ | |
622 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
623 int score=0;\ | |
624 score +=name8(s, dst , src , stride, 8);\ | |
625 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
626 if(h==16){\ | |
627 dst += 8*stride;\ | |
628 src += 8*stride;\ | |
629 score +=name8(s, dst , src , stride, 8);\ | |
630 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
631 }\ | |
632 return score;\ | |
936 | 633 } |
634 | |
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
|
635 #endif |