Mercurial > libavcodec.hg
annotate dsputil.h @ 8932:f647961ab2ac libavcodec
cosmetics: typo fix
author | diego |
---|---|
date | Sun, 15 Feb 2009 20:37:39 +0000 |
parents | bee83b3f9a6b |
children | e7d87561b42b |
rev | line source |
---|---|
429 | 1 /* |
2 * DSP utils | |
8629
04423b2f6e0b
cosmetics: Remove pointless period after copyright statement non-sentences.
diego
parents:
8590
diff
changeset
|
3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard |
1739
07a484280a82
copyright year update of the files i touched and remembered, things look annoyingly unmaintained otherwise
michael
parents:
1729
diff
changeset
|
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> |
429 | 5 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
6 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
7 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
8 * FFmpeg is free software; you can redistribute it and/or |
429 | 9 * modify it under the terms of the GNU Lesser General Public |
10 * License as published by the Free Software Foundation; either | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
11 * version 2.1 of the License, or (at your option) any later version. |
429 | 12 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
13 * FFmpeg is distributed in the hope that it will be useful, |
429 | 14 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
16 * Lesser General Public License for more details. | |
17 * | |
18 * You should have received a copy of the GNU Lesser General Public | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3807
diff
changeset
|
19 * License along with FFmpeg; if not, write to the Free Software |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
3029
diff
changeset
|
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
429 | 21 */ |
1102 | 22 |
23 /** | |
8718
e9d9d946f213
Use full internal pathname in doxygen @file directives.
diego
parents:
8694
diff
changeset
|
24 * @file libavcodec/dsputil.h |
1106 | 25 * DSP utils. |
1213 | 26 * note, many functions in here may use MMX which trashes the FPU state, it is |
27 * absolutely necessary to call emms_c() between dsp & float/double code | |
1102 | 28 */ |
29 | |
7760 | 30 #ifndef AVCODEC_DSPUTIL_H |
31 #define AVCODEC_DSPUTIL_H | |
0 | 32 |
8573
2acf0ae7b041
Fix build: Add intreadwrite.h and bswap.h #includes where necessary.
diego
parents:
8567
diff
changeset
|
33 #include "libavutil/intreadwrite.h" |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
34 #include "avcodec.h" |
0 | 35 |
1102 | 36 |
255 | 37 //#define DEBUG |
0 | 38 /* dct code */ |
39 typedef short DCTELEM; | |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
40 typedef int DWTELEM; |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5573
diff
changeset
|
41 typedef short IDWTELEM; |
0 | 42 |
474
11dbd00682fc
avoid name clash with libjpeg - added missing externs
bellard
parents:
429
diff
changeset
|
43 void fdct_ifast (DCTELEM *data); |
1571
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1567
diff
changeset
|
44 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
|
45 void ff_jpeg_fdct_islow (DCTELEM *data); |
1567 | 46 void ff_fdct248_islow (DCTELEM *data); |
0 | 47 |
48 void j_rev_dct (DCTELEM *data); | |
2256 | 49 void j_rev_dct4 (DCTELEM *data); |
2257 | 50 void j_rev_dct2 (DCTELEM *data); |
2259 | 51 void j_rev_dct1 (DCTELEM *data); |
5887 | 52 void ff_wmv2_idct_c(DCTELEM *data); |
0 | 53 |
687
9abb13c21fbe
fdct_mmx -> ff_fdct_mmx (renamed to avoid namespace conflict with xvid)
arpi_esp
parents:
675
diff
changeset
|
54 void ff_fdct_mmx(DCTELEM *block); |
1565 | 55 void ff_fdct_mmx2(DCTELEM *block); |
1765
e31754bc5b65
SSE2 fdct by (Balatoni Denes <pnis at coder dot hu>)
michael
parents:
1739
diff
changeset
|
56 void ff_fdct_sse2(DCTELEM *block); |
0 | 57 |
2755 | 58 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
|
59 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
|
60 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
|
61 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
|
62 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
|
63 void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block); |
8375
de2509cf3c44
H.264 idct functions that include the chroma, inter luma and intra16 luma loops
michael
parents:
8359
diff
changeset
|
64 void ff_h264_idct_add16_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]); |
de2509cf3c44
H.264 idct functions that include the chroma, inter luma and intra16 luma loops
michael
parents:
8359
diff
changeset
|
65 void ff_h264_idct_add16intra_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]); |
de2509cf3c44
H.264 idct functions that include the chroma, inter luma and intra16 luma loops
michael
parents:
8359
diff
changeset
|
66 void ff_h264_idct8_add4_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]); |
de2509cf3c44
H.264 idct functions that include the chroma, inter luma and intra16 luma loops
michael
parents:
8359
diff
changeset
|
67 void ff_h264_idct_add8_c(uint8_t **dest, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]); |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
68 |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
69 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
|
70 const float *src2, int src3, int blocksize, int step); |
7261 | 71 void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1, |
72 const float *win, float add_bias, int len); | |
7218 | 73 void ff_float_to_int16_c(int16_t *dst, const float *src, long len); |
7446 | 74 void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels); |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
75 |
34 | 76 /* encoding scans */ |
1064 | 77 extern const uint8_t ff_alternate_horizontal_scan[64]; |
78 extern const uint8_t ff_alternate_vertical_scan[64]; | |
79 extern const uint8_t ff_zigzag_direct[64]; | |
1567 | 80 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
|
81 |
0 | 82 /* pixel operations */ |
2090 | 83 #define MAX_NEG_CROP 1024 |
0 | 84 |
85 /* temporary */ | |
4179 | 86 extern uint32_t ff_squareTbl[512]; |
4176 | 87 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP]; |
0 | 88 |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
89 /* VP3 DSP functions */ |
2693 | 90 void ff_vp3_idct_c(DCTELEM *block/* align 16*/); |
91 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); | |
92 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); | |
0 | 93 |
7995 | 94 void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values); |
95 void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values); | |
96 | |
8785
bee83b3f9a6b
move vp6_filter_diag4() to a new vp6dsp.c file and use it throught dsputil
aurel
parents:
8760
diff
changeset
|
97 /* VP6 DSP functions */ |
bee83b3f9a6b
move vp6_filter_diag4() to a new vp6dsp.c file and use it throught dsputil
aurel
parents:
8760
diff
changeset
|
98 void ff_vp6_filter_diag4_c(uint8_t *dst, uint8_t *src, int stride, |
bee83b3f9a6b
move vp6_filter_diag4() to a new vp6dsp.c file and use it throught dsputil
aurel
parents:
8760
diff
changeset
|
99 const int16_t *h_weights, const int16_t *v_weights); |
bee83b3f9a6b
move vp6_filter_diag4() to a new vp6dsp.c file and use it throught dsputil
aurel
parents:
8760
diff
changeset
|
100 |
3245 | 101 /* 1/2^n downscaling functions from imgconvert.c */ |
102 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
103 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
104 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
105 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
106 | |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
107 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
|
108 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
|
109 |
675 | 110 /* minimum alignment rules ;) |
6486 | 111 If you notice errors in the align stuff, need more alignment for some ASM code |
112 for some CPU or need to use a function with less aligned data then send a mail | |
113 to the ffmpeg-devel mailing list, ... | |
675 | 114 |
6486 | 115 !warning These alignments might not match reality, (missing attribute((align)) |
116 stuff somewhere possible). | |
6488 | 117 I (Michael) did not check them, these are just the alignments which I think |
6486 | 118 could be reached easily ... |
675 | 119 |
120 !future video codecs might need functions with less strict alignment | |
121 */ | |
0 | 122 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
123 /* |
1064 | 124 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size); |
125 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride); | |
126 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
127 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
296 | 128 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
|
129 */ |
0 | 130 |
131 /* add and put pixel (decoding) */ | |
675 | 132 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 |
1709 | 133 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4 |
1064 | 134 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
|
135 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 | 136 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); |
1168 | 137 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 | 138 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset); |
3029 | 139 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset); |
0 | 140 |
984 | 141 #define DEF_OLD_QPEL(name)\ |
1064 | 142 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ |
143 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ | |
144 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
|
145 |
984 | 146 DEF_OLD_QPEL(qpel16_mc11_old_c) |
147 DEF_OLD_QPEL(qpel16_mc31_old_c) | |
148 DEF_OLD_QPEL(qpel16_mc12_old_c) | |
149 DEF_OLD_QPEL(qpel16_mc32_old_c) | |
150 DEF_OLD_QPEL(qpel16_mc13_old_c) | |
151 DEF_OLD_QPEL(qpel16_mc33_old_c) | |
152 DEF_OLD_QPEL(qpel8_mc11_old_c) | |
153 DEF_OLD_QPEL(qpel8_mc31_old_c) | |
154 DEF_OLD_QPEL(qpel8_mc12_old_c) | |
155 DEF_OLD_QPEL(qpel8_mc32_old_c) | |
156 DEF_OLD_QPEL(qpel8_mc13_old_c) | |
157 DEF_OLD_QPEL(qpel8_mc33_old_c) | |
651 | 158 |
159 #define CALL_2X_PIXELS(a, b, n)\ | |
160 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ | |
161 b(block , pixels , line_size, h);\ | |
162 b(block+n, pixels+n, line_size, h);\ | |
163 } | |
255 | 164 |
0 | 165 /* motion estimation */ |
1709 | 166 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2 |
5127 | 167 // although currently h<4 is not used as functions with width <8 are neither used nor implemented |
1708 | 168 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 | 169 |
1168 | 170 |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
171 // for snow slices |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
172 typedef struct slice_buffer_s slice_buffer; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
173 |
1102 | 174 /** |
6438 | 175 * Scantable. |
176 */ | |
177 typedef struct ScanTable{ | |
178 const uint8_t *scantable; | |
179 uint8_t permutated[64]; | |
180 uint8_t raster_end[64]; | |
8590 | 181 #if ARCH_PPC |
6438 | 182 /** Used by dct_quantize_altivec to find last-non-zero */ |
6440 | 183 DECLARE_ALIGNED(16, uint8_t, inverse[64]); |
6438 | 184 #endif |
185 } ScanTable; | |
186 | |
187 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable); | |
188 | |
6445 | 189 void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize, |
190 int block_w, int block_h, | |
191 int src_x, int src_y, int w, int h); | |
192 | |
6438 | 193 /** |
1102 | 194 * DSPContext. |
195 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
196 typedef struct DSPContext { |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
197 /* pixel ops : interface with DCT */ |
1064 | 198 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); |
199 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); | |
200 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
|
201 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
1064 | 202 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
2763 | 203 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size); |
204 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size); | |
4988
689490842cf5
factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents:
4962
diff
changeset
|
205 int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/); |
1136 | 206 /** |
207 * translational global motion compensation. | |
208 */ | |
1064 | 209 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); |
1136 | 210 /** |
211 * global motion compensation. | |
212 */ | |
1064 | 213 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy, |
2979 | 214 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); |
8288 | 215 void (*clear_block)(DCTELEM *block/*align 16*/); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
216 void (*clear_blocks)(DCTELEM *blocks/*align 16*/); |
1064 | 217 int (*pix_sum)(uint8_t * pix, int line_size); |
218 int (*pix_norm1)(uint8_t * pix, int line_size); | |
1708 | 219 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4 |
2967 | 220 |
1729 | 221 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */ |
222 me_cmp_func sse[5]; | |
223 me_cmp_func hadamard8_diff[5]; | |
224 me_cmp_func dct_sad[5]; | |
225 me_cmp_func quant_psnr[5]; | |
226 me_cmp_func bit[5]; | |
227 me_cmp_func rd[5]; | |
228 me_cmp_func vsad[5]; | |
229 me_cmp_func vsse[5]; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
230 me_cmp_func nsse[5]; |
2184 | 231 me_cmp_func w53[5]; |
232 me_cmp_func w97[5]; | |
2382 | 233 me_cmp_func dct_max[5]; |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
234 me_cmp_func dct264_sad[5]; |
936 | 235 |
1708 | 236 me_cmp_func me_pre_cmp[5]; |
237 me_cmp_func me_cmp[5]; | |
238 me_cmp_func me_sub_cmp[5]; | |
239 me_cmp_func mb_cmp[5]; | |
1729 | 240 me_cmp_func ildct_cmp[5]; //only width 16 used |
2382 | 241 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
|
242 |
5255 | 243 int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2, |
244 int size); | |
4749 | 245 |
1136 | 246 /** |
247 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
4751 | 248 * this is an array[4][4] of motion compensation functions for 4 |
1713 | 249 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 250 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 251 * @param block destination where the result is stored |
252 * @param pixels source | |
253 * @param line_size number of bytes in a horizontal line of block | |
254 * @param h height | |
255 */ | |
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
|
256 op_pixels_func put_pixels_tab[4][4]; |
1136 | 257 |
258 /** | |
259 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
2967 | 260 * This is an array[4][4] of motion compensation functions for 4 |
1713 | 261 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 262 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 263 * @param block destination into which the result is averaged (a+b+1)>>1 |
264 * @param pixels source | |
265 * @param line_size number of bytes in a horizontal line of block | |
266 * @param h height | |
267 */ | |
1319 | 268 op_pixels_func avg_pixels_tab[4][4]; |
1136 | 269 |
270 /** | |
271 * Halfpel motion compensation with no rounding (a+b)>>1. | |
4751 | 272 * this is an array[2][4] of motion compensation functions for 2 |
1225 | 273 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 274 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 275 * @param block destination where the result is stored |
276 * @param pixels source | |
277 * @param line_size number of bytes in a horizontal line of block | |
278 * @param h height | |
279 */ | |
2075 | 280 op_pixels_func put_no_rnd_pixels_tab[4][4]; |
1136 | 281 |
282 /** | |
283 * Halfpel motion compensation with no rounding (a+b)>>1. | |
4751 | 284 * this is an array[2][4] of motion compensation functions for 2 |
1225 | 285 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 286 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 287 * @param block destination into which the result is averaged (a+b)>>1 |
288 * @param pixels source | |
289 * @param line_size number of bytes in a horizontal line of block | |
290 * @param h height | |
291 */ | |
2075 | 292 op_pixels_func avg_no_rnd_pixels_tab[4][4]; |
2967 | 293 |
1864 | 294 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 | 295 |
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
|
296 /** |
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
|
297 * Thirdpel motion compensation with rounding (a+b+1)>>1. |
4751 | 298 * this is an array[12] of motion compensation functions for the 9 thirdpe |
299 * positions<br> | |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
300 * *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
|
301 * @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
|
302 * @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
|
303 * @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
|
304 * @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
|
305 */ |
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
|
306 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
1319 | 307 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
308 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
309 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
|
310 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
|
311 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
|
312 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; |
936 | 313 qpel_mc_func put_mspel_pixels_tab[8]; |
2967 | 314 |
1168 | 315 /** |
4751 | 316 * h264 Chroma MC |
1168 | 317 */ |
318 h264_chroma_mc_func put_h264_chroma_pixels_tab[3]; | |
3663 | 319 /* This is really one func used in VC-1 decoding */ |
320 h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3]; | |
1168 | 321 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
|
322 |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3010
diff
changeset
|
323 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
|
324 qpel_mc_func avg_h264_qpel_pixels_tab[4][16]; |
2967 | 325 |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3746
diff
changeset
|
326 qpel_mc_func put_2tap_qpel_pixels_tab[4][16]; |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3746
diff
changeset
|
327 qpel_mc_func avg_2tap_qpel_pixels_tab[4][16]; |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3746
diff
changeset
|
328 |
2415 | 329 h264_weight_func weight_h264_pixels_tab[10]; |
330 h264_biweight_func biweight_h264_pixels_tab[10]; | |
2967 | 331 |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
332 /* AVS specific */ |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
333 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
|
334 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
|
335 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
|
336 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
|
337 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
|
338 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
|
339 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
|
340 |
1708 | 341 me_cmp_func pix_abs[2][4]; |
2967 | 342 |
866 | 343 /* huffyuv specific */ |
344 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); | |
6384 | 345 void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w); |
936 | 346 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); |
1527 | 347 /** |
348 * subtract huffyuv's variant of median prediction | |
349 * note, this might read from src1[-1], src2[-1] | |
350 */ | |
351 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top); | |
8760 | 352 void (*add_hfyu_median_prediction)(uint8_t *dst, uint8_t *top, uint8_t *diff, int w, int *left, int *left_top); |
6384 | 353 /* this might write to dst[w] */ |
354 void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp); | |
6241 | 355 void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w); |
2633 | 356 |
7579 | 357 void (*h264_v_loop_filter_luma)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta, int8_t *tc0); |
358 void (*h264_h_loop_filter_luma)(uint8_t *pix/*align 4 */, int stride, int alpha, int beta, int8_t *tc0); | |
359 /* v/h_loop_filter_luma_intra: align 16 */ | |
8395
195cba8f6257
Move filter_luma_intra into dsputil for later addition of asm.
darkshikari
parents:
8375
diff
changeset
|
360 void (*h264_v_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta); |
195cba8f6257
Move filter_luma_intra into dsputil for later addition of asm.
darkshikari
parents:
8375
diff
changeset
|
361 void (*h264_h_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta); |
7579 | 362 void (*h264_v_loop_filter_chroma)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta, int8_t *tc0); |
363 void (*h264_h_loop_filter_chroma)(uint8_t *pix/*align 4*/, int stride, int alpha, int beta, int8_t *tc0); | |
364 void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta); | |
365 void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta); | |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3574
diff
changeset
|
366 // 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
|
367 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], |
7327
483421b11d98
Fix h264_loop_filter_strength_mmx2() so it works with PAFF.
michael
parents:
7286
diff
changeset
|
368 int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field); |
2967 | 369 |
1644 | 370 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale); |
371 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale); | |
372 | |
2045 | 373 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
|
374 |
5887 | 375 void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale); |
376 void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale); | |
377 | |
7995 | 378 void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values); |
379 void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values); | |
380 | |
8785
bee83b3f9a6b
move vp6_filter_diag4() to a new vp6dsp.c file and use it throught dsputil
aurel
parents:
8760
diff
changeset
|
381 void (*vp6_filter_diag4)(uint8_t *dst, uint8_t *src, int stride, |
bee83b3f9a6b
move vp6_filter_diag4() to a new vp6dsp.c file and use it throught dsputil
aurel
parents:
8760
diff
changeset
|
382 const int16_t *h_weights,const int16_t *v_weights); |
bee83b3f9a6b
move vp6_filter_diag4() to a new vp6dsp.c file and use it throught dsputil
aurel
parents:
8760
diff
changeset
|
383 |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
384 /* 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
|
385 void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize); |
7563 | 386 void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len); |
5737 | 387 /* no alignment needed */ |
388 void (*flac_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc); | |
3574 | 389 /* 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
|
390 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
|
391 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
|
392 /* 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
|
393 void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step); |
7261 | 394 /* assume len is a multiple of 4, and arrays are 16-byte aligned */ |
395 void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len); | |
7564 | 396 /* assume len is a multiple of 8, and arrays are 16-byte aligned */ |
397 void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len); | |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
398 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
399 /* 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
|
400 * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */ |
7218 | 401 void (*float_to_int16)(int16_t *dst, const float *src, long len); |
7286
e267f2519248
float_to_int16_interleave: change src to an array of pointers instead of assuming it's contiguous.
lorenm
parents:
7263
diff
changeset
|
402 void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels); |
3536
545a15c19c91
sse & sse2 implementations of vorbis channel coupling.
lorenm
parents:
3526
diff
changeset
|
403 |
1092 | 404 /* (I)DCT */ |
405 void (*fdct)(DCTELEM *block/* align 16*/); | |
1567 | 406 void (*fdct248)(DCTELEM *block/* align 16*/); |
2967 | 407 |
1324
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
408 /* IDCT really*/ |
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
409 void (*idct)(DCTELEM *block/* align 16*/); |
2967 | 410 |
1102 | 411 /** |
1104 | 412 * block -> idct -> clip to unsigned 8 bit -> dest. |
1102 | 413 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) |
5728 | 414 * @param line_size size in bytes of a horizontal line of dest |
1102 | 415 */ |
1092 | 416 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
2967 | 417 |
1102 | 418 /** |
419 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. | |
5728 | 420 * @param line_size size in bytes of a horizontal line of dest |
1102 | 421 */ |
1092 | 422 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
2967 | 423 |
1102 | 424 /** |
1104 | 425 * idct input permutation. |
1241 | 426 * several optimized IDCTs need a permutated input (relative to the normal order of the reference |
427 * IDCT) | |
428 * this permutation must be performed before the idct_put/add, note, normally this can be merged | |
429 * with the zigzag/alternate scan<br> | |
1102 | 430 * an example to avoid confusion: |
431 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...) | |
432 * - (x -> referece dct -> reference idct -> x) | |
433 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x) | |
434 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...) | |
435 */ | |
1092 | 436 uint8_t idct_permutation[64]; |
437 int idct_permutation_type; | |
438 #define FF_NO_IDCT_PERM 1 | |
439 #define FF_LIBMPEG2_IDCT_PERM 2 | |
440 #define FF_SIMPLE_IDCT_PERM 3 | |
441 #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
|
442 #define FF_PARTTRANS_IDCT_PERM 5 |
6600
c3213c91124c
Add a new IDCT permutation, used in xvid_sse2 and possibly future similar IDCTs.
astrange
parents:
6488
diff
changeset
|
443 #define FF_SSE2_IDCT_PERM 6 |
1092 | 444 |
1784 | 445 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale); |
446 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); | |
447 #define BASIS_SHIFT 16 | |
448 #define RECON_SHIFT 6 | |
2967 | 449 |
6437 | 450 void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w); |
6449
208074826b34
move EDGE_WIDTH definition allong with draw_edges where it belongs
aurel
parents:
6445
diff
changeset
|
451 #define EDGE_WIDTH 16 |
6437 | 452 |
4268 | 453 /* h264 functions */ |
8403 | 454 /* NOTE!!! if you implement any of h264_idct8_add, h264_idct8_add4 then you must implement all of them |
455 NOTE!!! if you implement any of h264_idct_add, h264_idct_add16, h264_idct_add16intra, h264_idct_add8 then you must implement all of them | |
456 The reason for above, is that no 2 out of one list may use a different permutation. | |
457 */ | |
7678 | 458 void (*h264_idct_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride); |
459 void (*h264_idct8_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride); | |
460 void (*h264_idct_dc_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride); | |
461 void (*h264_idct8_dc_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride); | |
4279 | 462 void (*h264_dct)(DCTELEM block[4][4]); |
8375
de2509cf3c44
H.264 idct functions that include the chroma, inter luma and intra16 luma loops
michael
parents:
8359
diff
changeset
|
463 void (*h264_idct_add16)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]); |
de2509cf3c44
H.264 idct functions that include the chroma, inter luma and intra16 luma loops
michael
parents:
8359
diff
changeset
|
464 void (*h264_idct8_add4)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]); |
de2509cf3c44
H.264 idct functions that include the chroma, inter luma and intra16 luma loops
michael
parents:
8359
diff
changeset
|
465 void (*h264_idct_add8)(uint8_t **dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]); |
de2509cf3c44
H.264 idct functions that include the chroma, inter luma and intra16 luma loops
michael
parents:
8359
diff
changeset
|
466 void (*h264_idct_add16intra)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]); |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
467 |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3175
diff
changeset
|
468 /* snow wavelet */ |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5573
diff
changeset
|
469 void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width); |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5573
diff
changeset
|
470 void (*horizontal_compose97i)(IDWTELEM *b, int width); |
4436
d3e389536b0a
Add the const specifier as needed to reduce the number of warnings.
takis
parents:
4311
diff
changeset
|
471 void (*inner_add_yblock)(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8); |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3198
diff
changeset
|
472 |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3198
diff
changeset
|
473 void (*prefetch)(void *mem, int stride, int h); |
3245 | 474 |
475 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
3526 | 476 |
477 /* vc1 functions */ | |
478 void (*vc1_inv_trans_8x8)(DCTELEM *b); | |
5997
90de28dfd8d6
Switch VC-1 decoder to output decoded residual immediately.
kostya
parents:
5907
diff
changeset
|
479 void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block); |
90de28dfd8d6
Switch VC-1 decoder to output decoded residual immediately.
kostya
parents:
5907
diff
changeset
|
480 void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block); |
90de28dfd8d6
Switch VC-1 decoder to output decoded residual immediately.
kostya
parents:
5907
diff
changeset
|
481 void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block); |
4239 | 482 void (*vc1_v_overlap)(uint8_t* src, int stride); |
483 void (*vc1_h_overlap)(uint8_t* src, int stride); | |
3526 | 484 /* put 8x8 block with bicubic interpolation and quarterpel precision |
485 * last argument is actually round value instead of height | |
486 */ | |
487 op_pixels_func put_vc1_mspel_pixels_tab[16]; | |
5887 | 488 |
489 /* intrax8 functions */ | |
5907
fbd10e6dfbe1
cosmetics: Fix spacial --> spatial typo in function names.
diego
parents:
5887
diff
changeset
|
490 void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize); |
fbd10e6dfbe1
cosmetics: Fix spacial --> spatial typo in function names.
diego
parents:
5887
diff
changeset
|
491 void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize, |
5887 | 492 int * range, int * sum, int edges); |
493 | |
7203
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7135
diff
changeset
|
494 /* ape functions */ |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7135
diff
changeset
|
495 /** |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7135
diff
changeset
|
496 * Add contents of the second vector to the first one. |
7232
cc55dd004819
Monkey's Audio decoder vector functions work on input with length
kostya
parents:
7218
diff
changeset
|
497 * @param len length of vectors, should be multiple of 16 |
7203
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7135
diff
changeset
|
498 */ |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7135
diff
changeset
|
499 void (*add_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len); |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7135
diff
changeset
|
500 /** |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7135
diff
changeset
|
501 * Add contents of the second vector to the first one. |
7232
cc55dd004819
Monkey's Audio decoder vector functions work on input with length
kostya
parents:
7218
diff
changeset
|
502 * @param len length of vectors, should be multiple of 16 |
7203
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7135
diff
changeset
|
503 */ |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7135
diff
changeset
|
504 void (*sub_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len); |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7135
diff
changeset
|
505 /** |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7135
diff
changeset
|
506 * Calculate scalar product of two vectors. |
7232
cc55dd004819
Monkey's Audio decoder vector functions work on input with length
kostya
parents:
7218
diff
changeset
|
507 * @param len length of vectors, should be multiple of 16 |
7203
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7135
diff
changeset
|
508 * @param shift number of bits to discard from product |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7135
diff
changeset
|
509 */ |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7135
diff
changeset
|
510 int32_t (*scalarproduct_int16)(int16_t *v1, int16_t *v2/*align 16*/, int len, int shift); |
8232 | 511 |
8233 | 512 /* rv30 functions */ |
513 qpel_mc_func put_rv30_tpel_pixels_tab[4][16]; | |
514 qpel_mc_func avg_rv30_tpel_pixels_tab[4][16]; | |
515 | |
8232 | 516 /* rv40 functions */ |
517 qpel_mc_func put_rv40_qpel_pixels_tab[4][16]; | |
518 qpel_mc_func avg_rv40_qpel_pixels_tab[4][16]; | |
519 h264_chroma_mc_func put_rv40_chroma_pixels_tab[3]; | |
520 h264_chroma_mc_func avg_rv40_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
|
521 } DSPContext; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
522 |
4197 | 523 void dsputil_static_init(void); |
1092 | 524 void dsputil_init(DSPContext* p, AVCodecContext *avctx); |
0 | 525 |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4279
diff
changeset
|
526 int ff_check_alignment(void); |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4279
diff
changeset
|
527 |
764 | 528 /** |
529 * permute block according to permuatation. | |
530 * @param last last non zero element in scantable order | |
531 */ | |
1064 | 532 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); |
34 | 533 |
1729 | 534 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type); |
535 | |
2979 | 536 #define BYTE_VEC32(c) ((c)*0x01010101UL) |
1264 | 537 |
538 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b) | |
539 { | |
540 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
541 } | |
542 | |
543 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b) | |
544 { | |
545 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
546 } | |
547 | |
2184 | 548 static inline int get_penalty_factor(int lambda, int lambda2, int type){ |
549 switch(type&0xFF){ | |
550 default: | |
551 case FF_CMP_SAD: | |
552 return lambda>>FF_LAMBDA_SHIFT; | |
553 case FF_CMP_DCT: | |
554 return (3*lambda)>>(FF_LAMBDA_SHIFT+1); | |
555 case FF_CMP_W53: | |
556 return (4*lambda)>>(FF_LAMBDA_SHIFT); | |
557 case FF_CMP_W97: | |
558 return (2*lambda)>>(FF_LAMBDA_SHIFT); | |
559 case FF_CMP_SATD: | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
560 case FF_CMP_DCT264: |
2184 | 561 return (2*lambda)>>FF_LAMBDA_SHIFT; |
562 case FF_CMP_RD: | |
563 case FF_CMP_PSNR: | |
564 case FF_CMP_SSE: | |
565 case FF_CMP_NSSE: | |
566 return lambda2>>FF_LAMBDA_SHIFT; | |
567 case FF_CMP_BIT: | |
568 return 1; | |
569 } | |
570 } | |
571 | |
1102 | 572 /** |
1104 | 573 * Empty mmx state. |
1102 | 574 * this must be called between any dsp function and float/double code. |
575 * for example sin(); dsp->idct_put(); emms_c(); cos() | |
576 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
577 #define emms_c() |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
578 |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
579 /* should be defined by architectures supporting |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
580 one or more MultiMedia extension */ |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
581 int mm_support(void); |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
582 |
5149 | 583 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx); |
8359 | 584 void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx); |
5149 | 585 void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx); |
586 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx); | |
587 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx); | |
588 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); | |
589 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx); | |
590 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx); | |
591 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx); | |
592 | |
4999
0c4bf6b7d1c6
use DECLARE_ALIGNED from mem.h (which is included through common.h)
alex
parents:
4988
diff
changeset
|
593 #define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v) |
1974
8c5489b2cf3e
move __align16 some place where non-MMX machines can see it
melanson
parents:
1972
diff
changeset
|
594 |
8590 | 595 #if HAVE_MMX |
0 | 596 |
862 | 597 #undef emms_c |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
598 |
4197 | 599 extern int mm_flags; |
600 | |
1064 | 601 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
602 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
|
603 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
0 | 604 |
605 static inline void emms(void) | |
606 { | |
8031 | 607 __asm__ volatile ("emms;":::"memory"); |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
608 } |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
609 |
936 | 610 |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
611 #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
|
612 {\ |
8104
0d108ec85620
Remove duplicated MM_* macros for CPU capabilities from dsputil.h.
rathann
parents:
8031
diff
changeset
|
613 if (mm_flags & FF_MM_MMX)\ |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
614 emms();\ |
0 | 615 } |
616 | |
1092 | 617 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx); |
1065 | 618 |
8590 | 619 #elif ARCH_ARM |
62 | 620 |
4197 | 621 extern int mm_flags; |
622 | |
8590 | 623 #if HAVE_NEON |
7687
e5b5a9af1b68
ARM: set STRIDE_ALIGN and DECLARE_ALIGNED_8 to 16 for NEON
mru
parents:
7678
diff
changeset
|
624 # define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v) |
e5b5a9af1b68
ARM: set STRIDE_ALIGN and DECLARE_ALIGNED_8 to 16 for NEON
mru
parents:
7678
diff
changeset
|
625 # define STRIDE_ALIGN 16 |
e5b5a9af1b68
ARM: set STRIDE_ALIGN and DECLARE_ALIGNED_8 to 16 for NEON
mru
parents:
7678
diff
changeset
|
626 #endif |
e5b5a9af1b68
ARM: set STRIDE_ALIGN and DECLARE_ALIGNED_8 to 16 for NEON
mru
parents:
7678
diff
changeset
|
627 |
8590 | 628 #elif ARCH_PPC |
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
|
629 |
4197 | 630 extern int mm_flags; |
631 | |
4999
0c4bf6b7d1c6
use DECLARE_ALIGNED from mem.h (which is included through common.h)
alex
parents:
4988
diff
changeset
|
632 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v) |
2324 | 633 #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
|
634 |
8590 | 635 #elif HAVE_MMI |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
636 |
4999
0c4bf6b7d1c6
use DECLARE_ALIGNED from mem.h (which is included through common.h)
alex
parents:
4988
diff
changeset
|
637 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v) |
2324 | 638 #define STRIDE_ALIGN 16 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
639 |
7135
6bd6a2da306e
Define mm_flags/support to be 0 on architectures where they don't exist.
astrange
parents:
7094
diff
changeset
|
640 #else |
6bd6a2da306e
Define mm_flags/support to be 0 on architectures where they don't exist.
astrange
parents:
7094
diff
changeset
|
641 |
6bd6a2da306e
Define mm_flags/support to be 0 on architectures where they don't exist.
astrange
parents:
7094
diff
changeset
|
642 #define mm_flags 0 |
6bd6a2da306e
Define mm_flags/support to be 0 on architectures where they don't exist.
astrange
parents:
7094
diff
changeset
|
643 #define mm_support() 0 |
6bd6a2da306e
Define mm_flags/support to be 0 on architectures where they don't exist.
astrange
parents:
7094
diff
changeset
|
644 |
6363
7ebd1cdb2142
clean up definition of DECLARE_ALIGNED_8 and STRIDE_ALIGN
mru
parents:
6241
diff
changeset
|
645 #endif |
3728 | 646 |
6363
7ebd1cdb2142
clean up definition of DECLARE_ALIGNED_8 and STRIDE_ALIGN
mru
parents:
6241
diff
changeset
|
647 #ifndef DECLARE_ALIGNED_8 |
7ebd1cdb2142
clean up definition of DECLARE_ALIGNED_8 and STRIDE_ALIGN
mru
parents:
6241
diff
changeset
|
648 # define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v) |
7ebd1cdb2142
clean up definition of DECLARE_ALIGNED_8 and STRIDE_ALIGN
mru
parents:
6241
diff
changeset
|
649 #endif |
3728 | 650 |
6363
7ebd1cdb2142
clean up definition of DECLARE_ALIGNED_8 and STRIDE_ALIGN
mru
parents:
6241
diff
changeset
|
651 #ifndef STRIDE_ALIGN |
7ebd1cdb2142
clean up definition of DECLARE_ALIGNED_8 and STRIDE_ALIGN
mru
parents:
6241
diff
changeset
|
652 # define STRIDE_ALIGN 8 |
0 | 653 #endif |
654 | |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
655 /* PSNR */ |
1064 | 656 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
|
657 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
|
658 AVCodecContext *avctx); |
781 | 659 |
660 /* FFT computation */ | |
661 | |
662 /* NOTE: soon integer code will be added, so you must use the | |
663 FFTSample type */ | |
664 typedef float FFTSample; | |
665 | |
3555 | 666 struct MDCTContext; |
667 | |
781 | 668 typedef struct FFTComplex { |
669 FFTSample re, im; | |
670 } FFTComplex; | |
671 | |
672 typedef struct FFTContext { | |
673 int nbits; | |
674 int inverse; | |
675 uint16_t *revtab; | |
676 FFTComplex *exptab; | |
677 FFTComplex *exptab1; /* only used by SSE code */ | |
7542 | 678 FFTComplex *tmp_buf; |
679 void (*fft_permute)(struct FFTContext *s, FFTComplex *z); | |
781 | 680 void (*fft_calc)(struct FFTContext *s, FFTComplex *z); |
7546 | 681 void (*imdct_calc)(struct MDCTContext *s, FFTSample *output, const FFTSample *input); |
682 void (*imdct_half)(struct MDCTContext *s, FFTSample *output, const FFTSample *input); | |
781 | 683 } FFTContext; |
684 | |
8694
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
685 extern FFTSample* ff_cos_tabs[13]; |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
686 |
8636 | 687 /** |
688 * Sets up a complex FFT. | |
689 * @param nbits log2 of the length of the input array | |
690 * @param inverse if 0 perform the forward transform, if 1 perform the inverse | |
691 */ | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
692 int ff_fft_init(FFTContext *s, int nbits, int inverse); |
7542 | 693 void ff_fft_permute_c(FFTContext *s, FFTComplex *z); |
694 void ff_fft_permute_sse(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
|
695 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
|
696 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z); |
3175 | 697 void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z); |
698 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
|
699 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
|
700 |
8636 | 701 /** |
702 * Do the permutation needed BEFORE calling ff_fft_calc(). | |
703 */ | |
7542 | 704 static inline void ff_fft_permute(FFTContext *s, FFTComplex *z) |
705 { | |
706 s->fft_permute(s, z); | |
707 } | |
8636 | 708 /** |
709 * Do a complex FFT with the parameters defined in ff_fft_init(). The | |
710 * input data must be permuted before. No 1.0/sqrt(n) normalization is done. | |
711 */ | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
712 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z) |
781 | 713 { |
714 s->fft_calc(s, z); | |
715 } | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
716 void ff_fft_end(FFTContext *s); |
781 | 717 |
718 /* MDCT computation */ | |
719 | |
720 typedef struct MDCTContext { | |
721 int n; /* size of MDCT (i.e. number of input data * 2) */ | |
722 int nbits; /* n = 2^nbits */ | |
723 /* pre/post rotation tables */ | |
724 FFTSample *tcos; | |
725 FFTSample *tsin; | |
726 FFTContext fft; | |
727 } MDCTContext; | |
728 | |
7547 | 729 static inline void ff_imdct_calc(MDCTContext *s, FFTSample *output, const FFTSample *input) |
730 { | |
731 s->fft.imdct_calc(s, output, input); | |
732 } | |
733 static inline void ff_imdct_half(MDCTContext *s, FFTSample *output, const FFTSample *input) | |
734 { | |
735 s->fft.imdct_half(s, output, input); | |
736 } | |
737 | |
6139
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
6056
diff
changeset
|
738 /** |
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
6056
diff
changeset
|
739 * Generate a Kaiser-Bessel Derived Window. |
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
6056
diff
changeset
|
740 * @param window pointer to half window |
6142
a35b838ab955
Add variable alpha and size of half window for Kaiser-Bessel Derived window
superdump
parents:
6139
diff
changeset
|
741 * @param alpha determines window shape |
a35b838ab955
Add variable alpha and size of half window for Kaiser-Bessel Derived window
superdump
parents:
6139
diff
changeset
|
742 * @param n size of half window |
6139
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
6056
diff
changeset
|
743 */ |
6142
a35b838ab955
Add variable alpha and size of half window for Kaiser-Bessel Derived window
superdump
parents:
6139
diff
changeset
|
744 void ff_kbd_window_init(float *window, float alpha, int n); |
6139
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
6056
diff
changeset
|
745 |
7094
b0820b8bd4dd
Add generic ff_sine_window_init function and implement in codecs appropriately
superdump
parents:
6600
diff
changeset
|
746 /** |
b0820b8bd4dd
Add generic ff_sine_window_init function and implement in codecs appropriately
superdump
parents:
6600
diff
changeset
|
747 * Generate a sine window. |
b0820b8bd4dd
Add generic ff_sine_window_init function and implement in codecs appropriately
superdump
parents:
6600
diff
changeset
|
748 * @param window pointer to half window |
b0820b8bd4dd
Add generic ff_sine_window_init function and implement in codecs appropriately
superdump
parents:
6600
diff
changeset
|
749 * @param n size of half window |
b0820b8bd4dd
Add generic ff_sine_window_init function and implement in codecs appropriately
superdump
parents:
6600
diff
changeset
|
750 */ |
b0820b8bd4dd
Add generic ff_sine_window_init function and implement in codecs appropriately
superdump
parents:
6600
diff
changeset
|
751 void ff_sine_window_init(float *window, int n); |
7573
7802295cae6f
Add declarations for the sine tables used in wma.c (half window sizes: 128,
superdump
parents:
7564
diff
changeset
|
752 extern float ff_sine_128 [ 128]; |
7802295cae6f
Add declarations for the sine tables used in wma.c (half window sizes: 128,
superdump
parents:
7564
diff
changeset
|
753 extern float ff_sine_256 [ 256]; |
7802295cae6f
Add declarations for the sine tables used in wma.c (half window sizes: 128,
superdump
parents:
7564
diff
changeset
|
754 extern float ff_sine_512 [ 512]; |
7802295cae6f
Add declarations for the sine tables used in wma.c (half window sizes: 128,
superdump
parents:
7564
diff
changeset
|
755 extern float ff_sine_1024[1024]; |
7802295cae6f
Add declarations for the sine tables used in wma.c (half window sizes: 128,
superdump
parents:
7564
diff
changeset
|
756 extern float ff_sine_2048[2048]; |
8567 | 757 extern float ff_sine_4096[4096]; |
758 extern float *ff_sine_windows[6]; | |
7094
b0820b8bd4dd
Add generic ff_sine_window_init function and implement in codecs appropriately
superdump
parents:
6600
diff
changeset
|
759 |
794 | 760 int ff_mdct_init(MDCTContext *s, int nbits, int inverse); |
7547 | 761 void ff_imdct_calc_c(MDCTContext *s, FFTSample *output, const FFTSample *input); |
762 void ff_imdct_half_c(MDCTContext *s, FFTSample *output, const FFTSample *input); | |
7546 | 763 void ff_imdct_calc_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input); |
7544 | 764 void ff_imdct_half_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input); |
7546 | 765 void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input); |
7544 | 766 void ff_imdct_half_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input); |
7546 | 767 void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output, const FFTSample *input); |
7544 | 768 void ff_imdct_half_sse(MDCTContext *s, FFTSample *output, const FFTSample *input); |
7546 | 769 void ff_mdct_calc(MDCTContext *s, FFTSample *out, const FFTSample *input); |
794 | 770 void ff_mdct_end(MDCTContext *s); |
781 | 771 |
8694
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
772 /* Real Discrete Fourier Transform */ |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
773 |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
774 enum RDFTransformType { |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
775 RDFT, |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
776 IRDFT, |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
777 RIDFT, |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
778 IRIDFT, |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
779 }; |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
780 |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
781 typedef struct { |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
782 int nbits; |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
783 int inverse; |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
784 int sign_convention; |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
785 |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
786 /* pre/post rotation tables */ |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
787 FFTSample *tcos; |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
788 FFTSample *tsin; |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
789 FFTContext fft; |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
790 } RDFTContext; |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
791 |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
792 /** |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
793 * Sets up a real FFT. |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
794 * @param nbits log2 of the length of the input array |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
795 * @param trans the type of transform |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
796 */ |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
797 int ff_rdft_init(RDFTContext *s, int nbits, enum RDFTransformType trans); |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
798 void ff_rdft_calc(RDFTContext *s, FFTSample *data); |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
799 void ff_rdft_end(RDFTContext *s); |
68fd157bab48
Add the rdft family of transforms (fft/ifft of an all real sequence) to dsputil.
alexc
parents:
8636
diff
changeset
|
800 |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
5997
diff
changeset
|
801 #define WRAPPER8_16(name8, name16)\ |
1708 | 802 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ |
803 return name8(s, dst , src , stride, h)\ | |
804 +name8(s, dst+8 , src+8 , stride, h);\ | |
805 } | |
806 | |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
5997
diff
changeset
|
807 #define WRAPPER8_16_SQ(name8, name16)\ |
1708 | 808 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ |
809 int score=0;\ | |
810 score +=name8(s, dst , src , stride, 8);\ | |
811 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
812 if(h==16){\ | |
813 dst += 8*stride;\ | |
814 src += 8*stride;\ | |
815 score +=name8(s, dst , src , stride, 8);\ | |
816 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
817 }\ | |
818 return score;\ | |
936 | 819 } |
820 | |
4240 | 821 |
822 static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
823 { | |
824 int i; | |
825 for(i=0; i<h; i++) | |
826 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
827 AV_WN16(dst , AV_RN16(src )); |
4240 | 828 dst+=dstStride; |
829 src+=srcStride; | |
830 } | |
831 } | |
832 | |
833 static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
834 { | |
835 int i; | |
836 for(i=0; i<h; i++) | |
837 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
838 AV_WN32(dst , AV_RN32(src )); |
4240 | 839 dst+=dstStride; |
840 src+=srcStride; | |
841 } | |
842 } | |
843 | |
844 static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
845 { | |
846 int i; | |
847 for(i=0; i<h; i++) | |
848 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
849 AV_WN32(dst , AV_RN32(src )); |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
850 AV_WN32(dst+4 , AV_RN32(src+4 )); |
4240 | 851 dst+=dstStride; |
852 src+=srcStride; | |
853 } | |
854 } | |
855 | |
856 static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
857 { | |
858 int i; | |
859 for(i=0; i<h; i++) | |
860 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
861 AV_WN32(dst , AV_RN32(src )); |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
862 AV_WN32(dst+4 , AV_RN32(src+4 )); |
4240 | 863 dst[8]= src[8]; |
864 dst+=dstStride; | |
865 src+=srcStride; | |
866 } | |
867 } | |
868 | |
869 static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
870 { | |
871 int i; | |
872 for(i=0; i<h; i++) | |
873 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
874 AV_WN32(dst , AV_RN32(src )); |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
875 AV_WN32(dst+4 , AV_RN32(src+4 )); |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
876 AV_WN32(dst+8 , AV_RN32(src+8 )); |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
877 AV_WN32(dst+12, AV_RN32(src+12)); |
4240 | 878 dst+=dstStride; |
879 src+=srcStride; | |
880 } | |
881 } | |
882 | |
883 static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) | |
884 { | |
885 int i; | |
886 for(i=0; i<h; i++) | |
887 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
888 AV_WN32(dst , AV_RN32(src )); |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
889 AV_WN32(dst+4 , AV_RN32(src+4 )); |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
890 AV_WN32(dst+8 , AV_RN32(src+8 )); |
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
891 AV_WN32(dst+12, AV_RN32(src+12)); |
4240 | 892 dst[16]= src[16]; |
893 dst+=dstStride; | |
894 src+=srcStride; | |
895 } | |
896 } | |
897 | |
7760 | 898 #endif /* AVCODEC_DSPUTIL_H */ |