Mercurial > libavcodec.hg
annotate dsputil.h @ 11486:1bf7af4db35a libavcodec
Add ff_ prefix to dwt functions
author | mru |
---|---|
date | Sun, 14 Mar 2010 17:50:16 +0000 |
parents | 0f0cd6b5791f |
children | 2a4dc3c0b012 |
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; | |
40 | |
474
11dbd00682fc
avoid name clash with libjpeg - added missing externs
bellard
parents:
429
diff
changeset
|
41 void fdct_ifast (DCTELEM *data); |
1571
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1567
diff
changeset
|
42 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
|
43 void ff_jpeg_fdct_islow (DCTELEM *data); |
1567 | 44 void ff_fdct248_islow (DCTELEM *data); |
0 | 45 |
46 void j_rev_dct (DCTELEM *data); | |
2256 | 47 void j_rev_dct4 (DCTELEM *data); |
2257 | 48 void j_rev_dct2 (DCTELEM *data); |
2259 | 49 void j_rev_dct1 (DCTELEM *data); |
5887 | 50 void ff_wmv2_idct_c(DCTELEM *data); |
0 | 51 |
687
9abb13c21fbe
fdct_mmx -> ff_fdct_mmx (renamed to avoid namespace conflict with xvid)
arpi_esp
parents:
675
diff
changeset
|
52 void ff_fdct_mmx(DCTELEM *block); |
1565 | 53 void ff_fdct_mmx2(DCTELEM *block); |
1765
e31754bc5b65
SSE2 fdct by (Balatoni Denes <pnis at coder dot hu>)
michael
parents:
1739
diff
changeset
|
54 void ff_fdct_sse2(DCTELEM *block); |
0 | 55 |
2755 | 56 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
|
57 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
|
58 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
|
59 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
|
60 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
|
61 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
|
62 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
|
63 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
|
64 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
|
65 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
|
66 |
7261 | 67 void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1, |
68 const float *win, float add_bias, int len); | |
7218 | 69 void ff_float_to_int16_c(int16_t *dst, const float *src, long len); |
7446 | 70 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
|
71 |
34 | 72 /* encoding scans */ |
1064 | 73 extern const uint8_t ff_alternate_horizontal_scan[64]; |
74 extern const uint8_t ff_alternate_vertical_scan[64]; | |
75 extern const uint8_t ff_zigzag_direct[64]; | |
1567 | 76 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
|
77 |
0 | 78 /* pixel operations */ |
2090 | 79 #define MAX_NEG_CROP 1024 |
0 | 80 |
81 /* temporary */ | |
4179 | 82 extern uint32_t ff_squareTbl[512]; |
4176 | 83 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP]; |
0 | 84 |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
85 /* VP3 DSP functions */ |
2693 | 86 void ff_vp3_idct_c(DCTELEM *block/* align 16*/); |
87 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); | |
88 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); | |
0 | 89 |
7995 | 90 void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values); |
91 void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values); | |
92 | |
8785
bee83b3f9a6b
move vp6_filter_diag4() to a new vp6dsp.c file and use it throught dsputil
aurel
parents:
8760
diff
changeset
|
93 /* VP6 DSP functions */ |
bee83b3f9a6b
move vp6_filter_diag4() to a new vp6dsp.c file and use it throught dsputil
aurel
parents:
8760
diff
changeset
|
94 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
|
95 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
|
96 |
11379
4b0cc1b97def
Move some dsp func prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11378
diff
changeset
|
97 /* Bink functions */ |
4b0cc1b97def
Move some dsp func prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11378
diff
changeset
|
98 void ff_bink_idct_c (DCTELEM *block); |
4b0cc1b97def
Move some dsp func prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11378
diff
changeset
|
99 void ff_bink_idct_add_c(uint8_t *dest, int linesize, DCTELEM *block); |
4b0cc1b97def
Move some dsp func prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11378
diff
changeset
|
100 void ff_bink_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block); |
4b0cc1b97def
Move some dsp func prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11378
diff
changeset
|
101 |
4b0cc1b97def
Move some dsp func prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11378
diff
changeset
|
102 /* CAVS functions */ |
4b0cc1b97def
Move some dsp func prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11378
diff
changeset
|
103 void ff_put_cavs_qpel8_mc00_c(uint8_t *dst, uint8_t *src, int stride); |
4b0cc1b97def
Move some dsp func prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11378
diff
changeset
|
104 void ff_avg_cavs_qpel8_mc00_c(uint8_t *dst, uint8_t *src, int stride); |
4b0cc1b97def
Move some dsp func prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11378
diff
changeset
|
105 void ff_put_cavs_qpel16_mc00_c(uint8_t *dst, uint8_t *src, int stride); |
4b0cc1b97def
Move some dsp func prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11378
diff
changeset
|
106 void ff_avg_cavs_qpel16_mc00_c(uint8_t *dst, uint8_t *src, int stride); |
4b0cc1b97def
Move some dsp func prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11378
diff
changeset
|
107 |
11378
f46b68960464
Move some VC1 dsp prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11376
diff
changeset
|
108 /* VC1 functions */ |
f46b68960464
Move some VC1 dsp prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11376
diff
changeset
|
109 void ff_put_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int rnd); |
f46b68960464
Move some VC1 dsp prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11376
diff
changeset
|
110 void ff_avg_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int rnd); |
f46b68960464
Move some VC1 dsp prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11376
diff
changeset
|
111 |
11379
4b0cc1b97def
Move some dsp func prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11378
diff
changeset
|
112 /* EA functions */ |
4b0cc1b97def
Move some dsp func prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11378
diff
changeset
|
113 void ff_ea_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block); |
4b0cc1b97def
Move some dsp func prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11378
diff
changeset
|
114 |
3245 | 115 /* 1/2^n downscaling functions from imgconvert.c */ |
116 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
117 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
118 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
119 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
120 | |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
121 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
|
122 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
|
123 |
675 | 124 /* minimum alignment rules ;) |
6486 | 125 If you notice errors in the align stuff, need more alignment for some ASM code |
126 for some CPU or need to use a function with less aligned data then send a mail | |
127 to the ffmpeg-devel mailing list, ... | |
675 | 128 |
6486 | 129 !warning These alignments might not match reality, (missing attribute((align)) |
130 stuff somewhere possible). | |
6488 | 131 I (Michael) did not check them, these are just the alignments which I think |
6486 | 132 could be reached easily ... |
675 | 133 |
134 !future video codecs might need functions with less strict alignment | |
135 */ | |
0 | 136 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
137 /* |
1064 | 138 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size); |
139 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride); | |
140 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
141 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
296 | 142 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
|
143 */ |
0 | 144 |
145 /* add and put pixel (decoding) */ | |
675 | 146 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 |
1709 | 147 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4 |
1064 | 148 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
|
149 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 | 150 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); |
1168 | 151 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 | 152 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset); |
3029 | 153 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset); |
0 | 154 |
11231 | 155 typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h); |
156 | |
984 | 157 #define DEF_OLD_QPEL(name)\ |
1064 | 158 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ |
159 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ | |
160 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
|
161 |
984 | 162 DEF_OLD_QPEL(qpel16_mc11_old_c) |
163 DEF_OLD_QPEL(qpel16_mc31_old_c) | |
164 DEF_OLD_QPEL(qpel16_mc12_old_c) | |
165 DEF_OLD_QPEL(qpel16_mc32_old_c) | |
166 DEF_OLD_QPEL(qpel16_mc13_old_c) | |
167 DEF_OLD_QPEL(qpel16_mc33_old_c) | |
168 DEF_OLD_QPEL(qpel8_mc11_old_c) | |
169 DEF_OLD_QPEL(qpel8_mc31_old_c) | |
170 DEF_OLD_QPEL(qpel8_mc12_old_c) | |
171 DEF_OLD_QPEL(qpel8_mc32_old_c) | |
172 DEF_OLD_QPEL(qpel8_mc13_old_c) | |
173 DEF_OLD_QPEL(qpel8_mc33_old_c) | |
651 | 174 |
175 #define CALL_2X_PIXELS(a, b, n)\ | |
176 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ | |
177 b(block , pixels , line_size, h);\ | |
178 b(block+n, pixels+n, line_size, h);\ | |
179 } | |
255 | 180 |
0 | 181 /* motion estimation */ |
1709 | 182 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2 |
5127 | 183 // although currently h<4 is not used as functions with width <8 are neither used nor implemented |
1708 | 184 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 | 185 |
1102 | 186 /** |
6438 | 187 * Scantable. |
188 */ | |
189 typedef struct ScanTable{ | |
190 const uint8_t *scantable; | |
191 uint8_t permutated[64]; | |
192 uint8_t raster_end[64]; | |
8590 | 193 #if ARCH_PPC |
6438 | 194 /** Used by dct_quantize_altivec to find last-non-zero */ |
10961
34a65026fa06
Move array specifiers outside DECLARE_ALIGNED() invocations
mru
parents:
10944
diff
changeset
|
195 DECLARE_ALIGNED(16, uint8_t, inverse)[64]; |
6438 | 196 #endif |
197 } ScanTable; | |
198 | |
199 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable); | |
200 | |
6445 | 201 void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize, |
202 int block_w, int block_h, | |
203 int src_x, int src_y, int w, int h); | |
204 | |
6438 | 205 /** |
1102 | 206 * DSPContext. |
207 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
208 typedef struct DSPContext { |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
209 /* pixel ops : interface with DCT */ |
1064 | 210 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); |
211 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); | |
212 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
|
213 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
11231 | 214 void (*put_pixels_nonclamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
1064 | 215 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
2763 | 216 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size); |
217 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
|
218 int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/); |
1136 | 219 /** |
220 * translational global motion compensation. | |
221 */ | |
1064 | 222 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); |
1136 | 223 /** |
224 * global motion compensation. | |
225 */ | |
1064 | 226 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy, |
2979 | 227 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); |
8288 | 228 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
|
229 void (*clear_blocks)(DCTELEM *blocks/*align 16*/); |
1064 | 230 int (*pix_sum)(uint8_t * pix, int line_size); |
231 int (*pix_norm1)(uint8_t * pix, int line_size); | |
1708 | 232 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4 |
2967 | 233 |
8976
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
234 me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */ |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
235 me_cmp_func sse[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
236 me_cmp_func hadamard8_diff[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
237 me_cmp_func dct_sad[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
238 me_cmp_func quant_psnr[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
239 me_cmp_func bit[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
240 me_cmp_func rd[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
241 me_cmp_func vsad[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
242 me_cmp_func vsse[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
243 me_cmp_func nsse[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
244 me_cmp_func w53[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
245 me_cmp_func w97[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
246 me_cmp_func dct_max[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
247 me_cmp_func dct264_sad[6]; |
936 | 248 |
8976
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
249 me_cmp_func me_pre_cmp[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
250 me_cmp_func me_cmp[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
251 me_cmp_func me_sub_cmp[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
252 me_cmp_func mb_cmp[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
253 me_cmp_func ildct_cmp[6]; //only width 16 used |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
254 me_cmp_func frame_skip_cmp[6]; //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
|
255 |
5255 | 256 int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2, |
257 int size); | |
4749 | 258 |
1136 | 259 /** |
260 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
4751 | 261 * this is an array[4][4] of motion compensation functions for 4 |
1713 | 262 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 263 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 264 * @param block destination where the result is stored |
265 * @param pixels source | |
266 * @param line_size number of bytes in a horizontal line of block | |
267 * @param h height | |
268 */ | |
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
|
269 op_pixels_func put_pixels_tab[4][4]; |
1136 | 270 |
271 /** | |
272 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
2967 | 273 * This is an array[4][4] of motion compensation functions for 4 |
1713 | 274 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 275 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 276 * @param block destination into which the result is averaged (a+b+1)>>1 |
277 * @param pixels source | |
278 * @param line_size number of bytes in a horizontal line of block | |
279 * @param h height | |
280 */ | |
1319 | 281 op_pixels_func avg_pixels_tab[4][4]; |
1136 | 282 |
283 /** | |
284 * Halfpel motion compensation with no rounding (a+b)>>1. | |
4751 | 285 * this is an array[2][4] of motion compensation functions for 2 |
1225 | 286 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 287 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 288 * @param block destination where the result is stored |
289 * @param pixels source | |
290 * @param line_size number of bytes in a horizontal line of block | |
291 * @param h height | |
292 */ | |
2075 | 293 op_pixels_func put_no_rnd_pixels_tab[4][4]; |
1136 | 294 |
295 /** | |
296 * Halfpel motion compensation with no rounding (a+b)>>1. | |
4751 | 297 * this is an array[2][4] of motion compensation functions for 2 |
1225 | 298 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 299 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 300 * @param block destination into which the result is averaged (a+b)>>1 |
301 * @param pixels source | |
302 * @param line_size number of bytes in a horizontal line of block | |
303 * @param h height | |
304 */ | |
2075 | 305 op_pixels_func avg_no_rnd_pixels_tab[4][4]; |
2967 | 306 |
1864 | 307 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 | 308 |
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
|
309 /** |
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
|
310 * Thirdpel motion compensation with rounding (a+b+1)>>1. |
4751 | 311 * this is an array[12] of motion compensation functions for the 9 thirdpe |
312 * 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
|
313 * *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
|
314 * @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
|
315 * @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
|
316 * @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
|
317 * @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
|
318 */ |
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
|
319 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
1319 | 320 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
321 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
322 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
|
323 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
|
324 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
|
325 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; |
936 | 326 qpel_mc_func put_mspel_pixels_tab[8]; |
2967 | 327 |
1168 | 328 /** |
4751 | 329 * h264 Chroma MC |
1168 | 330 */ |
331 h264_chroma_mc_func put_h264_chroma_pixels_tab[3]; | |
9439
ef3a7b711cc0
Rename put_no_rnd_h264_chroma* to reflect its usage in VC1 only
conrad
parents:
9437
diff
changeset
|
332 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3]; |
3663 | 333 /* This is really one func used in VC-1 decoding */ |
9439
ef3a7b711cc0
Rename put_no_rnd_h264_chroma* to reflect its usage in VC1 only
conrad
parents:
9437
diff
changeset
|
334 h264_chroma_mc_func put_no_rnd_vc1_chroma_pixels_tab[3]; |
9440 | 335 h264_chroma_mc_func avg_no_rnd_vc1_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
|
336 |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3010
diff
changeset
|
337 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
|
338 qpel_mc_func avg_h264_qpel_pixels_tab[4][16]; |
2967 | 339 |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3746
diff
changeset
|
340 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
|
341 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
|
342 |
2415 | 343 h264_weight_func weight_h264_pixels_tab[10]; |
344 h264_biweight_func biweight_h264_pixels_tab[10]; | |
2967 | 345 |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
346 /* AVS specific */ |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
347 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
|
348 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
|
349 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
|
350 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
|
351 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
|
352 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
|
353 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
|
354 |
1708 | 355 me_cmp_func pix_abs[2][4]; |
2967 | 356 |
866 | 357 /* huffyuv specific */ |
358 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); | |
6384 | 359 void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w); |
936 | 360 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); |
1527 | 361 /** |
362 * subtract huffyuv's variant of median prediction | |
363 * note, this might read from src1[-1], src2[-1] | |
364 */ | |
10431 | 365 void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top); |
366 void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top); | |
10430 | 367 int (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left); |
10878
a8620b001ed3
Implement alpha channel decoding for BGR HuffYUV.
astrange
parents:
10827
diff
changeset
|
368 void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha); |
6384 | 369 /* this might write to dst[w] */ |
370 void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp); | |
6241 | 371 void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w); |
2633 | 372 |
7579 | 373 void (*h264_v_loop_filter_luma)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta, int8_t *tc0); |
374 void (*h264_h_loop_filter_luma)(uint8_t *pix/*align 4 */, int stride, int alpha, int beta, int8_t *tc0); | |
375 /* 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
|
376 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
|
377 void (*h264_h_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta); |
7579 | 378 void (*h264_v_loop_filter_chroma)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta, int8_t *tc0); |
379 void (*h264_h_loop_filter_chroma)(uint8_t *pix/*align 4*/, int stride, int alpha, int beta, int8_t *tc0); | |
380 void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta); | |
381 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
|
382 // 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
|
383 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
|
384 int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field); |
2967 | 385 |
1644 | 386 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale); |
387 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale); | |
388 | |
2045 | 389 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
|
390 |
5887 | 391 void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale); |
392 void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale); | |
393 | |
7995 | 394 void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values); |
395 void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values); | |
396 | |
8785
bee83b3f9a6b
move vp6_filter_diag4() to a new vp6dsp.c file and use it throught dsputil
aurel
parents:
8760
diff
changeset
|
397 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
|
398 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
|
399 |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
400 /* 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
|
401 void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize); |
7563 | 402 void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len); |
5737 | 403 /* no alignment needed */ |
10424
94595d0e617c
Move autocorrelation function from flacenc.c to lpc.c. Also rename the
jbr
parents:
10420
diff
changeset
|
404 void (*lpc_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc); |
3574 | 405 /* 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
|
406 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
|
407 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
|
408 /* assume len is a multiple of 8, and src arrays are 16-byte aligned */ |
10300
4d1b9ca628fc
Drop unused args from vector_fmul_add_add, simpify code, and rename
mru
parents:
10219
diff
changeset
|
409 void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len); |
7261 | 410 /* assume len is a multiple of 4, and arrays are 16-byte aligned */ |
411 void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len); | |
7564 | 412 /* assume len is a multiple of 8, and arrays are 16-byte aligned */ |
413 void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len); | |
10105 | 414 void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */); |
10219 | 415 /** |
416 * Multiply a vector of floats by a scalar float. Source and | |
417 * destination vectors must overlap exactly or not at all. | |
418 * @param dst result vector, 16-byte aligned | |
419 * @param src input vector, 16-byte aligned | |
420 * @param mul scalar value | |
421 * @param len length of vector, multiple of 4 | |
422 */ | |
423 void (*vector_fmul_scalar)(float *dst, const float *src, float mul, | |
424 int len); | |
425 /** | |
426 * Multiply a vector of floats by concatenated short vectors of | |
427 * floats and by a scalar float. Source and destination vectors | |
428 * must overlap exactly or not at all. | |
429 * [0]: short vectors of length 2, 8-byte aligned | |
430 * [1]: short vectors of length 4, 16-byte aligned | |
431 * @param dst output vector, 16-byte aligned | |
432 * @param src input vector, 16-byte aligned | |
433 * @param sv array of pointers to short vectors | |
434 * @param mul scalar value | |
435 * @param len number of elements in src and dst, multiple of 4 | |
436 */ | |
437 void (*vector_fmul_sv_scalar[2])(float *dst, const float *src, | |
438 const float **sv, float mul, int len); | |
439 /** | |
440 * Multiply short vectors of floats by a scalar float, store | |
441 * concatenated result. | |
442 * [0]: short vectors of length 2, 8-byte aligned | |
443 * [1]: short vectors of length 4, 16-byte aligned | |
444 * @param dst output vector, 16-byte aligned | |
445 * @param sv array of pointers to short vectors | |
446 * @param mul scalar value | |
447 * @param len number of output elements, multiple of 4 | |
448 */ | |
449 void (*sv_fmul_scalar[2])(float *dst, const float **sv, | |
450 float mul, int len); | |
451 /** | |
452 * Calculate the scalar product of two vectors of floats. | |
453 * @param v1 first vector, 16-byte aligned | |
454 * @param v2 second vector, 16-byte aligned | |
455 * @param len length of vectors, multiple of 4 | |
456 */ | |
457 float (*scalarproduct_float)(const float *v1, const float *v2, int len); | |
458 /** | |
459 * Calculate the sum and difference of two vectors of floats. | |
460 * @param v1 first input vector, sum output, 16-byte aligned | |
461 * @param v2 second input vector, difference output, 16-byte aligned | |
462 * @param len length of vectors, multiple of 4 | |
463 */ | |
464 void (*butterflies_float)(float *restrict v1, float *restrict v2, int len); | |
3568
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
465 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
466 /* 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
|
467 * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */ |
7218 | 468 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
|
469 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
|
470 |
1092 | 471 /* (I)DCT */ |
472 void (*fdct)(DCTELEM *block/* align 16*/); | |
1567 | 473 void (*fdct248)(DCTELEM *block/* align 16*/); |
2967 | 474 |
1324
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
475 /* IDCT really*/ |
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
476 void (*idct)(DCTELEM *block/* align 16*/); |
2967 | 477 |
1102 | 478 /** |
1104 | 479 * block -> idct -> clip to unsigned 8 bit -> dest. |
1102 | 480 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) |
5728 | 481 * @param line_size size in bytes of a horizontal line of dest |
1102 | 482 */ |
1092 | 483 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
2967 | 484 |
1102 | 485 /** |
486 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. | |
5728 | 487 * @param line_size size in bytes of a horizontal line of dest |
1102 | 488 */ |
1092 | 489 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
2967 | 490 |
1102 | 491 /** |
1104 | 492 * idct input permutation. |
1241 | 493 * several optimized IDCTs need a permutated input (relative to the normal order of the reference |
494 * IDCT) | |
495 * this permutation must be performed before the idct_put/add, note, normally this can be merged | |
496 * with the zigzag/alternate scan<br> | |
1102 | 497 * an example to avoid confusion: |
498 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...) | |
499 * - (x -> referece dct -> reference idct -> x) | |
500 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x) | |
501 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...) | |
502 */ | |
1092 | 503 uint8_t idct_permutation[64]; |
504 int idct_permutation_type; | |
505 #define FF_NO_IDCT_PERM 1 | |
506 #define FF_LIBMPEG2_IDCT_PERM 2 | |
507 #define FF_SIMPLE_IDCT_PERM 3 | |
508 #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
|
509 #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
|
510 #define FF_SSE2_IDCT_PERM 6 |
1092 | 511 |
1784 | 512 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale); |
513 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); | |
514 #define BASIS_SHIFT 16 | |
515 #define RECON_SHIFT 6 | |
2967 | 516 |
6437 | 517 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
|
518 #define EDGE_WIDTH 16 |
6437 | 519 |
4268 | 520 /* h264 functions */ |
8403 | 521 /* NOTE!!! if you implement any of h264_idct8_add, h264_idct8_add4 then you must implement all of them |
522 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 | |
523 The reason for above, is that no 2 out of one list may use a different permutation. | |
524 */ | |
7678 | 525 void (*h264_idct_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride); |
526 void (*h264_idct8_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride); | |
527 void (*h264_idct_dc_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride); | |
528 void (*h264_idct8_dc_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride); | |
4279 | 529 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
|
530 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
|
531 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
|
532 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
|
533 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
|
534 |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3198
diff
changeset
|
535 void (*prefetch)(void *mem, int stride, int h); |
3245 | 536 |
537 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
3526 | 538 |
9585 | 539 /* mlp/truehd functions */ |
9647
d0fe5dc427f0
mlp: Simplify adressing of state and coeffs arrays for both filters by making
ramiro
parents:
9585
diff
changeset
|
540 void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff, |
d0fe5dc427f0
mlp: Simplify adressing of state and coeffs arrays for both filters by making
ramiro
parents:
9585
diff
changeset
|
541 int firorder, int iirorder, |
9585 | 542 unsigned int filter_shift, int32_t mask, int blocksize, |
543 int32_t *sample_buffer); | |
544 | |
3526 | 545 /* vc1 functions */ |
546 void (*vc1_inv_trans_8x8)(DCTELEM *b); | |
5997
90de28dfd8d6
Switch VC-1 decoder to output decoded residual immediately.
kostya
parents:
5907
diff
changeset
|
547 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
|
548 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
|
549 void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block); |
9859
7a116de63777
idct_dc for VC-1/WMV3 decoder; ~11% faster decoding overall.
darkshikari
parents:
9658
diff
changeset
|
550 void (*vc1_inv_trans_8x8_dc)(uint8_t *dest, int line_size, DCTELEM *block); |
7a116de63777
idct_dc for VC-1/WMV3 decoder; ~11% faster decoding overall.
darkshikari
parents:
9658
diff
changeset
|
551 void (*vc1_inv_trans_8x4_dc)(uint8_t *dest, int line_size, DCTELEM *block); |
7a116de63777
idct_dc for VC-1/WMV3 decoder; ~11% faster decoding overall.
darkshikari
parents:
9658
diff
changeset
|
552 void (*vc1_inv_trans_4x8_dc)(uint8_t *dest, int line_size, DCTELEM *block); |
7a116de63777
idct_dc for VC-1/WMV3 decoder; ~11% faster decoding overall.
darkshikari
parents:
9658
diff
changeset
|
553 void (*vc1_inv_trans_4x4_dc)(uint8_t *dest, int line_size, DCTELEM *block); |
4239 | 554 void (*vc1_v_overlap)(uint8_t* src, int stride); |
555 void (*vc1_h_overlap)(uint8_t* src, int stride); | |
9443
3970fe47fea3
Split VC1 loop filter into separate functions for h/v and size
conrad
parents:
9442
diff
changeset
|
556 void (*vc1_v_loop_filter4)(uint8_t *src, int stride, int pq); |
3970fe47fea3
Split VC1 loop filter into separate functions for h/v and size
conrad
parents:
9442
diff
changeset
|
557 void (*vc1_h_loop_filter4)(uint8_t *src, int stride, int pq); |
3970fe47fea3
Split VC1 loop filter into separate functions for h/v and size
conrad
parents:
9442
diff
changeset
|
558 void (*vc1_v_loop_filter8)(uint8_t *src, int stride, int pq); |
3970fe47fea3
Split VC1 loop filter into separate functions for h/v and size
conrad
parents:
9442
diff
changeset
|
559 void (*vc1_h_loop_filter8)(uint8_t *src, int stride, int pq); |
3970fe47fea3
Split VC1 loop filter into separate functions for h/v and size
conrad
parents:
9442
diff
changeset
|
560 void (*vc1_v_loop_filter16)(uint8_t *src, int stride, int pq); |
3970fe47fea3
Split VC1 loop filter into separate functions for h/v and size
conrad
parents:
9442
diff
changeset
|
561 void (*vc1_h_loop_filter16)(uint8_t *src, int stride, int pq); |
3526 | 562 /* put 8x8 block with bicubic interpolation and quarterpel precision |
563 * last argument is actually round value instead of height | |
564 */ | |
565 op_pixels_func put_vc1_mspel_pixels_tab[16]; | |
9437 | 566 op_pixels_func avg_vc1_mspel_pixels_tab[16]; |
5887 | 567 |
568 /* intrax8 functions */ | |
5907
fbd10e6dfbe1
cosmetics: Fix spacial --> spatial typo in function names.
diego
parents:
5887
diff
changeset
|
569 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
|
570 void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize, |
5887 | 571 int * range, int * sum, int edges); |
572 | |
7203
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7135
diff
changeset
|
573 /** |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7135
diff
changeset
|
574 * Calculate scalar product of two vectors. |
7232
cc55dd004819
Monkey's Audio decoder vector functions work on input with length
kostya
parents:
7218
diff
changeset
|
575 * @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
|
576 * @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
|
577 */ |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7135
diff
changeset
|
578 int32_t (*scalarproduct_int16)(int16_t *v1, int16_t *v2/*align 16*/, int len, int shift); |
10644 | 579 /* ape functions */ |
580 /** | |
581 * Calculate scalar product of v1 and v2, | |
582 * and v1[i] += v3[i] * mul | |
583 * @param len length of vectors, should be multiple of 16 | |
584 */ | |
585 int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, int16_t *v2, int16_t *v3, int len, int mul); | |
8232 | 586 |
8233 | 587 /* rv30 functions */ |
588 qpel_mc_func put_rv30_tpel_pixels_tab[4][16]; | |
589 qpel_mc_func avg_rv30_tpel_pixels_tab[4][16]; | |
590 | |
8232 | 591 /* rv40 functions */ |
592 qpel_mc_func put_rv40_qpel_pixels_tab[4][16]; | |
593 qpel_mc_func avg_rv40_qpel_pixels_tab[4][16]; | |
594 h264_chroma_mc_func put_rv40_chroma_pixels_tab[3]; | |
595 h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3]; | |
11231 | 596 |
597 /* bink functions */ | |
598 op_fill_func fill_block_tab[2]; | |
599 void (*scale_block)(const uint8_t src[64]/*align 8*/, uint8_t *dst/*align 8*/, int linesize); | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
600 } DSPContext; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
601 |
4197 | 602 void dsputil_static_init(void); |
1092 | 603 void dsputil_init(DSPContext* p, AVCodecContext *avctx); |
0 | 604 |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4279
diff
changeset
|
605 int ff_check_alignment(void); |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4279
diff
changeset
|
606 |
764 | 607 /** |
608 * permute block according to permuatation. | |
609 * @param last last non zero element in scantable order | |
610 */ | |
1064 | 611 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); |
34 | 612 |
1729 | 613 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type); |
614 | |
2979 | 615 #define BYTE_VEC32(c) ((c)*0x01010101UL) |
1264 | 616 |
617 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b) | |
618 { | |
619 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
620 } | |
621 | |
622 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b) | |
623 { | |
624 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
625 } | |
626 | |
2184 | 627 static inline int get_penalty_factor(int lambda, int lambda2, int type){ |
628 switch(type&0xFF){ | |
629 default: | |
630 case FF_CMP_SAD: | |
631 return lambda>>FF_LAMBDA_SHIFT; | |
632 case FF_CMP_DCT: | |
633 return (3*lambda)>>(FF_LAMBDA_SHIFT+1); | |
634 case FF_CMP_W53: | |
635 return (4*lambda)>>(FF_LAMBDA_SHIFT); | |
636 case FF_CMP_W97: | |
637 return (2*lambda)>>(FF_LAMBDA_SHIFT); | |
638 case FF_CMP_SATD: | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
639 case FF_CMP_DCT264: |
2184 | 640 return (2*lambda)>>FF_LAMBDA_SHIFT; |
641 case FF_CMP_RD: | |
642 case FF_CMP_PSNR: | |
643 case FF_CMP_SSE: | |
644 case FF_CMP_NSSE: | |
645 return lambda2>>FF_LAMBDA_SHIFT; | |
646 case FF_CMP_BIT: | |
647 return 1; | |
648 } | |
649 } | |
650 | |
1102 | 651 /** |
1104 | 652 * Empty mmx state. |
1102 | 653 * this must be called between any dsp function and float/double code. |
654 * for example sin(); dsp->idct_put(); emms_c(); cos() | |
655 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
656 #define emms_c() |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
657 |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
658 /* should be defined by architectures supporting |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
659 one or more MultiMedia extension */ |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
660 int mm_support(void); |
10115 | 661 extern int mm_flags; |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
662 |
5149 | 663 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx); |
8359 | 664 void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx); |
5149 | 665 void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx); |
666 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx); | |
667 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx); | |
668 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); | |
669 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx); | |
670 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx); | |
671 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx); | |
672 | |
11485 | 673 void ff_dsputil_init_dwt(DSPContext *c); |
11376
0b220468ba0d
Move prototypes for various dsputil init functions to dsputil.h
mru
parents:
11370
diff
changeset
|
674 void ff_cavsdsp_init(DSPContext* c, AVCodecContext *avctx); |
0b220468ba0d
Move prototypes for various dsputil init functions to dsputil.h
mru
parents:
11370
diff
changeset
|
675 void ff_rv30dsp_init(DSPContext* c, AVCodecContext *avctx); |
0b220468ba0d
Move prototypes for various dsputil init functions to dsputil.h
mru
parents:
11370
diff
changeset
|
676 void ff_rv40dsp_init(DSPContext* c, AVCodecContext *avctx); |
0b220468ba0d
Move prototypes for various dsputil init functions to dsputil.h
mru
parents:
11370
diff
changeset
|
677 void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx); |
0b220468ba0d
Move prototypes for various dsputil init functions to dsputil.h
mru
parents:
11370
diff
changeset
|
678 void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx); |
0b220468ba0d
Move prototypes for various dsputil init functions to dsputil.h
mru
parents:
11370
diff
changeset
|
679 void ff_mlp_init(DSPContext* c, AVCodecContext *avctx); |
0b220468ba0d
Move prototypes for various dsputil init functions to dsputil.h
mru
parents:
11370
diff
changeset
|
680 void ff_mlp_init_x86(DSPContext* c, AVCodecContext *avctx); |
0b220468ba0d
Move prototypes for various dsputil init functions to dsputil.h
mru
parents:
11370
diff
changeset
|
681 |
8590 | 682 #if HAVE_MMX |
0 | 683 |
862 | 684 #undef emms_c |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
685 |
0 | 686 static inline void emms(void) |
687 { | |
8031 | 688 __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
|
689 } |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
690 |
936 | 691 |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
692 #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
|
693 {\ |
8104
0d108ec85620
Remove duplicated MM_* macros for CPU capabilities from dsputil.h.
rathann
parents:
8031
diff
changeset
|
694 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
|
695 emms();\ |
0 | 696 } |
697 | |
8590 | 698 #elif ARCH_ARM |
62 | 699 |
8590 | 700 #if HAVE_NEON |
7687
e5b5a9af1b68
ARM: set STRIDE_ALIGN and DECLARE_ALIGNED_8 to 16 for NEON
mru
parents:
7678
diff
changeset
|
701 # define STRIDE_ALIGN 16 |
e5b5a9af1b68
ARM: set STRIDE_ALIGN and DECLARE_ALIGNED_8 to 16 for NEON
mru
parents:
7678
diff
changeset
|
702 #endif |
e5b5a9af1b68
ARM: set STRIDE_ALIGN and DECLARE_ALIGNED_8 to 16 for NEON
mru
parents:
7678
diff
changeset
|
703 |
8590 | 704 #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
|
705 |
2324 | 706 #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
|
707 |
8590 | 708 #elif HAVE_MMI |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
709 |
2324 | 710 #define STRIDE_ALIGN 16 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
711 |
7135
6bd6a2da306e
Define mm_flags/support to be 0 on architectures where they don't exist.
astrange
parents:
7094
diff
changeset
|
712 #else |
6bd6a2da306e
Define mm_flags/support to be 0 on architectures where they don't exist.
astrange
parents:
7094
diff
changeset
|
713 |
6bd6a2da306e
Define mm_flags/support to be 0 on architectures where they don't exist.
astrange
parents:
7094
diff
changeset
|
714 #define mm_flags 0 |
6bd6a2da306e
Define mm_flags/support to be 0 on architectures where they don't exist.
astrange
parents:
7094
diff
changeset
|
715 #define mm_support() 0 |
6bd6a2da306e
Define mm_flags/support to be 0 on architectures where they don't exist.
astrange
parents:
7094
diff
changeset
|
716 |
6363
7ebd1cdb2142
clean up definition of DECLARE_ALIGNED_8 and STRIDE_ALIGN
mru
parents:
6241
diff
changeset
|
717 #endif |
3728 | 718 |
6363
7ebd1cdb2142
clean up definition of DECLARE_ALIGNED_8 and STRIDE_ALIGN
mru
parents:
6241
diff
changeset
|
719 #ifndef STRIDE_ALIGN |
7ebd1cdb2142
clean up definition of DECLARE_ALIGNED_8 and STRIDE_ALIGN
mru
parents:
6241
diff
changeset
|
720 # define STRIDE_ALIGN 8 |
0 | 721 #endif |
722 | |
11194
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
723 #define LOCAL_ALIGNED(a, t, v, s, ...) \ |
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
724 uint8_t la_##v[sizeof(t s __VA_ARGS__) + (a)]; \ |
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
725 t (*v) __VA_ARGS__ = (void *)FFALIGN((uintptr_t)la_##v, a) |
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
726 |
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
727 #if HAVE_LOCAL_ALIGNED_8 |
11369 | 728 # define LOCAL_ALIGNED_8(t, v, s, ...) DECLARE_ALIGNED(8, t, v) s __VA_ARGS__ |
11194
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
729 #else |
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
730 # define LOCAL_ALIGNED_8(t, v, s, ...) LOCAL_ALIGNED(8, t, v, s, __VA_ARGS__) |
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
731 #endif |
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
732 |
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
733 #if HAVE_LOCAL_ALIGNED_16 |
11369 | 734 # define LOCAL_ALIGNED_16(t, v, s, ...) DECLARE_ALIGNED(16, t, v) s __VA_ARGS__ |
11194
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
735 #else |
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
736 # define LOCAL_ALIGNED_16(t, v, s, ...) LOCAL_ALIGNED(16, t, v, s, __VA_ARGS__) |
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
737 #endif |
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
738 |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
739 /* PSNR */ |
1064 | 740 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
|
741 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
|
742 AVCodecContext *avctx); |
781 | 743 |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
5997
diff
changeset
|
744 #define WRAPPER8_16(name8, name16)\ |
1708 | 745 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ |
746 return name8(s, dst , src , stride, h)\ | |
747 +name8(s, dst+8 , src+8 , stride, h);\ | |
748 } | |
749 | |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
5997
diff
changeset
|
750 #define WRAPPER8_16_SQ(name8, name16)\ |
1708 | 751 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ |
752 int score=0;\ | |
753 score +=name8(s, dst , src , stride, 8);\ | |
754 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
755 if(h==16){\ | |
756 dst += 8*stride;\ | |
757 src += 8*stride;\ | |
758 score +=name8(s, dst , src , stride, 8);\ | |
759 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
760 }\ | |
761 return score;\ | |
936 | 762 } |
763 | |
4240 | 764 |
9436 | 765 static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) |
4240 | 766 { |
767 int i; | |
768 for(i=0; i<h; i++) | |
769 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
770 AV_WN16(dst , AV_RN16(src )); |
4240 | 771 dst+=dstStride; |
772 src+=srcStride; | |
773 } | |
774 } | |
775 | |
9436 | 776 static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) |
4240 | 777 { |
778 int i; | |
779 for(i=0; i<h; i++) | |
780 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
781 AV_WN32(dst , AV_RN32(src )); |
4240 | 782 dst+=dstStride; |
783 src+=srcStride; | |
784 } | |
785 } | |
786 | |
9436 | 787 static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) |
4240 | 788 { |
789 int i; | |
790 for(i=0; i<h; i++) | |
791 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
792 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
|
793 AV_WN32(dst+4 , AV_RN32(src+4 )); |
4240 | 794 dst+=dstStride; |
795 src+=srcStride; | |
796 } | |
797 } | |
798 | |
9436 | 799 static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) |
4240 | 800 { |
801 int i; | |
802 for(i=0; i<h; i++) | |
803 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
804 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
|
805 AV_WN32(dst+4 , AV_RN32(src+4 )); |
4240 | 806 dst[8]= src[8]; |
807 dst+=dstStride; | |
808 src+=srcStride; | |
809 } | |
810 } | |
811 | |
9436 | 812 static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) |
4240 | 813 { |
814 int i; | |
815 for(i=0; i<h; i++) | |
816 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
817 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
|
818 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
|
819 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
|
820 AV_WN32(dst+12, AV_RN32(src+12)); |
4240 | 821 dst+=dstStride; |
822 src+=srcStride; | |
823 } | |
824 } | |
825 | |
9436 | 826 static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) |
4240 | 827 { |
828 int i; | |
829 for(i=0; i<h; i++) | |
830 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
831 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
|
832 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
|
833 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
|
834 AV_WN32(dst+12, AV_RN32(src+12)); |
4240 | 835 dst[16]= src[16]; |
836 dst+=dstStride; | |
837 src+=srcStride; | |
838 } | |
839 } | |
840 | |
7760 | 841 #endif /* AVCODEC_DSPUTIL_H */ |