Mercurial > libavcodec.hg
annotate dsputil.h @ 12197:fbf4d5b1b664 libavcodec
Remove FF_MM_SSE2/3 flags for CPUs where this is generally not faster than
regular MMX code. Examples of this are the Core1 CPU. Instead, set a new flag,
FF_MM_SSE2/3SLOW, which can be checked for particular SSE2/3 functions that
have been checked specifically on such CPUs and are actually faster than
their MMX counterparts.
In addition, use this flag to enable particular VP8 and LPC SSE2 functions
that are faster than their MMX counterparts.
Based on a patch by Loren Merritt <lorenm AT u washington edu>.
author | rbultje |
---|---|
date | Mon, 19 Jul 2010 22:38:23 +0000 |
parents | d3551fcf4c1c |
children | 1241c824de46 |
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 /** | |
11644
7dd2a45249a9
Remove explicit filename from Doxygen @file commands.
diego
parents:
11637
diff
changeset
|
24 * @file |
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*/); | |
11637 | 89 void ff_vp3_idct_dc_add_c(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/); |
0 | 90 |
7995 | 91 void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values); |
92 void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values); | |
93 | |
8785
bee83b3f9a6b
move vp6_filter_diag4() to a new vp6dsp.c file and use it throught dsputil
aurel
parents:
8760
diff
changeset
|
94 /* VP6 DSP functions */ |
bee83b3f9a6b
move vp6_filter_diag4() to a new vp6dsp.c file and use it throught dsputil
aurel
parents:
8760
diff
changeset
|
95 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
|
96 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
|
97 |
11379
4b0cc1b97def
Move some dsp func prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11378
diff
changeset
|
98 /* Bink functions */ |
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_c (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_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
|
101 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
|
102 |
4b0cc1b97def
Move some dsp func prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11378
diff
changeset
|
103 /* CAVS functions */ |
4b0cc1b97def
Move some dsp func prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11378
diff
changeset
|
104 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
|
105 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
|
106 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
|
107 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
|
108 |
11378
f46b68960464
Move some VC1 dsp prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11376
diff
changeset
|
109 /* VC1 functions */ |
f46b68960464
Move some VC1 dsp prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11376
diff
changeset
|
110 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
|
111 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
|
112 |
11379
4b0cc1b97def
Move some dsp func prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11378
diff
changeset
|
113 /* EA functions */ |
4b0cc1b97def
Move some dsp func prototypes to dsputil.h; they are defined in dsputil.c
mru
parents:
11378
diff
changeset
|
114 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
|
115 |
3245 | 116 /* 1/2^n downscaling functions from imgconvert.c */ |
117 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
118 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
119 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
120 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
121 | |
3248
7aa9f80e7954
mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents:
3245
diff
changeset
|
122 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
|
123 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
|
124 |
675 | 125 /* minimum alignment rules ;) |
6486 | 126 If you notice errors in the align stuff, need more alignment for some ASM code |
127 for some CPU or need to use a function with less aligned data then send a mail | |
128 to the ffmpeg-devel mailing list, ... | |
675 | 129 |
6486 | 130 !warning These alignments might not match reality, (missing attribute((align)) |
131 stuff somewhere possible). | |
6488 | 132 I (Michael) did not check them, these are just the alignments which I think |
6486 | 133 could be reached easily ... |
675 | 134 |
135 !future video codecs might need functions with less strict alignment | |
136 */ | |
0 | 137 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
138 /* |
1064 | 139 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size); |
140 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride); | |
141 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
142 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
296 | 143 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
|
144 */ |
0 | 145 |
146 /* add and put pixel (decoding) */ | |
675 | 147 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 |
1709 | 148 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4 |
1064 | 149 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
|
150 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 | 151 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); |
1168 | 152 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y); |
0 | 153 |
11231 | 154 typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h); |
155 | |
984 | 156 #define DEF_OLD_QPEL(name)\ |
1064 | 157 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ |
158 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ | |
159 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
|
160 |
984 | 161 DEF_OLD_QPEL(qpel16_mc11_old_c) |
162 DEF_OLD_QPEL(qpel16_mc31_old_c) | |
163 DEF_OLD_QPEL(qpel16_mc12_old_c) | |
164 DEF_OLD_QPEL(qpel16_mc32_old_c) | |
165 DEF_OLD_QPEL(qpel16_mc13_old_c) | |
166 DEF_OLD_QPEL(qpel16_mc33_old_c) | |
167 DEF_OLD_QPEL(qpel8_mc11_old_c) | |
168 DEF_OLD_QPEL(qpel8_mc31_old_c) | |
169 DEF_OLD_QPEL(qpel8_mc12_old_c) | |
170 DEF_OLD_QPEL(qpel8_mc32_old_c) | |
171 DEF_OLD_QPEL(qpel8_mc13_old_c) | |
172 DEF_OLD_QPEL(qpel8_mc33_old_c) | |
651 | 173 |
174 #define CALL_2X_PIXELS(a, b, n)\ | |
175 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ | |
176 b(block , pixels , line_size, h);\ | |
177 b(block+n, pixels+n, line_size, h);\ | |
178 } | |
255 | 179 |
0 | 180 /* motion estimation */ |
1709 | 181 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2 |
5127 | 182 // although currently h<4 is not used as functions with width <8 are neither used nor implemented |
1708 | 183 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 | 184 |
1102 | 185 /** |
6438 | 186 * Scantable. |
187 */ | |
188 typedef struct ScanTable{ | |
189 const uint8_t *scantable; | |
190 uint8_t permutated[64]; | |
191 uint8_t raster_end[64]; | |
8590 | 192 #if ARCH_PPC |
6438 | 193 /** Used by dct_quantize_altivec to find last-non-zero */ |
10961
34a65026fa06
Move array specifiers outside DECLARE_ALIGNED() invocations
mru
parents:
10944
diff
changeset
|
194 DECLARE_ALIGNED(16, uint8_t, inverse)[64]; |
6438 | 195 #endif |
196 } ScanTable; | |
197 | |
198 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable); | |
199 | |
11784 | 200 void ff_emulated_edge_mc(uint8_t *buf, const uint8_t *src, int linesize, |
6445 | 201 int block_w, int block_h, |
202 int src_x, int src_y, int w, int h); | |
203 | |
6438 | 204 /** |
1102 | 205 * DSPContext. |
206 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
207 typedef struct DSPContext { |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
208 /* pixel ops : interface with DCT */ |
1064 | 209 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); |
210 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); | |
211 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
|
212 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
11231 | 213 void (*put_pixels_nonclamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
1064 | 214 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
2763 | 215 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size); |
216 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
|
217 int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/); |
1136 | 218 /** |
219 * translational global motion compensation. | |
220 */ | |
1064 | 221 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); |
1136 | 222 /** |
223 * global motion compensation. | |
224 */ | |
1064 | 225 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy, |
2979 | 226 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); |
8288 | 227 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
|
228 void (*clear_blocks)(DCTELEM *blocks/*align 16*/); |
1064 | 229 int (*pix_sum)(uint8_t * pix, int line_size); |
230 int (*pix_norm1)(uint8_t * pix, int line_size); | |
1708 | 231 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4 |
2967 | 232 |
8976
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
233 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
|
234 me_cmp_func sse[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
235 me_cmp_func hadamard8_diff[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
236 me_cmp_func dct_sad[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
237 me_cmp_func quant_psnr[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
238 me_cmp_func bit[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
239 me_cmp_func rd[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
240 me_cmp_func vsad[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
241 me_cmp_func vsse[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
242 me_cmp_func nsse[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
243 me_cmp_func w53[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
244 me_cmp_func w97[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
245 me_cmp_func dct_max[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
246 me_cmp_func dct264_sad[6]; |
936 | 247 |
8976
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
248 me_cmp_func me_pre_cmp[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
249 me_cmp_func me_cmp[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
250 me_cmp_func me_sub_cmp[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
251 me_cmp_func mb_cmp[6]; |
e7d87561b42b
Making the arrays accomodate an extra intra 8x8 cmp function
romansh
parents:
8785
diff
changeset
|
252 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
|
253 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
|
254 |
5255 | 255 int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2, |
256 int size); | |
4749 | 257 |
1136 | 258 /** |
259 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
4751 | 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 where the result is stored |
264 * @param pixels source | |
265 * @param line_size number of bytes in a horizontal line of block | |
266 * @param h height | |
267 */ | |
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
|
268 op_pixels_func put_pixels_tab[4][4]; |
1136 | 269 |
270 /** | |
271 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
2967 | 272 * This is an array[4][4] of motion compensation functions for 4 |
1713 | 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 into which the result is averaged (a+b+1)>>1 |
276 * @param pixels source | |
277 * @param line_size number of bytes in a horizontal line of block | |
278 * @param h height | |
279 */ | |
1319 | 280 op_pixels_func avg_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 where the result is stored |
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 put_no_rnd_pixels_tab[4][4]; |
1136 | 293 |
294 /** | |
295 * Halfpel motion compensation with no rounding (a+b)>>1. | |
4751 | 296 * this is an array[2][4] of motion compensation functions for 2 |
1225 | 297 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 298 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 299 * @param block destination into which the result is averaged (a+b)>>1 |
300 * @param pixels source | |
301 * @param line_size number of bytes in a horizontal line of block | |
302 * @param h height | |
303 */ | |
2075 | 304 op_pixels_func avg_no_rnd_pixels_tab[4][4]; |
2967 | 305 |
1864 | 306 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 | 307 |
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
|
308 /** |
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 * Thirdpel motion compensation with rounding (a+b+1)>>1. |
4751 | 310 * this is an array[12] of motion compensation functions for the 9 thirdpe |
311 * 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
|
312 * *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
|
313 * @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
|
314 * @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
|
315 * @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
|
316 * @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
|
317 */ |
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 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
1319 | 319 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
320 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
321 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
|
322 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
|
323 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
|
324 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; |
936 | 325 qpel_mc_func put_mspel_pixels_tab[8]; |
2967 | 326 |
1168 | 327 /** |
4751 | 328 * h264 Chroma MC |
1168 | 329 */ |
330 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
|
331 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3]; |
3663 | 332 /* 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
|
333 h264_chroma_mc_func put_no_rnd_vc1_chroma_pixels_tab[3]; |
9440 | 334 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
|
335 |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3010
diff
changeset
|
336 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
|
337 qpel_mc_func avg_h264_qpel_pixels_tab[4][16]; |
2967 | 338 |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3746
diff
changeset
|
339 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
|
340 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
|
341 |
3395
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
342 /* AVS specific */ |
adccbf4a1040
CAVS decoder by (Stefan Gehrer stefan.gehrer gmx.de)
michael
parents:
3279
diff
changeset
|
343 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
|
344 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
|
345 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
|
346 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
|
347 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
|
348 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
|
349 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
|
350 |
1708 | 351 me_cmp_func pix_abs[2][4]; |
2967 | 352 |
866 | 353 /* huffyuv specific */ |
354 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); | |
6384 | 355 void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w); |
936 | 356 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); |
1527 | 357 /** |
358 * subtract huffyuv's variant of median prediction | |
359 * note, this might read from src1[-1], src2[-1] | |
360 */ | |
10431 | 361 void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top); |
362 void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top); | |
10430 | 363 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
|
364 void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha); |
6384 | 365 /* this might write to dst[w] */ |
366 void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp); | |
6241 | 367 void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w); |
2633 | 368 |
1644 | 369 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale); |
370 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale); | |
371 | |
2045 | 372 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
|
373 |
5887 | 374 void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale); |
375 void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale); | |
376 | |
11637 | 377 void (*vp3_idct_dc_add)(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/); |
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 */ |
10424
94595d0e617c
Move autocorrelation function from flacenc.c to lpc.c. Also rename the
jbr
parents:
10420
diff
changeset
|
388 void (*lpc_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 */ |
10300
4d1b9ca628fc
Drop unused args from vector_fmul_add_add, simpify code, and rename
mru
parents:
10219
diff
changeset
|
393 void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len); |
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); | |
10105 | 398 void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */); |
10219 | 399 /** |
400 * Multiply a vector of floats by a scalar float. Source and | |
401 * destination vectors must overlap exactly or not at all. | |
402 * @param dst result vector, 16-byte aligned | |
403 * @param src input vector, 16-byte aligned | |
404 * @param mul scalar value | |
405 * @param len length of vector, multiple of 4 | |
406 */ | |
407 void (*vector_fmul_scalar)(float *dst, const float *src, float mul, | |
408 int len); | |
409 /** | |
410 * Multiply a vector of floats by concatenated short vectors of | |
411 * floats and by a scalar float. Source and destination vectors | |
412 * must overlap exactly or not at all. | |
413 * [0]: short vectors of length 2, 8-byte aligned | |
414 * [1]: short vectors of length 4, 16-byte aligned | |
415 * @param dst output vector, 16-byte aligned | |
416 * @param src input vector, 16-byte aligned | |
417 * @param sv array of pointers to short vectors | |
418 * @param mul scalar value | |
419 * @param len number of elements in src and dst, multiple of 4 | |
420 */ | |
421 void (*vector_fmul_sv_scalar[2])(float *dst, const float *src, | |
422 const float **sv, float mul, int len); | |
423 /** | |
424 * Multiply short vectors of floats by a scalar float, store | |
425 * concatenated result. | |
426 * [0]: short vectors of length 2, 8-byte aligned | |
427 * [1]: short vectors of length 4, 16-byte aligned | |
428 * @param dst output vector, 16-byte aligned | |
429 * @param sv array of pointers to short vectors | |
430 * @param mul scalar value | |
431 * @param len number of output elements, multiple of 4 | |
432 */ | |
433 void (*sv_fmul_scalar[2])(float *dst, const float **sv, | |
434 float mul, int len); | |
435 /** | |
436 * Calculate the scalar product of two vectors of floats. | |
437 * @param v1 first vector, 16-byte aligned | |
438 * @param v2 second vector, 16-byte aligned | |
439 * @param len length of vectors, multiple of 4 | |
440 */ | |
441 float (*scalarproduct_float)(const float *v1, const float *v2, int len); | |
442 /** | |
443 * Calculate the sum and difference of two vectors of floats. | |
444 * @param v1 first input vector, sum output, 16-byte aligned | |
445 * @param v2 second input vector, difference output, 16-byte aligned | |
446 * @param len length of vectors, multiple of 4 | |
447 */ | |
448 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
|
449 |
945caa35ee9a
sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents:
3555
diff
changeset
|
450 /* 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
|
451 * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */ |
7218 | 452 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
|
453 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
|
454 |
1092 | 455 /* (I)DCT */ |
456 void (*fdct)(DCTELEM *block/* align 16*/); | |
1567 | 457 void (*fdct248)(DCTELEM *block/* align 16*/); |
2967 | 458 |
1324
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
459 /* IDCT really*/ |
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
460 void (*idct)(DCTELEM *block/* align 16*/); |
2967 | 461 |
1102 | 462 /** |
1104 | 463 * block -> idct -> clip to unsigned 8 bit -> dest. |
1102 | 464 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) |
5728 | 465 * @param line_size size in bytes of a horizontal line of dest |
1102 | 466 */ |
1092 | 467 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
2967 | 468 |
1102 | 469 /** |
470 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. | |
5728 | 471 * @param line_size size in bytes of a horizontal line of dest |
1102 | 472 */ |
1092 | 473 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
2967 | 474 |
1102 | 475 /** |
1104 | 476 * idct input permutation. |
1241 | 477 * several optimized IDCTs need a permutated input (relative to the normal order of the reference |
478 * IDCT) | |
479 * this permutation must be performed before the idct_put/add, note, normally this can be merged | |
480 * with the zigzag/alternate scan<br> | |
1102 | 481 * an example to avoid confusion: |
482 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...) | |
483 * - (x -> referece dct -> reference idct -> x) | |
484 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x) | |
485 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...) | |
486 */ | |
1092 | 487 uint8_t idct_permutation[64]; |
488 int idct_permutation_type; | |
489 #define FF_NO_IDCT_PERM 1 | |
490 #define FF_LIBMPEG2_IDCT_PERM 2 | |
491 #define FF_SIMPLE_IDCT_PERM 3 | |
492 #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
|
493 #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
|
494 #define FF_SSE2_IDCT_PERM 6 |
1092 | 495 |
1784 | 496 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale); |
497 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); | |
498 #define BASIS_SHIFT 16 | |
499 #define RECON_SHIFT 6 | |
2967 | 500 |
6437 | 501 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
|
502 #define EDGE_WIDTH 16 |
6437 | 503 |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3198
diff
changeset
|
504 void (*prefetch)(void *mem, int stride, int h); |
3245 | 505 |
506 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); | |
3526 | 507 |
9585 | 508 /* mlp/truehd functions */ |
9647
d0fe5dc427f0
mlp: Simplify adressing of state and coeffs arrays for both filters by making
ramiro
parents:
9585
diff
changeset
|
509 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
|
510 int firorder, int iirorder, |
9585 | 511 unsigned int filter_shift, int32_t mask, int blocksize, |
512 int32_t *sample_buffer); | |
513 | |
3526 | 514 /* vc1 functions */ |
515 void (*vc1_inv_trans_8x8)(DCTELEM *b); | |
5997
90de28dfd8d6
Switch VC-1 decoder to output decoded residual immediately.
kostya
parents:
5907
diff
changeset
|
516 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
|
517 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
|
518 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
|
519 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
|
520 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
|
521 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
|
522 void (*vc1_inv_trans_4x4_dc)(uint8_t *dest, int line_size, DCTELEM *block); |
4239 | 523 void (*vc1_v_overlap)(uint8_t* src, int stride); |
524 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
|
525 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
|
526 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
|
527 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
|
528 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
|
529 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
|
530 void (*vc1_h_loop_filter16)(uint8_t *src, int stride, int pq); |
3526 | 531 /* put 8x8 block with bicubic interpolation and quarterpel precision |
532 * last argument is actually round value instead of height | |
533 */ | |
534 op_pixels_func put_vc1_mspel_pixels_tab[16]; | |
9437 | 535 op_pixels_func avg_vc1_mspel_pixels_tab[16]; |
5887 | 536 |
537 /* intrax8 functions */ | |
5907
fbd10e6dfbe1
cosmetics: Fix spacial --> spatial typo in function names.
diego
parents:
5887
diff
changeset
|
538 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
|
539 void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize, |
5887 | 540 int * range, int * sum, int edges); |
541 | |
7203
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7135
diff
changeset
|
542 /** |
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7135
diff
changeset
|
543 * Calculate scalar product of two vectors. |
7232
cc55dd004819
Monkey's Audio decoder vector functions work on input with length
kostya
parents:
7218
diff
changeset
|
544 * @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
|
545 * @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
|
546 */ |
11981 | 547 int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len, int shift); |
10644 | 548 /* ape functions */ |
549 /** | |
550 * Calculate scalar product of v1 and v2, | |
551 * and v1[i] += v3[i] * mul | |
552 * @param len length of vectors, should be multiple of 16 | |
553 */ | |
11981 | 554 int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul); |
8232 | 555 |
8233 | 556 /* rv30 functions */ |
557 qpel_mc_func put_rv30_tpel_pixels_tab[4][16]; | |
558 qpel_mc_func avg_rv30_tpel_pixels_tab[4][16]; | |
559 | |
8232 | 560 /* rv40 functions */ |
561 qpel_mc_func put_rv40_qpel_pixels_tab[4][16]; | |
562 qpel_mc_func avg_rv40_qpel_pixels_tab[4][16]; | |
563 h264_chroma_mc_func put_rv40_chroma_pixels_tab[3]; | |
564 h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3]; | |
11231 | 565 |
566 /* bink functions */ | |
567 op_fill_func fill_block_tab[2]; | |
568 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
|
569 } DSPContext; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
570 |
4197 | 571 void dsputil_static_init(void); |
1092 | 572 void dsputil_init(DSPContext* p, AVCodecContext *avctx); |
0 | 573 |
4281
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4279
diff
changeset
|
574 int ff_check_alignment(void); |
de525a2b41db
ff_check_alignment to warn the user about a missaligned stack
michael
parents:
4279
diff
changeset
|
575 |
764 | 576 /** |
577 * permute block according to permuatation. | |
578 * @param last last non zero element in scantable order | |
579 */ | |
1064 | 580 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); |
34 | 581 |
1729 | 582 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type); |
583 | |
2979 | 584 #define BYTE_VEC32(c) ((c)*0x01010101UL) |
1264 | 585 |
586 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b) | |
587 { | |
588 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
589 } | |
590 | |
591 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b) | |
592 { | |
593 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
594 } | |
595 | |
2184 | 596 static inline int get_penalty_factor(int lambda, int lambda2, int type){ |
597 switch(type&0xFF){ | |
598 default: | |
599 case FF_CMP_SAD: | |
600 return lambda>>FF_LAMBDA_SHIFT; | |
601 case FF_CMP_DCT: | |
602 return (3*lambda)>>(FF_LAMBDA_SHIFT+1); | |
603 case FF_CMP_W53: | |
604 return (4*lambda)>>(FF_LAMBDA_SHIFT); | |
605 case FF_CMP_W97: | |
606 return (2*lambda)>>(FF_LAMBDA_SHIFT); | |
607 case FF_CMP_SATD: | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
608 case FF_CMP_DCT264: |
2184 | 609 return (2*lambda)>>FF_LAMBDA_SHIFT; |
610 case FF_CMP_RD: | |
611 case FF_CMP_PSNR: | |
612 case FF_CMP_SSE: | |
613 case FF_CMP_NSSE: | |
614 return lambda2>>FF_LAMBDA_SHIFT; | |
615 case FF_CMP_BIT: | |
616 return 1; | |
617 } | |
618 } | |
619 | |
1102 | 620 /** |
1104 | 621 * Empty mmx state. |
1102 | 622 * this must be called between any dsp function and float/double code. |
623 * for example sin(); dsp->idct_put(); emms_c(); cos() | |
624 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
625 #define emms_c() |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
626 |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
627 /* should be defined by architectures supporting |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
628 one or more MultiMedia extension */ |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
629 int mm_support(void); |
10115 | 630 extern int mm_flags; |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
631 |
5149 | 632 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx); |
8359 | 633 void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx); |
5149 | 634 void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx); |
635 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx); | |
636 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx); | |
637 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); | |
638 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx); | |
639 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx); | |
640 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx); | |
641 | |
11485 | 642 void ff_dsputil_init_dwt(DSPContext *c); |
11376
0b220468ba0d
Move prototypes for various dsputil init functions to dsputil.h
mru
parents:
11370
diff
changeset
|
643 void ff_cavsdsp_init(DSPContext* c, AVCodecContext *avctx); |
0b220468ba0d
Move prototypes for various dsputil init functions to dsputil.h
mru
parents:
11370
diff
changeset
|
644 void ff_rv30dsp_init(DSPContext* c, AVCodecContext *avctx); |
0b220468ba0d
Move prototypes for various dsputil init functions to dsputil.h
mru
parents:
11370
diff
changeset
|
645 void ff_rv40dsp_init(DSPContext* c, AVCodecContext *avctx); |
0b220468ba0d
Move prototypes for various dsputil init functions to dsputil.h
mru
parents:
11370
diff
changeset
|
646 void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx); |
0b220468ba0d
Move prototypes for various dsputil init functions to dsputil.h
mru
parents:
11370
diff
changeset
|
647 void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx); |
0b220468ba0d
Move prototypes for various dsputil init functions to dsputil.h
mru
parents:
11370
diff
changeset
|
648 void ff_mlp_init(DSPContext* c, AVCodecContext *avctx); |
0b220468ba0d
Move prototypes for various dsputil init functions to dsputil.h
mru
parents:
11370
diff
changeset
|
649 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
|
650 |
8590 | 651 #if HAVE_MMX |
0 | 652 |
862 | 653 #undef emms_c |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
654 |
0 | 655 static inline void emms(void) |
656 { | |
8031 | 657 __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
|
658 } |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
659 |
936 | 660 |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
661 #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
|
662 {\ |
8104
0d108ec85620
Remove duplicated MM_* macros for CPU capabilities from dsputil.h.
rathann
parents:
8031
diff
changeset
|
663 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
|
664 emms();\ |
0 | 665 } |
666 | |
8590 | 667 #elif ARCH_ARM |
62 | 668 |
8590 | 669 #if HAVE_NEON |
7687
e5b5a9af1b68
ARM: set STRIDE_ALIGN and DECLARE_ALIGNED_8 to 16 for NEON
mru
parents:
7678
diff
changeset
|
670 # define STRIDE_ALIGN 16 |
e5b5a9af1b68
ARM: set STRIDE_ALIGN and DECLARE_ALIGNED_8 to 16 for NEON
mru
parents:
7678
diff
changeset
|
671 #endif |
e5b5a9af1b68
ARM: set STRIDE_ALIGN and DECLARE_ALIGNED_8 to 16 for NEON
mru
parents:
7678
diff
changeset
|
672 |
8590 | 673 #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
|
674 |
2324 | 675 #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
|
676 |
8590 | 677 #elif HAVE_MMI |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
678 |
2324 | 679 #define STRIDE_ALIGN 16 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
680 |
7135
6bd6a2da306e
Define mm_flags/support to be 0 on architectures where they don't exist.
astrange
parents:
7094
diff
changeset
|
681 #else |
6bd6a2da306e
Define mm_flags/support to be 0 on architectures where they don't exist.
astrange
parents:
7094
diff
changeset
|
682 |
6bd6a2da306e
Define mm_flags/support to be 0 on architectures where they don't exist.
astrange
parents:
7094
diff
changeset
|
683 #define mm_flags 0 |
6bd6a2da306e
Define mm_flags/support to be 0 on architectures where they don't exist.
astrange
parents:
7094
diff
changeset
|
684 #define mm_support() 0 |
6bd6a2da306e
Define mm_flags/support to be 0 on architectures where they don't exist.
astrange
parents:
7094
diff
changeset
|
685 |
6363
7ebd1cdb2142
clean up definition of DECLARE_ALIGNED_8 and STRIDE_ALIGN
mru
parents:
6241
diff
changeset
|
686 #endif |
3728 | 687 |
6363
7ebd1cdb2142
clean up definition of DECLARE_ALIGNED_8 and STRIDE_ALIGN
mru
parents:
6241
diff
changeset
|
688 #ifndef STRIDE_ALIGN |
7ebd1cdb2142
clean up definition of DECLARE_ALIGNED_8 and STRIDE_ALIGN
mru
parents:
6241
diff
changeset
|
689 # define STRIDE_ALIGN 8 |
0 | 690 #endif |
691 | |
11194
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
692 #define LOCAL_ALIGNED(a, t, v, s, ...) \ |
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
693 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
|
694 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
|
695 |
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
696 #if HAVE_LOCAL_ALIGNED_8 |
11369 | 697 # 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
|
698 #else |
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
699 # 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
|
700 #endif |
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
701 |
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
702 #if HAVE_LOCAL_ALIGNED_16 |
11369 | 703 # 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
|
704 #else |
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
705 # 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
|
706 #endif |
7b3f6955462b
Add LOCAL_ALIGNED() macro for declaring aligned local arrays
mru
parents:
10961
diff
changeset
|
707 |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
708 /* PSNR */ |
1064 | 709 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
|
710 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
|
711 AVCodecContext *avctx); |
781 | 712 |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
5997
diff
changeset
|
713 #define WRAPPER8_16(name8, name16)\ |
1708 | 714 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ |
715 return name8(s, dst , src , stride, h)\ | |
716 +name8(s, dst+8 , src+8 , stride, h);\ | |
717 } | |
718 | |
6056
558c1fd0ee72
Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents:
5997
diff
changeset
|
719 #define WRAPPER8_16_SQ(name8, name16)\ |
1708 | 720 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ |
721 int score=0;\ | |
722 score +=name8(s, dst , src , stride, 8);\ | |
723 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
724 if(h==16){\ | |
725 dst += 8*stride;\ | |
726 src += 8*stride;\ | |
727 score +=name8(s, dst , src , stride, 8);\ | |
728 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
729 }\ | |
730 return score;\ | |
936 | 731 } |
732 | |
4240 | 733 |
9436 | 734 static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) |
4240 | 735 { |
736 int i; | |
737 for(i=0; i<h; i++) | |
738 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
739 AV_WN16(dst , AV_RN16(src )); |
4240 | 740 dst+=dstStride; |
741 src+=srcStride; | |
742 } | |
743 } | |
744 | |
9436 | 745 static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) |
4240 | 746 { |
747 int i; | |
748 for(i=0; i<h; i++) | |
749 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
750 AV_WN32(dst , AV_RN32(src )); |
4240 | 751 dst+=dstStride; |
752 src+=srcStride; | |
753 } | |
754 } | |
755 | |
9436 | 756 static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) |
4240 | 757 { |
758 int i; | |
759 for(i=0; i<h; i++) | |
760 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
761 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
|
762 AV_WN32(dst+4 , AV_RN32(src+4 )); |
4240 | 763 dst+=dstStride; |
764 src+=srcStride; | |
765 } | |
766 } | |
767 | |
9436 | 768 static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) |
4240 | 769 { |
770 int i; | |
771 for(i=0; i<h; i++) | |
772 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
773 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
|
774 AV_WN32(dst+4 , AV_RN32(src+4 )); |
4240 | 775 dst[8]= src[8]; |
776 dst+=dstStride; | |
777 src+=srcStride; | |
778 } | |
779 } | |
780 | |
9436 | 781 static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) |
4240 | 782 { |
783 int i; | |
784 for(i=0; i<h; i++) | |
785 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
786 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
|
787 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
|
788 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
|
789 AV_WN32(dst+12, AV_RN32(src+12)); |
4240 | 790 dst+=dstStride; |
791 src+=srcStride; | |
792 } | |
793 } | |
794 | |
9436 | 795 static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) |
4240 | 796 { |
797 int i; | |
798 for(i=0; i<h; i++) | |
799 { | |
5520
c16a59ef6a86
* renaming (ST|LD)(16|32|64) -> AV_(R|W)N(16|32|64)
romansh
parents:
5291
diff
changeset
|
800 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
|
801 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
|
802 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
|
803 AV_WN32(dst+12, AV_RN32(src+12)); |
4240 | 804 dst[16]= src[16]; |
805 dst+=dstStride; | |
806 src+=srcStride; | |
807 } | |
808 } | |
809 | |
7760 | 810 #endif /* AVCODEC_DSPUTIL_H */ |