Correct rounding values in overlap filtering
[libav.git] / libavcodec / dsputil.h
CommitLineData
ff4ec49e
FB
1/*
2 * DSP utils
3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard.
8f2ab833 4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
ff4ec49e 5 *
b78e7197
DB
6 * This file is part of FFmpeg.
7 *
8 * FFmpeg is free software; you can redistribute it and/or
ff4ec49e
FB
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
b78e7197 11 * version 2.1 of the License, or (at your option) any later version.
ff4ec49e 12 *
b78e7197 13 * FFmpeg is distributed in the hope that it will be useful,
ff4ec49e
FB
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
b78e7197 19 * License along with FFmpeg; if not, write to the Free Software
5509bffa 20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
ff4ec49e 21 */
24641185
MN
22
23/**
24 * @file dsputil.h
983e3246 25 * DSP utils.
5755c27f
MN
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
24641185
MN
28 */
29
de6d9b64
FB
30#ifndef DSPUTIL_H
31#define DSPUTIL_H
32
33#include "common.h"
43f1708f 34#include "avcodec.h"
de6d9b64 35
24641185 36
44eb4951 37//#define DEBUG
de6d9b64
FB
38/* dct code */
39typedef short DCTELEM;
059715a4 40typedef int DWTELEM;
de6d9b64 41
03c94ede 42void fdct_ifast (DCTELEM *data);
48b1f800 43void fdct_ifast248 (DCTELEM *data);
28db7fce 44void ff_jpeg_fdct_islow (DCTELEM *data);
10acc479 45void ff_fdct248_islow (DCTELEM *data);
de6d9b64
FB
46
47void j_rev_dct (DCTELEM *data);
178fcca8 48void j_rev_dct4 (DCTELEM *data);
9ca358b9 49void j_rev_dct2 (DCTELEM *data);
1aa8c57b 50void j_rev_dct1 (DCTELEM *data);
de6d9b64 51
3f09f52a 52void ff_fdct_mmx(DCTELEM *block);
cf3bf5bb 53void ff_fdct_mmx2(DCTELEM *block);
8fd19ab2 54void ff_fdct_sse2(DCTELEM *block);
de6d9b64 55
43efd19a 56void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
0fa8158d 57void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
ef9d1d15
LM
58void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
59void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
0fa8158d
MN
60void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
61void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
62
eb4825b5
LM
63void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1,
64 const float *src2, int src3, int blocksize, int step);
65void ff_float_to_int16_c(int16_t *dst, const float *src, int len);
66
e0eac44e 67/* encoding scans */
0c1a9eda
ZK
68extern const uint8_t ff_alternate_horizontal_scan[64];
69extern const uint8_t ff_alternate_vertical_scan[64];
70extern const uint8_t ff_zigzag_direct[64];
10acc479 71extern const uint8_t ff_zigzag248_direct[64];
5a240838 72
de6d9b64 73/* pixel operations */
f2e92ef2 74#define MAX_NEG_CROP 1024
de6d9b64
FB
75
76/* temporary */
1d503957 77extern uint32_t ff_squareTbl[512];
55fde95e 78extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
de6d9b64 79
44cb64ee 80/* VP3 DSP functions */
8b6103da
MN
81void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
82void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
83void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
de6d9b64 84
54009d42
MN
85/* 1/2^n downscaling functions from imgconvert.c */
86void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
87void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
88void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
89void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
703c8195
LM
90
91void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
92 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
54009d42 93
b7c27ee6 94/* minimum alignment rules ;)
eb4b3dd3 95if u notice errors in the align stuff, need more alignment for some asm code for some cpu
b7c27ee6
MN
96or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
97
98!warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
99i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
de6d9b64 100
b7c27ee6
MN
101!future video codecs might need functions with less strict alignment
102*/
103
eb4b3dd3 104/*
0c1a9eda
ZK
105void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
106void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
107void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
108void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
649c00c9 109void clear_blocks_c(DCTELEM *blocks);
eb4b3dd3 110*/
de6d9b64
FB
111
112/* add and put pixel (decoding) */
b7c27ee6 113// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
7d67aa9b 114//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
0c1a9eda 115typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
669ac79c 116typedef 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);
0c1a9eda 117typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
0da71265 118typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
9f2d1b4f 119typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
e8b56208 120typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
b3184779 121
db794953 122#define DEF_OLD_QPEL(name)\
0c1a9eda
ZK
123void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
124void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
125void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
db794953
MN
126
127DEF_OLD_QPEL(qpel16_mc11_old_c)
128DEF_OLD_QPEL(qpel16_mc31_old_c)
129DEF_OLD_QPEL(qpel16_mc12_old_c)
130DEF_OLD_QPEL(qpel16_mc32_old_c)
131DEF_OLD_QPEL(qpel16_mc13_old_c)
132DEF_OLD_QPEL(qpel16_mc33_old_c)
133DEF_OLD_QPEL(qpel8_mc11_old_c)
134DEF_OLD_QPEL(qpel8_mc31_old_c)
135DEF_OLD_QPEL(qpel8_mc12_old_c)
136DEF_OLD_QPEL(qpel8_mc32_old_c)
137DEF_OLD_QPEL(qpel8_mc13_old_c)
138DEF_OLD_QPEL(qpel8_mc33_old_c)
b3184779
MN
139
140#define CALL_2X_PIXELS(a, b, n)\
141static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
142 b(block , pixels , line_size, h);\
143 b(block+n, pixels+n, line_size, h);\
144}
44eb4951 145
de6d9b64 146/* motion estimation */
7d67aa9b
MN
147// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
148// allthough currently h<4 is not used as functions with width <8 are not used and neither implemented
bb198e19 149typedef 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))*/;
1457ab52 150
0da71265 151
059715a4
RE
152// for snow slices
153typedef struct slice_buffer_s slice_buffer;
154
24641185
MN
155/**
156 * DSPContext.
157 */
eb4b3dd3
ZK
158typedef struct DSPContext {
159 /* pixel ops : interface with DCT */
0c1a9eda
ZK
160 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
161 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
162 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
f9ed9d85 163 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
0c1a9eda 164 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
36940eca
LM
165 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
166 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
d518aebd
MN
167 /**
168 * translational global motion compensation.
169 */
0c1a9eda 170 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
d518aebd
MN
171 /**
172 * global motion compensation.
173 */
0c1a9eda 174 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
bb270c08 175 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
eb4b3dd3 176 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
0c1a9eda
ZK
177 int (*pix_sum)(uint8_t * pix, int line_size);
178 int (*pix_norm1)(uint8_t * pix, int line_size);
bb198e19 179// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
115329f1 180
622348f9
MN
181 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
182 me_cmp_func sse[5];
183 me_cmp_func hadamard8_diff[5];
184 me_cmp_func dct_sad[5];
185 me_cmp_func quant_psnr[5];
186 me_cmp_func bit[5];
187 me_cmp_func rd[5];
188 me_cmp_func vsad[5];
189 me_cmp_func vsse[5];
e6a2ac34 190 me_cmp_func nsse[5];
26efc54e
MN
191 me_cmp_func w53[5];
192 me_cmp_func w97[5];
0fd6aea1 193 me_cmp_func dct_max[5];
27c61ac5 194 me_cmp_func dct264_sad[5];
1457ab52 195
bb198e19
MN
196 me_cmp_func me_pre_cmp[5];
197 me_cmp_func me_cmp[5];
198 me_cmp_func me_sub_cmp[5];
199 me_cmp_func mb_cmp[5];
622348f9 200 me_cmp_func ildct_cmp[5]; //only width 16 used
0fd6aea1 201 me_cmp_func frame_skip_cmp[5]; //only width 8 used
eb4b3dd3 202
d518aebd
MN
203 /**
204 * Halfpel motion compensation with rounding (a+b+1)>>1.
115329f1 205 * this is an array[4][4] of motion compensation funcions for 4
e5771f4f 206 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 207 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
208 * @param block destination where the result is stored
209 * @param pixels source
210 * @param line_size number of bytes in a horizontal line of block
211 * @param h height
212 */
669ac79c 213 op_pixels_func put_pixels_tab[4][4];
d518aebd
MN
214
215 /**
216 * Halfpel motion compensation with rounding (a+b+1)>>1.
115329f1 217 * This is an array[4][4] of motion compensation functions for 4
e5771f4f 218 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 219 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
220 * @param block destination into which the result is averaged (a+b+1)>>1
221 * @param pixels source
222 * @param line_size number of bytes in a horizontal line of block
223 * @param h height
224 */
da3b9756 225 op_pixels_func avg_pixels_tab[4][4];
d518aebd
MN
226
227 /**
228 * Halfpel motion compensation with no rounding (a+b)>>1.
115329f1 229 * this is an array[2][4] of motion compensation funcions for 2
eb14c713 230 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 231 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
232 * @param block destination where the result is stored
233 * @param pixels source
234 * @param line_size number of bytes in a horizontal line of block
235 * @param h height
236 */
dbc56b39 237 op_pixels_func put_no_rnd_pixels_tab[4][4];
d518aebd
MN
238
239 /**
240 * Halfpel motion compensation with no rounding (a+b)>>1.
115329f1 241 * this is an array[2][4] of motion compensation funcions for 2
eb14c713 242 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 243 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
244 * @param block destination into which the result is averaged (a+b)>>1
245 * @param pixels source
246 * @param line_size number of bytes in a horizontal line of block
247 * @param h height
248 */
dbc56b39 249 op_pixels_func avg_no_rnd_pixels_tab[4][4];
115329f1 250
c0a0170c 251 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);
115329f1 252
669ac79c
MN
253 /**
254 * Thirdpel motion compensation with rounding (a+b+1)>>1.
255 * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
256 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
257 * @param block destination where the result is stored
258 * @param pixels source
259 * @param line_size number of bytes in a horizontal line of block
260 * @param h height
261 */
262 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
da3b9756
MM
263 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
264
eb4b3dd3
ZK
265 qpel_mc_func put_qpel_pixels_tab[2][16];
266 qpel_mc_func avg_qpel_pixels_tab[2][16];
267 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
268 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
1457ab52 269 qpel_mc_func put_mspel_pixels_tab[8];
115329f1 270
0da71265
MN
271 /**
272 * h264 Chram MC
273 */
274 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
e34350a3
KS
275 /* This is really one func used in VC-1 decoding */
276 h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
0da71265 277 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
eb4b3dd3 278
80e44bc3
MN
279 qpel_mc_func put_h264_qpel_pixels_tab[4][16];
280 qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
115329f1 281
2833fc46
LM
282 qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
283 qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
284
9f2d1b4f
LM
285 h264_weight_func weight_h264_pixels_tab[10];
286 h264_biweight_func biweight_h264_pixels_tab[10];
115329f1 287
b482e2d1
MN
288 /* AVS specific */
289 qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
290 qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
291 void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
292 void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
293 void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
294 void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
295 void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
296
bb198e19 297 me_cmp_func pix_abs[2][4];
115329f1 298
11f18faf 299 /* huffyuv specific */
11f18faf 300 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
1457ab52 301 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
84705403
MN
302 /**
303 * subtract huffyuv's variant of median prediction
304 * note, this might read from src1[-1], src2[-1]
305 */
306 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
3d2e8cce 307 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
42251a2a 308
5cf08f23
LM
309 void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
310 void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
311 void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
312 void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
313 void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
314 void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
3e20143e
LM
315 // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
316 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],
317 int bidir, int edges, int step, int mask_mv0, int mask_mv1);
115329f1 318
332f9ac4
MN
319 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
320 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
321
fdbbf2e0 322 void (*h261_loop_filter)(uint8_t *src, int stride);
c6148de2 323
eb4825b5 324 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
2dac4acf 325 void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
eb4825b5 326 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
06972056 327 void (*vector_fmul)(float *dst, const float *src, int len);
eb4825b5
LM
328 void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
329 /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
330 void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
331
332 /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
98145875 333 * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
eb4825b5 334 void (*float_to_int16)(int16_t *dst, const float *src, int len);
2dac4acf 335
b0368839
MN
336 /* (I)DCT */
337 void (*fdct)(DCTELEM *block/* align 16*/);
10acc479 338 void (*fdct248)(DCTELEM *block/* align 16*/);
115329f1 339
4fb518c3
MN
340 /* IDCT really*/
341 void (*idct)(DCTELEM *block/* align 16*/);
115329f1 342
24641185 343 /**
77c92c2d 344 * block -> idct -> clip to unsigned 8 bit -> dest.
24641185 345 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
a1adf436 346 * @param line_size size in bytes of a horizotal line of dest
24641185 347 */
b0368839 348 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
115329f1 349
24641185
MN
350 /**
351 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
a1adf436 352 * @param line_size size in bytes of a horizotal line of dest
24641185 353 */
b0368839 354 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
115329f1 355
24641185 356 /**
77c92c2d 357 * idct input permutation.
05493021
MN
358 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
359 * IDCT)
360 * this permutation must be performed before the idct_put/add, note, normally this can be merged
361 * with the zigzag/alternate scan<br>
24641185
MN
362 * an example to avoid confusion:
363 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
364 * - (x -> referece dct -> reference idct -> x)
365 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
366 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
367 */
b0368839
MN
368 uint8_t idct_permutation[64];
369 int idct_permutation_type;
370#define FF_NO_IDCT_PERM 1
371#define FF_LIBMPEG2_IDCT_PERM 2
372#define FF_SIMPLE_IDCT_PERM 3
373#define FF_TRANSPOSE_IDCT_PERM 4
5773a746 374#define FF_PARTTRANS_IDCT_PERM 5
b0368839 375
364a1797
MN
376 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
377 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
378#define BASIS_SHIFT 16
379#define RECON_SHIFT 6
115329f1 380
0fa8158d 381 void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
43efd19a 382 void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
ef9d1d15
LM
383 void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
384 void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
059715a4
RE
385
386 /* snow wavelet */
387 void (*vertical_compose97i)(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width);
388 void (*horizontal_compose97i)(DWTELEM *b, int width);
389 void (*inner_add_yblock)(uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
513fbd8e
LM
390
391 void (*prefetch)(void *mem, int stride, int h);
54009d42
MN
392
393 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
64db55ae
KS
394
395 /* vc1 functions */
396 void (*vc1_inv_trans_8x8)(DCTELEM *b);
397 void (*vc1_inv_trans_8x4)(DCTELEM *b, int n);
398 void (*vc1_inv_trans_4x8)(DCTELEM *b, int n);
399 void (*vc1_inv_trans_4x4)(DCTELEM *b, int n);
61f5b14a
KS
400 void (*vc1_v_overlap)(uint8_t* src, int stride);
401 void (*vc1_h_overlap)(uint8_t* src, int stride);
64db55ae
KS
402 /* put 8x8 block with bicubic interpolation and quarterpel precision
403 * last argument is actually round value instead of height
404 */
405 op_pixels_func put_vc1_mspel_pixels_tab[16];
eb4b3dd3
ZK
406} DSPContext;
407
486497e0 408void dsputil_static_init(void);
b0368839 409void dsputil_init(DSPContext* p, AVCodecContext *avctx);
de6d9b64 410
7801d21d
MN
411/**
412 * permute block according to permuatation.
413 * @param last last non zero element in scantable order
414 */
0c1a9eda 415void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
e0eac44e 416
622348f9
MN
417void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
418
bb270c08 419#define BYTE_VEC32(c) ((c)*0x01010101UL)
d8085ea7
MN
420
421static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
422{
423 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
424}
425
426static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
427{
428 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
429}
430
26efc54e
MN
431static inline int get_penalty_factor(int lambda, int lambda2, int type){
432 switch(type&0xFF){
433 default:
434 case FF_CMP_SAD:
435 return lambda>>FF_LAMBDA_SHIFT;
436 case FF_CMP_DCT:
437 return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
438 case FF_CMP_W53:
439 return (4*lambda)>>(FF_LAMBDA_SHIFT);
440 case FF_CMP_W97:
441 return (2*lambda)>>(FF_LAMBDA_SHIFT);
442 case FF_CMP_SATD:
27c61ac5 443 case FF_CMP_DCT264:
26efc54e
MN
444 return (2*lambda)>>FF_LAMBDA_SHIFT;
445 case FF_CMP_RD:
446 case FF_CMP_PSNR:
447 case FF_CMP_SSE:
448 case FF_CMP_NSSE:
449 return lambda2>>FF_LAMBDA_SHIFT;
450 case FF_CMP_BIT:
451 return 1;
452 }
453}
454
24641185 455/**
77c92c2d 456 * Empty mmx state.
24641185
MN
457 * this must be called between any dsp function and float/double code.
458 * for example sin(); dsp->idct_put(); emms_c(); cos()
459 */
eb4b3dd3
ZK
460#define emms_c()
461
e629ab68
RD
462/* should be defined by architectures supporting
463 one or more MultiMedia extension */
464int mm_support(void);
465
68b51e58
SH
466#ifdef __GNUC__
467 #define DECLARE_ALIGNED_16(t,v) t v __attribute__ ((aligned (16)))
468#else
469 #define DECLARE_ALIGNED_16(t,v) __declspec(align(16)) t v
470#endif
92a69cf8 471
3d03c0a2 472#if defined(HAVE_MMX)
de6d9b64 473
18f77016 474#undef emms_c
eb4b3dd3 475
de6d9b64
FB
476#define MM_MMX 0x0001 /* standard MMX */
477#define MM_3DNOW 0x0004 /* AMD 3DNOW */
478#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
479#define MM_SSE 0x0008 /* SSE functions */
480#define MM_SSE2 0x0010 /* PIV SSE2 functions */
e42a152b 481#define MM_3DNOWEXT 0x0020 /* AMD 3DNowExt */
392f6da8 482#define MM_SSE3 0x0040 /* Prescott SSE3 functions */
de6d9b64 483
486497e0
MR
484extern int mm_flags;
485
0c1a9eda
ZK
486void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
487void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
f9ed9d85 488void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
de6d9b64
FB
489
490static inline void emms(void)
491{
fb16b7e7
FB
492 __asm __volatile ("emms;":::"memory");
493}
494
1457ab52 495
fb16b7e7
FB
496#define emms_c() \
497{\
486497e0 498 if (mm_flags & MM_MMX)\
fb16b7e7 499 emms();\
de6d9b64
FB
500}
501
68b51e58
SH
502#ifdef __GNUC__
503 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
504#else
505 #define DECLARE_ALIGNED_8(t,v) __declspec(align(8)) t v
506#endif
507
3237f731 508#define STRIDE_ALIGN 8
de6d9b64 509
b0368839
MN
510void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
511void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
2720569a 512
3d03c0a2
FB
513#elif defined(ARCH_ARMV4L)
514
3d03c0a2 515/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
92a69cf8 516 line optimizations */
68b51e58 517#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (4)))
3237f731 518#define STRIDE_ALIGN 4
3d03c0a2 519
eba9ae3c
GB
520#define MM_IWMMXT 0x0100 /* XScale IWMMXT */
521
486497e0
MR
522extern int mm_flags;
523
b0368839 524void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
3d03c0a2 525
c34270f5 526#elif defined(HAVE_MLIB)
c34270f5
FB
527
528/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
68b51e58 529#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
3237f731 530#define STRIDE_ALIGN 8
c34270f5 531
b0368839 532void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
c34270f5 533
44f54ceb
MN
534#elif defined(ARCH_SPARC)
535
536/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
68b51e58 537#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
3237f731 538#define STRIDE_ALIGN 8
44f54ceb
MN
539void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
540
1e98dffb
NK
541#elif defined(ARCH_ALPHA)
542
68b51e58 543#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
3237f731 544#define STRIDE_ALIGN 8
1e98dffb 545
b0368839 546void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
1e98dffb 547
59925ef2
BF
548#elif defined(ARCH_POWERPC)
549
404d2241
BF
550#define MM_ALTIVEC 0x0001 /* standard AltiVec */
551
486497e0
MR
552extern int mm_flags;
553
3b991c54 554#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
9a197a24 555#define pixel altivec_pixel
3b991c54 556#include <altivec.h>
9a197a24 557#undef pixel
3b991c54
RD
558#endif
559
68b51e58 560#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16)))
3237f731 561#define STRIDE_ALIGN 16
59925ef2 562
b0368839 563void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
59925ef2 564
d46aba26
LS
565#elif defined(HAVE_MMI)
566
68b51e58 567#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16)))
3237f731 568#define STRIDE_ALIGN 16
d46aba26 569
b0368839 570void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
d46aba26 571
0c6bd2ea
B
572#elif defined(ARCH_SH4)
573
68b51e58 574#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
3237f731 575#define STRIDE_ALIGN 8
0c6bd2ea
B
576
577void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
578
5ca030fe
MB
579#elif defined(ARCH_BFIN)
580
581#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
582#define STRIDE_ALIGN 8
583
584void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
585
de6d9b64
FB
586#else
587
68b51e58 588#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
3237f731 589#define STRIDE_ALIGN 8
de6d9b64
FB
590
591#endif
592
6d4985bb
FB
593#ifdef __GNUC__
594
595struct unaligned_64 { uint64_t l; } __attribute__((packed));
596struct unaligned_32 { uint32_t l; } __attribute__((packed));
669ac79c 597struct unaligned_16 { uint16_t l; } __attribute__((packed));
6d4985bb 598
669ac79c 599#define LD16(a) (((const struct unaligned_16 *) (a))->l)
6d4985bb
FB
600#define LD32(a) (((const struct unaligned_32 *) (a))->l)
601#define LD64(a) (((const struct unaligned_64 *) (a))->l)
602
80e44bc3 603#define ST16(a, b) (((struct unaligned_16 *) (a))->l) = (b)
6d4985bb
FB
604#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
605
606#else /* __GNUC__ */
607
669ac79c 608#define LD16(a) (*((uint16_t*)(a)))
6d4985bb
FB
609#define LD32(a) (*((uint32_t*)(a)))
610#define LD64(a) (*((uint64_t*)(a)))
611
dc2bf20e 612#define ST16(a, b) *((uint16_t*)(a)) = (b)
6d4985bb
FB
613#define ST32(a, b) *((uint32_t*)(a)) = (b)
614
615#endif /* !__GNUC__ */
616
43f1708f 617/* PSNR */
0c1a9eda 618void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
43f1708f
J
619 int orig_linesize[3], int coded_linesize,
620 AVCodecContext *avctx);
bb6f5690
FB
621
622/* FFT computation */
623
624/* NOTE: soon integer code will be added, so you must use the
625 FFTSample type */
626typedef float FFTSample;
627
bcfa3e58
LM
628struct MDCTContext;
629
bb6f5690
FB
630typedef struct FFTComplex {
631 FFTSample re, im;
632} FFTComplex;
633
634typedef struct FFTContext {
635 int nbits;
636 int inverse;
637 uint16_t *revtab;
638 FFTComplex *exptab;
639 FFTComplex *exptab1; /* only used by SSE code */
640 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
bcfa3e58
LM
641 void (*imdct_calc)(struct MDCTContext *s, FFTSample *output,
642 const FFTSample *input, FFTSample *tmp);
bb6f5690
FB
643} FFTContext;
644
68951ecf
GB
645int ff_fft_init(FFTContext *s, int nbits, int inverse);
646void ff_fft_permute(FFTContext *s, FFTComplex *z);
647void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
648void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
82eb4b0f
ZM
649void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
650void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
68951ecf 651void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
8d268a7d 652
68951ecf 653static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
bb6f5690
FB
654{
655 s->fft_calc(s, z);
656}
68951ecf 657void ff_fft_end(FFTContext *s);
bb6f5690
FB
658
659/* MDCT computation */
660
661typedef struct MDCTContext {
662 int n; /* size of MDCT (i.e. number of input data * 2) */
663 int nbits; /* n = 2^nbits */
664 /* pre/post rotation tables */
665 FFTSample *tcos;
666 FFTSample *tsin;
667 FFTContext fft;
668} MDCTContext;
669
82696bee 670int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
eb4b3dd3 671void ff_imdct_calc(MDCTContext *s, FFTSample *output,
bb6f5690 672 const FFTSample *input, FFTSample *tmp);
bcfa3e58
LM
673void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output,
674 const FFTSample *input, FFTSample *tmp);
ebbafcb4
LM
675void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output,
676 const FFTSample *input, FFTSample *tmp);
eb4b3dd3 677void ff_mdct_calc(MDCTContext *s, FFTSample *out,
bb6f5690 678 const FFTSample *input, FFTSample *tmp);
82696bee 679void ff_mdct_end(MDCTContext *s);
bb6f5690 680
bb198e19
MN
681#define WARPER8_16(name8, name16)\
682static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
683 return name8(s, dst , src , stride, h)\
684 +name8(s, dst+8 , src+8 , stride, h);\
685}
686
687#define WARPER8_16_SQ(name8, name16)\
688static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
689 int score=0;\
690 score +=name8(s, dst , src , stride, 8);\
691 score +=name8(s, dst+8 , src+8 , stride, 8);\
692 if(h==16){\
693 dst += 8*stride;\
694 src += 8*stride;\
695 score +=name8(s, dst , src , stride, 8);\
696 score +=name8(s, dst+8 , src+8 , stride, 8);\
697 }\
698 return score;\
1457ab52
MN
699}
700
de6d9b64 701#endif