Typos in comments
[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
59006372
LM
203 int (*ssd_int8_vs_int16)(int8_t *pix1, int16_t *pix2, int size);
204
d518aebd
MN
205 /**
206 * Halfpel motion compensation with rounding (a+b+1)>>1.
30f15053 207 * this is an array[4][4] of motion compensation functions for 4
e5771f4f 208 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 209 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
210 * @param block destination where the result is stored
211 * @param pixels source
212 * @param line_size number of bytes in a horizontal line of block
213 * @param h height
214 */
669ac79c 215 op_pixels_func put_pixels_tab[4][4];
d518aebd
MN
216
217 /**
218 * Halfpel motion compensation with rounding (a+b+1)>>1.
115329f1 219 * This is an array[4][4] of motion compensation functions for 4
e5771f4f 220 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 221 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
222 * @param block destination into which the result is averaged (a+b+1)>>1
223 * @param pixels source
224 * @param line_size number of bytes in a horizontal line of block
225 * @param h height
226 */
da3b9756 227 op_pixels_func avg_pixels_tab[4][4];
d518aebd
MN
228
229 /**
230 * Halfpel motion compensation with no rounding (a+b)>>1.
30f15053 231 * this is an array[2][4] of motion compensation functions for 2
eb14c713 232 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 233 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
234 * @param block destination where the result is stored
235 * @param pixels source
236 * @param line_size number of bytes in a horizontal line of block
237 * @param h height
238 */
dbc56b39 239 op_pixels_func put_no_rnd_pixels_tab[4][4];
d518aebd
MN
240
241 /**
242 * Halfpel motion compensation with no rounding (a+b)>>1.
30f15053 243 * this is an array[2][4] of motion compensation functions for 2
eb14c713 244 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 245 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
246 * @param block destination into which the result is averaged (a+b)>>1
247 * @param pixels source
248 * @param line_size number of bytes in a horizontal line of block
249 * @param h height
250 */
dbc56b39 251 op_pixels_func avg_no_rnd_pixels_tab[4][4];
115329f1 252
c0a0170c 253 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 254
669ac79c
MN
255 /**
256 * Thirdpel motion compensation with rounding (a+b+1)>>1.
30f15053
LB
257 * this is an array[12] of motion compensation functions for the 9 thirdpe
258 * positions<br>
669ac79c
MN
259 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
260 * @param block destination where the result is stored
261 * @param pixels source
262 * @param line_size number of bytes in a horizontal line of block
263 * @param h height
264 */
265 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
da3b9756
MM
266 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
267
eb4b3dd3
ZK
268 qpel_mc_func put_qpel_pixels_tab[2][16];
269 qpel_mc_func avg_qpel_pixels_tab[2][16];
270 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
271 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
1457ab52 272 qpel_mc_func put_mspel_pixels_tab[8];
115329f1 273
0da71265 274 /**
30f15053 275 * h264 Chroma MC
0da71265
MN
276 */
277 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
e34350a3
KS
278 /* This is really one func used in VC-1 decoding */
279 h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
0da71265 280 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
eb4b3dd3 281
80e44bc3
MN
282 qpel_mc_func put_h264_qpel_pixels_tab[4][16];
283 qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
115329f1 284
2833fc46
LM
285 qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
286 qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
287
9f2d1b4f
LM
288 h264_weight_func weight_h264_pixels_tab[10];
289 h264_biweight_func biweight_h264_pixels_tab[10];
115329f1 290
b482e2d1
MN
291 /* AVS specific */
292 qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
293 qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
294 void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
295 void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
296 void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
297 void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
298 void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
299
bb198e19 300 me_cmp_func pix_abs[2][4];
115329f1 301
11f18faf 302 /* huffyuv specific */
11f18faf 303 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
1457ab52 304 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
84705403
MN
305 /**
306 * subtract huffyuv's variant of median prediction
307 * note, this might read from src1[-1], src2[-1]
308 */
309 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
3d2e8cce 310 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
42251a2a 311
5cf08f23
LM
312 void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
313 void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
314 void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
315 void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
316 void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
317 void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
3e20143e
LM
318 // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
319 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],
320 int bidir, int edges, int step, int mask_mv0, int mask_mv1);
115329f1 321
332f9ac4
MN
322 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
323 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
324
fdbbf2e0 325 void (*h261_loop_filter)(uint8_t *src, int stride);
c6148de2 326
eb4825b5 327 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
2dac4acf 328 void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
eb4825b5 329 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
06972056 330 void (*vector_fmul)(float *dst, const float *src, int len);
eb4825b5
LM
331 void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
332 /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
333 void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
334
335 /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
98145875 336 * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
eb4825b5 337 void (*float_to_int16)(int16_t *dst, const float *src, int len);
2dac4acf 338
b0368839
MN
339 /* (I)DCT */
340 void (*fdct)(DCTELEM *block/* align 16*/);
10acc479 341 void (*fdct248)(DCTELEM *block/* align 16*/);
115329f1 342
4fb518c3
MN
343 /* IDCT really*/
344 void (*idct)(DCTELEM *block/* align 16*/);
115329f1 345
24641185 346 /**
77c92c2d 347 * block -> idct -> clip to unsigned 8 bit -> dest.
24641185 348 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
a1adf436 349 * @param line_size size in bytes of a horizotal line of dest
24641185 350 */
b0368839 351 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
115329f1 352
24641185
MN
353 /**
354 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
a1adf436 355 * @param line_size size in bytes of a horizotal line of dest
24641185 356 */
b0368839 357 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
115329f1 358
24641185 359 /**
77c92c2d 360 * idct input permutation.
05493021
MN
361 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
362 * IDCT)
363 * this permutation must be performed before the idct_put/add, note, normally this can be merged
364 * with the zigzag/alternate scan<br>
24641185
MN
365 * an example to avoid confusion:
366 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
367 * - (x -> referece dct -> reference idct -> x)
368 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
369 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
370 */
b0368839
MN
371 uint8_t idct_permutation[64];
372 int idct_permutation_type;
373#define FF_NO_IDCT_PERM 1
374#define FF_LIBMPEG2_IDCT_PERM 2
375#define FF_SIMPLE_IDCT_PERM 3
376#define FF_TRANSPOSE_IDCT_PERM 4
5773a746 377#define FF_PARTTRANS_IDCT_PERM 5
b0368839 378
364a1797
MN
379 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
380 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
381#define BASIS_SHIFT 16
382#define RECON_SHIFT 6
115329f1 383
ea9f5d6f 384 /* h264 functions */
0fa8158d 385 void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
43efd19a 386 void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
ef9d1d15
LM
387 void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
388 void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
f1862127 389 void (*h264_dct)(DCTELEM block[4][4]);
059715a4
RE
390
391 /* snow wavelet */
392 void (*vertical_compose97i)(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width);
393 void (*horizontal_compose97i)(DWTELEM *b, int width);
9dd6c804 394 void (*inner_add_yblock)(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
513fbd8e
LM
395
396 void (*prefetch)(void *mem, int stride, int h);
54009d42
MN
397
398 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
64db55ae
KS
399
400 /* vc1 functions */
401 void (*vc1_inv_trans_8x8)(DCTELEM *b);
402 void (*vc1_inv_trans_8x4)(DCTELEM *b, int n);
403 void (*vc1_inv_trans_4x8)(DCTELEM *b, int n);
404 void (*vc1_inv_trans_4x4)(DCTELEM *b, int n);
61f5b14a
KS
405 void (*vc1_v_overlap)(uint8_t* src, int stride);
406 void (*vc1_h_overlap)(uint8_t* src, int stride);
64db55ae
KS
407 /* put 8x8 block with bicubic interpolation and quarterpel precision
408 * last argument is actually round value instead of height
409 */
410 op_pixels_func put_vc1_mspel_pixels_tab[16];
eb4b3dd3
ZK
411} DSPContext;
412
486497e0 413void dsputil_static_init(void);
b0368839 414void dsputil_init(DSPContext* p, AVCodecContext *avctx);
de6d9b64 415
6dc7d5da
MN
416int ff_check_alignment(void);
417
7801d21d
MN
418/**
419 * permute block according to permuatation.
420 * @param last last non zero element in scantable order
421 */
0c1a9eda 422void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
e0eac44e 423
622348f9
MN
424void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
425
bb270c08 426#define BYTE_VEC32(c) ((c)*0x01010101UL)
d8085ea7
MN
427
428static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
429{
430 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
431}
432
433static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
434{
435 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
436}
437
26efc54e
MN
438static inline int get_penalty_factor(int lambda, int lambda2, int type){
439 switch(type&0xFF){
440 default:
441 case FF_CMP_SAD:
442 return lambda>>FF_LAMBDA_SHIFT;
443 case FF_CMP_DCT:
444 return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
445 case FF_CMP_W53:
446 return (4*lambda)>>(FF_LAMBDA_SHIFT);
447 case FF_CMP_W97:
448 return (2*lambda)>>(FF_LAMBDA_SHIFT);
449 case FF_CMP_SATD:
27c61ac5 450 case FF_CMP_DCT264:
26efc54e
MN
451 return (2*lambda)>>FF_LAMBDA_SHIFT;
452 case FF_CMP_RD:
453 case FF_CMP_PSNR:
454 case FF_CMP_SSE:
455 case FF_CMP_NSSE:
456 return lambda2>>FF_LAMBDA_SHIFT;
457 case FF_CMP_BIT:
458 return 1;
459 }
460}
461
24641185 462/**
77c92c2d 463 * Empty mmx state.
24641185
MN
464 * this must be called between any dsp function and float/double code.
465 * for example sin(); dsp->idct_put(); emms_c(); cos()
466 */
eb4b3dd3
ZK
467#define emms_c()
468
e629ab68
RD
469/* should be defined by architectures supporting
470 one or more MultiMedia extension */
471int mm_support(void);
472
68b51e58
SH
473#ifdef __GNUC__
474 #define DECLARE_ALIGNED_16(t,v) t v __attribute__ ((aligned (16)))
475#else
476 #define DECLARE_ALIGNED_16(t,v) __declspec(align(16)) t v
477#endif
92a69cf8 478
3d03c0a2 479#if defined(HAVE_MMX)
de6d9b64 480
18f77016 481#undef emms_c
eb4b3dd3 482
de6d9b64
FB
483#define MM_MMX 0x0001 /* standard MMX */
484#define MM_3DNOW 0x0004 /* AMD 3DNOW */
485#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
486#define MM_SSE 0x0008 /* SSE functions */
487#define MM_SSE2 0x0010 /* PIV SSE2 functions */
e42a152b 488#define MM_3DNOWEXT 0x0020 /* AMD 3DNowExt */
392f6da8 489#define MM_SSE3 0x0040 /* Prescott SSE3 functions */
5a5c770d 490#define MM_SSSE3 0x0080 /* Conroe SSSE3 functions */
de6d9b64 491
486497e0
MR
492extern int mm_flags;
493
0c1a9eda
ZK
494void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
495void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
f9ed9d85 496void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
de6d9b64
FB
497
498static inline void emms(void)
499{
fb16b7e7
FB
500 __asm __volatile ("emms;":::"memory");
501}
502
1457ab52 503
fb16b7e7
FB
504#define emms_c() \
505{\
486497e0 506 if (mm_flags & MM_MMX)\
fb16b7e7 507 emms();\
de6d9b64
FB
508}
509
68b51e58
SH
510#ifdef __GNUC__
511 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
512#else
513 #define DECLARE_ALIGNED_8(t,v) __declspec(align(8)) t v
514#endif
515
3237f731 516#define STRIDE_ALIGN 8
de6d9b64 517
b0368839
MN
518void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
519void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
2720569a 520
3d03c0a2
FB
521#elif defined(ARCH_ARMV4L)
522
3d03c0a2 523/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
92a69cf8 524 line optimizations */
68b51e58 525#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (4)))
3237f731 526#define STRIDE_ALIGN 4
3d03c0a2 527
eba9ae3c
GB
528#define MM_IWMMXT 0x0100 /* XScale IWMMXT */
529
486497e0
MR
530extern int mm_flags;
531
b0368839 532void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
3d03c0a2 533
c34270f5 534#elif defined(HAVE_MLIB)
c34270f5
FB
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
c34270f5 539
b0368839 540void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
c34270f5 541
44f54ceb
MN
542#elif defined(ARCH_SPARC)
543
544/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
68b51e58 545#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
3237f731 546#define STRIDE_ALIGN 8
44f54ceb
MN
547void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
548
1e98dffb
NK
549#elif defined(ARCH_ALPHA)
550
68b51e58 551#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
3237f731 552#define STRIDE_ALIGN 8
1e98dffb 553
b0368839 554void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
1e98dffb 555
59925ef2
BF
556#elif defined(ARCH_POWERPC)
557
404d2241
BF
558#define MM_ALTIVEC 0x0001 /* standard AltiVec */
559
486497e0
MR
560extern int mm_flags;
561
3b991c54 562#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
9a197a24 563#define pixel altivec_pixel
3b991c54 564#include <altivec.h>
9a197a24 565#undef pixel
3b991c54
RD
566#endif
567
68b51e58 568#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16)))
3237f731 569#define STRIDE_ALIGN 16
59925ef2 570
b0368839 571void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
59925ef2 572
d46aba26
LS
573#elif defined(HAVE_MMI)
574
68b51e58 575#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16)))
3237f731 576#define STRIDE_ALIGN 16
d46aba26 577
b0368839 578void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
d46aba26 579
0c6bd2ea
B
580#elif defined(ARCH_SH4)
581
68b51e58 582#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
3237f731 583#define STRIDE_ALIGN 8
0c6bd2ea
B
584
585void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
586
5ca030fe
MB
587#elif defined(ARCH_BFIN)
588
589#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
590#define STRIDE_ALIGN 8
591
592void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
593
de6d9b64
FB
594#else
595
68b51e58 596#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
3237f731 597#define STRIDE_ALIGN 8
de6d9b64
FB
598
599#endif
600
43f1708f 601/* PSNR */
0c1a9eda 602void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
43f1708f
J
603 int orig_linesize[3], int coded_linesize,
604 AVCodecContext *avctx);
bb6f5690
FB
605
606/* FFT computation */
607
608/* NOTE: soon integer code will be added, so you must use the
609 FFTSample type */
610typedef float FFTSample;
611
bcfa3e58
LM
612struct MDCTContext;
613
bb6f5690
FB
614typedef struct FFTComplex {
615 FFTSample re, im;
616} FFTComplex;
617
618typedef struct FFTContext {
619 int nbits;
620 int inverse;
621 uint16_t *revtab;
622 FFTComplex *exptab;
623 FFTComplex *exptab1; /* only used by SSE code */
624 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
bcfa3e58
LM
625 void (*imdct_calc)(struct MDCTContext *s, FFTSample *output,
626 const FFTSample *input, FFTSample *tmp);
bb6f5690
FB
627} FFTContext;
628
68951ecf
GB
629int ff_fft_init(FFTContext *s, int nbits, int inverse);
630void ff_fft_permute(FFTContext *s, FFTComplex *z);
631void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
632void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
82eb4b0f
ZM
633void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
634void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
68951ecf 635void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
8d268a7d 636
68951ecf 637static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
bb6f5690
FB
638{
639 s->fft_calc(s, z);
640}
68951ecf 641void ff_fft_end(FFTContext *s);
bb6f5690
FB
642
643/* MDCT computation */
644
645typedef struct MDCTContext {
646 int n; /* size of MDCT (i.e. number of input data * 2) */
647 int nbits; /* n = 2^nbits */
648 /* pre/post rotation tables */
649 FFTSample *tcos;
650 FFTSample *tsin;
651 FFTContext fft;
652} MDCTContext;
653
82696bee 654int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
eb4b3dd3 655void ff_imdct_calc(MDCTContext *s, FFTSample *output,
bb6f5690 656 const FFTSample *input, FFTSample *tmp);
bcfa3e58
LM
657void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output,
658 const FFTSample *input, FFTSample *tmp);
ebbafcb4
LM
659void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output,
660 const FFTSample *input, FFTSample *tmp);
eb4b3dd3 661void ff_mdct_calc(MDCTContext *s, FFTSample *out,
bb6f5690 662 const FFTSample *input, FFTSample *tmp);
82696bee 663void ff_mdct_end(MDCTContext *s);
bb6f5690 664
bb198e19
MN
665#define WARPER8_16(name8, name16)\
666static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
667 return name8(s, dst , src , stride, h)\
668 +name8(s, dst+8 , src+8 , stride, h);\
669}
670
671#define WARPER8_16_SQ(name8, name16)\
672static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
673 int score=0;\
674 score +=name8(s, dst , src , stride, 8);\
675 score +=name8(s, dst+8 , src+8 , stride, 8);\
676 if(h==16){\
677 dst += 8*stride;\
678 src += 8*stride;\
679 score +=name8(s, dst , src , stride, 8);\
680 score +=name8(s, dst+8 , src+8 , stride, 8);\
681 }\
682 return score;\
1457ab52
MN
683}
684
49cef744
BF
685
686static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
687{
688 int i;
689 for(i=0; i<h; i++)
690 {
691 ST16(dst , LD16(src ));
692 dst+=dstStride;
693 src+=srcStride;
694 }
695}
696
697static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
698{
699 int i;
700 for(i=0; i<h; i++)
701 {
702 ST32(dst , LD32(src ));
703 dst+=dstStride;
704 src+=srcStride;
705 }
706}
707
708static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
709{
710 int i;
711 for(i=0; i<h; i++)
712 {
713 ST32(dst , LD32(src ));
714 ST32(dst+4 , LD32(src+4 ));
715 dst+=dstStride;
716 src+=srcStride;
717 }
718}
719
720static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
721{
722 int i;
723 for(i=0; i<h; i++)
724 {
725 ST32(dst , LD32(src ));
726 ST32(dst+4 , LD32(src+4 ));
727 dst[8]= src[8];
728 dst+=dstStride;
729 src+=srcStride;
730 }
731}
732
733static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
734{
735 int i;
736 for(i=0; i<h; i++)
737 {
738 ST32(dst , LD32(src ));
739 ST32(dst+4 , LD32(src+4 ));
740 ST32(dst+8 , LD32(src+8 ));
741 ST32(dst+12, LD32(src+12));
742 dst+=dstStride;
743 src+=srcStride;
744 }
745}
746
747static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
748{
749 int i;
750 for(i=0; i<h; i++)
751 {
752 ST32(dst , LD32(src ));
753 ST32(dst+4 , LD32(src+4 ));
754 ST32(dst+8 , LD32(src+8 ));
755 ST32(dst+12, LD32(src+12));
756 dst[16]= src[16];
757 dst+=dstStride;
758 src+=srcStride;
759 }
760}
761
de6d9b64 762#endif