r5954 broke fft on cpus with 3dnow but without mm3dnow.h
[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
FB
5 *
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
10 *
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
5509bffa 18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
ff4ec49e 19 */
24641185
MN
20
21/**
22 * @file dsputil.h
983e3246 23 * DSP utils.
5755c27f
MN
24 * note, many functions in here may use MMX which trashes the FPU state, it is
25 * absolutely necessary to call emms_c() between dsp & float/double code
24641185
MN
26 */
27
de6d9b64
FB
28#ifndef DSPUTIL_H
29#define DSPUTIL_H
30
31#include "common.h"
43f1708f 32#include "avcodec.h"
de6d9b64 33
24641185 34
44eb4951 35//#define DEBUG
de6d9b64
FB
36/* dct code */
37typedef short DCTELEM;
059715a4 38typedef int DWTELEM;
de6d9b64 39
03c94ede 40void fdct_ifast (DCTELEM *data);
48b1f800 41void fdct_ifast248 (DCTELEM *data);
28db7fce 42void ff_jpeg_fdct_islow (DCTELEM *data);
10acc479 43void ff_fdct248_islow (DCTELEM *data);
de6d9b64
FB
44
45void j_rev_dct (DCTELEM *data);
178fcca8 46void j_rev_dct4 (DCTELEM *data);
9ca358b9 47void j_rev_dct2 (DCTELEM *data);
1aa8c57b 48void j_rev_dct1 (DCTELEM *data);
de6d9b64 49
3f09f52a 50void ff_fdct_mmx(DCTELEM *block);
cf3bf5bb 51void ff_fdct_mmx2(DCTELEM *block);
8fd19ab2 52void ff_fdct_sse2(DCTELEM *block);
de6d9b64 53
43efd19a 54void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
0fa8158d 55void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
ef9d1d15
LM
56void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
57void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
0fa8158d
MN
58void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
59void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
60
e0eac44e 61/* encoding scans */
0c1a9eda
ZK
62extern const uint8_t ff_alternate_horizontal_scan[64];
63extern const uint8_t ff_alternate_vertical_scan[64];
64extern const uint8_t ff_zigzag_direct[64];
10acc479 65extern const uint8_t ff_zigzag248_direct[64];
5a240838 66
de6d9b64 67/* pixel operations */
f2e92ef2 68#define MAX_NEG_CROP 1024
de6d9b64
FB
69
70/* temporary */
0c1a9eda
ZK
71extern uint32_t squareTbl[512];
72extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
de6d9b64 73
44cb64ee 74/* VP3 DSP functions */
8b6103da
MN
75void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
76void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
77void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
de6d9b64 78
54009d42
MN
79/* 1/2^n downscaling functions from imgconvert.c */
80void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
81void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
82void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
83void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
703c8195
LM
84
85void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
86 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
54009d42 87
b7c27ee6 88/* minimum alignment rules ;)
eb4b3dd3 89if u notice errors in the align stuff, need more alignment for some asm code for some cpu
b7c27ee6
MN
90or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
91
92!warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
93i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
de6d9b64 94
b7c27ee6
MN
95!future video codecs might need functions with less strict alignment
96*/
97
eb4b3dd3 98/*
0c1a9eda
ZK
99void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
100void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
101void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
102void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
649c00c9 103void clear_blocks_c(DCTELEM *blocks);
eb4b3dd3 104*/
de6d9b64
FB
105
106/* add and put pixel (decoding) */
b7c27ee6 107// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
7d67aa9b 108//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
0c1a9eda 109typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
669ac79c 110typedef 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 111typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
0da71265 112typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
9f2d1b4f 113typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
e8b56208 114typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
b3184779 115
db794953 116#define DEF_OLD_QPEL(name)\
0c1a9eda
ZK
117void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
118void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
119void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
db794953
MN
120
121DEF_OLD_QPEL(qpel16_mc11_old_c)
122DEF_OLD_QPEL(qpel16_mc31_old_c)
123DEF_OLD_QPEL(qpel16_mc12_old_c)
124DEF_OLD_QPEL(qpel16_mc32_old_c)
125DEF_OLD_QPEL(qpel16_mc13_old_c)
126DEF_OLD_QPEL(qpel16_mc33_old_c)
127DEF_OLD_QPEL(qpel8_mc11_old_c)
128DEF_OLD_QPEL(qpel8_mc31_old_c)
129DEF_OLD_QPEL(qpel8_mc12_old_c)
130DEF_OLD_QPEL(qpel8_mc32_old_c)
131DEF_OLD_QPEL(qpel8_mc13_old_c)
132DEF_OLD_QPEL(qpel8_mc33_old_c)
b3184779
MN
133
134#define CALL_2X_PIXELS(a, b, n)\
135static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
136 b(block , pixels , line_size, h);\
137 b(block+n, pixels+n, line_size, h);\
138}
44eb4951 139
de6d9b64 140/* motion estimation */
7d67aa9b
MN
141// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
142// allthough currently h<4 is not used as functions with width <8 are not used and neither implemented
bb198e19 143typedef 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 144
0da71265 145
059715a4
RE
146// for snow slices
147typedef struct slice_buffer_s slice_buffer;
148
24641185
MN
149/**
150 * DSPContext.
151 */
eb4b3dd3
ZK
152typedef struct DSPContext {
153 /* pixel ops : interface with DCT */
0c1a9eda
ZK
154 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
155 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
156 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
f9ed9d85 157 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
0c1a9eda 158 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
36940eca
LM
159 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
160 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
d518aebd
MN
161 /**
162 * translational global motion compensation.
163 */
0c1a9eda 164 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
d518aebd
MN
165 /**
166 * global motion compensation.
167 */
0c1a9eda 168 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
bb270c08 169 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
eb4b3dd3 170 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
0c1a9eda
ZK
171 int (*pix_sum)(uint8_t * pix, int line_size);
172 int (*pix_norm1)(uint8_t * pix, int line_size);
bb198e19 173// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
115329f1 174
622348f9
MN
175 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
176 me_cmp_func sse[5];
177 me_cmp_func hadamard8_diff[5];
178 me_cmp_func dct_sad[5];
179 me_cmp_func quant_psnr[5];
180 me_cmp_func bit[5];
181 me_cmp_func rd[5];
182 me_cmp_func vsad[5];
183 me_cmp_func vsse[5];
e6a2ac34 184 me_cmp_func nsse[5];
26efc54e
MN
185 me_cmp_func w53[5];
186 me_cmp_func w97[5];
0fd6aea1 187 me_cmp_func dct_max[5];
27c61ac5 188 me_cmp_func dct264_sad[5];
1457ab52 189
bb198e19
MN
190 me_cmp_func me_pre_cmp[5];
191 me_cmp_func me_cmp[5];
192 me_cmp_func me_sub_cmp[5];
193 me_cmp_func mb_cmp[5];
622348f9 194 me_cmp_func ildct_cmp[5]; //only width 16 used
0fd6aea1 195 me_cmp_func frame_skip_cmp[5]; //only width 8 used
eb4b3dd3 196
d518aebd
MN
197 /**
198 * Halfpel motion compensation with rounding (a+b+1)>>1.
115329f1 199 * this is an array[4][4] of motion compensation funcions for 4
e5771f4f 200 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 201 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
202 * @param block destination where the result is stored
203 * @param pixels source
204 * @param line_size number of bytes in a horizontal line of block
205 * @param h height
206 */
669ac79c 207 op_pixels_func put_pixels_tab[4][4];
d518aebd
MN
208
209 /**
210 * Halfpel motion compensation with rounding (a+b+1)>>1.
115329f1 211 * This is an array[4][4] of motion compensation functions for 4
e5771f4f 212 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 213 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
214 * @param block destination into which the result is averaged (a+b+1)>>1
215 * @param pixels source
216 * @param line_size number of bytes in a horizontal line of block
217 * @param h height
218 */
da3b9756 219 op_pixels_func avg_pixels_tab[4][4];
d518aebd
MN
220
221 /**
222 * Halfpel motion compensation with no rounding (a+b)>>1.
115329f1 223 * this is an array[2][4] of motion compensation funcions for 2
eb14c713 224 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 225 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
226 * @param block destination where the result is stored
227 * @param pixels source
228 * @param line_size number of bytes in a horizontal line of block
229 * @param h height
230 */
dbc56b39 231 op_pixels_func put_no_rnd_pixels_tab[4][4];
d518aebd
MN
232
233 /**
234 * Halfpel motion compensation with no rounding (a+b)>>1.
115329f1 235 * this is an array[2][4] of motion compensation funcions for 2
eb14c713 236 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 237 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
238 * @param block destination into which the result is averaged (a+b)>>1
239 * @param pixels source
240 * @param line_size number of bytes in a horizontal line of block
241 * @param h height
242 */
dbc56b39 243 op_pixels_func avg_no_rnd_pixels_tab[4][4];
115329f1 244
c0a0170c 245 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 246
669ac79c
MN
247 /**
248 * Thirdpel motion compensation with rounding (a+b+1)>>1.
249 * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
250 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
251 * @param block destination where the result is stored
252 * @param pixels source
253 * @param line_size number of bytes in a horizontal line of block
254 * @param h height
255 */
256 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
da3b9756
MM
257 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
258
eb4b3dd3
ZK
259 qpel_mc_func put_qpel_pixels_tab[2][16];
260 qpel_mc_func avg_qpel_pixels_tab[2][16];
261 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
262 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
1457ab52 263 qpel_mc_func put_mspel_pixels_tab[8];
115329f1 264
0da71265
MN
265 /**
266 * h264 Chram MC
267 */
268 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
269 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
eb4b3dd3 270
80e44bc3
MN
271 qpel_mc_func put_h264_qpel_pixels_tab[4][16];
272 qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
115329f1 273
9f2d1b4f
LM
274 h264_weight_func weight_h264_pixels_tab[10];
275 h264_biweight_func biweight_h264_pixels_tab[10];
115329f1 276
b482e2d1
MN
277 /* AVS specific */
278 qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
279 qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
280 void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
281 void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
282 void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
283 void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
284 void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
285
bb198e19 286 me_cmp_func pix_abs[2][4];
115329f1 287
11f18faf 288 /* huffyuv specific */
11f18faf 289 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
1457ab52 290 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
84705403
MN
291 /**
292 * subtract huffyuv's variant of median prediction
293 * note, this might read from src1[-1], src2[-1]
294 */
295 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
3d2e8cce 296 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
42251a2a 297
5cf08f23
LM
298 void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
299 void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
300 void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
301 void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
302 void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
303 void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
115329f1 304
332f9ac4
MN
305 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
306 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
307
fdbbf2e0 308 void (*h261_loop_filter)(uint8_t *src, int stride);
c6148de2 309
2dac4acf
LM
310 void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
311
b0368839
MN
312 /* (I)DCT */
313 void (*fdct)(DCTELEM *block/* align 16*/);
10acc479 314 void (*fdct248)(DCTELEM *block/* align 16*/);
115329f1 315
4fb518c3
MN
316 /* IDCT really*/
317 void (*idct)(DCTELEM *block/* align 16*/);
115329f1 318
24641185 319 /**
77c92c2d 320 * block -> idct -> clip to unsigned 8 bit -> dest.
24641185 321 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
a1adf436 322 * @param line_size size in bytes of a horizotal line of dest
24641185 323 */
b0368839 324 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
115329f1 325
24641185
MN
326 /**
327 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
a1adf436 328 * @param line_size size in bytes of a horizotal line of dest
24641185 329 */
b0368839 330 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
115329f1 331
24641185 332 /**
77c92c2d 333 * idct input permutation.
05493021
MN
334 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
335 * IDCT)
336 * this permutation must be performed before the idct_put/add, note, normally this can be merged
337 * with the zigzag/alternate scan<br>
24641185
MN
338 * an example to avoid confusion:
339 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
340 * - (x -> referece dct -> reference idct -> x)
341 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
342 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
343 */
b0368839
MN
344 uint8_t idct_permutation[64];
345 int idct_permutation_type;
346#define FF_NO_IDCT_PERM 1
347#define FF_LIBMPEG2_IDCT_PERM 2
348#define FF_SIMPLE_IDCT_PERM 3
349#define FF_TRANSPOSE_IDCT_PERM 4
5773a746 350#define FF_PARTTRANS_IDCT_PERM 5
b0368839 351
364a1797
MN
352 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
353 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
354#define BASIS_SHIFT 16
355#define RECON_SHIFT 6
115329f1 356
0fa8158d 357 void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
43efd19a 358 void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
ef9d1d15
LM
359 void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
360 void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
059715a4
RE
361
362 /* snow wavelet */
363 void (*vertical_compose97i)(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width);
364 void (*horizontal_compose97i)(DWTELEM *b, int width);
365 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
366
367 void (*prefetch)(void *mem, int stride, int h);
54009d42
MN
368
369 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
64db55ae
KS
370
371 /* vc1 functions */
372 void (*vc1_inv_trans_8x8)(DCTELEM *b);
373 void (*vc1_inv_trans_8x4)(DCTELEM *b, int n);
374 void (*vc1_inv_trans_4x8)(DCTELEM *b, int n);
375 void (*vc1_inv_trans_4x4)(DCTELEM *b, int n);
376 void (*vc1_v_overlap)(uint8_t* src, int stride, int rnd);
377 void (*vc1_h_overlap)(uint8_t* src, int stride, int rnd);
378 /* put 8x8 block with bicubic interpolation and quarterpel precision
379 * last argument is actually round value instead of height
380 */
381 op_pixels_func put_vc1_mspel_pixels_tab[16];
eb4b3dd3
ZK
382} DSPContext;
383
59cf08ce 384void dsputil_static_init(void);
b0368839 385void dsputil_init(DSPContext* p, AVCodecContext *avctx);
de6d9b64 386
7801d21d
MN
387/**
388 * permute block according to permuatation.
389 * @param last last non zero element in scantable order
390 */
0c1a9eda 391void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
e0eac44e 392
622348f9
MN
393void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
394
bb270c08 395#define BYTE_VEC32(c) ((c)*0x01010101UL)
d8085ea7
MN
396
397static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
398{
399 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
400}
401
402static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
403{
404 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
405}
406
26efc54e
MN
407static inline int get_penalty_factor(int lambda, int lambda2, int type){
408 switch(type&0xFF){
409 default:
410 case FF_CMP_SAD:
411 return lambda>>FF_LAMBDA_SHIFT;
412 case FF_CMP_DCT:
413 return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
414 case FF_CMP_W53:
415 return (4*lambda)>>(FF_LAMBDA_SHIFT);
416 case FF_CMP_W97:
417 return (2*lambda)>>(FF_LAMBDA_SHIFT);
418 case FF_CMP_SATD:
27c61ac5 419 case FF_CMP_DCT264:
26efc54e
MN
420 return (2*lambda)>>FF_LAMBDA_SHIFT;
421 case FF_CMP_RD:
422 case FF_CMP_PSNR:
423 case FF_CMP_SSE:
424 case FF_CMP_NSSE:
425 return lambda2>>FF_LAMBDA_SHIFT;
426 case FF_CMP_BIT:
427 return 1;
428 }
429}
430
24641185 431/**
77c92c2d 432 * Empty mmx state.
24641185
MN
433 * this must be called between any dsp function and float/double code.
434 * for example sin(); dsp->idct_put(); emms_c(); cos()
435 */
eb4b3dd3
ZK
436#define emms_c()
437
e629ab68
RD
438/* should be defined by architectures supporting
439 one or more MultiMedia extension */
440int mm_support(void);
441
68b51e58
SH
442#ifdef __GNUC__
443 #define DECLARE_ALIGNED_16(t,v) t v __attribute__ ((aligned (16)))
444#else
445 #define DECLARE_ALIGNED_16(t,v) __declspec(align(16)) t v
446#endif
92a69cf8 447
3d03c0a2 448#if defined(HAVE_MMX)
de6d9b64 449
18f77016 450#undef emms_c
eb4b3dd3 451
de6d9b64
FB
452#define MM_MMX 0x0001 /* standard MMX */
453#define MM_3DNOW 0x0004 /* AMD 3DNOW */
454#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
455#define MM_SSE 0x0008 /* SSE functions */
456#define MM_SSE2 0x0010 /* PIV SSE2 functions */
e42a152b 457#define MM_3DNOWEXT 0x0020 /* AMD 3DNowExt */
392f6da8 458#define MM_SSE3 0x0040 /* Prescott SSE3 functions */
de6d9b64
FB
459
460extern int mm_flags;
461
0c1a9eda
ZK
462void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
463void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
f9ed9d85 464void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
de6d9b64
FB
465
466static inline void emms(void)
467{
fb16b7e7
FB
468 __asm __volatile ("emms;":::"memory");
469}
470
1457ab52 471
fb16b7e7
FB
472#define emms_c() \
473{\
474 if (mm_flags & MM_MMX)\
475 emms();\
de6d9b64
FB
476}
477
68b51e58
SH
478#ifdef __GNUC__
479 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
480#else
481 #define DECLARE_ALIGNED_8(t,v) __declspec(align(8)) t v
482#endif
483
3237f731 484#define STRIDE_ALIGN 8
de6d9b64 485
b0368839
MN
486void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
487void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
2720569a 488
3d03c0a2
FB
489#elif defined(ARCH_ARMV4L)
490
3d03c0a2 491/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
92a69cf8 492 line optimizations */
68b51e58 493#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (4)))
3237f731 494#define STRIDE_ALIGN 4
3d03c0a2 495
eba9ae3c
GB
496#define MM_IWMMXT 0x0100 /* XScale IWMMXT */
497
498extern int mm_flags;
499
b0368839 500void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
3d03c0a2 501
c34270f5 502#elif defined(HAVE_MLIB)
c34270f5
FB
503
504/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
68b51e58 505#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
3237f731 506#define STRIDE_ALIGN 8
c34270f5 507
b0368839 508void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
c34270f5 509
44f54ceb
MN
510#elif defined(ARCH_SPARC)
511
512/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
68b51e58 513#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
3237f731 514#define STRIDE_ALIGN 8
44f54ceb
MN
515void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
516
1e98dffb
NK
517#elif defined(ARCH_ALPHA)
518
68b51e58 519#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
3237f731 520#define STRIDE_ALIGN 8
1e98dffb 521
b0368839 522void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
1e98dffb 523
59925ef2
BF
524#elif defined(ARCH_POWERPC)
525
404d2241
BF
526#define MM_ALTIVEC 0x0001 /* standard AltiVec */
527
528extern int mm_flags;
529
3b991c54 530#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
9a197a24 531#define pixel altivec_pixel
3b991c54 532#include <altivec.h>
9a197a24 533#undef pixel
3b991c54
RD
534#endif
535
68b51e58 536#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16)))
3237f731 537#define STRIDE_ALIGN 16
59925ef2 538
b0368839 539void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
59925ef2 540
d46aba26
LS
541#elif defined(HAVE_MMI)
542
68b51e58 543#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16)))
3237f731 544#define STRIDE_ALIGN 16
d46aba26 545
b0368839 546void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
d46aba26 547
0c6bd2ea
B
548#elif defined(ARCH_SH4)
549
68b51e58 550#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
3237f731 551#define STRIDE_ALIGN 8
0c6bd2ea
B
552
553void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
554
de6d9b64
FB
555#else
556
68b51e58 557#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
3237f731 558#define STRIDE_ALIGN 8
de6d9b64
FB
559
560#endif
561
6d4985bb
FB
562#ifdef __GNUC__
563
564struct unaligned_64 { uint64_t l; } __attribute__((packed));
565struct unaligned_32 { uint32_t l; } __attribute__((packed));
669ac79c 566struct unaligned_16 { uint16_t l; } __attribute__((packed));
6d4985bb 567
669ac79c 568#define LD16(a) (((const struct unaligned_16 *) (a))->l)
6d4985bb
FB
569#define LD32(a) (((const struct unaligned_32 *) (a))->l)
570#define LD64(a) (((const struct unaligned_64 *) (a))->l)
571
80e44bc3 572#define ST16(a, b) (((struct unaligned_16 *) (a))->l) = (b)
6d4985bb
FB
573#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
574
575#else /* __GNUC__ */
576
669ac79c 577#define LD16(a) (*((uint16_t*)(a)))
6d4985bb
FB
578#define LD32(a) (*((uint32_t*)(a)))
579#define LD64(a) (*((uint64_t*)(a)))
580
dc2bf20e 581#define ST16(a, b) *((uint16_t*)(a)) = (b)
6d4985bb
FB
582#define ST32(a, b) *((uint32_t*)(a)) = (b)
583
584#endif /* !__GNUC__ */
585
43f1708f 586/* PSNR */
0c1a9eda 587void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
43f1708f
J
588 int orig_linesize[3], int coded_linesize,
589 AVCodecContext *avctx);
bb6f5690
FB
590
591/* FFT computation */
592
593/* NOTE: soon integer code will be added, so you must use the
594 FFTSample type */
595typedef float FFTSample;
596
bcfa3e58
LM
597struct MDCTContext;
598
bb6f5690
FB
599typedef struct FFTComplex {
600 FFTSample re, im;
601} FFTComplex;
602
603typedef struct FFTContext {
604 int nbits;
605 int inverse;
606 uint16_t *revtab;
607 FFTComplex *exptab;
608 FFTComplex *exptab1; /* only used by SSE code */
609 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
bcfa3e58
LM
610 void (*imdct_calc)(struct MDCTContext *s, FFTSample *output,
611 const FFTSample *input, FFTSample *tmp);
bb6f5690
FB
612} FFTContext;
613
68951ecf
GB
614int ff_fft_init(FFTContext *s, int nbits, int inverse);
615void ff_fft_permute(FFTContext *s, FFTComplex *z);
616void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
617void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
82eb4b0f
ZM
618void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
619void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
68951ecf 620void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
8d268a7d 621
68951ecf 622static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
bb6f5690
FB
623{
624 s->fft_calc(s, z);
625}
68951ecf 626void ff_fft_end(FFTContext *s);
bb6f5690
FB
627
628/* MDCT computation */
629
630typedef struct MDCTContext {
631 int n; /* size of MDCT (i.e. number of input data * 2) */
632 int nbits; /* n = 2^nbits */
633 /* pre/post rotation tables */
634 FFTSample *tcos;
635 FFTSample *tsin;
636 FFTContext fft;
637} MDCTContext;
638
82696bee 639int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
eb4b3dd3 640void ff_imdct_calc(MDCTContext *s, FFTSample *output,
bb6f5690 641 const FFTSample *input, FFTSample *tmp);
bcfa3e58
LM
642void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output,
643 const FFTSample *input, FFTSample *tmp);
eb4b3dd3 644void ff_mdct_calc(MDCTContext *s, FFTSample *out,
bb6f5690 645 const FFTSample *input, FFTSample *tmp);
82696bee 646void ff_mdct_end(MDCTContext *s);
bb6f5690 647
bb198e19
MN
648#define WARPER8_16(name8, name16)\
649static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
650 return name8(s, dst , src , stride, h)\
651 +name8(s, dst+8 , src+8 , stride, h);\
652}
653
654#define WARPER8_16_SQ(name8, name16)\
655static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
656 int score=0;\
657 score +=name8(s, dst , src , stride, 8);\
658 score +=name8(s, dst+8 , src+8 , stride, 8);\
659 if(h==16){\
660 dst += 8*stride;\
661 src += 8*stride;\
662 score +=name8(s, dst , src , stride, 8);\
663 score +=name8(s, dst+8 , src+8 , stride, 8);\
664 }\
665 return score;\
1457ab52
MN
666}
667
de6d9b64 668#endif