Correct the relation between floors of type 0 and block sizes.
[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;
38
03c94ede 39void fdct_ifast (DCTELEM *data);
48b1f800 40void fdct_ifast248 (DCTELEM *data);
28db7fce 41void ff_jpeg_fdct_islow (DCTELEM *data);
10acc479 42void ff_fdct248_islow (DCTELEM *data);
de6d9b64
FB
43
44void j_rev_dct (DCTELEM *data);
178fcca8 45void j_rev_dct4 (DCTELEM *data);
9ca358b9 46void j_rev_dct2 (DCTELEM *data);
1aa8c57b 47void j_rev_dct1 (DCTELEM *data);
de6d9b64 48
3f09f52a 49void ff_fdct_mmx(DCTELEM *block);
cf3bf5bb 50void ff_fdct_mmx2(DCTELEM *block);
8fd19ab2 51void ff_fdct_sse2(DCTELEM *block);
de6d9b64 52
43efd19a 53void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
0fa8158d
MN
54void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
55void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
56void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
57
e0eac44e 58/* encoding scans */
0c1a9eda
ZK
59extern const uint8_t ff_alternate_horizontal_scan[64];
60extern const uint8_t ff_alternate_vertical_scan[64];
61extern const uint8_t ff_zigzag_direct[64];
10acc479 62extern const uint8_t ff_zigzag248_direct[64];
5a240838 63
de6d9b64 64/* pixel operations */
f2e92ef2 65#define MAX_NEG_CROP 1024
de6d9b64
FB
66
67/* temporary */
0c1a9eda
ZK
68extern uint32_t squareTbl[512];
69extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
de6d9b64 70
44cb64ee 71/* VP3 DSP functions */
8b6103da
MN
72void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
73void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
74void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
de6d9b64 75
b7c27ee6 76/* minimum alignment rules ;)
eb4b3dd3 77if u notice errors in the align stuff, need more alignment for some asm code for some cpu
b7c27ee6
MN
78or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
79
80!warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
81i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
de6d9b64 82
b7c27ee6
MN
83!future video codecs might need functions with less strict alignment
84*/
85
eb4b3dd3 86/*
0c1a9eda
ZK
87void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
88void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
89void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
90void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
649c00c9 91void clear_blocks_c(DCTELEM *blocks);
eb4b3dd3 92*/
de6d9b64
FB
93
94/* add and put pixel (decoding) */
b7c27ee6 95// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
7d67aa9b 96//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
0c1a9eda 97typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
669ac79c 98typedef 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 99typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
0da71265 100typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
9f2d1b4f 101typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
e8b56208 102typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
b3184779 103
db794953 104#define DEF_OLD_QPEL(name)\
0c1a9eda
ZK
105void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
106void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
107void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
db794953
MN
108
109DEF_OLD_QPEL(qpel16_mc11_old_c)
110DEF_OLD_QPEL(qpel16_mc31_old_c)
111DEF_OLD_QPEL(qpel16_mc12_old_c)
112DEF_OLD_QPEL(qpel16_mc32_old_c)
113DEF_OLD_QPEL(qpel16_mc13_old_c)
114DEF_OLD_QPEL(qpel16_mc33_old_c)
115DEF_OLD_QPEL(qpel8_mc11_old_c)
116DEF_OLD_QPEL(qpel8_mc31_old_c)
117DEF_OLD_QPEL(qpel8_mc12_old_c)
118DEF_OLD_QPEL(qpel8_mc32_old_c)
119DEF_OLD_QPEL(qpel8_mc13_old_c)
120DEF_OLD_QPEL(qpel8_mc33_old_c)
b3184779
MN
121
122#define CALL_2X_PIXELS(a, b, n)\
123static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
124 b(block , pixels , line_size, h);\
125 b(block+n, pixels+n, line_size, h);\
126}
44eb4951 127
de6d9b64 128/* motion estimation */
7d67aa9b
MN
129// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
130// allthough currently h<4 is not used as functions with width <8 are not used and neither implemented
bb198e19 131typedef 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 132
0da71265 133
24641185
MN
134/**
135 * DSPContext.
136 */
eb4b3dd3
ZK
137typedef struct DSPContext {
138 /* pixel ops : interface with DCT */
0c1a9eda
ZK
139 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
140 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
141 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
f9ed9d85 142 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
0c1a9eda 143 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
36940eca
LM
144 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
145 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
d518aebd
MN
146 /**
147 * translational global motion compensation.
148 */
0c1a9eda 149 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
d518aebd
MN
150 /**
151 * global motion compensation.
152 */
0c1a9eda 153 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
bb270c08 154 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
eb4b3dd3 155 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
0c1a9eda
ZK
156 int (*pix_sum)(uint8_t * pix, int line_size);
157 int (*pix_norm1)(uint8_t * pix, int line_size);
bb198e19 158// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
115329f1 159
622348f9
MN
160 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
161 me_cmp_func sse[5];
162 me_cmp_func hadamard8_diff[5];
163 me_cmp_func dct_sad[5];
164 me_cmp_func quant_psnr[5];
165 me_cmp_func bit[5];
166 me_cmp_func rd[5];
167 me_cmp_func vsad[5];
168 me_cmp_func vsse[5];
e6a2ac34 169 me_cmp_func nsse[5];
26efc54e
MN
170 me_cmp_func w53[5];
171 me_cmp_func w97[5];
0fd6aea1 172 me_cmp_func dct_max[5];
27c61ac5 173 me_cmp_func dct264_sad[5];
1457ab52 174
bb198e19
MN
175 me_cmp_func me_pre_cmp[5];
176 me_cmp_func me_cmp[5];
177 me_cmp_func me_sub_cmp[5];
178 me_cmp_func mb_cmp[5];
622348f9 179 me_cmp_func ildct_cmp[5]; //only width 16 used
0fd6aea1 180 me_cmp_func frame_skip_cmp[5]; //only width 8 used
eb4b3dd3 181
d518aebd
MN
182 /**
183 * Halfpel motion compensation with rounding (a+b+1)>>1.
115329f1 184 * this is an array[4][4] of motion compensation funcions for 4
e5771f4f 185 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 186 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
187 * @param block destination where the result is stored
188 * @param pixels source
189 * @param line_size number of bytes in a horizontal line of block
190 * @param h height
191 */
669ac79c 192 op_pixels_func put_pixels_tab[4][4];
d518aebd
MN
193
194 /**
195 * Halfpel motion compensation with rounding (a+b+1)>>1.
115329f1 196 * This is an array[4][4] of motion compensation functions for 4
e5771f4f 197 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 198 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
199 * @param block destination into which the result is averaged (a+b+1)>>1
200 * @param pixels source
201 * @param line_size number of bytes in a horizontal line of block
202 * @param h height
203 */
da3b9756 204 op_pixels_func avg_pixels_tab[4][4];
d518aebd
MN
205
206 /**
207 * Halfpel motion compensation with no rounding (a+b)>>1.
115329f1 208 * this is an array[2][4] of motion compensation funcions for 2
eb14c713 209 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 210 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
211 * @param block destination where the result is stored
212 * @param pixels source
213 * @param line_size number of bytes in a horizontal line of block
214 * @param h height
215 */
dbc56b39 216 op_pixels_func put_no_rnd_pixels_tab[4][4];
d518aebd
MN
217
218 /**
219 * Halfpel motion compensation with no rounding (a+b)>>1.
115329f1 220 * this is an array[2][4] of motion compensation funcions for 2
eb14c713 221 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 222 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
223 * @param block destination into which the result is averaged (a+b)>>1
224 * @param pixels source
225 * @param line_size number of bytes in a horizontal line of block
226 * @param h height
227 */
dbc56b39 228 op_pixels_func avg_no_rnd_pixels_tab[4][4];
115329f1 229
c0a0170c 230 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 231
669ac79c
MN
232 /**
233 * Thirdpel motion compensation with rounding (a+b+1)>>1.
234 * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
235 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
236 * @param block destination where the result is stored
237 * @param pixels source
238 * @param line_size number of bytes in a horizontal line of block
239 * @param h height
240 */
241 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
da3b9756
MM
242 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
243
eb4b3dd3
ZK
244 qpel_mc_func put_qpel_pixels_tab[2][16];
245 qpel_mc_func avg_qpel_pixels_tab[2][16];
246 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
247 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
1457ab52 248 qpel_mc_func put_mspel_pixels_tab[8];
115329f1 249
0da71265
MN
250 /**
251 * h264 Chram MC
252 */
253 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
254 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
eb4b3dd3 255
80e44bc3
MN
256 qpel_mc_func put_h264_qpel_pixels_tab[4][16];
257 qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
115329f1 258
9f2d1b4f
LM
259 h264_weight_func weight_h264_pixels_tab[10];
260 h264_biweight_func biweight_h264_pixels_tab[10];
115329f1 261
bb198e19 262 me_cmp_func pix_abs[2][4];
115329f1 263
11f18faf 264 /* huffyuv specific */
11f18faf 265 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
1457ab52 266 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
84705403
MN
267 /**
268 * subtract huffyuv's variant of median prediction
269 * note, this might read from src1[-1], src2[-1]
270 */
271 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
3d2e8cce 272 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
42251a2a 273
5cf08f23
LM
274 void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
275 void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
276 void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
277 void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
278 void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
279 void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
115329f1 280
332f9ac4
MN
281 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
282 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
283
fdbbf2e0 284 void (*h261_loop_filter)(uint8_t *src, int stride);
c6148de2 285
b0368839
MN
286 /* (I)DCT */
287 void (*fdct)(DCTELEM *block/* align 16*/);
10acc479 288 void (*fdct248)(DCTELEM *block/* align 16*/);
115329f1 289
4fb518c3
MN
290 /* IDCT really*/
291 void (*idct)(DCTELEM *block/* align 16*/);
115329f1 292
24641185 293 /**
77c92c2d 294 * block -> idct -> clip to unsigned 8 bit -> dest.
24641185 295 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
a1adf436 296 * @param line_size size in bytes of a horizotal line of dest
24641185 297 */
b0368839 298 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
115329f1 299
24641185
MN
300 /**
301 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
a1adf436 302 * @param line_size size in bytes of a horizotal line of dest
24641185 303 */
b0368839 304 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
115329f1 305
24641185 306 /**
77c92c2d 307 * idct input permutation.
05493021
MN
308 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
309 * IDCT)
310 * this permutation must be performed before the idct_put/add, note, normally this can be merged
311 * with the zigzag/alternate scan<br>
24641185
MN
312 * an example to avoid confusion:
313 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
314 * - (x -> referece dct -> reference idct -> x)
315 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
316 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
317 */
b0368839
MN
318 uint8_t idct_permutation[64];
319 int idct_permutation_type;
320#define FF_NO_IDCT_PERM 1
321#define FF_LIBMPEG2_IDCT_PERM 2
322#define FF_SIMPLE_IDCT_PERM 3
323#define FF_TRANSPOSE_IDCT_PERM 4
5773a746 324#define FF_PARTTRANS_IDCT_PERM 5
b0368839 325
364a1797
MN
326 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
327 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
328#define BASIS_SHIFT 16
329#define RECON_SHIFT 6
115329f1 330
0fa8158d 331 void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
43efd19a 332 void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
eb4b3dd3
ZK
333} DSPContext;
334
59cf08ce 335void dsputil_static_init(void);
b0368839 336void dsputil_init(DSPContext* p, AVCodecContext *avctx);
de6d9b64 337
7801d21d
MN
338/**
339 * permute block according to permuatation.
340 * @param last last non zero element in scantable order
341 */
0c1a9eda 342void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
e0eac44e 343
622348f9
MN
344void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
345
bb270c08 346#define BYTE_VEC32(c) ((c)*0x01010101UL)
d8085ea7
MN
347
348static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
349{
350 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
351}
352
353static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
354{
355 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
356}
357
26efc54e
MN
358static inline int get_penalty_factor(int lambda, int lambda2, int type){
359 switch(type&0xFF){
360 default:
361 case FF_CMP_SAD:
362 return lambda>>FF_LAMBDA_SHIFT;
363 case FF_CMP_DCT:
364 return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
365 case FF_CMP_W53:
366 return (4*lambda)>>(FF_LAMBDA_SHIFT);
367 case FF_CMP_W97:
368 return (2*lambda)>>(FF_LAMBDA_SHIFT);
369 case FF_CMP_SATD:
27c61ac5 370 case FF_CMP_DCT264:
26efc54e
MN
371 return (2*lambda)>>FF_LAMBDA_SHIFT;
372 case FF_CMP_RD:
373 case FF_CMP_PSNR:
374 case FF_CMP_SSE:
375 case FF_CMP_NSSE:
376 return lambda2>>FF_LAMBDA_SHIFT;
377 case FF_CMP_BIT:
378 return 1;
379 }
380}
381
24641185 382/**
77c92c2d 383 * Empty mmx state.
24641185
MN
384 * this must be called between any dsp function and float/double code.
385 * for example sin(); dsp->idct_put(); emms_c(); cos()
386 */
eb4b3dd3
ZK
387#define emms_c()
388
e629ab68
RD
389/* should be defined by architectures supporting
390 one or more MultiMedia extension */
391int mm_support(void);
392
92a69cf8
MM
393#define __align16 __attribute__ ((aligned (16)))
394
3d03c0a2 395#if defined(HAVE_MMX)
de6d9b64 396
18f77016 397#undef emms_c
eb4b3dd3 398
de6d9b64
FB
399#define MM_MMX 0x0001 /* standard MMX */
400#define MM_3DNOW 0x0004 /* AMD 3DNOW */
401#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
402#define MM_SSE 0x0008 /* SSE functions */
403#define MM_SSE2 0x0010 /* PIV SSE2 functions */
e42a152b 404#define MM_3DNOWEXT 0x0020 /* AMD 3DNowExt */
de6d9b64
FB
405
406extern int mm_flags;
407
0c1a9eda
ZK
408void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
409void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
f9ed9d85 410void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
de6d9b64
FB
411
412static inline void emms(void)
413{
fb16b7e7
FB
414 __asm __volatile ("emms;":::"memory");
415}
416
1457ab52 417
fb16b7e7
FB
418#define emms_c() \
419{\
420 if (mm_flags & MM_MMX)\
421 emms();\
de6d9b64
FB
422}
423
424#define __align8 __attribute__ ((aligned (8)))
3237f731 425#define STRIDE_ALIGN 8
de6d9b64 426
b0368839
MN
427void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
428void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
2720569a 429
3d03c0a2
FB
430#elif defined(ARCH_ARMV4L)
431
3d03c0a2 432/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
92a69cf8 433 line optimizations */
3d03c0a2 434#define __align8 __attribute__ ((aligned (4)))
3237f731 435#define STRIDE_ALIGN 4
3d03c0a2 436
eba9ae3c
GB
437#define MM_IWMMXT 0x0100 /* XScale IWMMXT */
438
439extern int mm_flags;
440
b0368839 441void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
3d03c0a2 442
c34270f5 443#elif defined(HAVE_MLIB)
c34270f5
FB
444
445/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
446#define __align8 __attribute__ ((aligned (8)))
3237f731 447#define STRIDE_ALIGN 8
c34270f5 448
b0368839 449void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
c34270f5 450
44f54ceb
MN
451#elif defined(ARCH_SPARC)
452
453/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
454#define __align8 __attribute__ ((aligned (8)))
3237f731 455#define STRIDE_ALIGN 8
44f54ceb
MN
456void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
457
1e98dffb
NK
458#elif defined(ARCH_ALPHA)
459
1e98dffb 460#define __align8 __attribute__ ((aligned (8)))
3237f731 461#define STRIDE_ALIGN 8
1e98dffb 462
b0368839 463void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
1e98dffb 464
59925ef2
BF
465#elif defined(ARCH_POWERPC)
466
404d2241
BF
467#define MM_ALTIVEC 0x0001 /* standard AltiVec */
468
469extern int mm_flags;
470
3b991c54 471#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
9a197a24 472#define pixel altivec_pixel
3b991c54 473#include <altivec.h>
9a197a24 474#undef pixel
3b991c54
RD
475#endif
476
59925ef2 477#define __align8 __attribute__ ((aligned (16)))
3237f731 478#define STRIDE_ALIGN 16
59925ef2 479
b0368839 480void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
59925ef2 481
d46aba26
LS
482#elif defined(HAVE_MMI)
483
d46aba26 484#define __align8 __attribute__ ((aligned (16)))
3237f731 485#define STRIDE_ALIGN 16
d46aba26 486
b0368839 487void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
d46aba26 488
0c6bd2ea
B
489#elif defined(ARCH_SH4)
490
491#define __align8 __attribute__ ((aligned (8)))
3237f731 492#define STRIDE_ALIGN 8
0c6bd2ea
B
493
494void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
495
de6d9b64
FB
496#else
497
3237f731
MN
498#define __align8 __attribute__ ((aligned (8)))
499#define STRIDE_ALIGN 8
de6d9b64
FB
500
501#endif
502
6d4985bb
FB
503#ifdef __GNUC__
504
505struct unaligned_64 { uint64_t l; } __attribute__((packed));
506struct unaligned_32 { uint32_t l; } __attribute__((packed));
669ac79c 507struct unaligned_16 { uint16_t l; } __attribute__((packed));
6d4985bb 508
669ac79c 509#define LD16(a) (((const struct unaligned_16 *) (a))->l)
6d4985bb
FB
510#define LD32(a) (((const struct unaligned_32 *) (a))->l)
511#define LD64(a) (((const struct unaligned_64 *) (a))->l)
512
80e44bc3 513#define ST16(a, b) (((struct unaligned_16 *) (a))->l) = (b)
6d4985bb
FB
514#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
515
516#else /* __GNUC__ */
517
669ac79c 518#define LD16(a) (*((uint16_t*)(a)))
6d4985bb
FB
519#define LD32(a) (*((uint32_t*)(a)))
520#define LD64(a) (*((uint64_t*)(a)))
521
522#define ST32(a, b) *((uint32_t*)(a)) = (b)
523
524#endif /* !__GNUC__ */
525
43f1708f 526/* PSNR */
0c1a9eda 527void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
43f1708f
J
528 int orig_linesize[3], int coded_linesize,
529 AVCodecContext *avctx);
bb6f5690
FB
530
531/* FFT computation */
532
533/* NOTE: soon integer code will be added, so you must use the
534 FFTSample type */
535typedef float FFTSample;
536
537typedef struct FFTComplex {
538 FFTSample re, im;
539} FFTComplex;
540
541typedef struct FFTContext {
542 int nbits;
543 int inverse;
544 uint16_t *revtab;
545 FFTComplex *exptab;
546 FFTComplex *exptab1; /* only used by SSE code */
547 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
548} FFTContext;
549
68951ecf
GB
550int ff_fft_init(FFTContext *s, int nbits, int inverse);
551void ff_fft_permute(FFTContext *s, FFTComplex *z);
552void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
553void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
554void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
8d268a7d 555
68951ecf 556static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
bb6f5690
FB
557{
558 s->fft_calc(s, z);
559}
68951ecf 560void ff_fft_end(FFTContext *s);
bb6f5690
FB
561
562/* MDCT computation */
563
564typedef struct MDCTContext {
565 int n; /* size of MDCT (i.e. number of input data * 2) */
566 int nbits; /* n = 2^nbits */
567 /* pre/post rotation tables */
568 FFTSample *tcos;
569 FFTSample *tsin;
570 FFTContext fft;
571} MDCTContext;
572
82696bee 573int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
eb4b3dd3 574void ff_imdct_calc(MDCTContext *s, FFTSample *output,
bb6f5690 575 const FFTSample *input, FFTSample *tmp);
eb4b3dd3 576void ff_mdct_calc(MDCTContext *s, FFTSample *out,
bb6f5690 577 const FFTSample *input, FFTSample *tmp);
82696bee 578void ff_mdct_end(MDCTContext *s);
bb6f5690 579
bb198e19
MN
580#define WARPER8_16(name8, name16)\
581static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
582 return name8(s, dst , src , stride, h)\
583 +name8(s, dst+8 , src+8 , stride, h);\
584}
585
586#define WARPER8_16_SQ(name8, name16)\
587static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
588 int score=0;\
589 score +=name8(s, dst , src , stride, 8);\
590 score +=name8(s, dst+8 , src+8 , stride, 8);\
591 if(h==16){\
592 dst += 8*stride;\
593 src += 8*stride;\
594 score +=name8(s, dst , src , stride, 8);\
595 score +=name8(s, dst+8 , src+8 , stride, 8);\
596 }\
597 return score;\
1457ab52
MN
598}
599
de6d9b64 600#endif