* adding Michael's explanation to the FAQ
[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
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
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);
45
3f09f52a 46void ff_fdct_mmx(DCTELEM *block);
cf3bf5bb 47void ff_fdct_mmx2(DCTELEM *block);
8fd19ab2 48void ff_fdct_sse2(DCTELEM *block);
de6d9b64 49
e0eac44e 50/* encoding scans */
0c1a9eda
ZK
51extern const uint8_t ff_alternate_horizontal_scan[64];
52extern const uint8_t ff_alternate_vertical_scan[64];
53extern const uint8_t ff_zigzag_direct[64];
10acc479 54extern const uint8_t ff_zigzag248_direct[64];
5a240838 55
de6d9b64 56/* pixel operations */
f2e92ef2 57#define MAX_NEG_CROP 1024
de6d9b64
FB
58
59/* temporary */
0c1a9eda
ZK
60extern uint32_t squareTbl[512];
61extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
de6d9b64 62
44cb64ee
MM
63/* VP3 DSP functions */
64void vp3_dsp_init_c(void);
116824d0
MM
65void vp3_idct_c(int16_t *input_data, int16_t *dequant_matrix,
66 int coeff_count, DCTELEM *output_data);
44cb64ee
MM
67
68void vp3_dsp_init_mmx(void);
116824d0
MM
69void vp3_idct_mmx(int16_t *input_data, int16_t *dequant_matrix,
70 int coeff_count, DCTELEM *output_data);
44cb64ee 71
38acbc3c 72void vp3_dsp_init_sse2(void);
116824d0
MM
73void vp3_idct_sse2(int16_t *input_data, int16_t *dequant_matrix,
74 int coeff_count, DCTELEM *output_data);
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);
b3184779 101
db794953 102#define DEF_OLD_QPEL(name)\
0c1a9eda
ZK
103void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
104void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
105void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
db794953
MN
106
107DEF_OLD_QPEL(qpel16_mc11_old_c)
108DEF_OLD_QPEL(qpel16_mc31_old_c)
109DEF_OLD_QPEL(qpel16_mc12_old_c)
110DEF_OLD_QPEL(qpel16_mc32_old_c)
111DEF_OLD_QPEL(qpel16_mc13_old_c)
112DEF_OLD_QPEL(qpel16_mc33_old_c)
113DEF_OLD_QPEL(qpel8_mc11_old_c)
114DEF_OLD_QPEL(qpel8_mc31_old_c)
115DEF_OLD_QPEL(qpel8_mc12_old_c)
116DEF_OLD_QPEL(qpel8_mc32_old_c)
117DEF_OLD_QPEL(qpel8_mc13_old_c)
118DEF_OLD_QPEL(qpel8_mc33_old_c)
b3184779
MN
119
120#define CALL_2X_PIXELS(a, b, n)\
121static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
122 b(block , pixels , line_size, h);\
123 b(block+n, pixels+n, line_size, h);\
124}
44eb4951 125
de6d9b64 126/* motion estimation */
7d67aa9b
MN
127// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
128// allthough currently h<4 is not used as functions with width <8 are not used and neither implemented
bb198e19 129typedef 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 130
0da71265 131
24641185
MN
132/**
133 * DSPContext.
134 */
eb4b3dd3
ZK
135typedef struct DSPContext {
136 /* pixel ops : interface with DCT */
0c1a9eda
ZK
137 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
138 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
139 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
f9ed9d85 140 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
0c1a9eda 141 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
d518aebd
MN
142 /**
143 * translational global motion compensation.
144 */
0c1a9eda 145 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
d518aebd
MN
146 /**
147 * global motion compensation.
148 */
0c1a9eda 149 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
eb4b3dd3
ZK
150 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
151 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
0c1a9eda
ZK
152 int (*pix_sum)(uint8_t * pix, int line_size);
153 int (*pix_norm1)(uint8_t * pix, int line_size);
bb198e19
MN
154// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
155
622348f9
MN
156 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
157 me_cmp_func sse[5];
158 me_cmp_func hadamard8_diff[5];
159 me_cmp_func dct_sad[5];
160 me_cmp_func quant_psnr[5];
161 me_cmp_func bit[5];
162 me_cmp_func rd[5];
163 me_cmp_func vsad[5];
164 me_cmp_func vsse[5];
e6a2ac34 165 me_cmp_func nsse[5];
26efc54e
MN
166 me_cmp_func w53[5];
167 me_cmp_func w97[5];
1457ab52 168
bb198e19
MN
169 me_cmp_func me_pre_cmp[5];
170 me_cmp_func me_cmp[5];
171 me_cmp_func me_sub_cmp[5];
172 me_cmp_func mb_cmp[5];
622348f9 173 me_cmp_func ildct_cmp[5]; //only width 16 used
eb4b3dd3 174
d518aebd
MN
175 /**
176 * Halfpel motion compensation with rounding (a+b+1)>>1.
669ac79c 177 * this is an array[4][4] of motion compensation funcions for 4
e5771f4f 178 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 179 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
180 * @param block destination where the result is stored
181 * @param pixels source
182 * @param line_size number of bytes in a horizontal line of block
183 * @param h height
184 */
669ac79c 185 op_pixels_func put_pixels_tab[4][4];
d518aebd
MN
186
187 /**
188 * Halfpel motion compensation with rounding (a+b+1)>>1.
4e8eed2f 189 * This is an array[4][4] of motion compensation functions for 4
e5771f4f 190 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 191 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
192 * @param block destination into which the result is averaged (a+b+1)>>1
193 * @param pixels source
194 * @param line_size number of bytes in a horizontal line of block
195 * @param h height
196 */
da3b9756 197 op_pixels_func avg_pixels_tab[4][4];
d518aebd
MN
198
199 /**
200 * Halfpel motion compensation with no rounding (a+b)>>1.
eb14c713
MN
201 * this is an array[2][4] of motion compensation funcions for 2
202 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 203 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
204 * @param block destination where the result is stored
205 * @param pixels source
206 * @param line_size number of bytes in a horizontal line of block
207 * @param h height
208 */
dbc56b39 209 op_pixels_func put_no_rnd_pixels_tab[4][4];
d518aebd
MN
210
211 /**
212 * Halfpel motion compensation with no rounding (a+b)>>1.
eb14c713
MN
213 * this is an array[2][4] of motion compensation funcions for 2
214 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 215 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
216 * @param block destination into which the result is averaged (a+b)>>1
217 * @param pixels source
218 * @param line_size number of bytes in a horizontal line of block
219 * @param h height
220 */
dbc56b39 221 op_pixels_func avg_no_rnd_pixels_tab[4][4];
669ac79c 222
c0a0170c
MN
223 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);
224
669ac79c
MN
225 /**
226 * Thirdpel motion compensation with rounding (a+b+1)>>1.
227 * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
228 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
229 * @param block destination where the result is stored
230 * @param pixels source
231 * @param line_size number of bytes in a horizontal line of block
232 * @param h height
233 */
234 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
da3b9756
MM
235 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
236
eb4b3dd3
ZK
237 qpel_mc_func put_qpel_pixels_tab[2][16];
238 qpel_mc_func avg_qpel_pixels_tab[2][16];
239 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
240 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
1457ab52 241 qpel_mc_func put_mspel_pixels_tab[8];
0da71265
MN
242
243 /**
244 * h264 Chram MC
245 */
246 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
247 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
eb4b3dd3 248
0da71265
MN
249 qpel_mc_func put_h264_qpel_pixels_tab[3][16];
250 qpel_mc_func avg_h264_qpel_pixels_tab[3][16];
251
bb198e19 252 me_cmp_func pix_abs[2][4];
11f18faf
MN
253
254 /* huffyuv specific */
11f18faf 255 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
1457ab52 256 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
84705403
MN
257 /**
258 * subtract huffyuv's variant of median prediction
259 * note, this might read from src1[-1], src2[-1]
260 */
261 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
3d2e8cce 262 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
b0368839 263
332f9ac4
MN
264 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
265 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
266
fdbbf2e0 267 void (*h261_loop_filter)(uint8_t *src, int stride);
c6148de2 268
b0368839
MN
269 /* (I)DCT */
270 void (*fdct)(DCTELEM *block/* align 16*/);
10acc479 271 void (*fdct248)(DCTELEM *block/* align 16*/);
4fb518c3
MN
272
273 /* IDCT really*/
274 void (*idct)(DCTELEM *block/* align 16*/);
24641185
MN
275
276 /**
77c92c2d 277 * block -> idct -> clip to unsigned 8 bit -> dest.
24641185 278 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
a1adf436 279 * @param line_size size in bytes of a horizotal line of dest
24641185 280 */
b0368839 281 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
24641185
MN
282
283 /**
284 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
a1adf436 285 * @param line_size size in bytes of a horizotal line of dest
24641185 286 */
b0368839 287 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
24641185
MN
288
289 /**
77c92c2d 290 * idct input permutation.
05493021
MN
291 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
292 * IDCT)
293 * this permutation must be performed before the idct_put/add, note, normally this can be merged
294 * with the zigzag/alternate scan<br>
24641185
MN
295 * an example to avoid confusion:
296 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
297 * - (x -> referece dct -> reference idct -> x)
298 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
299 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
300 */
b0368839
MN
301 uint8_t idct_permutation[64];
302 int idct_permutation_type;
303#define FF_NO_IDCT_PERM 1
304#define FF_LIBMPEG2_IDCT_PERM 2
305#define FF_SIMPLE_IDCT_PERM 3
306#define FF_TRANSPOSE_IDCT_PERM 4
307
364a1797
MN
308 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
309 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
310#define BASIS_SHIFT 16
311#define RECON_SHIFT 6
312
44cb64ee
MM
313 /**
314 * This function handles any initialization for the VP3 DSP functions.
315 */
316 void (*vp3_dsp_init)(void);
317
318 /**
319 * This function is responsible for taking a block of zigzag'd,
116824d0
MM
320 * quantized DCT coefficients and reconstructing the original block of
321 * samples.
44cb64ee
MM
322 * @param input_data 64 zigzag'd, quantized DCT coefficients
323 * @param dequant_matrix 64 zigzag'd quantizer coefficients
324 * @param coeff_count index of the last coefficient
116824d0
MM
325 * @param output_samples space for 64 DCTELEMs where the transformed
326 * samples will be stored
44cb64ee 327 */
116824d0
MM
328 void (*vp3_idct)(int16_t *input_data, int16_t *dequant_matrix,
329 int coeff_count, DCTELEM *output_samples);
44cb64ee 330
eb4b3dd3
ZK
331} DSPContext;
332
59cf08ce 333void dsputil_static_init(void);
b0368839 334void dsputil_init(DSPContext* p, AVCodecContext *avctx);
de6d9b64 335
7801d21d
MN
336/**
337 * permute block according to permuatation.
338 * @param last last non zero element in scantable order
339 */
0c1a9eda 340void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
e0eac44e 341
622348f9
MN
342void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
343
d8085ea7
MN
344#define BYTE_VEC32(c) ((c)*0x01010101UL)
345
346static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
347{
348 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
349}
350
351static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
352{
353 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
354}
355
26efc54e
MN
356static inline int get_penalty_factor(int lambda, int lambda2, int type){
357 switch(type&0xFF){
358 default:
359 case FF_CMP_SAD:
360 return lambda>>FF_LAMBDA_SHIFT;
361 case FF_CMP_DCT:
362 return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
363 case FF_CMP_W53:
364 return (4*lambda)>>(FF_LAMBDA_SHIFT);
365 case FF_CMP_W97:
366 return (2*lambda)>>(FF_LAMBDA_SHIFT);
367 case FF_CMP_SATD:
368 return (2*lambda)>>FF_LAMBDA_SHIFT;
369 case FF_CMP_RD:
370 case FF_CMP_PSNR:
371 case FF_CMP_SSE:
372 case FF_CMP_NSSE:
373 return lambda2>>FF_LAMBDA_SHIFT;
374 case FF_CMP_BIT:
375 return 1;
376 }
377}
378
24641185 379/**
77c92c2d 380 * Empty mmx state.
24641185
MN
381 * this must be called between any dsp function and float/double code.
382 * for example sin(); dsp->idct_put(); emms_c(); cos()
383 */
eb4b3dd3
ZK
384#define emms_c()
385
e629ab68
RD
386/* should be defined by architectures supporting
387 one or more MultiMedia extension */
388int mm_support(void);
389
92a69cf8
MM
390#define __align16 __attribute__ ((aligned (16)))
391
3d03c0a2 392#if defined(HAVE_MMX)
de6d9b64 393
18f77016 394#undef emms_c
eb4b3dd3 395
de6d9b64
FB
396#define MM_MMX 0x0001 /* standard MMX */
397#define MM_3DNOW 0x0004 /* AMD 3DNOW */
398#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
399#define MM_SSE 0x0008 /* SSE functions */
400#define MM_SSE2 0x0010 /* PIV SSE2 functions */
401
402extern int mm_flags;
403
0c1a9eda
ZK
404void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
405void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
f9ed9d85 406void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
de6d9b64
FB
407
408static inline void emms(void)
409{
fb16b7e7
FB
410 __asm __volatile ("emms;":::"memory");
411}
412
1457ab52 413
fb16b7e7
FB
414#define emms_c() \
415{\
416 if (mm_flags & MM_MMX)\
417 emms();\
de6d9b64
FB
418}
419
420#define __align8 __attribute__ ((aligned (8)))
421
b0368839
MN
422void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
423void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
2720569a 424
3d03c0a2
FB
425#elif defined(ARCH_ARMV4L)
426
3d03c0a2 427/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
92a69cf8 428 line optimizations */
3d03c0a2
FB
429#define __align8 __attribute__ ((aligned (4)))
430
b0368839 431void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
3d03c0a2 432
c34270f5 433#elif defined(HAVE_MLIB)
c34270f5
FB
434
435/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
436#define __align8 __attribute__ ((aligned (8)))
437
b0368839 438void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
c34270f5 439
44f54ceb
MN
440#elif defined(ARCH_SPARC)
441
442/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
443#define __align8 __attribute__ ((aligned (8)))
444void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
445
1e98dffb
NK
446#elif defined(ARCH_ALPHA)
447
1e98dffb
NK
448#define __align8 __attribute__ ((aligned (8)))
449
b0368839 450void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
1e98dffb 451
59925ef2
BF
452#elif defined(ARCH_POWERPC)
453
404d2241
BF
454#define MM_ALTIVEC 0x0001 /* standard AltiVec */
455
456extern int mm_flags;
457
3b991c54 458#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
9a197a24 459#define pixel altivec_pixel
3b991c54 460#include <altivec.h>
9a197a24 461#undef pixel
3b991c54
RD
462#endif
463
59925ef2
BF
464#define __align8 __attribute__ ((aligned (16)))
465
b0368839 466void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
59925ef2 467
d46aba26
LS
468#elif defined(HAVE_MMI)
469
d46aba26
LS
470#define __align8 __attribute__ ((aligned (16)))
471
b0368839 472void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
d46aba26 473
0c6bd2ea
B
474#elif defined(ARCH_SH4)
475
476#define __align8 __attribute__ ((aligned (8)))
477
478void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
479
de6d9b64
FB
480#else
481
482#define __align8
483
484#endif
485
6d4985bb
FB
486#ifdef __GNUC__
487
488struct unaligned_64 { uint64_t l; } __attribute__((packed));
489struct unaligned_32 { uint32_t l; } __attribute__((packed));
669ac79c 490struct unaligned_16 { uint16_t l; } __attribute__((packed));
6d4985bb 491
669ac79c 492#define LD16(a) (((const struct unaligned_16 *) (a))->l)
6d4985bb
FB
493#define LD32(a) (((const struct unaligned_32 *) (a))->l)
494#define LD64(a) (((const struct unaligned_64 *) (a))->l)
495
496#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
497
498#else /* __GNUC__ */
499
669ac79c 500#define LD16(a) (*((uint16_t*)(a)))
6d4985bb
FB
501#define LD32(a) (*((uint32_t*)(a)))
502#define LD64(a) (*((uint64_t*)(a)))
503
504#define ST32(a, b) *((uint32_t*)(a)) = (b)
505
506#endif /* !__GNUC__ */
507
43f1708f 508/* PSNR */
0c1a9eda 509void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
43f1708f
J
510 int orig_linesize[3], int coded_linesize,
511 AVCodecContext *avctx);
bb6f5690
FB
512
513/* FFT computation */
514
515/* NOTE: soon integer code will be added, so you must use the
516 FFTSample type */
517typedef float FFTSample;
518
519typedef struct FFTComplex {
520 FFTSample re, im;
521} FFTComplex;
522
523typedef struct FFTContext {
524 int nbits;
525 int inverse;
526 uint16_t *revtab;
527 FFTComplex *exptab;
528 FFTComplex *exptab1; /* only used by SSE code */
529 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
530} FFTContext;
531
68951ecf
GB
532int ff_fft_init(FFTContext *s, int nbits, int inverse);
533void ff_fft_permute(FFTContext *s, FFTComplex *z);
534void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
535void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
536void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
8d268a7d 537
68951ecf 538static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
bb6f5690
FB
539{
540 s->fft_calc(s, z);
541}
68951ecf 542void ff_fft_end(FFTContext *s);
bb6f5690
FB
543
544/* MDCT computation */
545
546typedef struct MDCTContext {
547 int n; /* size of MDCT (i.e. number of input data * 2) */
548 int nbits; /* n = 2^nbits */
549 /* pre/post rotation tables */
550 FFTSample *tcos;
551 FFTSample *tsin;
552 FFTContext fft;
553} MDCTContext;
554
82696bee 555int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
eb4b3dd3 556void ff_imdct_calc(MDCTContext *s, FFTSample *output,
bb6f5690 557 const FFTSample *input, FFTSample *tmp);
eb4b3dd3 558void ff_mdct_calc(MDCTContext *s, FFTSample *out,
bb6f5690 559 const FFTSample *input, FFTSample *tmp);
82696bee 560void ff_mdct_end(MDCTContext *s);
bb6f5690 561
bb198e19
MN
562#define WARPER8_16(name8, name16)\
563static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
564 return name8(s, dst , src , stride, h)\
565 +name8(s, dst+8 , src+8 , stride, h);\
566}
567
568#define WARPER8_16_SQ(name8, name16)\
569static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
570 int score=0;\
571 score +=name8(s, dst , src , stride, 8);\
572 score +=name8(s, dst+8 , src+8 , stride, 8);\
573 if(h==16){\
574 dst += 8*stride;\
575 src += 8*stride;\
576 score +=name8(s, dst , src , stride, 8);\
577 score +=name8(s, dst+8 , src+8 , stride, 8);\
578 }\
579 return score;\
1457ab52
MN
580}
581
95e2ce4a 582#ifndef HAVE_LRINTF
9d85cbd9
FB
583/* XXX: add ISOC specific test to avoid specific BSD testing. */
584/* better than nothing implementation. */
6234d753 585/* btw, rintf() is existing on fbsd too -- alex */
9d85cbd9
FB
586static inline long int lrintf(float x)
587{
ea937d01
FB
588#ifdef CONFIG_WIN32
589 /* XXX: incorrect, but make it compile */
590 return (int)(x);
591#else
9d85cbd9 592 return (int)(rint(x));
ea937d01 593#endif
9d85cbd9 594}
e96682e6
MN
595#else
596#ifndef _ISOC9X_SOURCE
597#define _ISOC9X_SOURCE
598#endif
599#include <math.h>
9d85cbd9
FB
600#endif
601
de6d9b64 602#endif