first pass at a new VP3 video decoder
[libav.git] / libavcodec / dsputil.h
CommitLineData
ff4ec49e
FB
1/*
2 * DSP utils
3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard.
4 *
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
9 *
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
14 *
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 */
24641185
MN
19
20/**
21 * @file dsputil.h
983e3246 22 * DSP utils.
5755c27f
MN
23 * note, many functions in here may use MMX which trashes the FPU state, it is
24 * absolutely necessary to call emms_c() between dsp & float/double code
24641185
MN
25 */
26
de6d9b64
FB
27#ifndef DSPUTIL_H
28#define DSPUTIL_H
29
30#include "common.h"
43f1708f 31#include "avcodec.h"
de6d9b64 32
24641185 33
44eb4951 34//#define DEBUG
de6d9b64
FB
35/* dct code */
36typedef short DCTELEM;
0e15384d 37//typedef int DCTELEM;
de6d9b64 38
03c94ede 39void fdct_ifast (DCTELEM *data);
28db7fce 40void ff_jpeg_fdct_islow (DCTELEM *data);
de6d9b64
FB
41
42void j_rev_dct (DCTELEM *data);
43
3f09f52a 44void ff_fdct_mmx(DCTELEM *block);
de6d9b64 45
e0eac44e 46/* encoding scans */
0c1a9eda
ZK
47extern const uint8_t ff_alternate_horizontal_scan[64];
48extern const uint8_t ff_alternate_vertical_scan[64];
49extern const uint8_t ff_zigzag_direct[64];
5a240838 50
de6d9b64
FB
51/* pixel operations */
52#define MAX_NEG_CROP 384
53
54/* temporary */
0c1a9eda
ZK
55extern uint32_t squareTbl[512];
56extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
de6d9b64 57
de6d9b64 58
b7c27ee6 59/* minimum alignment rules ;)
eb4b3dd3 60if u notice errors in the align stuff, need more alignment for some asm code for some cpu
b7c27ee6
MN
61or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
62
63!warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
64i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
de6d9b64 65
b7c27ee6
MN
66!future video codecs might need functions with less strict alignment
67*/
68
eb4b3dd3 69/*
0c1a9eda
ZK
70void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
71void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
72void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
73void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
649c00c9 74void clear_blocks_c(DCTELEM *blocks);
eb4b3dd3 75*/
de6d9b64
FB
76
77/* add and put pixel (decoding) */
b7c27ee6 78// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
0c1a9eda
ZK
79typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
80typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
0da71265 81typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
b3184779 82
db794953 83#define DEF_OLD_QPEL(name)\
0c1a9eda
ZK
84void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
85void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
86void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
db794953
MN
87
88DEF_OLD_QPEL(qpel16_mc11_old_c)
89DEF_OLD_QPEL(qpel16_mc31_old_c)
90DEF_OLD_QPEL(qpel16_mc12_old_c)
91DEF_OLD_QPEL(qpel16_mc32_old_c)
92DEF_OLD_QPEL(qpel16_mc13_old_c)
93DEF_OLD_QPEL(qpel16_mc33_old_c)
94DEF_OLD_QPEL(qpel8_mc11_old_c)
95DEF_OLD_QPEL(qpel8_mc31_old_c)
96DEF_OLD_QPEL(qpel8_mc12_old_c)
97DEF_OLD_QPEL(qpel8_mc32_old_c)
98DEF_OLD_QPEL(qpel8_mc13_old_c)
99DEF_OLD_QPEL(qpel8_mc33_old_c)
b3184779
MN
100
101#define CALL_2X_PIXELS(a, b, n)\
102static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
103 b(block , pixels , line_size, h);\
104 b(block+n, pixels+n, line_size, h);\
105}
44eb4951 106
de6d9b64
FB
107/* motion estimation */
108
0c1a9eda 109typedef int (*op_pixels_abs_func)(uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
1457ab52 110
0c1a9eda 111typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
1457ab52 112
0da71265 113
24641185
MN
114/**
115 * DSPContext.
116 */
eb4b3dd3
ZK
117typedef struct DSPContext {
118 /* pixel ops : interface with DCT */
0c1a9eda
ZK
119 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
120 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
121 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
122 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
d518aebd
MN
123 /**
124 * translational global motion compensation.
125 */
0c1a9eda 126 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
d518aebd
MN
127 /**
128 * global motion compensation.
129 */
0c1a9eda 130 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
eb4b3dd3
ZK
131 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
132 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
0c1a9eda
ZK
133 int (*pix_sum)(uint8_t * pix, int line_size);
134 int (*pix_norm1)(uint8_t * pix, int line_size);
1457ab52
MN
135 me_cmp_func sad[2]; /* identical to pix_absAxA except additional void * */
136 me_cmp_func sse[2];
137 me_cmp_func hadamard8_diff[2];
138 me_cmp_func dct_sad[2];
139 me_cmp_func quant_psnr[2];
3a87ac94
MN
140 me_cmp_func bit[2];
141 me_cmp_func rd[2];
1457ab52
MN
142 int (*hadamard8_abs )(uint8_t *src, int stride, int mean);
143
826f429a 144 me_cmp_func me_pre_cmp[11];
1457ab52
MN
145 me_cmp_func me_cmp[11];
146 me_cmp_func me_sub_cmp[11];
147 me_cmp_func mb_cmp[11];
eb4b3dd3
ZK
148
149 /* maybe create an array for 16/8 functions */
d518aebd
MN
150 /**
151 * Halfpel motion compensation with rounding (a+b+1)>>1.
5755c27f 152 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
153 * @param block destination where the result is stored
154 * @param pixels source
155 * @param line_size number of bytes in a horizontal line of block
156 * @param h height
157 */
eb4b3dd3 158 op_pixels_func put_pixels_tab[2][4];
d518aebd
MN
159
160 /**
161 * Halfpel motion compensation with rounding (a+b+1)>>1.
5755c27f 162 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
163 * @param block destination into which the result is averaged (a+b+1)>>1
164 * @param pixels source
165 * @param line_size number of bytes in a horizontal line of block
166 * @param h height
167 */
eb4b3dd3 168 op_pixels_func avg_pixels_tab[2][4];
d518aebd
MN
169
170 /**
171 * Halfpel motion compensation with no rounding (a+b)>>1.
5755c27f 172 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
173 * @param block destination where the result is stored
174 * @param pixels source
175 * @param line_size number of bytes in a horizontal line of block
176 * @param h height
177 */
eb4b3dd3 178 op_pixels_func put_no_rnd_pixels_tab[2][4];
d518aebd
MN
179
180 /**
181 * Halfpel motion compensation with no rounding (a+b)>>1.
5755c27f 182 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
183 * @param block destination into which the result is averaged (a+b)>>1
184 * @param pixels source
185 * @param line_size number of bytes in a horizontal line of block
186 * @param h height
187 */
eb4b3dd3
ZK
188 op_pixels_func avg_no_rnd_pixels_tab[2][4];
189 qpel_mc_func put_qpel_pixels_tab[2][16];
190 qpel_mc_func avg_qpel_pixels_tab[2][16];
191 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
192 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
1457ab52 193 qpel_mc_func put_mspel_pixels_tab[8];
0da71265
MN
194
195 /**
196 * h264 Chram MC
197 */
198 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
199 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
eb4b3dd3 200
0da71265
MN
201 qpel_mc_func put_h264_qpel_pixels_tab[3][16];
202 qpel_mc_func avg_h264_qpel_pixels_tab[3][16];
203
eb4b3dd3
ZK
204 op_pixels_abs_func pix_abs16x16;
205 op_pixels_abs_func pix_abs16x16_x2;
206 op_pixels_abs_func pix_abs16x16_y2;
207 op_pixels_abs_func pix_abs16x16_xy2;
208 op_pixels_abs_func pix_abs8x8;
209 op_pixels_abs_func pix_abs8x8_x2;
210 op_pixels_abs_func pix_abs8x8_y2;
211 op_pixels_abs_func pix_abs8x8_xy2;
11f18faf
MN
212
213 /* huffyuv specific */
11f18faf 214 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
1457ab52 215 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
b0368839
MN
216
217 /* (I)DCT */
218 void (*fdct)(DCTELEM *block/* align 16*/);
24641185
MN
219
220 /**
77c92c2d 221 * block -> idct -> clip to unsigned 8 bit -> dest.
24641185 222 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
a1adf436 223 * @param line_size size in bytes of a horizotal line of dest
24641185 224 */
b0368839 225 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
24641185
MN
226
227 /**
228 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
a1adf436 229 * @param line_size size in bytes of a horizotal line of dest
24641185 230 */
b0368839 231 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
24641185
MN
232
233 /**
77c92c2d 234 * idct input permutation.
24641185
MN
235 * an example to avoid confusion:
236 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
237 * - (x -> referece dct -> reference idct -> x)
238 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
239 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
240 */
b0368839
MN
241 uint8_t idct_permutation[64];
242 int idct_permutation_type;
243#define FF_NO_IDCT_PERM 1
244#define FF_LIBMPEG2_IDCT_PERM 2
245#define FF_SIMPLE_IDCT_PERM 3
246#define FF_TRANSPOSE_IDCT_PERM 4
247
eb4b3dd3
ZK
248} DSPContext;
249
59cf08ce 250void dsputil_static_init(void);
b0368839 251void dsputil_init(DSPContext* p, AVCodecContext *avctx);
de6d9b64 252
7801d21d
MN
253/**
254 * permute block according to permuatation.
255 * @param last last non zero element in scantable order
256 */
0c1a9eda 257void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
e0eac44e 258
24641185 259/**
77c92c2d 260 * Empty mmx state.
24641185
MN
261 * this must be called between any dsp function and float/double code.
262 * for example sin(); dsp->idct_put(); emms_c(); cos()
263 */
eb4b3dd3
ZK
264#define emms_c()
265
e629ab68
RD
266/* should be defined by architectures supporting
267 one or more MultiMedia extension */
268int mm_support(void);
269
3d03c0a2 270#if defined(HAVE_MMX)
de6d9b64 271
18f77016 272#undef emms_c
eb4b3dd3 273
de6d9b64
FB
274#define MM_MMX 0x0001 /* standard MMX */
275#define MM_3DNOW 0x0004 /* AMD 3DNOW */
276#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
277#define MM_SSE 0x0008 /* SSE functions */
278#define MM_SSE2 0x0010 /* PIV SSE2 functions */
279
280extern int mm_flags;
281
0c1a9eda
ZK
282void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
283void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
de6d9b64
FB
284
285static inline void emms(void)
286{
fb16b7e7
FB
287 __asm __volatile ("emms;":::"memory");
288}
289
1457ab52 290
fb16b7e7
FB
291#define emms_c() \
292{\
293 if (mm_flags & MM_MMX)\
294 emms();\
de6d9b64
FB
295}
296
297#define __align8 __attribute__ ((aligned (8)))
298
b0368839
MN
299void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
300void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
2720569a 301
3d03c0a2
FB
302#elif defined(ARCH_ARMV4L)
303
3d03c0a2
FB
304/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
305 line ptimizations */
306#define __align8 __attribute__ ((aligned (4)))
307
b0368839 308void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
3d03c0a2 309
c34270f5 310#elif defined(HAVE_MLIB)
c34270f5
FB
311
312/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
313#define __align8 __attribute__ ((aligned (8)))
314
b0368839 315void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
c34270f5 316
1e98dffb
NK
317#elif defined(ARCH_ALPHA)
318
1e98dffb
NK
319#define __align8 __attribute__ ((aligned (8)))
320
b0368839 321void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
1e98dffb 322
59925ef2
BF
323#elif defined(ARCH_POWERPC)
324
404d2241
BF
325#define MM_ALTIVEC 0x0001 /* standard AltiVec */
326
327extern int mm_flags;
328
3b991c54
RD
329#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
330#include <altivec.h>
331#endif
332
59925ef2
BF
333#define __align8 __attribute__ ((aligned (16)))
334
b0368839 335void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
59925ef2 336
d46aba26
LS
337#elif defined(HAVE_MMI)
338
d46aba26
LS
339#define __align8 __attribute__ ((aligned (16)))
340
b0368839 341void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
d46aba26 342
de6d9b64
FB
343#else
344
345#define __align8
346
347#endif
348
6d4985bb
FB
349#ifdef __GNUC__
350
351struct unaligned_64 { uint64_t l; } __attribute__((packed));
352struct unaligned_32 { uint32_t l; } __attribute__((packed));
353
354#define LD32(a) (((const struct unaligned_32 *) (a))->l)
355#define LD64(a) (((const struct unaligned_64 *) (a))->l)
356
357#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
358
359#else /* __GNUC__ */
360
361#define LD32(a) (*((uint32_t*)(a)))
362#define LD64(a) (*((uint64_t*)(a)))
363
364#define ST32(a, b) *((uint32_t*)(a)) = (b)
365
366#endif /* !__GNUC__ */
367
43f1708f 368/* PSNR */
0c1a9eda 369void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
43f1708f
J
370 int orig_linesize[3], int coded_linesize,
371 AVCodecContext *avctx);
bb6f5690
FB
372
373/* FFT computation */
374
375/* NOTE: soon integer code will be added, so you must use the
376 FFTSample type */
377typedef float FFTSample;
378
379typedef struct FFTComplex {
380 FFTSample re, im;
381} FFTComplex;
382
383typedef struct FFTContext {
384 int nbits;
385 int inverse;
386 uint16_t *revtab;
387 FFTComplex *exptab;
388 FFTComplex *exptab1; /* only used by SSE code */
389 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
390} FFTContext;
391
392int fft_init(FFTContext *s, int nbits, int inverse);
393void fft_permute(FFTContext *s, FFTComplex *z);
394void fft_calc_c(FFTContext *s, FFTComplex *z);
395void fft_calc_sse(FFTContext *s, FFTComplex *z);
8d268a7d
FB
396void fft_calc_altivec(FFTContext *s, FFTComplex *z);
397
bb6f5690
FB
398static inline void fft_calc(FFTContext *s, FFTComplex *z)
399{
400 s->fft_calc(s, z);
401}
402void fft_end(FFTContext *s);
403
404/* MDCT computation */
405
406typedef struct MDCTContext {
407 int n; /* size of MDCT (i.e. number of input data * 2) */
408 int nbits; /* n = 2^nbits */
409 /* pre/post rotation tables */
410 FFTSample *tcos;
411 FFTSample *tsin;
412 FFTContext fft;
413} MDCTContext;
414
82696bee 415int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
eb4b3dd3 416void ff_imdct_calc(MDCTContext *s, FFTSample *output,
bb6f5690 417 const FFTSample *input, FFTSample *tmp);
eb4b3dd3 418void ff_mdct_calc(MDCTContext *s, FFTSample *out,
bb6f5690 419 const FFTSample *input, FFTSample *tmp);
82696bee 420void ff_mdct_end(MDCTContext *s);
bb6f5690 421
1457ab52
MN
422#define WARPER88_1616(name8, name16)\
423static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride){\
424 return name8(s, dst , src , stride)\
425 +name8(s, dst+8 , src+8 , stride)\
426 +name8(s, dst +8*stride, src +8*stride, stride)\
427 +name8(s, dst+8+8*stride, src+8+8*stride, stride);\
428}
429
95e2ce4a 430#ifndef HAVE_LRINTF
9d85cbd9
FB
431/* XXX: add ISOC specific test to avoid specific BSD testing. */
432/* better than nothing implementation. */
6234d753 433/* btw, rintf() is existing on fbsd too -- alex */
9d85cbd9
FB
434static inline long int lrintf(float x)
435{
ea937d01
FB
436#ifdef CONFIG_WIN32
437 /* XXX: incorrect, but make it compile */
438 return (int)(x);
439#else
9d85cbd9 440 return (int)(rint(x));
ea937d01 441#endif
9d85cbd9
FB
442}
443#endif
444
de6d9b64 445#endif