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