10l
[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);
b3184779 79
db794953 80#define DEF_OLD_QPEL(name)\
0c1a9eda
ZK
81void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
82void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
83void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
db794953
MN
84
85DEF_OLD_QPEL(qpel16_mc11_old_c)
86DEF_OLD_QPEL(qpel16_mc31_old_c)
87DEF_OLD_QPEL(qpel16_mc12_old_c)
88DEF_OLD_QPEL(qpel16_mc32_old_c)
89DEF_OLD_QPEL(qpel16_mc13_old_c)
90DEF_OLD_QPEL(qpel16_mc33_old_c)
91DEF_OLD_QPEL(qpel8_mc11_old_c)
92DEF_OLD_QPEL(qpel8_mc31_old_c)
93DEF_OLD_QPEL(qpel8_mc12_old_c)
94DEF_OLD_QPEL(qpel8_mc32_old_c)
95DEF_OLD_QPEL(qpel8_mc13_old_c)
96DEF_OLD_QPEL(qpel8_mc33_old_c)
b3184779
MN
97
98#define CALL_2X_PIXELS(a, b, n)\
99static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
100 b(block , pixels , line_size, h);\
101 b(block+n, pixels+n, line_size, h);\
102}
44eb4951 103
de6d9b64
FB
104/* motion estimation */
105
0c1a9eda 106typedef int (*op_pixels_abs_func)(uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
1457ab52 107
0c1a9eda 108typedef 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 109
24641185
MN
110/**
111 * DSPContext.
112 */
eb4b3dd3
ZK
113typedef struct DSPContext {
114 /* pixel ops : interface with DCT */
0c1a9eda
ZK
115 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
116 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
117 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
118 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
119 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
120 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
eb4b3dd3
ZK
121 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
122 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
0c1a9eda
ZK
123 int (*pix_sum)(uint8_t * pix, int line_size);
124 int (*pix_norm1)(uint8_t * pix, int line_size);
1457ab52
MN
125 me_cmp_func sad[2]; /* identical to pix_absAxA except additional void * */
126 me_cmp_func sse[2];
127 me_cmp_func hadamard8_diff[2];
128 me_cmp_func dct_sad[2];
129 me_cmp_func quant_psnr[2];
3a87ac94
MN
130 me_cmp_func bit[2];
131 me_cmp_func rd[2];
1457ab52
MN
132 int (*hadamard8_abs )(uint8_t *src, int stride, int mean);
133
826f429a 134 me_cmp_func me_pre_cmp[11];
1457ab52
MN
135 me_cmp_func me_cmp[11];
136 me_cmp_func me_sub_cmp[11];
137 me_cmp_func mb_cmp[11];
eb4b3dd3
ZK
138
139 /* maybe create an array for 16/8 functions */
140 op_pixels_func put_pixels_tab[2][4];
141 op_pixels_func avg_pixels_tab[2][4];
142 op_pixels_func put_no_rnd_pixels_tab[2][4];
143 op_pixels_func avg_no_rnd_pixels_tab[2][4];
144 qpel_mc_func put_qpel_pixels_tab[2][16];
145 qpel_mc_func avg_qpel_pixels_tab[2][16];
146 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
147 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
1457ab52 148 qpel_mc_func put_mspel_pixels_tab[8];
eb4b3dd3
ZK
149
150 op_pixels_abs_func pix_abs16x16;
151 op_pixels_abs_func pix_abs16x16_x2;
152 op_pixels_abs_func pix_abs16x16_y2;
153 op_pixels_abs_func pix_abs16x16_xy2;
154 op_pixels_abs_func pix_abs8x8;
155 op_pixels_abs_func pix_abs8x8_x2;
156 op_pixels_abs_func pix_abs8x8_y2;
157 op_pixels_abs_func pix_abs8x8_xy2;
11f18faf
MN
158
159 /* huffyuv specific */
11f18faf 160 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
1457ab52 161 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
b0368839
MN
162
163 /* (I)DCT */
164 void (*fdct)(DCTELEM *block/* align 16*/);
24641185
MN
165
166 /**
77c92c2d 167 * block -> idct -> clip to unsigned 8 bit -> dest.
24641185 168 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
a1adf436 169 * @param line_size size in bytes of a horizotal line of dest
24641185 170 */
b0368839 171 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
24641185
MN
172
173 /**
174 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
a1adf436 175 * @param line_size size in bytes of a horizotal line of dest
24641185 176 */
b0368839 177 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
24641185
MN
178
179 /**
77c92c2d 180 * idct input permutation.
24641185
MN
181 * an example to avoid confusion:
182 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
183 * - (x -> referece dct -> reference idct -> x)
184 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
185 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
186 */
b0368839
MN
187 uint8_t idct_permutation[64];
188 int idct_permutation_type;
189#define FF_NO_IDCT_PERM 1
190#define FF_LIBMPEG2_IDCT_PERM 2
191#define FF_SIMPLE_IDCT_PERM 3
192#define FF_TRANSPOSE_IDCT_PERM 4
193
eb4b3dd3
ZK
194} DSPContext;
195
b0368839 196void dsputil_init(DSPContext* p, AVCodecContext *avctx);
de6d9b64 197
7801d21d
MN
198/**
199 * permute block according to permuatation.
200 * @param last last non zero element in scantable order
201 */
0c1a9eda 202void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
e0eac44e 203
24641185 204/**
77c92c2d 205 * Empty mmx state.
24641185
MN
206 * this must be called between any dsp function and float/double code.
207 * for example sin(); dsp->idct_put(); emms_c(); cos()
208 */
eb4b3dd3
ZK
209#define emms_c()
210
e629ab68
RD
211/* should be defined by architectures supporting
212 one or more MultiMedia extension */
213int mm_support(void);
214
3d03c0a2 215#if defined(HAVE_MMX)
de6d9b64 216
18f77016 217#undef emms_c
eb4b3dd3 218
de6d9b64
FB
219#define MM_MMX 0x0001 /* standard MMX */
220#define MM_3DNOW 0x0004 /* AMD 3DNOW */
221#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
222#define MM_SSE 0x0008 /* SSE functions */
223#define MM_SSE2 0x0010 /* PIV SSE2 functions */
224
225extern int mm_flags;
226
0c1a9eda
ZK
227void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
228void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
de6d9b64
FB
229
230static inline void emms(void)
231{
fb16b7e7
FB
232 __asm __volatile ("emms;":::"memory");
233}
234
1457ab52 235
fb16b7e7
FB
236#define emms_c() \
237{\
238 if (mm_flags & MM_MMX)\
239 emms();\
de6d9b64
FB
240}
241
242#define __align8 __attribute__ ((aligned (8)))
243
b0368839
MN
244void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
245void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
2720569a 246
3d03c0a2
FB
247#elif defined(ARCH_ARMV4L)
248
3d03c0a2
FB
249/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
250 line ptimizations */
251#define __align8 __attribute__ ((aligned (4)))
252
b0368839 253void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
3d03c0a2 254
c34270f5 255#elif defined(HAVE_MLIB)
c34270f5
FB
256
257/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
258#define __align8 __attribute__ ((aligned (8)))
259
b0368839 260void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
c34270f5 261
1e98dffb
NK
262#elif defined(ARCH_ALPHA)
263
1e98dffb
NK
264#define __align8 __attribute__ ((aligned (8)))
265
b0368839 266void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
1e98dffb 267
59925ef2
BF
268#elif defined(ARCH_POWERPC)
269
404d2241
BF
270#define MM_ALTIVEC 0x0001 /* standard AltiVec */
271
272extern int mm_flags;
273
3b991c54
RD
274#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
275#include <altivec.h>
276#endif
277
59925ef2
BF
278#define __align8 __attribute__ ((aligned (16)))
279
b0368839 280void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
59925ef2 281
d46aba26
LS
282#elif defined(HAVE_MMI)
283
d46aba26
LS
284#define __align8 __attribute__ ((aligned (16)))
285
b0368839 286void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
d46aba26 287
de6d9b64
FB
288#else
289
290#define __align8
291
292#endif
293
6d4985bb
FB
294#ifdef __GNUC__
295
296struct unaligned_64 { uint64_t l; } __attribute__((packed));
297struct unaligned_32 { uint32_t l; } __attribute__((packed));
298
299#define LD32(a) (((const struct unaligned_32 *) (a))->l)
300#define LD64(a) (((const struct unaligned_64 *) (a))->l)
301
302#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
303
304#else /* __GNUC__ */
305
306#define LD32(a) (*((uint32_t*)(a)))
307#define LD64(a) (*((uint64_t*)(a)))
308
309#define ST32(a, b) *((uint32_t*)(a)) = (b)
310
311#endif /* !__GNUC__ */
312
43f1708f 313/* PSNR */
0c1a9eda 314void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
43f1708f
J
315 int orig_linesize[3], int coded_linesize,
316 AVCodecContext *avctx);
bb6f5690
FB
317
318/* FFT computation */
319
320/* NOTE: soon integer code will be added, so you must use the
321 FFTSample type */
322typedef float FFTSample;
323
324typedef struct FFTComplex {
325 FFTSample re, im;
326} FFTComplex;
327
328typedef struct FFTContext {
329 int nbits;
330 int inverse;
331 uint16_t *revtab;
332 FFTComplex *exptab;
333 FFTComplex *exptab1; /* only used by SSE code */
334 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
335} FFTContext;
336
337int fft_init(FFTContext *s, int nbits, int inverse);
338void fft_permute(FFTContext *s, FFTComplex *z);
339void fft_calc_c(FFTContext *s, FFTComplex *z);
340void fft_calc_sse(FFTContext *s, FFTComplex *z);
8d268a7d
FB
341void fft_calc_altivec(FFTContext *s, FFTComplex *z);
342
bb6f5690
FB
343static inline void fft_calc(FFTContext *s, FFTComplex *z)
344{
345 s->fft_calc(s, z);
346}
347void fft_end(FFTContext *s);
348
349/* MDCT computation */
350
351typedef struct MDCTContext {
352 int n; /* size of MDCT (i.e. number of input data * 2) */
353 int nbits; /* n = 2^nbits */
354 /* pre/post rotation tables */
355 FFTSample *tcos;
356 FFTSample *tsin;
357 FFTContext fft;
358} MDCTContext;
359
82696bee 360int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
eb4b3dd3 361void ff_imdct_calc(MDCTContext *s, FFTSample *output,
bb6f5690 362 const FFTSample *input, FFTSample *tmp);
eb4b3dd3 363void ff_mdct_calc(MDCTContext *s, FFTSample *out,
bb6f5690 364 const FFTSample *input, FFTSample *tmp);
82696bee 365void ff_mdct_end(MDCTContext *s);
bb6f5690 366
1457ab52
MN
367#define WARPER88_1616(name8, name16)\
368static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride){\
369 return name8(s, dst , src , stride)\
370 +name8(s, dst+8 , src+8 , stride)\
371 +name8(s, dst +8*stride, src +8*stride, stride)\
372 +name8(s, dst+8+8*stride, src+8+8*stride, stride);\
373}
374
95e2ce4a 375#ifndef HAVE_LRINTF
9d85cbd9
FB
376/* XXX: add ISOC specific test to avoid specific BSD testing. */
377/* better than nothing implementation. */
6234d753 378/* btw, rintf() is existing on fbsd too -- alex */
9d85cbd9
FB
379static inline long int lrintf(float x)
380{
ea937d01
FB
381#ifdef CONFIG_WIN32
382 /* XXX: incorrect, but make it compile */
383 return (int)(x);
384#else
9d85cbd9 385 return (int)(rint(x));
ea937d01 386#endif
9d85cbd9
FB
387}
388#endif
389
de6d9b64 390#endif