removed unneeded mallocs
[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 */
de6d9b64
FB
19#ifndef DSPUTIL_H
20#define DSPUTIL_H
21
22#include "common.h"
43f1708f 23#include "avcodec.h"
de6d9b64 24
44eb4951 25//#define DEBUG
de6d9b64
FB
26/* dct code */
27typedef short DCTELEM;
0e15384d 28//typedef int DCTELEM;
de6d9b64 29
03c94ede 30void fdct_ifast (DCTELEM *data);
28db7fce 31void ff_jpeg_fdct_islow (DCTELEM *data);
de6d9b64
FB
32
33void j_rev_dct (DCTELEM *data);
34
3f09f52a 35void ff_fdct_mmx(DCTELEM *block);
de6d9b64 36
e0eac44e 37/* encoding scans */
2ad1516a
MN
38extern const UINT8 ff_alternate_horizontal_scan[64];
39extern const UINT8 ff_alternate_vertical_scan[64];
40extern const UINT8 ff_zigzag_direct[64];
5a240838 41
de6d9b64
FB
42/* pixel operations */
43#define MAX_NEG_CROP 384
44
45/* temporary */
46extern UINT32 squareTbl[512];
0cfa9713 47extern UINT8 cropTbl[256 + 2 * MAX_NEG_CROP];
de6d9b64 48
de6d9b64 49
b7c27ee6 50/* minimum alignment rules ;)
eb4b3dd3 51if u notice errors in the align stuff, need more alignment for some asm code for some cpu
b7c27ee6
MN
52or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
53
54!warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
55i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
de6d9b64 56
b7c27ee6
MN
57!future video codecs might need functions with less strict alignment
58*/
59
eb4b3dd3 60/*
de6d9b64 61void get_pixels_c(DCTELEM *block, const UINT8 *pixels, int line_size);
9dbcbd92 62void diff_pixels_c(DCTELEM *block, const UINT8 *s1, const UINT8 *s2, int stride);
de6d9b64
FB
63void put_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size);
64void add_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size);
649c00c9 65void clear_blocks_c(DCTELEM *blocks);
eb4b3dd3 66*/
de6d9b64
FB
67
68/* add and put pixel (decoding) */
b7c27ee6
MN
69// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
70typedef void (*op_pixels_func)(UINT8 *block/*align width (8 or 16)*/, const UINT8 *pixels/*align 1*/, int line_size, int h);
71typedef void (*qpel_mc_func)(UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);
b3184779 72
db794953
MN
73#define DEF_OLD_QPEL(name)\
74void ff_put_ ## name (UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);\
75void ff_put_no_rnd_ ## name (UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);\
76void ff_avg_ ## name (UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);
77
78DEF_OLD_QPEL(qpel16_mc11_old_c)
79DEF_OLD_QPEL(qpel16_mc31_old_c)
80DEF_OLD_QPEL(qpel16_mc12_old_c)
81DEF_OLD_QPEL(qpel16_mc32_old_c)
82DEF_OLD_QPEL(qpel16_mc13_old_c)
83DEF_OLD_QPEL(qpel16_mc33_old_c)
84DEF_OLD_QPEL(qpel8_mc11_old_c)
85DEF_OLD_QPEL(qpel8_mc31_old_c)
86DEF_OLD_QPEL(qpel8_mc12_old_c)
87DEF_OLD_QPEL(qpel8_mc32_old_c)
88DEF_OLD_QPEL(qpel8_mc13_old_c)
89DEF_OLD_QPEL(qpel8_mc33_old_c)
b3184779
MN
90
91#define CALL_2X_PIXELS(a, b, n)\
92static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
93 b(block , pixels , line_size, h);\
94 b(block+n, pixels+n, line_size, h);\
95}
44eb4951 96
de6d9b64
FB
97/* motion estimation */
98
1457ab52
MN
99typedef int (*op_pixels_abs_func)(UINT8 *blk1/*align width (8 or 16)*/, UINT8 *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
100
101typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, UINT8 *blk1/*align width (8 or 16)*/, UINT8 *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
102
eb4b3dd3
ZK
103typedef struct DSPContext {
104 /* pixel ops : interface with DCT */
105 void (*get_pixels)(DCTELEM *block/*align 16*/, const UINT8 *pixels/*align 8*/, int line_size);
106 void (*diff_pixels)(DCTELEM *block/*align 16*/, const UINT8 *s1/*align 8*/, const UINT8 *s2/*align 8*/, int stride);
107 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
108 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
109 void (*gmc1)(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
110 void (*gmc )(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int stride, int h, int ox, int oy,
111 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
112 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
113 int (*pix_sum)(UINT8 * pix, int line_size);
114 int (*pix_norm1)(UINT8 * pix, int line_size);
1457ab52
MN
115 me_cmp_func sad[2]; /* identical to pix_absAxA except additional void * */
116 me_cmp_func sse[2];
117 me_cmp_func hadamard8_diff[2];
118 me_cmp_func dct_sad[2];
119 me_cmp_func quant_psnr[2];
3a87ac94
MN
120 me_cmp_func bit[2];
121 me_cmp_func rd[2];
1457ab52
MN
122 int (*hadamard8_abs )(uint8_t *src, int stride, int mean);
123
826f429a 124 me_cmp_func me_pre_cmp[11];
1457ab52
MN
125 me_cmp_func me_cmp[11];
126 me_cmp_func me_sub_cmp[11];
127 me_cmp_func mb_cmp[11];
eb4b3dd3
ZK
128
129 /* maybe create an array for 16/8 functions */
130 op_pixels_func put_pixels_tab[2][4];
131 op_pixels_func avg_pixels_tab[2][4];
132 op_pixels_func put_no_rnd_pixels_tab[2][4];
133 op_pixels_func avg_no_rnd_pixels_tab[2][4];
134 qpel_mc_func put_qpel_pixels_tab[2][16];
135 qpel_mc_func avg_qpel_pixels_tab[2][16];
136 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
137 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
1457ab52 138 qpel_mc_func put_mspel_pixels_tab[8];
eb4b3dd3
ZK
139
140 op_pixels_abs_func pix_abs16x16;
141 op_pixels_abs_func pix_abs16x16_x2;
142 op_pixels_abs_func pix_abs16x16_y2;
143 op_pixels_abs_func pix_abs16x16_xy2;
144 op_pixels_abs_func pix_abs8x8;
145 op_pixels_abs_func pix_abs8x8_x2;
146 op_pixels_abs_func pix_abs8x8_y2;
147 op_pixels_abs_func pix_abs8x8_xy2;
11f18faf
MN
148
149 /* huffyuv specific */
11f18faf 150 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
1457ab52 151 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
eb4b3dd3
ZK
152} DSPContext;
153
154void dsputil_init(DSPContext* p, unsigned mask);
de6d9b64 155
7801d21d
MN
156/**
157 * permute block according to permuatation.
158 * @param last last non zero element in scantable order
159 */
0e15384d 160void ff_block_permute(DCTELEM *block, UINT8 *permutation, const UINT8 *scantable, int last);
e0eac44e 161
eb4b3dd3
ZK
162#define emms_c()
163
e629ab68
RD
164/* should be defined by architectures supporting
165 one or more MultiMedia extension */
166int mm_support(void);
167
3d03c0a2 168#if defined(HAVE_MMX)
de6d9b64 169
18f77016 170#undef emms_c
eb4b3dd3 171
de6d9b64
FB
172#define MM_MMX 0x0001 /* standard MMX */
173#define MM_3DNOW 0x0004 /* AMD 3DNOW */
174#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
175#define MM_SSE 0x0008 /* SSE functions */
176#define MM_SSE2 0x0010 /* PIV SSE2 functions */
177
178extern int mm_flags;
179
eb4b3dd3
ZK
180void add_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size);
181void put_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size);
de6d9b64
FB
182
183static inline void emms(void)
184{
fb16b7e7
FB
185 __asm __volatile ("emms;":::"memory");
186}
187
1457ab52 188
fb16b7e7
FB
189#define emms_c() \
190{\
191 if (mm_flags & MM_MMX)\
192 emms();\
de6d9b64
FB
193}
194
195#define __align8 __attribute__ ((aligned (8)))
196
eb4b3dd3
ZK
197void dsputil_init_mmx(DSPContext* c, unsigned mask);
198void dsputil_set_bit_exact_mmx(DSPContext* c, unsigned mask);
de6d9b64 199
3d03c0a2
FB
200#elif defined(ARCH_ARMV4L)
201
3d03c0a2
FB
202/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
203 line ptimizations */
204#define __align8 __attribute__ ((aligned (4)))
205
eb4b3dd3 206void dsputil_init_armv4l(DSPContext* c, unsigned mask);
3d03c0a2 207
c34270f5 208#elif defined(HAVE_MLIB)
c34270f5
FB
209
210/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
211#define __align8 __attribute__ ((aligned (8)))
212
eb4b3dd3 213void dsputil_init_mlib(DSPContext* c, unsigned mask);
c34270f5 214
1e98dffb
NK
215#elif defined(ARCH_ALPHA)
216
1e98dffb
NK
217#define __align8 __attribute__ ((aligned (8)))
218
eb4b3dd3 219void dsputil_init_alpha(DSPContext* c, unsigned mask);
1e98dffb 220
59925ef2
BF
221#elif defined(ARCH_POWERPC)
222
404d2241
BF
223#define MM_ALTIVEC 0x0001 /* standard AltiVec */
224
225extern int mm_flags;
226
59925ef2
BF
227#define __align8 __attribute__ ((aligned (16)))
228
eb4b3dd3 229void dsputil_init_ppc(DSPContext* c, unsigned mask);
59925ef2 230
d46aba26
LS
231#elif defined(HAVE_MMI)
232
d46aba26
LS
233#define __align8 __attribute__ ((aligned (16)))
234
eb4b3dd3 235void dsputil_init_mmi(DSPContext* c, unsigned mask);
d46aba26 236
de6d9b64
FB
237#else
238
239#define __align8
240
241#endif
242
6d4985bb
FB
243#ifdef __GNUC__
244
245struct unaligned_64 { uint64_t l; } __attribute__((packed));
246struct unaligned_32 { uint32_t l; } __attribute__((packed));
247
248#define LD32(a) (((const struct unaligned_32 *) (a))->l)
249#define LD64(a) (((const struct unaligned_64 *) (a))->l)
250
251#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
252
253#else /* __GNUC__ */
254
255#define LD32(a) (*((uint32_t*)(a)))
256#define LD64(a) (*((uint64_t*)(a)))
257
258#define ST32(a, b) *((uint32_t*)(a)) = (b)
259
260#endif /* !__GNUC__ */
261
43f1708f
J
262/* PSNR */
263void get_psnr(UINT8 *orig_image[3], UINT8 *coded_image[3],
264 int orig_linesize[3], int coded_linesize,
265 AVCodecContext *avctx);
bb6f5690
FB
266
267/* FFT computation */
268
269/* NOTE: soon integer code will be added, so you must use the
270 FFTSample type */
271typedef float FFTSample;
272
273typedef struct FFTComplex {
274 FFTSample re, im;
275} FFTComplex;
276
277typedef struct FFTContext {
278 int nbits;
279 int inverse;
280 uint16_t *revtab;
281 FFTComplex *exptab;
282 FFTComplex *exptab1; /* only used by SSE code */
283 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
284} FFTContext;
285
286int fft_init(FFTContext *s, int nbits, int inverse);
287void fft_permute(FFTContext *s, FFTComplex *z);
288void fft_calc_c(FFTContext *s, FFTComplex *z);
289void fft_calc_sse(FFTContext *s, FFTComplex *z);
8d268a7d
FB
290void fft_calc_altivec(FFTContext *s, FFTComplex *z);
291
bb6f5690
FB
292static inline void fft_calc(FFTContext *s, FFTComplex *z)
293{
294 s->fft_calc(s, z);
295}
296void fft_end(FFTContext *s);
297
298/* MDCT computation */
299
300typedef struct MDCTContext {
301 int n; /* size of MDCT (i.e. number of input data * 2) */
302 int nbits; /* n = 2^nbits */
303 /* pre/post rotation tables */
304 FFTSample *tcos;
305 FFTSample *tsin;
306 FFTContext fft;
307} MDCTContext;
308
82696bee 309int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
eb4b3dd3 310void ff_imdct_calc(MDCTContext *s, FFTSample *output,
bb6f5690 311 const FFTSample *input, FFTSample *tmp);
eb4b3dd3 312void ff_mdct_calc(MDCTContext *s, FFTSample *out,
bb6f5690 313 const FFTSample *input, FFTSample *tmp);
82696bee 314void ff_mdct_end(MDCTContext *s);
bb6f5690 315
1457ab52
MN
316#define WARPER88_1616(name8, name16)\
317static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride){\
318 return name8(s, dst , src , stride)\
319 +name8(s, dst+8 , src+8 , stride)\
320 +name8(s, dst +8*stride, src +8*stride, stride)\
321 +name8(s, dst+8+8*stride, src+8+8*stride, stride);\
322}
323
95e2ce4a 324#ifndef HAVE_LRINTF
9d85cbd9
FB
325/* XXX: add ISOC specific test to avoid specific BSD testing. */
326/* better than nothing implementation. */
6234d753 327/* btw, rintf() is existing on fbsd too -- alex */
9d85cbd9
FB
328static inline long int lrintf(float x)
329{
330 return (int)(rint(x));
331}
332#endif
333
de6d9b64 334#endif