fixed wmv2 slices
[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;
28
03c94ede 29void fdct_ifast (DCTELEM *data);
28db7fce 30void ff_jpeg_fdct_islow (DCTELEM *data);
de6d9b64
FB
31
32void j_rev_dct (DCTELEM *data);
33
3f09f52a 34void ff_fdct_mmx(DCTELEM *block);
de6d9b64 35
e0eac44e 36/* encoding scans */
2ad1516a
MN
37extern const UINT8 ff_alternate_horizontal_scan[64];
38extern const UINT8 ff_alternate_vertical_scan[64];
39extern const UINT8 ff_zigzag_direct[64];
5a240838 40
de6d9b64
FB
41/* pixel operations */
42#define MAX_NEG_CROP 384
43
44/* temporary */
45extern UINT32 squareTbl[512];
0cfa9713 46extern UINT8 cropTbl[256 + 2 * MAX_NEG_CROP];
de6d9b64 47
de6d9b64 48
b7c27ee6 49/* minimum alignment rules ;)
eb4b3dd3 50if u notice errors in the align stuff, need more alignment for some asm code for some cpu
b7c27ee6
MN
51or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
52
53!warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
54i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
de6d9b64 55
b7c27ee6
MN
56!future video codecs might need functions with less strict alignment
57*/
58
eb4b3dd3 59/*
de6d9b64 60void get_pixels_c(DCTELEM *block, const UINT8 *pixels, int line_size);
9dbcbd92 61void diff_pixels_c(DCTELEM *block, const UINT8 *s1, const UINT8 *s2, int stride);
de6d9b64
FB
62void put_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size);
63void add_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size);
649c00c9 64void clear_blocks_c(DCTELEM *blocks);
eb4b3dd3 65*/
de6d9b64
FB
66
67/* add and put pixel (decoding) */
b7c27ee6
MN
68// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
69typedef void (*op_pixels_func)(UINT8 *block/*align width (8 or 16)*/, const UINT8 *pixels/*align 1*/, int line_size, int h);
70typedef void (*qpel_mc_func)(UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);
b3184779 71
eb4b3dd3 72
b3184779
MN
73
74#define CALL_2X_PIXELS(a, b, n)\
75static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
76 b(block , pixels , line_size, h);\
77 b(block+n, pixels+n, line_size, h);\
78}
44eb4951 79
de6d9b64
FB
80/* motion estimation */
81
1457ab52
MN
82typedef int (*op_pixels_abs_func)(UINT8 *blk1/*align width (8 or 16)*/, UINT8 *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
83
84typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, UINT8 *blk1/*align width (8 or 16)*/, UINT8 *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
85
eb4b3dd3
ZK
86typedef struct DSPContext {
87 /* pixel ops : interface with DCT */
88 void (*get_pixels)(DCTELEM *block/*align 16*/, const UINT8 *pixels/*align 8*/, int line_size);
89 void (*diff_pixels)(DCTELEM *block/*align 16*/, const UINT8 *s1/*align 8*/, const UINT8 *s2/*align 8*/, int stride);
90 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
91 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
92 void (*gmc1)(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
93 void (*gmc )(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int stride, int h, int ox, int oy,
94 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
95 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
96 int (*pix_sum)(UINT8 * pix, int line_size);
97 int (*pix_norm1)(UINT8 * pix, int line_size);
1457ab52
MN
98 me_cmp_func sad[2]; /* identical to pix_absAxA except additional void * */
99 me_cmp_func sse[2];
100 me_cmp_func hadamard8_diff[2];
101 me_cmp_func dct_sad[2];
102 me_cmp_func quant_psnr[2];
103 int (*hadamard8_abs )(uint8_t *src, int stride, int mean);
104
826f429a 105 me_cmp_func me_pre_cmp[11];
1457ab52
MN
106 me_cmp_func me_cmp[11];
107 me_cmp_func me_sub_cmp[11];
108 me_cmp_func mb_cmp[11];
eb4b3dd3
ZK
109
110 /* maybe create an array for 16/8 functions */
111 op_pixels_func put_pixels_tab[2][4];
112 op_pixels_func avg_pixels_tab[2][4];
113 op_pixels_func put_no_rnd_pixels_tab[2][4];
114 op_pixels_func avg_no_rnd_pixels_tab[2][4];
115 qpel_mc_func put_qpel_pixels_tab[2][16];
116 qpel_mc_func avg_qpel_pixels_tab[2][16];
117 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
118 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
1457ab52 119 qpel_mc_func put_mspel_pixels_tab[8];
eb4b3dd3
ZK
120
121 op_pixels_abs_func pix_abs16x16;
122 op_pixels_abs_func pix_abs16x16_x2;
123 op_pixels_abs_func pix_abs16x16_y2;
124 op_pixels_abs_func pix_abs16x16_xy2;
125 op_pixels_abs_func pix_abs8x8;
126 op_pixels_abs_func pix_abs8x8_x2;
127 op_pixels_abs_func pix_abs8x8_y2;
128 op_pixels_abs_func pix_abs8x8_xy2;
11f18faf
MN
129
130 /* huffyuv specific */
11f18faf 131 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
1457ab52 132 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
eb4b3dd3
ZK
133} DSPContext;
134
135void dsputil_init(DSPContext* p, unsigned mask);
de6d9b64 136
7801d21d
MN
137/**
138 * permute block according to permuatation.
139 * @param last last non zero element in scantable order
140 */
141void ff_block_permute(INT16 *block, UINT8 *permutation, const UINT8 *scantable, int last);
e0eac44e 142
eb4b3dd3
ZK
143#define emms_c()
144
3d03c0a2 145#if defined(HAVE_MMX)
de6d9b64 146
18f77016 147#undef emms_c
eb4b3dd3 148
de6d9b64
FB
149#define MM_MMX 0x0001 /* standard MMX */
150#define MM_3DNOW 0x0004 /* AMD 3DNOW */
151#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
152#define MM_SSE 0x0008 /* SSE functions */
153#define MM_SSE2 0x0010 /* PIV SSE2 functions */
154
155extern int mm_flags;
156
157int mm_support(void);
eb4b3dd3
ZK
158void add_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size);
159void put_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size);
de6d9b64
FB
160
161static inline void emms(void)
162{
fb16b7e7
FB
163 __asm __volatile ("emms;":::"memory");
164}
165
1457ab52 166
fb16b7e7
FB
167#define emms_c() \
168{\
169 if (mm_flags & MM_MMX)\
170 emms();\
de6d9b64
FB
171}
172
173#define __align8 __attribute__ ((aligned (8)))
174
eb4b3dd3
ZK
175void dsputil_init_mmx(DSPContext* c, unsigned mask);
176void dsputil_set_bit_exact_mmx(DSPContext* c, unsigned mask);
de6d9b64 177
3d03c0a2
FB
178#elif defined(ARCH_ARMV4L)
179
3d03c0a2
FB
180/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
181 line ptimizations */
182#define __align8 __attribute__ ((aligned (4)))
183
eb4b3dd3 184void dsputil_init_armv4l(DSPContext* c, unsigned mask);
3d03c0a2 185
c34270f5 186#elif defined(HAVE_MLIB)
c34270f5
FB
187
188/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
189#define __align8 __attribute__ ((aligned (8)))
190
eb4b3dd3 191void dsputil_init_mlib(DSPContext* c, unsigned mask);
c34270f5 192
1e98dffb
NK
193#elif defined(ARCH_ALPHA)
194
1e98dffb
NK
195#define __align8 __attribute__ ((aligned (8)))
196
eb4b3dd3 197void dsputil_init_alpha(DSPContext* c, unsigned mask);
1e98dffb 198
59925ef2
BF
199#elif defined(ARCH_POWERPC)
200
404d2241
BF
201#define MM_ALTIVEC 0x0001 /* standard AltiVec */
202
203extern int mm_flags;
204
59925ef2
BF
205#define __align8 __attribute__ ((aligned (16)))
206
eb4b3dd3 207void dsputil_init_ppc(DSPContext* c, unsigned mask);
59925ef2 208
d46aba26
LS
209#elif defined(HAVE_MMI)
210
d46aba26
LS
211#define __align8 __attribute__ ((aligned (16)))
212
eb4b3dd3 213void dsputil_init_mmi(DSPContext* c, unsigned mask);
d46aba26 214
de6d9b64
FB
215#else
216
217#define __align8
218
219#endif
220
6d4985bb
FB
221#ifdef __GNUC__
222
223struct unaligned_64 { uint64_t l; } __attribute__((packed));
224struct unaligned_32 { uint32_t l; } __attribute__((packed));
225
226#define LD32(a) (((const struct unaligned_32 *) (a))->l)
227#define LD64(a) (((const struct unaligned_64 *) (a))->l)
228
229#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
230
231#else /* __GNUC__ */
232
233#define LD32(a) (*((uint32_t*)(a)))
234#define LD64(a) (*((uint64_t*)(a)))
235
236#define ST32(a, b) *((uint32_t*)(a)) = (b)
237
238#endif /* !__GNUC__ */
239
43f1708f
J
240/* PSNR */
241void get_psnr(UINT8 *orig_image[3], UINT8 *coded_image[3],
242 int orig_linesize[3], int coded_linesize,
243 AVCodecContext *avctx);
bb6f5690
FB
244
245/* FFT computation */
246
247/* NOTE: soon integer code will be added, so you must use the
248 FFTSample type */
249typedef float FFTSample;
250
251typedef struct FFTComplex {
252 FFTSample re, im;
253} FFTComplex;
254
255typedef struct FFTContext {
256 int nbits;
257 int inverse;
258 uint16_t *revtab;
259 FFTComplex *exptab;
260 FFTComplex *exptab1; /* only used by SSE code */
261 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
262} FFTContext;
263
264int fft_init(FFTContext *s, int nbits, int inverse);
265void fft_permute(FFTContext *s, FFTComplex *z);
266void fft_calc_c(FFTContext *s, FFTComplex *z);
267void fft_calc_sse(FFTContext *s, FFTComplex *z);
8d268a7d
FB
268void fft_calc_altivec(FFTContext *s, FFTComplex *z);
269
bb6f5690
FB
270static inline void fft_calc(FFTContext *s, FFTComplex *z)
271{
272 s->fft_calc(s, z);
273}
274void fft_end(FFTContext *s);
275
276/* MDCT computation */
277
278typedef struct MDCTContext {
279 int n; /* size of MDCT (i.e. number of input data * 2) */
280 int nbits; /* n = 2^nbits */
281 /* pre/post rotation tables */
282 FFTSample *tcos;
283 FFTSample *tsin;
284 FFTContext fft;
285} MDCTContext;
286
82696bee 287int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
eb4b3dd3 288void ff_imdct_calc(MDCTContext *s, FFTSample *output,
bb6f5690 289 const FFTSample *input, FFTSample *tmp);
eb4b3dd3 290void ff_mdct_calc(MDCTContext *s, FFTSample *out,
bb6f5690 291 const FFTSample *input, FFTSample *tmp);
82696bee 292void ff_mdct_end(MDCTContext *s);
bb6f5690 293
1457ab52
MN
294#define WARPER88_1616(name8, name16)\
295static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride){\
296 return name8(s, dst , src , stride)\
297 +name8(s, dst+8 , src+8 , stride)\
298 +name8(s, dst +8*stride, src +8*stride, stride)\
299 +name8(s, dst+8+8*stride, src+8+8*stride, stride);\
300}
301
95e2ce4a 302#ifndef HAVE_LRINTF
9d85cbd9
FB
303/* XXX: add ISOC specific test to avoid specific BSD testing. */
304/* better than nothing implementation. */
6234d753 305/* btw, rintf() is existing on fbsd too -- alex */
9d85cbd9
FB
306static inline long int lrintf(float x)
307{
308 return (int)(rint(x));
309}
310#endif
311
de6d9b64 312#endif