os2 support patch by ("Slavik Gnatenko" <miracle9 at newmail dot ru>)
[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
db794953
MN
72#define DEF_OLD_QPEL(name)\
73void ff_put_ ## name (UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);\
74void ff_put_no_rnd_ ## name (UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);\
75void ff_avg_ ## name (UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);
76
77DEF_OLD_QPEL(qpel16_mc11_old_c)
78DEF_OLD_QPEL(qpel16_mc31_old_c)
79DEF_OLD_QPEL(qpel16_mc12_old_c)
80DEF_OLD_QPEL(qpel16_mc32_old_c)
81DEF_OLD_QPEL(qpel16_mc13_old_c)
82DEF_OLD_QPEL(qpel16_mc33_old_c)
83DEF_OLD_QPEL(qpel8_mc11_old_c)
84DEF_OLD_QPEL(qpel8_mc31_old_c)
85DEF_OLD_QPEL(qpel8_mc12_old_c)
86DEF_OLD_QPEL(qpel8_mc32_old_c)
87DEF_OLD_QPEL(qpel8_mc13_old_c)
88DEF_OLD_QPEL(qpel8_mc33_old_c)
b3184779
MN
89
90#define CALL_2X_PIXELS(a, b, n)\
91static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
92 b(block , pixels , line_size, h);\
93 b(block+n, pixels+n, line_size, h);\
94}
44eb4951 95
de6d9b64
FB
96/* motion estimation */
97
1457ab52
MN
98typedef int (*op_pixels_abs_func)(UINT8 *blk1/*align width (8 or 16)*/, UINT8 *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
99
100typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, UINT8 *blk1/*align width (8 or 16)*/, UINT8 *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
101
eb4b3dd3
ZK
102typedef struct DSPContext {
103 /* pixel ops : interface with DCT */
104 void (*get_pixels)(DCTELEM *block/*align 16*/, const UINT8 *pixels/*align 8*/, int line_size);
105 void (*diff_pixels)(DCTELEM *block/*align 16*/, const UINT8 *s1/*align 8*/, const UINT8 *s2/*align 8*/, int stride);
106 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
107 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
108 void (*gmc1)(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
109 void (*gmc )(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int stride, int h, int ox, int oy,
110 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
111 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
112 int (*pix_sum)(UINT8 * pix, int line_size);
113 int (*pix_norm1)(UINT8 * pix, int line_size);
1457ab52
MN
114 me_cmp_func sad[2]; /* identical to pix_absAxA except additional void * */
115 me_cmp_func sse[2];
116 me_cmp_func hadamard8_diff[2];
117 me_cmp_func dct_sad[2];
118 me_cmp_func quant_psnr[2];
119 int (*hadamard8_abs )(uint8_t *src, int stride, int mean);
120
826f429a 121 me_cmp_func me_pre_cmp[11];
1457ab52
MN
122 me_cmp_func me_cmp[11];
123 me_cmp_func me_sub_cmp[11];
124 me_cmp_func mb_cmp[11];
eb4b3dd3
ZK
125
126 /* maybe create an array for 16/8 functions */
127 op_pixels_func put_pixels_tab[2][4];
128 op_pixels_func avg_pixels_tab[2][4];
129 op_pixels_func put_no_rnd_pixels_tab[2][4];
130 op_pixels_func avg_no_rnd_pixels_tab[2][4];
131 qpel_mc_func put_qpel_pixels_tab[2][16];
132 qpel_mc_func avg_qpel_pixels_tab[2][16];
133 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
134 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
1457ab52 135 qpel_mc_func put_mspel_pixels_tab[8];
eb4b3dd3
ZK
136
137 op_pixels_abs_func pix_abs16x16;
138 op_pixels_abs_func pix_abs16x16_x2;
139 op_pixels_abs_func pix_abs16x16_y2;
140 op_pixels_abs_func pix_abs16x16_xy2;
141 op_pixels_abs_func pix_abs8x8;
142 op_pixels_abs_func pix_abs8x8_x2;
143 op_pixels_abs_func pix_abs8x8_y2;
144 op_pixels_abs_func pix_abs8x8_xy2;
11f18faf
MN
145
146 /* huffyuv specific */
11f18faf 147 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
1457ab52 148 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
eb4b3dd3
ZK
149} DSPContext;
150
151void dsputil_init(DSPContext* p, unsigned mask);
de6d9b64 152
7801d21d
MN
153/**
154 * permute block according to permuatation.
155 * @param last last non zero element in scantable order
156 */
157void ff_block_permute(INT16 *block, UINT8 *permutation, const UINT8 *scantable, int last);
e0eac44e 158
eb4b3dd3
ZK
159#define emms_c()
160
3d03c0a2 161#if defined(HAVE_MMX)
de6d9b64 162
18f77016 163#undef emms_c
eb4b3dd3 164
de6d9b64
FB
165#define MM_MMX 0x0001 /* standard MMX */
166#define MM_3DNOW 0x0004 /* AMD 3DNOW */
167#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
168#define MM_SSE 0x0008 /* SSE functions */
169#define MM_SSE2 0x0010 /* PIV SSE2 functions */
170
171extern int mm_flags;
172
173int mm_support(void);
eb4b3dd3
ZK
174void add_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size);
175void put_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size);
de6d9b64
FB
176
177static inline void emms(void)
178{
fb16b7e7
FB
179 __asm __volatile ("emms;":::"memory");
180}
181
1457ab52 182
fb16b7e7
FB
183#define emms_c() \
184{\
185 if (mm_flags & MM_MMX)\
186 emms();\
de6d9b64
FB
187}
188
189#define __align8 __attribute__ ((aligned (8)))
190
eb4b3dd3
ZK
191void dsputil_init_mmx(DSPContext* c, unsigned mask);
192void dsputil_set_bit_exact_mmx(DSPContext* c, unsigned mask);
de6d9b64 193
3d03c0a2
FB
194#elif defined(ARCH_ARMV4L)
195
3d03c0a2
FB
196/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
197 line ptimizations */
198#define __align8 __attribute__ ((aligned (4)))
199
eb4b3dd3 200void dsputil_init_armv4l(DSPContext* c, unsigned mask);
3d03c0a2 201
c34270f5 202#elif defined(HAVE_MLIB)
c34270f5
FB
203
204/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
205#define __align8 __attribute__ ((aligned (8)))
206
eb4b3dd3 207void dsputil_init_mlib(DSPContext* c, unsigned mask);
c34270f5 208
1e98dffb
NK
209#elif defined(ARCH_ALPHA)
210
1e98dffb
NK
211#define __align8 __attribute__ ((aligned (8)))
212
eb4b3dd3 213void dsputil_init_alpha(DSPContext* c, unsigned mask);
1e98dffb 214
59925ef2
BF
215#elif defined(ARCH_POWERPC)
216
404d2241
BF
217#define MM_ALTIVEC 0x0001 /* standard AltiVec */
218
219extern int mm_flags;
220
59925ef2
BF
221#define __align8 __attribute__ ((aligned (16)))
222
eb4b3dd3 223void dsputil_init_ppc(DSPContext* c, unsigned mask);
59925ef2 224
d46aba26
LS
225#elif defined(HAVE_MMI)
226
d46aba26
LS
227#define __align8 __attribute__ ((aligned (16)))
228
eb4b3dd3 229void dsputil_init_mmi(DSPContext* c, unsigned mask);
d46aba26 230
de6d9b64
FB
231#else
232
233#define __align8
234
235#endif
236
6d4985bb
FB
237#ifdef __GNUC__
238
239struct unaligned_64 { uint64_t l; } __attribute__((packed));
240struct unaligned_32 { uint32_t l; } __attribute__((packed));
241
242#define LD32(a) (((const struct unaligned_32 *) (a))->l)
243#define LD64(a) (((const struct unaligned_64 *) (a))->l)
244
245#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
246
247#else /* __GNUC__ */
248
249#define LD32(a) (*((uint32_t*)(a)))
250#define LD64(a) (*((uint64_t*)(a)))
251
252#define ST32(a, b) *((uint32_t*)(a)) = (b)
253
254#endif /* !__GNUC__ */
255
43f1708f
J
256/* PSNR */
257void get_psnr(UINT8 *orig_image[3], UINT8 *coded_image[3],
258 int orig_linesize[3], int coded_linesize,
259 AVCodecContext *avctx);
bb6f5690
FB
260
261/* FFT computation */
262
263/* NOTE: soon integer code will be added, so you must use the
264 FFTSample type */
265typedef float FFTSample;
266
267typedef struct FFTComplex {
268 FFTSample re, im;
269} FFTComplex;
270
271typedef struct FFTContext {
272 int nbits;
273 int inverse;
274 uint16_t *revtab;
275 FFTComplex *exptab;
276 FFTComplex *exptab1; /* only used by SSE code */
277 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
278} FFTContext;
279
280int fft_init(FFTContext *s, int nbits, int inverse);
281void fft_permute(FFTContext *s, FFTComplex *z);
282void fft_calc_c(FFTContext *s, FFTComplex *z);
283void fft_calc_sse(FFTContext *s, FFTComplex *z);
8d268a7d
FB
284void fft_calc_altivec(FFTContext *s, FFTComplex *z);
285
bb6f5690
FB
286static inline void fft_calc(FFTContext *s, FFTComplex *z)
287{
288 s->fft_calc(s, z);
289}
290void fft_end(FFTContext *s);
291
292/* MDCT computation */
293
294typedef struct MDCTContext {
295 int n; /* size of MDCT (i.e. number of input data * 2) */
296 int nbits; /* n = 2^nbits */
297 /* pre/post rotation tables */
298 FFTSample *tcos;
299 FFTSample *tsin;
300 FFTContext fft;
301} MDCTContext;
302
82696bee 303int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
eb4b3dd3 304void ff_imdct_calc(MDCTContext *s, FFTSample *output,
bb6f5690 305 const FFTSample *input, FFTSample *tmp);
eb4b3dd3 306void ff_mdct_calc(MDCTContext *s, FFTSample *out,
bb6f5690 307 const FFTSample *input, FFTSample *tmp);
82696bee 308void ff_mdct_end(MDCTContext *s);
bb6f5690 309
1457ab52
MN
310#define WARPER88_1616(name8, name16)\
311static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride){\
312 return name8(s, dst , src , stride)\
313 +name8(s, dst+8 , src+8 , stride)\
314 +name8(s, dst +8*stride, src +8*stride, stride)\
315 +name8(s, dst+8+8*stride, src+8+8*stride, stride);\
316}
317
95e2ce4a 318#ifndef HAVE_LRINTF
9d85cbd9
FB
319/* XXX: add ISOC specific test to avoid specific BSD testing. */
320/* better than nothing implementation. */
6234d753 321/* btw, rintf() is existing on fbsd too -- alex */
9d85cbd9
FB
322static inline long int lrintf(float x)
323{
324 return (int)(rint(x));
325}
326#endif
327
de6d9b64 328#endif