huffyuv
[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
b7c27ee6 82typedef int (*op_pixels_abs_func)(UINT8 *blk1/*align width (8 or 16)*/, UINT8 *blk2/*align 1*/, int line_size);
eb4b3dd3 83/*
ba6802de
MN
84int pix_abs16x16_c(UINT8 *blk1, UINT8 *blk2, int lx);
85int pix_abs16x16_x2_c(UINT8 *blk1, UINT8 *blk2, int lx);
86int pix_abs16x16_y2_c(UINT8 *blk1, UINT8 *blk2, int lx);
87int pix_abs16x16_xy2_c(UINT8 *blk1, UINT8 *blk2, int lx);
eb4b3dd3
ZK
88*/
89typedef struct DSPContext {
90 /* pixel ops : interface with DCT */
91 void (*get_pixels)(DCTELEM *block/*align 16*/, const UINT8 *pixels/*align 8*/, int line_size);
92 void (*diff_pixels)(DCTELEM *block/*align 16*/, const UINT8 *s1/*align 8*/, const UINT8 *s2/*align 8*/, int stride);
93 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
94 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
95 void (*gmc1)(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
96 void (*gmc )(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int stride, int h, int ox, int oy,
97 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
98 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
99 int (*pix_sum)(UINT8 * pix, int line_size);
100 int (*pix_norm1)(UINT8 * pix, int line_size);
101
102 /* maybe create an array for 16/8 functions */
103 op_pixels_func put_pixels_tab[2][4];
104 op_pixels_func avg_pixels_tab[2][4];
105 op_pixels_func put_no_rnd_pixels_tab[2][4];
106 op_pixels_func avg_no_rnd_pixels_tab[2][4];
107 qpel_mc_func put_qpel_pixels_tab[2][16];
108 qpel_mc_func avg_qpel_pixels_tab[2][16];
109 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
110 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
111
112 op_pixels_abs_func pix_abs16x16;
113 op_pixels_abs_func pix_abs16x16_x2;
114 op_pixels_abs_func pix_abs16x16_y2;
115 op_pixels_abs_func pix_abs16x16_xy2;
116 op_pixels_abs_func pix_abs8x8;
117 op_pixels_abs_func pix_abs8x8_x2;
118 op_pixels_abs_func pix_abs8x8_y2;
119 op_pixels_abs_func pix_abs8x8_xy2;
11f18faf
MN
120
121 /* huffyuv specific */
122 //FIXME note: alignment isnt guranteed currently but could be if needed
123 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
124 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/,int w);
eb4b3dd3
ZK
125} DSPContext;
126
127void dsputil_init(DSPContext* p, unsigned mask);
de6d9b64 128
7801d21d
MN
129/**
130 * permute block according to permuatation.
131 * @param last last non zero element in scantable order
132 */
133void ff_block_permute(INT16 *block, UINT8 *permutation, const UINT8 *scantable, int last);
e0eac44e 134
eb4b3dd3
ZK
135#define emms_c()
136
3d03c0a2 137#if defined(HAVE_MMX)
de6d9b64 138
18f77016 139#undef emms_c
eb4b3dd3 140
de6d9b64
FB
141#define MM_MMX 0x0001 /* standard MMX */
142#define MM_3DNOW 0x0004 /* AMD 3DNOW */
143#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
144#define MM_SSE 0x0008 /* SSE functions */
145#define MM_SSE2 0x0010 /* PIV SSE2 functions */
146
147extern int mm_flags;
148
149int mm_support(void);
eb4b3dd3
ZK
150void add_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size);
151void put_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size);
de6d9b64
FB
152
153static inline void emms(void)
154{
fb16b7e7
FB
155 __asm __volatile ("emms;":::"memory");
156}
157
158#define emms_c() \
159{\
160 if (mm_flags & MM_MMX)\
161 emms();\
de6d9b64
FB
162}
163
164#define __align8 __attribute__ ((aligned (8)))
165
eb4b3dd3
ZK
166void dsputil_init_mmx(DSPContext* c, unsigned mask);
167void dsputil_set_bit_exact_mmx(DSPContext* c, unsigned mask);
de6d9b64 168
3d03c0a2
FB
169#elif defined(ARCH_ARMV4L)
170
3d03c0a2
FB
171/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
172 line ptimizations */
173#define __align8 __attribute__ ((aligned (4)))
174
eb4b3dd3 175void dsputil_init_armv4l(DSPContext* c, unsigned mask);
3d03c0a2 176
c34270f5 177#elif defined(HAVE_MLIB)
c34270f5
FB
178
179/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
180#define __align8 __attribute__ ((aligned (8)))
181
eb4b3dd3 182void dsputil_init_mlib(DSPContext* c, unsigned mask);
c34270f5 183
1e98dffb
NK
184#elif defined(ARCH_ALPHA)
185
1e98dffb
NK
186#define __align8 __attribute__ ((aligned (8)))
187
eb4b3dd3 188void dsputil_init_alpha(DSPContext* c, unsigned mask);
1e98dffb 189
59925ef2
BF
190#elif defined(ARCH_POWERPC)
191
59925ef2
BF
192#define __align8 __attribute__ ((aligned (16)))
193
eb4b3dd3 194void dsputil_init_ppc(DSPContext* c, unsigned mask);
59925ef2 195
d46aba26
LS
196#elif defined(HAVE_MMI)
197
d46aba26
LS
198#define __align8 __attribute__ ((aligned (16)))
199
eb4b3dd3 200void dsputil_init_mmi(DSPContext* c, unsigned mask);
d46aba26 201
de6d9b64
FB
202#else
203
204#define __align8
205
206#endif
207
6d4985bb
FB
208#ifdef __GNUC__
209
210struct unaligned_64 { uint64_t l; } __attribute__((packed));
211struct unaligned_32 { uint32_t l; } __attribute__((packed));
212
213#define LD32(a) (((const struct unaligned_32 *) (a))->l)
214#define LD64(a) (((const struct unaligned_64 *) (a))->l)
215
216#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
217
218#else /* __GNUC__ */
219
220#define LD32(a) (*((uint32_t*)(a)))
221#define LD64(a) (*((uint64_t*)(a)))
222
223#define ST32(a, b) *((uint32_t*)(a)) = (b)
224
225#endif /* !__GNUC__ */
226
43f1708f
J
227/* PSNR */
228void get_psnr(UINT8 *orig_image[3], UINT8 *coded_image[3],
229 int orig_linesize[3], int coded_linesize,
230 AVCodecContext *avctx);
bb6f5690
FB
231
232/* FFT computation */
233
234/* NOTE: soon integer code will be added, so you must use the
235 FFTSample type */
236typedef float FFTSample;
237
238typedef struct FFTComplex {
239 FFTSample re, im;
240} FFTComplex;
241
242typedef struct FFTContext {
243 int nbits;
244 int inverse;
245 uint16_t *revtab;
246 FFTComplex *exptab;
247 FFTComplex *exptab1; /* only used by SSE code */
248 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
249} FFTContext;
250
251int fft_init(FFTContext *s, int nbits, int inverse);
252void fft_permute(FFTContext *s, FFTComplex *z);
253void fft_calc_c(FFTContext *s, FFTComplex *z);
254void fft_calc_sse(FFTContext *s, FFTComplex *z);
255static inline void fft_calc(FFTContext *s, FFTComplex *z)
256{
257 s->fft_calc(s, z);
258}
259void fft_end(FFTContext *s);
260
261/* MDCT computation */
262
263typedef struct MDCTContext {
264 int n; /* size of MDCT (i.e. number of input data * 2) */
265 int nbits; /* n = 2^nbits */
266 /* pre/post rotation tables */
267 FFTSample *tcos;
268 FFTSample *tsin;
269 FFTContext fft;
270} MDCTContext;
271
82696bee 272int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
eb4b3dd3 273void ff_imdct_calc(MDCTContext *s, FFTSample *output,
bb6f5690 274 const FFTSample *input, FFTSample *tmp);
eb4b3dd3 275void ff_mdct_calc(MDCTContext *s, FFTSample *out,
bb6f5690 276 const FFTSample *input, FFTSample *tmp);
82696bee 277void ff_mdct_end(MDCTContext *s);
bb6f5690 278
95e2ce4a 279#ifndef HAVE_LRINTF
9d85cbd9
FB
280/* XXX: add ISOC specific test to avoid specific BSD testing. */
281/* better than nothing implementation. */
6234d753 282/* btw, rintf() is existing on fbsd too -- alex */
9d85cbd9
FB
283static inline long int lrintf(float x)
284{
285 return (int)(rint(x));
286}
287#endif
288
de6d9b64 289#endif