new generic FFT/MDCT code for audio codecs
[libav.git] / libavcodec / dsputil.h
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 */
19 #ifndef DSPUTIL_H
20 #define DSPUTIL_H
21
22 #include "common.h"
23 #include "avcodec.h"
24
25 //#define DEBUG
26 /* dct code */
27 typedef short DCTELEM;
28
29 void fdct_ifast (DCTELEM *data);
30 void ff_jpeg_fdct_islow (DCTELEM *data);
31
32 void j_rev_dct (DCTELEM *data);
33
34 void ff_fdct_mmx(DCTELEM *block);
35
36 /* encoding scans */
37 extern const UINT8 ff_alternate_horizontal_scan[64];
38 extern const UINT8 ff_alternate_vertical_scan[64];
39 extern const UINT8 ff_zigzag_direct[64];
40
41 /* pixel operations */
42 #define MAX_NEG_CROP 384
43
44 /* temporary */
45 extern UINT32 squareTbl[512];
46 extern UINT8 cropTbl[256 + 2 * MAX_NEG_CROP];
47
48 void dsputil_init(void);
49
50 /* minimum alignment rules ;)
51 if u notice errors in the align stuff, need more alignment for some asm code for some cpu
52 or 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)
55 i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
56
57 !future video codecs might need functions with less strict alignment
58 */
59
60 /* pixel ops : interface with DCT */
61 extern void (*get_pixels)(DCTELEM *block/*align 16*/, const UINT8 *pixels/*align 8*/, int line_size);
62 extern void (*diff_pixels)(DCTELEM *block/*align 16*/, const UINT8 *s1/*align 8*/, const UINT8 *s2/*align 8*/, int stride);
63 extern void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
64 extern void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
65 extern void (*ff_gmc1)(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
66 extern void (*ff_gmc )(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int stride, int h, int ox, int oy,
67 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
68 extern void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
69 extern int (*pix_sum)(UINT8 * pix, int line_size);
70 extern int (*pix_norm1)(UINT8 * pix, int line_size);
71
72
73
74 void get_pixels_c(DCTELEM *block, const UINT8 *pixels, int line_size);
75 void diff_pixels_c(DCTELEM *block, const UINT8 *s1, const UINT8 *s2, int stride);
76 void put_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size);
77 void add_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size);
78 void clear_blocks_c(DCTELEM *blocks);
79
80 /* add and put pixel (decoding) */
81 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
82 typedef void (*op_pixels_func)(UINT8 *block/*align width (8 or 16)*/, const UINT8 *pixels/*align 1*/, int line_size, int h);
83 typedef void (*qpel_mc_func)(UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);
84
85 extern op_pixels_func put_pixels_tab[2][4];
86 extern op_pixels_func avg_pixels_tab[2][4];
87 extern op_pixels_func put_no_rnd_pixels_tab[2][4];
88 extern op_pixels_func avg_no_rnd_pixels_tab[2][4];
89 extern qpel_mc_func put_qpel_pixels_tab[2][16];
90 extern qpel_mc_func avg_qpel_pixels_tab[2][16];
91 extern qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
92 extern qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
93
94 #define CALL_2X_PIXELS(a, b, n)\
95 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
96 b(block , pixels , line_size, h);\
97 b(block+n, pixels+n, line_size, h);\
98 }
99
100 /* motion estimation */
101
102 typedef int (*op_pixels_abs_func)(UINT8 *blk1/*align width (8 or 16)*/, UINT8 *blk2/*align 1*/, int line_size);
103
104 extern op_pixels_abs_func pix_abs16x16;
105 extern op_pixels_abs_func pix_abs16x16_x2;
106 extern op_pixels_abs_func pix_abs16x16_y2;
107 extern op_pixels_abs_func pix_abs16x16_xy2;
108 extern op_pixels_abs_func pix_abs8x8;
109 extern op_pixels_abs_func pix_abs8x8_x2;
110 extern op_pixels_abs_func pix_abs8x8_y2;
111 extern op_pixels_abs_func pix_abs8x8_xy2;
112
113 int pix_abs16x16_c(UINT8 *blk1, UINT8 *blk2, int lx);
114 int pix_abs16x16_x2_c(UINT8 *blk1, UINT8 *blk2, int lx);
115 int pix_abs16x16_y2_c(UINT8 *blk1, UINT8 *blk2, int lx);
116 int pix_abs16x16_xy2_c(UINT8 *blk1, UINT8 *blk2, int lx);
117
118 /**
119 * permute block according to permuatation.
120 * @param last last non zero element in scantable order
121 */
122 void ff_block_permute(INT16 *block, UINT8 *permutation, const UINT8 *scantable, int last);
123
124 #if defined(HAVE_MMX)
125
126 #define MM_MMX 0x0001 /* standard MMX */
127 #define MM_3DNOW 0x0004 /* AMD 3DNOW */
128 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
129 #define MM_SSE 0x0008 /* SSE functions */
130 #define MM_SSE2 0x0010 /* PIV SSE2 functions */
131
132 extern int mm_flags;
133
134 int mm_support(void);
135
136 static inline void emms(void)
137 {
138 __asm __volatile ("emms;":::"memory");
139 }
140
141 #define emms_c() \
142 {\
143 if (mm_flags & MM_MMX)\
144 emms();\
145 }
146
147 #define __align8 __attribute__ ((aligned (8)))
148
149 void dsputil_init_mmx(void);
150 void dsputil_set_bit_exact_mmx(void);
151
152 #elif defined(ARCH_ARMV4L)
153
154 #define emms_c()
155
156 /* This is to use 4 bytes read to the IDCT pointers for some 'zero'
157 line ptimizations */
158 #define __align8 __attribute__ ((aligned (4)))
159
160 void dsputil_init_armv4l(void);
161
162 #elif defined(HAVE_MLIB)
163
164 #define emms_c()
165
166 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
167 #define __align8 __attribute__ ((aligned (8)))
168
169 void dsputil_init_mlib(void);
170
171 #elif defined(ARCH_ALPHA)
172
173 #define emms_c()
174 #define __align8 __attribute__ ((aligned (8)))
175
176 void dsputil_init_alpha(void);
177
178 #elif defined(ARCH_POWERPC)
179
180 #define emms_c()
181 #define __align8 __attribute__ ((aligned (16)))
182
183 void dsputil_init_ppc(void);
184
185 #elif defined(HAVE_MMI)
186
187 #define emms_c()
188
189 #define __align8 __attribute__ ((aligned (16)))
190
191 void dsputil_init_mmi(void);
192
193 #else
194
195 #define emms_c()
196
197 #define __align8
198
199 #endif
200
201 #ifdef __GNUC__
202
203 struct unaligned_64 { uint64_t l; } __attribute__((packed));
204 struct unaligned_32 { uint32_t l; } __attribute__((packed));
205
206 #define LD32(a) (((const struct unaligned_32 *) (a))->l)
207 #define LD64(a) (((const struct unaligned_64 *) (a))->l)
208
209 #define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
210
211 #else /* __GNUC__ */
212
213 #define LD32(a) (*((uint32_t*)(a)))
214 #define LD64(a) (*((uint64_t*)(a)))
215
216 #define ST32(a, b) *((uint32_t*)(a)) = (b)
217
218 #endif /* !__GNUC__ */
219
220 /* PSNR */
221 void get_psnr(UINT8 *orig_image[3], UINT8 *coded_image[3],
222 int orig_linesize[3], int coded_linesize,
223 AVCodecContext *avctx);
224
225 /* FFT computation */
226
227 /* NOTE: soon integer code will be added, so you must use the
228 FFTSample type */
229 typedef float FFTSample;
230
231 typedef struct FFTComplex {
232 FFTSample re, im;
233 } FFTComplex;
234
235 typedef struct FFTContext {
236 int nbits;
237 int inverse;
238 uint16_t *revtab;
239 FFTComplex *exptab;
240 FFTComplex *exptab1; /* only used by SSE code */
241 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
242 } FFTContext;
243
244 int fft_init(FFTContext *s, int nbits, int inverse);
245 void fft_permute(FFTContext *s, FFTComplex *z);
246 void fft_calc_c(FFTContext *s, FFTComplex *z);
247 void fft_calc_sse(FFTContext *s, FFTComplex *z);
248 static inline void fft_calc(FFTContext *s, FFTComplex *z)
249 {
250 s->fft_calc(s, z);
251 }
252 void fft_end(FFTContext *s);
253
254 /* MDCT computation */
255
256 typedef struct MDCTContext {
257 int n; /* size of MDCT (i.e. number of input data * 2) */
258 int nbits; /* n = 2^nbits */
259 /* pre/post rotation tables */
260 FFTSample *tcos;
261 FFTSample *tsin;
262 FFTContext fft;
263 } MDCTContext;
264
265 int mdct_init(MDCTContext *s, int nbits, int inverse);
266 void imdct_calc(MDCTContext *s, FFTSample *output,
267 const FFTSample *input, FFTSample *tmp);
268 void mdct_calc(MDCTContext *s, FFTSample *out,
269 const FFTSample *input, FFTSample *tmp);
270 void mdct_end(MDCTContext *s);
271
272 #endif