Altivec on non darwin systems patch by Romain Dolbeau
[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 //typedef int DCTELEM;
29
30 void fdct_ifast (DCTELEM *data);
31 void ff_jpeg_fdct_islow (DCTELEM *data);
32
33 void j_rev_dct (DCTELEM *data);
34
35 void ff_fdct_mmx(DCTELEM *block);
36
37 /* encoding scans */
38 extern const UINT8 ff_alternate_horizontal_scan[64];
39 extern const UINT8 ff_alternate_vertical_scan[64];
40 extern const UINT8 ff_zigzag_direct[64];
41
42 /* pixel operations */
43 #define MAX_NEG_CROP 384
44
45 /* temporary */
46 extern UINT32 squareTbl[512];
47 extern UINT8 cropTbl[256 + 2 * MAX_NEG_CROP];
48
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 /*
61 void get_pixels_c(DCTELEM *block, const UINT8 *pixels, int line_size);
62 void diff_pixels_c(DCTELEM *block, const UINT8 *s1, const UINT8 *s2, int stride);
63 void put_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size);
64 void add_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size);
65 void clear_blocks_c(DCTELEM *blocks);
66 */
67
68 /* add and put pixel (decoding) */
69 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
70 typedef void (*op_pixels_func)(UINT8 *block/*align width (8 or 16)*/, const UINT8 *pixels/*align 1*/, int line_size, int h);
71 typedef void (*qpel_mc_func)(UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);
72
73 #define DEF_OLD_QPEL(name)\
74 void ff_put_ ## name (UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);\
75 void ff_put_no_rnd_ ## name (UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);\
76 void ff_avg_ ## name (UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);
77
78 DEF_OLD_QPEL(qpel16_mc11_old_c)
79 DEF_OLD_QPEL(qpel16_mc31_old_c)
80 DEF_OLD_QPEL(qpel16_mc12_old_c)
81 DEF_OLD_QPEL(qpel16_mc32_old_c)
82 DEF_OLD_QPEL(qpel16_mc13_old_c)
83 DEF_OLD_QPEL(qpel16_mc33_old_c)
84 DEF_OLD_QPEL(qpel8_mc11_old_c)
85 DEF_OLD_QPEL(qpel8_mc31_old_c)
86 DEF_OLD_QPEL(qpel8_mc12_old_c)
87 DEF_OLD_QPEL(qpel8_mc32_old_c)
88 DEF_OLD_QPEL(qpel8_mc13_old_c)
89 DEF_OLD_QPEL(qpel8_mc33_old_c)
90
91 #define CALL_2X_PIXELS(a, b, n)\
92 static 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 }
96
97 /* motion estimation */
98
99 typedef int (*op_pixels_abs_func)(UINT8 *blk1/*align width (8 or 16)*/, UINT8 *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
100
101 typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, UINT8 *blk1/*align width (8 or 16)*/, UINT8 *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
102
103 typedef 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);
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];
120 me_cmp_func bit[2];
121 me_cmp_func rd[2];
122 int (*hadamard8_abs )(uint8_t *src, int stride, int mean);
123
124 me_cmp_func me_pre_cmp[11];
125 me_cmp_func me_cmp[11];
126 me_cmp_func me_sub_cmp[11];
127 me_cmp_func mb_cmp[11];
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];
138 qpel_mc_func put_mspel_pixels_tab[8];
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;
148
149 /* huffyuv specific */
150 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
151 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
152 } DSPContext;
153
154 void dsputil_init(DSPContext* p, unsigned mask);
155
156 /**
157 * permute block according to permuatation.
158 * @param last last non zero element in scantable order
159 */
160 void ff_block_permute(DCTELEM *block, UINT8 *permutation, const UINT8 *scantable, int last);
161
162 #define emms_c()
163
164 /* should be defined by architectures supporting
165 one or more MultiMedia extension */
166 int mm_support(void);
167
168 #if defined(HAVE_MMX)
169
170 #undef emms_c
171
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
178 extern int mm_flags;
179
180 void add_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size);
181 void put_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size);
182
183 static inline void emms(void)
184 {
185 __asm __volatile ("emms;":::"memory");
186 }
187
188
189 #define emms_c() \
190 {\
191 if (mm_flags & MM_MMX)\
192 emms();\
193 }
194
195 #define __align8 __attribute__ ((aligned (8)))
196
197 void dsputil_init_mmx(DSPContext* c, unsigned mask);
198 void dsputil_set_bit_exact_mmx(DSPContext* c, unsigned mask);
199
200 #elif defined(ARCH_ARMV4L)
201
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
206 void dsputil_init_armv4l(DSPContext* c, unsigned mask);
207
208 #elif defined(HAVE_MLIB)
209
210 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
211 #define __align8 __attribute__ ((aligned (8)))
212
213 void dsputil_init_mlib(DSPContext* c, unsigned mask);
214
215 #elif defined(ARCH_ALPHA)
216
217 #define __align8 __attribute__ ((aligned (8)))
218
219 void dsputil_init_alpha(DSPContext* c, unsigned mask);
220
221 #elif defined(ARCH_POWERPC)
222
223 #define MM_ALTIVEC 0x0001 /* standard AltiVec */
224
225 extern int mm_flags;
226
227 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
228 #include <altivec.h>
229 #endif
230
231 #define __align8 __attribute__ ((aligned (16)))
232
233 void dsputil_init_ppc(DSPContext* c, unsigned mask);
234
235 #elif defined(HAVE_MMI)
236
237 #define __align8 __attribute__ ((aligned (16)))
238
239 void dsputil_init_mmi(DSPContext* c, unsigned mask);
240
241 #else
242
243 #define __align8
244
245 #endif
246
247 #ifdef __GNUC__
248
249 struct unaligned_64 { uint64_t l; } __attribute__((packed));
250 struct unaligned_32 { uint32_t l; } __attribute__((packed));
251
252 #define LD32(a) (((const struct unaligned_32 *) (a))->l)
253 #define LD64(a) (((const struct unaligned_64 *) (a))->l)
254
255 #define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
256
257 #else /* __GNUC__ */
258
259 #define LD32(a) (*((uint32_t*)(a)))
260 #define LD64(a) (*((uint64_t*)(a)))
261
262 #define ST32(a, b) *((uint32_t*)(a)) = (b)
263
264 #endif /* !__GNUC__ */
265
266 /* PSNR */
267 void get_psnr(UINT8 *orig_image[3], UINT8 *coded_image[3],
268 int orig_linesize[3], int coded_linesize,
269 AVCodecContext *avctx);
270
271 /* FFT computation */
272
273 /* NOTE: soon integer code will be added, so you must use the
274 FFTSample type */
275 typedef float FFTSample;
276
277 typedef struct FFTComplex {
278 FFTSample re, im;
279 } FFTComplex;
280
281 typedef struct FFTContext {
282 int nbits;
283 int inverse;
284 uint16_t *revtab;
285 FFTComplex *exptab;
286 FFTComplex *exptab1; /* only used by SSE code */
287 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
288 } FFTContext;
289
290 int fft_init(FFTContext *s, int nbits, int inverse);
291 void fft_permute(FFTContext *s, FFTComplex *z);
292 void fft_calc_c(FFTContext *s, FFTComplex *z);
293 void fft_calc_sse(FFTContext *s, FFTComplex *z);
294 void fft_calc_altivec(FFTContext *s, FFTComplex *z);
295
296 static inline void fft_calc(FFTContext *s, FFTComplex *z)
297 {
298 s->fft_calc(s, z);
299 }
300 void fft_end(FFTContext *s);
301
302 /* MDCT computation */
303
304 typedef struct MDCTContext {
305 int n; /* size of MDCT (i.e. number of input data * 2) */
306 int nbits; /* n = 2^nbits */
307 /* pre/post rotation tables */
308 FFTSample *tcos;
309 FFTSample *tsin;
310 FFTContext fft;
311 } MDCTContext;
312
313 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
314 void ff_imdct_calc(MDCTContext *s, FFTSample *output,
315 const FFTSample *input, FFTSample *tmp);
316 void ff_mdct_calc(MDCTContext *s, FFTSample *out,
317 const FFTSample *input, FFTSample *tmp);
318 void ff_mdct_end(MDCTContext *s);
319
320 #define WARPER88_1616(name8, name16)\
321 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride){\
322 return name8(s, dst , src , stride)\
323 +name8(s, dst+8 , src+8 , stride)\
324 +name8(s, dst +8*stride, src +8*stride, stride)\
325 +name8(s, dst+8+8*stride, src+8+8*stride, stride);\
326 }
327
328 #ifndef HAVE_LRINTF
329 /* XXX: add ISOC specific test to avoid specific BSD testing. */
330 /* better than nothing implementation. */
331 /* btw, rintf() is existing on fbsd too -- alex */
332 static inline long int lrintf(float x)
333 {
334 return (int)(rint(x));
335 }
336 #endif
337
338 #endif