a7c83f00e55daad8ce0d79be62db3fe11166097c
[libav.git] / libavcodec / dsputil.h
1 /*
2 * DSP utils
3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard.
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5 *
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
10 *
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21 /**
22 * @file dsputil.h
23 * DSP utils.
24 * note, many functions in here may use MMX which trashes the FPU state, it is
25 * absolutely necessary to call emms_c() between dsp & float/double code
26 */
27
28 #ifndef DSPUTIL_H
29 #define DSPUTIL_H
30
31 #include "common.h"
32 #include "avcodec.h"
33
34
35 //#define DEBUG
36 /* dct code */
37 typedef short DCTELEM;
38 typedef int DWTELEM;
39
40 void fdct_ifast (DCTELEM *data);
41 void fdct_ifast248 (DCTELEM *data);
42 void ff_jpeg_fdct_islow (DCTELEM *data);
43 void ff_fdct248_islow (DCTELEM *data);
44
45 void j_rev_dct (DCTELEM *data);
46 void j_rev_dct4 (DCTELEM *data);
47 void j_rev_dct2 (DCTELEM *data);
48 void j_rev_dct1 (DCTELEM *data);
49
50 void ff_fdct_mmx(DCTELEM *block);
51 void ff_fdct_mmx2(DCTELEM *block);
52 void ff_fdct_sse2(DCTELEM *block);
53
54 void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
55 void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
56 void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
57 void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
58 void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
59 void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
60
61 /* encoding scans */
62 extern const uint8_t ff_alternate_horizontal_scan[64];
63 extern const uint8_t ff_alternate_vertical_scan[64];
64 extern const uint8_t ff_zigzag_direct[64];
65 extern const uint8_t ff_zigzag248_direct[64];
66
67 /* pixel operations */
68 #define MAX_NEG_CROP 1024
69
70 /* temporary */
71 extern uint32_t squareTbl[512];
72 extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
73
74 /* VP3 DSP functions */
75 void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
76 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
77 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
78
79 /* minimum alignment rules ;)
80 if u notice errors in the align stuff, need more alignment for some asm code for some cpu
81 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
82
83 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
84 i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
85
86 !future video codecs might need functions with less strict alignment
87 */
88
89 /*
90 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
91 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
92 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
93 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
94 void clear_blocks_c(DCTELEM *blocks);
95 */
96
97 /* add and put pixel (decoding) */
98 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
99 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
100 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
101 typedef void (*tpel_mc_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int w, int h);
102 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
103 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
104 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
105 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
106
107 #define DEF_OLD_QPEL(name)\
108 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
109 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
110 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
111
112 DEF_OLD_QPEL(qpel16_mc11_old_c)
113 DEF_OLD_QPEL(qpel16_mc31_old_c)
114 DEF_OLD_QPEL(qpel16_mc12_old_c)
115 DEF_OLD_QPEL(qpel16_mc32_old_c)
116 DEF_OLD_QPEL(qpel16_mc13_old_c)
117 DEF_OLD_QPEL(qpel16_mc33_old_c)
118 DEF_OLD_QPEL(qpel8_mc11_old_c)
119 DEF_OLD_QPEL(qpel8_mc31_old_c)
120 DEF_OLD_QPEL(qpel8_mc12_old_c)
121 DEF_OLD_QPEL(qpel8_mc32_old_c)
122 DEF_OLD_QPEL(qpel8_mc13_old_c)
123 DEF_OLD_QPEL(qpel8_mc33_old_c)
124
125 #define CALL_2X_PIXELS(a, b, n)\
126 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
127 b(block , pixels , line_size, h);\
128 b(block+n, pixels+n, line_size, h);\
129 }
130
131 /* motion estimation */
132 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
133 // allthough currently h<4 is not used as functions with width <8 are not used and neither implemented
134 typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size, int h)/* __attribute__ ((const))*/;
135
136
137 // for snow slices
138 typedef struct slice_buffer_s slice_buffer;
139
140 /**
141 * DSPContext.
142 */
143 typedef struct DSPContext {
144 /* pixel ops : interface with DCT */
145 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
146 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
147 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
148 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
149 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
150 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
151 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
152 /**
153 * translational global motion compensation.
154 */
155 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
156 /**
157 * global motion compensation.
158 */
159 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
160 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
161 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
162 int (*pix_sum)(uint8_t * pix, int line_size);
163 int (*pix_norm1)(uint8_t * pix, int line_size);
164 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
165
166 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
167 me_cmp_func sse[5];
168 me_cmp_func hadamard8_diff[5];
169 me_cmp_func dct_sad[5];
170 me_cmp_func quant_psnr[5];
171 me_cmp_func bit[5];
172 me_cmp_func rd[5];
173 me_cmp_func vsad[5];
174 me_cmp_func vsse[5];
175 me_cmp_func nsse[5];
176 me_cmp_func w53[5];
177 me_cmp_func w97[5];
178 me_cmp_func dct_max[5];
179 me_cmp_func dct264_sad[5];
180
181 me_cmp_func me_pre_cmp[5];
182 me_cmp_func me_cmp[5];
183 me_cmp_func me_sub_cmp[5];
184 me_cmp_func mb_cmp[5];
185 me_cmp_func ildct_cmp[5]; //only width 16 used
186 me_cmp_func frame_skip_cmp[5]; //only width 8 used
187
188 /**
189 * Halfpel motion compensation with rounding (a+b+1)>>1.
190 * this is an array[4][4] of motion compensation funcions for 4
191 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
192 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
193 * @param block destination where the result is stored
194 * @param pixels source
195 * @param line_size number of bytes in a horizontal line of block
196 * @param h height
197 */
198 op_pixels_func put_pixels_tab[4][4];
199
200 /**
201 * Halfpel motion compensation with rounding (a+b+1)>>1.
202 * This is an array[4][4] of motion compensation functions for 4
203 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
204 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
205 * @param block destination into which the result is averaged (a+b+1)>>1
206 * @param pixels source
207 * @param line_size number of bytes in a horizontal line of block
208 * @param h height
209 */
210 op_pixels_func avg_pixels_tab[4][4];
211
212 /**
213 * Halfpel motion compensation with no rounding (a+b)>>1.
214 * this is an array[2][4] of motion compensation funcions for 2
215 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
216 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
217 * @param block destination where the result is stored
218 * @param pixels source
219 * @param line_size number of bytes in a horizontal line of block
220 * @param h height
221 */
222 op_pixels_func put_no_rnd_pixels_tab[4][4];
223
224 /**
225 * Halfpel motion compensation with no rounding (a+b)>>1.
226 * this is an array[2][4] of motion compensation funcions for 2
227 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
228 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
229 * @param block destination into which the result is averaged (a+b)>>1
230 * @param pixels source
231 * @param line_size number of bytes in a horizontal line of block
232 * @param h height
233 */
234 op_pixels_func avg_no_rnd_pixels_tab[4][4];
235
236 void (*put_no_rnd_pixels_l2[2])(uint8_t *block/*align width (8 or 16)*/, const uint8_t *a/*align 1*/, const uint8_t *b/*align 1*/, int line_size, int h);
237
238 /**
239 * Thirdpel motion compensation with rounding (a+b+1)>>1.
240 * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
241 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
242 * @param block destination where the result is stored
243 * @param pixels source
244 * @param line_size number of bytes in a horizontal line of block
245 * @param h height
246 */
247 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
248 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
249
250 qpel_mc_func put_qpel_pixels_tab[2][16];
251 qpel_mc_func avg_qpel_pixels_tab[2][16];
252 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
253 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
254 qpel_mc_func put_mspel_pixels_tab[8];
255
256 /**
257 * h264 Chram MC
258 */
259 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
260 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
261
262 qpel_mc_func put_h264_qpel_pixels_tab[4][16];
263 qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
264
265 h264_weight_func weight_h264_pixels_tab[10];
266 h264_biweight_func biweight_h264_pixels_tab[10];
267
268 me_cmp_func pix_abs[2][4];
269
270 /* huffyuv specific */
271 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
272 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
273 /**
274 * subtract huffyuv's variant of median prediction
275 * note, this might read from src1[-1], src2[-1]
276 */
277 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
278 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
279
280 void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
281 void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
282 void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
283 void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
284 void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
285 void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
286
287 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
288 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
289
290 void (*h261_loop_filter)(uint8_t *src, int stride);
291
292 /* (I)DCT */
293 void (*fdct)(DCTELEM *block/* align 16*/);
294 void (*fdct248)(DCTELEM *block/* align 16*/);
295
296 /* IDCT really*/
297 void (*idct)(DCTELEM *block/* align 16*/);
298
299 /**
300 * block -> idct -> clip to unsigned 8 bit -> dest.
301 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
302 * @param line_size size in bytes of a horizotal line of dest
303 */
304 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
305
306 /**
307 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
308 * @param line_size size in bytes of a horizotal line of dest
309 */
310 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
311
312 /**
313 * idct input permutation.
314 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
315 * IDCT)
316 * this permutation must be performed before the idct_put/add, note, normally this can be merged
317 * with the zigzag/alternate scan<br>
318 * an example to avoid confusion:
319 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
320 * - (x -> referece dct -> reference idct -> x)
321 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
322 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
323 */
324 uint8_t idct_permutation[64];
325 int idct_permutation_type;
326 #define FF_NO_IDCT_PERM 1
327 #define FF_LIBMPEG2_IDCT_PERM 2
328 #define FF_SIMPLE_IDCT_PERM 3
329 #define FF_TRANSPOSE_IDCT_PERM 4
330 #define FF_PARTTRANS_IDCT_PERM 5
331
332 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
333 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
334 #define BASIS_SHIFT 16
335 #define RECON_SHIFT 6
336
337 void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
338 void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
339 void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
340 void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
341
342 /* snow wavelet */
343 void (*vertical_compose97i)(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width);
344 void (*horizontal_compose97i)(DWTELEM *b, int width);
345 void (*inner_add_yblock)(uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
346
347 void (*prefetch)(void *mem, int stride, int h);
348 } DSPContext;
349
350 void dsputil_static_init(void);
351 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
352
353 /**
354 * permute block according to permuatation.
355 * @param last last non zero element in scantable order
356 */
357 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
358
359 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
360
361 #define BYTE_VEC32(c) ((c)*0x01010101UL)
362
363 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
364 {
365 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
366 }
367
368 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
369 {
370 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
371 }
372
373 static inline int get_penalty_factor(int lambda, int lambda2, int type){
374 switch(type&0xFF){
375 default:
376 case FF_CMP_SAD:
377 return lambda>>FF_LAMBDA_SHIFT;
378 case FF_CMP_DCT:
379 return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
380 case FF_CMP_W53:
381 return (4*lambda)>>(FF_LAMBDA_SHIFT);
382 case FF_CMP_W97:
383 return (2*lambda)>>(FF_LAMBDA_SHIFT);
384 case FF_CMP_SATD:
385 case FF_CMP_DCT264:
386 return (2*lambda)>>FF_LAMBDA_SHIFT;
387 case FF_CMP_RD:
388 case FF_CMP_PSNR:
389 case FF_CMP_SSE:
390 case FF_CMP_NSSE:
391 return lambda2>>FF_LAMBDA_SHIFT;
392 case FF_CMP_BIT:
393 return 1;
394 }
395 }
396
397 /**
398 * Empty mmx state.
399 * this must be called between any dsp function and float/double code.
400 * for example sin(); dsp->idct_put(); emms_c(); cos()
401 */
402 #define emms_c()
403
404 /* should be defined by architectures supporting
405 one or more MultiMedia extension */
406 int mm_support(void);
407
408 #ifdef __GNUC__
409 #define DECLARE_ALIGNED_16(t,v) t v __attribute__ ((aligned (16)))
410 #else
411 #define DECLARE_ALIGNED_16(t,v) __declspec(align(16)) t v
412 #endif
413
414 #if defined(HAVE_MMX)
415
416 #undef emms_c
417
418 #define MM_MMX 0x0001 /* standard MMX */
419 #define MM_3DNOW 0x0004 /* AMD 3DNOW */
420 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
421 #define MM_SSE 0x0008 /* SSE functions */
422 #define MM_SSE2 0x0010 /* PIV SSE2 functions */
423 #define MM_3DNOWEXT 0x0020 /* AMD 3DNowExt */
424
425 extern int mm_flags;
426
427 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
428 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
429 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
430
431 static inline void emms(void)
432 {
433 __asm __volatile ("emms;":::"memory");
434 }
435
436
437 #define emms_c() \
438 {\
439 if (mm_flags & MM_MMX)\
440 emms();\
441 }
442
443 #ifdef __GNUC__
444 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
445 #else
446 #define DECLARE_ALIGNED_8(t,v) __declspec(align(8)) t v
447 #endif
448
449 #define STRIDE_ALIGN 8
450
451 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
452 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
453
454 #elif defined(ARCH_ARMV4L)
455
456 /* This is to use 4 bytes read to the IDCT pointers for some 'zero'
457 line optimizations */
458 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (4)))
459 #define STRIDE_ALIGN 4
460
461 #define MM_IWMMXT 0x0100 /* XScale IWMMXT */
462
463 extern int mm_flags;
464
465 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
466
467 #elif defined(HAVE_MLIB)
468
469 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
470 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
471 #define STRIDE_ALIGN 8
472
473 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
474
475 #elif defined(ARCH_SPARC)
476
477 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
478 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
479 #define STRIDE_ALIGN 8
480 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
481
482 #elif defined(ARCH_ALPHA)
483
484 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
485 #define STRIDE_ALIGN 8
486
487 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
488
489 #elif defined(ARCH_POWERPC)
490
491 #define MM_ALTIVEC 0x0001 /* standard AltiVec */
492
493 extern int mm_flags;
494
495 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
496 #define pixel altivec_pixel
497 #include <altivec.h>
498 #undef pixel
499 #endif
500
501 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16)))
502 #define STRIDE_ALIGN 16
503
504 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
505
506 #elif defined(HAVE_MMI)
507
508 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16)))
509 #define STRIDE_ALIGN 16
510
511 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
512
513 #elif defined(ARCH_SH4)
514
515 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
516 #define STRIDE_ALIGN 8
517
518 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
519
520 #else
521
522 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
523 #define STRIDE_ALIGN 8
524
525 #endif
526
527 #ifdef __GNUC__
528
529 struct unaligned_64 { uint64_t l; } __attribute__((packed));
530 struct unaligned_32 { uint32_t l; } __attribute__((packed));
531 struct unaligned_16 { uint16_t l; } __attribute__((packed));
532
533 #define LD16(a) (((const struct unaligned_16 *) (a))->l)
534 #define LD32(a) (((const struct unaligned_32 *) (a))->l)
535 #define LD64(a) (((const struct unaligned_64 *) (a))->l)
536
537 #define ST16(a, b) (((struct unaligned_16 *) (a))->l) = (b)
538 #define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
539
540 #else /* __GNUC__ */
541
542 #define LD16(a) (*((uint16_t*)(a)))
543 #define LD32(a) (*((uint32_t*)(a)))
544 #define LD64(a) (*((uint64_t*)(a)))
545
546 #define ST16(a, b) *((uint16_t*)(a)) = (b)
547 #define ST32(a, b) *((uint32_t*)(a)) = (b)
548
549 #endif /* !__GNUC__ */
550
551 /* PSNR */
552 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
553 int orig_linesize[3], int coded_linesize,
554 AVCodecContext *avctx);
555
556 /* FFT computation */
557
558 /* NOTE: soon integer code will be added, so you must use the
559 FFTSample type */
560 typedef float FFTSample;
561
562 typedef struct FFTComplex {
563 FFTSample re, im;
564 } FFTComplex;
565
566 typedef struct FFTContext {
567 int nbits;
568 int inverse;
569 uint16_t *revtab;
570 FFTComplex *exptab;
571 FFTComplex *exptab1; /* only used by SSE code */
572 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
573 } FFTContext;
574
575 int ff_fft_init(FFTContext *s, int nbits, int inverse);
576 void ff_fft_permute(FFTContext *s, FFTComplex *z);
577 void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
578 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
579 void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
580 void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
581 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
582
583 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
584 {
585 s->fft_calc(s, z);
586 }
587 void ff_fft_end(FFTContext *s);
588
589 /* MDCT computation */
590
591 typedef struct MDCTContext {
592 int n; /* size of MDCT (i.e. number of input data * 2) */
593 int nbits; /* n = 2^nbits */
594 /* pre/post rotation tables */
595 FFTSample *tcos;
596 FFTSample *tsin;
597 FFTContext fft;
598 } MDCTContext;
599
600 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
601 void ff_imdct_calc(MDCTContext *s, FFTSample *output,
602 const FFTSample *input, FFTSample *tmp);
603 void ff_mdct_calc(MDCTContext *s, FFTSample *out,
604 const FFTSample *input, FFTSample *tmp);
605 void ff_mdct_end(MDCTContext *s);
606
607 #define WARPER8_16(name8, name16)\
608 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
609 return name8(s, dst , src , stride, h)\
610 +name8(s, dst+8 , src+8 , stride, h);\
611 }
612
613 #define WARPER8_16_SQ(name8, name16)\
614 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
615 int score=0;\
616 score +=name8(s, dst , src , stride, 8);\
617 score +=name8(s, dst+8 , src+8 , stride, 8);\
618 if(h==16){\
619 dst += 8*stride;\
620 src += 8*stride;\
621 score +=name8(s, dst , src , stride, 8);\
622 score +=name8(s, dst+8 , src+8 , stride, 8);\
623 }\
624 return score;\
625 }
626
627 #endif