remove mpegvideo.c img resample dependancy
[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 /* 1/2^n downscaling functions from imgconvert.c */
80 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
81 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
82 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
83 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
84
85 /* minimum alignment rules ;)
86 if u notice errors in the align stuff, need more alignment for some asm code for some cpu
87 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
88
89 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
90 i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
91
92 !future video codecs might need functions with less strict alignment
93 */
94
95 /*
96 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
97 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
98 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
99 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
100 void clear_blocks_c(DCTELEM *blocks);
101 */
102
103 /* add and put pixel (decoding) */
104 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
105 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
106 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
107 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);
108 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
109 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
110 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
111 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
112
113 #define DEF_OLD_QPEL(name)\
114 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
115 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
116 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
117
118 DEF_OLD_QPEL(qpel16_mc11_old_c)
119 DEF_OLD_QPEL(qpel16_mc31_old_c)
120 DEF_OLD_QPEL(qpel16_mc12_old_c)
121 DEF_OLD_QPEL(qpel16_mc32_old_c)
122 DEF_OLD_QPEL(qpel16_mc13_old_c)
123 DEF_OLD_QPEL(qpel16_mc33_old_c)
124 DEF_OLD_QPEL(qpel8_mc11_old_c)
125 DEF_OLD_QPEL(qpel8_mc31_old_c)
126 DEF_OLD_QPEL(qpel8_mc12_old_c)
127 DEF_OLD_QPEL(qpel8_mc32_old_c)
128 DEF_OLD_QPEL(qpel8_mc13_old_c)
129 DEF_OLD_QPEL(qpel8_mc33_old_c)
130
131 #define CALL_2X_PIXELS(a, b, n)\
132 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
133 b(block , pixels , line_size, h);\
134 b(block+n, pixels+n, line_size, h);\
135 }
136
137 /* motion estimation */
138 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
139 // allthough currently h<4 is not used as functions with width <8 are not used and neither implemented
140 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))*/;
141
142
143 // for snow slices
144 typedef struct slice_buffer_s slice_buffer;
145
146 /**
147 * DSPContext.
148 */
149 typedef struct DSPContext {
150 /* pixel ops : interface with DCT */
151 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
152 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
153 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
154 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
155 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
156 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
157 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
158 /**
159 * translational global motion compensation.
160 */
161 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
162 /**
163 * global motion compensation.
164 */
165 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
166 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
167 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
168 int (*pix_sum)(uint8_t * pix, int line_size);
169 int (*pix_norm1)(uint8_t * pix, int line_size);
170 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
171
172 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
173 me_cmp_func sse[5];
174 me_cmp_func hadamard8_diff[5];
175 me_cmp_func dct_sad[5];
176 me_cmp_func quant_psnr[5];
177 me_cmp_func bit[5];
178 me_cmp_func rd[5];
179 me_cmp_func vsad[5];
180 me_cmp_func vsse[5];
181 me_cmp_func nsse[5];
182 me_cmp_func w53[5];
183 me_cmp_func w97[5];
184 me_cmp_func dct_max[5];
185 me_cmp_func dct264_sad[5];
186
187 me_cmp_func me_pre_cmp[5];
188 me_cmp_func me_cmp[5];
189 me_cmp_func me_sub_cmp[5];
190 me_cmp_func mb_cmp[5];
191 me_cmp_func ildct_cmp[5]; //only width 16 used
192 me_cmp_func frame_skip_cmp[5]; //only width 8 used
193
194 /**
195 * Halfpel motion compensation with rounding (a+b+1)>>1.
196 * this is an array[4][4] of motion compensation funcions for 4
197 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
198 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
199 * @param block destination where the result is stored
200 * @param pixels source
201 * @param line_size number of bytes in a horizontal line of block
202 * @param h height
203 */
204 op_pixels_func put_pixels_tab[4][4];
205
206 /**
207 * Halfpel motion compensation with rounding (a+b+1)>>1.
208 * This is an array[4][4] of motion compensation functions for 4
209 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
210 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
211 * @param block destination into which the result is averaged (a+b+1)>>1
212 * @param pixels source
213 * @param line_size number of bytes in a horizontal line of block
214 * @param h height
215 */
216 op_pixels_func avg_pixels_tab[4][4];
217
218 /**
219 * Halfpel motion compensation with no rounding (a+b)>>1.
220 * this is an array[2][4] of motion compensation funcions for 2
221 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
222 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
223 * @param block destination where the result is stored
224 * @param pixels source
225 * @param line_size number of bytes in a horizontal line of block
226 * @param h height
227 */
228 op_pixels_func put_no_rnd_pixels_tab[4][4];
229
230 /**
231 * Halfpel motion compensation with no rounding (a+b)>>1.
232 * this is an array[2][4] of motion compensation funcions for 2
233 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
234 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
235 * @param block destination into which the result is averaged (a+b)>>1
236 * @param pixels source
237 * @param line_size number of bytes in a horizontal line of block
238 * @param h height
239 */
240 op_pixels_func avg_no_rnd_pixels_tab[4][4];
241
242 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);
243
244 /**
245 * Thirdpel motion compensation with rounding (a+b+1)>>1.
246 * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
247 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
248 * @param block destination where the result is stored
249 * @param pixels source
250 * @param line_size number of bytes in a horizontal line of block
251 * @param h height
252 */
253 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
254 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
255
256 qpel_mc_func put_qpel_pixels_tab[2][16];
257 qpel_mc_func avg_qpel_pixels_tab[2][16];
258 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
259 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
260 qpel_mc_func put_mspel_pixels_tab[8];
261
262 /**
263 * h264 Chram MC
264 */
265 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
266 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
267
268 qpel_mc_func put_h264_qpel_pixels_tab[4][16];
269 qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
270
271 h264_weight_func weight_h264_pixels_tab[10];
272 h264_biweight_func biweight_h264_pixels_tab[10];
273
274 me_cmp_func pix_abs[2][4];
275
276 /* huffyuv specific */
277 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
278 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
279 /**
280 * subtract huffyuv's variant of median prediction
281 * note, this might read from src1[-1], src2[-1]
282 */
283 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
284 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
285
286 void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
287 void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
288 void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
289 void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
290 void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
291 void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
292
293 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
294 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
295
296 void (*h261_loop_filter)(uint8_t *src, int stride);
297
298 /* (I)DCT */
299 void (*fdct)(DCTELEM *block/* align 16*/);
300 void (*fdct248)(DCTELEM *block/* align 16*/);
301
302 /* IDCT really*/
303 void (*idct)(DCTELEM *block/* align 16*/);
304
305 /**
306 * block -> idct -> clip to unsigned 8 bit -> dest.
307 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
308 * @param line_size size in bytes of a horizotal line of dest
309 */
310 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
311
312 /**
313 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
314 * @param line_size size in bytes of a horizotal line of dest
315 */
316 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
317
318 /**
319 * idct input permutation.
320 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
321 * IDCT)
322 * this permutation must be performed before the idct_put/add, note, normally this can be merged
323 * with the zigzag/alternate scan<br>
324 * an example to avoid confusion:
325 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
326 * - (x -> referece dct -> reference idct -> x)
327 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
328 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
329 */
330 uint8_t idct_permutation[64];
331 int idct_permutation_type;
332 #define FF_NO_IDCT_PERM 1
333 #define FF_LIBMPEG2_IDCT_PERM 2
334 #define FF_SIMPLE_IDCT_PERM 3
335 #define FF_TRANSPOSE_IDCT_PERM 4
336 #define FF_PARTTRANS_IDCT_PERM 5
337
338 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
339 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
340 #define BASIS_SHIFT 16
341 #define RECON_SHIFT 6
342
343 void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
344 void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
345 void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
346 void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
347
348 /* snow wavelet */
349 void (*vertical_compose97i)(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width);
350 void (*horizontal_compose97i)(DWTELEM *b, int width);
351 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);
352
353 void (*prefetch)(void *mem, int stride, int h);
354
355 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
356 } DSPContext;
357
358 void dsputil_static_init(void);
359 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
360
361 /**
362 * permute block according to permuatation.
363 * @param last last non zero element in scantable order
364 */
365 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
366
367 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
368
369 #define BYTE_VEC32(c) ((c)*0x01010101UL)
370
371 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
372 {
373 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
374 }
375
376 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
377 {
378 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
379 }
380
381 static inline int get_penalty_factor(int lambda, int lambda2, int type){
382 switch(type&0xFF){
383 default:
384 case FF_CMP_SAD:
385 return lambda>>FF_LAMBDA_SHIFT;
386 case FF_CMP_DCT:
387 return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
388 case FF_CMP_W53:
389 return (4*lambda)>>(FF_LAMBDA_SHIFT);
390 case FF_CMP_W97:
391 return (2*lambda)>>(FF_LAMBDA_SHIFT);
392 case FF_CMP_SATD:
393 case FF_CMP_DCT264:
394 return (2*lambda)>>FF_LAMBDA_SHIFT;
395 case FF_CMP_RD:
396 case FF_CMP_PSNR:
397 case FF_CMP_SSE:
398 case FF_CMP_NSSE:
399 return lambda2>>FF_LAMBDA_SHIFT;
400 case FF_CMP_BIT:
401 return 1;
402 }
403 }
404
405 /**
406 * Empty mmx state.
407 * this must be called between any dsp function and float/double code.
408 * for example sin(); dsp->idct_put(); emms_c(); cos()
409 */
410 #define emms_c()
411
412 /* should be defined by architectures supporting
413 one or more MultiMedia extension */
414 int mm_support(void);
415
416 #ifdef __GNUC__
417 #define DECLARE_ALIGNED_16(t,v) t v __attribute__ ((aligned (16)))
418 #else
419 #define DECLARE_ALIGNED_16(t,v) __declspec(align(16)) t v
420 #endif
421
422 #if defined(HAVE_MMX)
423
424 #undef emms_c
425
426 #define MM_MMX 0x0001 /* standard MMX */
427 #define MM_3DNOW 0x0004 /* AMD 3DNOW */
428 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
429 #define MM_SSE 0x0008 /* SSE functions */
430 #define MM_SSE2 0x0010 /* PIV SSE2 functions */
431 #define MM_3DNOWEXT 0x0020 /* AMD 3DNowExt */
432
433 extern int mm_flags;
434
435 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
436 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
437 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
438
439 static inline void emms(void)
440 {
441 __asm __volatile ("emms;":::"memory");
442 }
443
444
445 #define emms_c() \
446 {\
447 if (mm_flags & MM_MMX)\
448 emms();\
449 }
450
451 #ifdef __GNUC__
452 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
453 #else
454 #define DECLARE_ALIGNED_8(t,v) __declspec(align(8)) t v
455 #endif
456
457 #define STRIDE_ALIGN 8
458
459 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
460 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
461
462 #elif defined(ARCH_ARMV4L)
463
464 /* This is to use 4 bytes read to the IDCT pointers for some 'zero'
465 line optimizations */
466 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (4)))
467 #define STRIDE_ALIGN 4
468
469 #define MM_IWMMXT 0x0100 /* XScale IWMMXT */
470
471 extern int mm_flags;
472
473 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
474
475 #elif defined(HAVE_MLIB)
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
481 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
482
483 #elif defined(ARCH_SPARC)
484
485 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
486 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
487 #define STRIDE_ALIGN 8
488 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
489
490 #elif defined(ARCH_ALPHA)
491
492 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
493 #define STRIDE_ALIGN 8
494
495 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
496
497 #elif defined(ARCH_POWERPC)
498
499 #define MM_ALTIVEC 0x0001 /* standard AltiVec */
500
501 extern int mm_flags;
502
503 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
504 #define pixel altivec_pixel
505 #include <altivec.h>
506 #undef pixel
507 #endif
508
509 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16)))
510 #define STRIDE_ALIGN 16
511
512 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
513
514 #elif defined(HAVE_MMI)
515
516 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16)))
517 #define STRIDE_ALIGN 16
518
519 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
520
521 #elif defined(ARCH_SH4)
522
523 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
524 #define STRIDE_ALIGN 8
525
526 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
527
528 #else
529
530 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
531 #define STRIDE_ALIGN 8
532
533 #endif
534
535 #ifdef __GNUC__
536
537 struct unaligned_64 { uint64_t l; } __attribute__((packed));
538 struct unaligned_32 { uint32_t l; } __attribute__((packed));
539 struct unaligned_16 { uint16_t l; } __attribute__((packed));
540
541 #define LD16(a) (((const struct unaligned_16 *) (a))->l)
542 #define LD32(a) (((const struct unaligned_32 *) (a))->l)
543 #define LD64(a) (((const struct unaligned_64 *) (a))->l)
544
545 #define ST16(a, b) (((struct unaligned_16 *) (a))->l) = (b)
546 #define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
547
548 #else /* __GNUC__ */
549
550 #define LD16(a) (*((uint16_t*)(a)))
551 #define LD32(a) (*((uint32_t*)(a)))
552 #define LD64(a) (*((uint64_t*)(a)))
553
554 #define ST16(a, b) *((uint16_t*)(a)) = (b)
555 #define ST32(a, b) *((uint32_t*)(a)) = (b)
556
557 #endif /* !__GNUC__ */
558
559 /* PSNR */
560 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
561 int orig_linesize[3], int coded_linesize,
562 AVCodecContext *avctx);
563
564 /* FFT computation */
565
566 /* NOTE: soon integer code will be added, so you must use the
567 FFTSample type */
568 typedef float FFTSample;
569
570 typedef struct FFTComplex {
571 FFTSample re, im;
572 } FFTComplex;
573
574 typedef struct FFTContext {
575 int nbits;
576 int inverse;
577 uint16_t *revtab;
578 FFTComplex *exptab;
579 FFTComplex *exptab1; /* only used by SSE code */
580 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
581 } FFTContext;
582
583 int ff_fft_init(FFTContext *s, int nbits, int inverse);
584 void ff_fft_permute(FFTContext *s, FFTComplex *z);
585 void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
586 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
587 void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
588 void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
589 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
590
591 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
592 {
593 s->fft_calc(s, z);
594 }
595 void ff_fft_end(FFTContext *s);
596
597 /* MDCT computation */
598
599 typedef struct MDCTContext {
600 int n; /* size of MDCT (i.e. number of input data * 2) */
601 int nbits; /* n = 2^nbits */
602 /* pre/post rotation tables */
603 FFTSample *tcos;
604 FFTSample *tsin;
605 FFTContext fft;
606 } MDCTContext;
607
608 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
609 void ff_imdct_calc(MDCTContext *s, FFTSample *output,
610 const FFTSample *input, FFTSample *tmp);
611 void ff_mdct_calc(MDCTContext *s, FFTSample *out,
612 const FFTSample *input, FFTSample *tmp);
613 void ff_mdct_end(MDCTContext *s);
614
615 #define WARPER8_16(name8, name16)\
616 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
617 return name8(s, dst , src , stride, h)\
618 +name8(s, dst+8 , src+8 , stride, h);\
619 }
620
621 #define WARPER8_16_SQ(name8, name16)\
622 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
623 int score=0;\
624 score +=name8(s, dst , src , stride, 8);\
625 score +=name8(s, dst+8 , src+8 , stride, 8);\
626 if(h==16){\
627 dst += 8*stride;\
628 src += 8*stride;\
629 score +=name8(s, dst , src , stride, 8);\
630 score +=name8(s, dst+8 , src+8 , stride, 8);\
631 }\
632 return score;\
633 }
634
635 #endif