Remove duplicated MM_* macros for CPU capabilities from dsputil.h.
[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 file is part of FFmpeg.
7 *
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
12 *
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22
23 /**
24 * @file dsputil.h
25 * DSP utils.
26 * note, many functions in here may use MMX which trashes the FPU state, it is
27 * absolutely necessary to call emms_c() between dsp & float/double code
28 */
29
30 #ifndef AVCODEC_DSPUTIL_H
31 #define AVCODEC_DSPUTIL_H
32
33 #include "avcodec.h"
34
35
36 //#define DEBUG
37 /* dct code */
38 typedef short DCTELEM;
39 typedef int DWTELEM;
40 typedef short IDWTELEM;
41
42 void fdct_ifast (DCTELEM *data);
43 void fdct_ifast248 (DCTELEM *data);
44 void ff_jpeg_fdct_islow (DCTELEM *data);
45 void ff_fdct248_islow (DCTELEM *data);
46
47 void j_rev_dct (DCTELEM *data);
48 void j_rev_dct4 (DCTELEM *data);
49 void j_rev_dct2 (DCTELEM *data);
50 void j_rev_dct1 (DCTELEM *data);
51 void ff_wmv2_idct_c(DCTELEM *data);
52
53 void ff_fdct_mmx(DCTELEM *block);
54 void ff_fdct_mmx2(DCTELEM *block);
55 void ff_fdct_sse2(DCTELEM *block);
56
57 void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
58 void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
59 void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
60 void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
61 void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
62 void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
63
64 void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1,
65 const float *src2, int src3, int blocksize, int step);
66 void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1,
67 const float *win, float add_bias, int len);
68 void ff_float_to_int16_c(int16_t *dst, const float *src, long len);
69 void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels);
70
71 /* encoding scans */
72 extern const uint8_t ff_alternate_horizontal_scan[64];
73 extern const uint8_t ff_alternate_vertical_scan[64];
74 extern const uint8_t ff_zigzag_direct[64];
75 extern const uint8_t ff_zigzag248_direct[64];
76
77 /* pixel operations */
78 #define MAX_NEG_CROP 1024
79
80 /* temporary */
81 extern uint32_t ff_squareTbl[512];
82 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
83
84 /* VP3 DSP functions */
85 void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
86 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
87 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
88
89 void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
90 void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
91
92 /* 1/2^n downscaling functions from imgconvert.c */
93 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
94 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
95 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
96 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
97
98 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
99 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
100
101 /* minimum alignment rules ;)
102 If you notice errors in the align stuff, need more alignment for some ASM code
103 for some CPU or need to use a function with less aligned data then send a mail
104 to the ffmpeg-devel mailing list, ...
105
106 !warning These alignments might not match reality, (missing attribute((align))
107 stuff somewhere possible).
108 I (Michael) did not check them, these are just the alignments which I think
109 could be reached easily ...
110
111 !future video codecs might need functions with less strict alignment
112 */
113
114 /*
115 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
116 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
117 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
118 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
119 void clear_blocks_c(DCTELEM *blocks);
120 */
121
122 /* add and put pixel (decoding) */
123 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
124 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
125 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
126 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);
127 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
128 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
129 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
130 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
131
132 #define DEF_OLD_QPEL(name)\
133 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
134 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
135 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
136
137 DEF_OLD_QPEL(qpel16_mc11_old_c)
138 DEF_OLD_QPEL(qpel16_mc31_old_c)
139 DEF_OLD_QPEL(qpel16_mc12_old_c)
140 DEF_OLD_QPEL(qpel16_mc32_old_c)
141 DEF_OLD_QPEL(qpel16_mc13_old_c)
142 DEF_OLD_QPEL(qpel16_mc33_old_c)
143 DEF_OLD_QPEL(qpel8_mc11_old_c)
144 DEF_OLD_QPEL(qpel8_mc31_old_c)
145 DEF_OLD_QPEL(qpel8_mc12_old_c)
146 DEF_OLD_QPEL(qpel8_mc32_old_c)
147 DEF_OLD_QPEL(qpel8_mc13_old_c)
148 DEF_OLD_QPEL(qpel8_mc33_old_c)
149
150 #define CALL_2X_PIXELS(a, b, n)\
151 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
152 b(block , pixels , line_size, h);\
153 b(block+n, pixels+n, line_size, h);\
154 }
155
156 /* motion estimation */
157 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
158 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
159 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))*/;
160
161
162 // for snow slices
163 typedef struct slice_buffer_s slice_buffer;
164
165 /**
166 * Scantable.
167 */
168 typedef struct ScanTable{
169 const uint8_t *scantable;
170 uint8_t permutated[64];
171 uint8_t raster_end[64];
172 #ifdef ARCH_POWERPC
173 /** Used by dct_quantize_altivec to find last-non-zero */
174 DECLARE_ALIGNED(16, uint8_t, inverse[64]);
175 #endif
176 } ScanTable;
177
178 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
179
180 void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize,
181 int block_w, int block_h,
182 int src_x, int src_y, int w, int h);
183
184 /**
185 * DSPContext.
186 */
187 typedef struct DSPContext {
188 /* pixel ops : interface with DCT */
189 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
190 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
191 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
192 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
193 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
194 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
195 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
196 int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
197 /**
198 * translational global motion compensation.
199 */
200 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
201 /**
202 * global motion compensation.
203 */
204 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
205 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
206 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
207 int (*pix_sum)(uint8_t * pix, int line_size);
208 int (*pix_norm1)(uint8_t * pix, int line_size);
209 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
210
211 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
212 me_cmp_func sse[5];
213 me_cmp_func hadamard8_diff[5];
214 me_cmp_func dct_sad[5];
215 me_cmp_func quant_psnr[5];
216 me_cmp_func bit[5];
217 me_cmp_func rd[5];
218 me_cmp_func vsad[5];
219 me_cmp_func vsse[5];
220 me_cmp_func nsse[5];
221 me_cmp_func w53[5];
222 me_cmp_func w97[5];
223 me_cmp_func dct_max[5];
224 me_cmp_func dct264_sad[5];
225
226 me_cmp_func me_pre_cmp[5];
227 me_cmp_func me_cmp[5];
228 me_cmp_func me_sub_cmp[5];
229 me_cmp_func mb_cmp[5];
230 me_cmp_func ildct_cmp[5]; //only width 16 used
231 me_cmp_func frame_skip_cmp[5]; //only width 8 used
232
233 int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
234 int size);
235
236 /**
237 * Halfpel motion compensation with rounding (a+b+1)>>1.
238 * this is an array[4][4] of motion compensation functions for 4
239 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
240 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
241 * @param block destination where the result is stored
242 * @param pixels source
243 * @param line_size number of bytes in a horizontal line of block
244 * @param h height
245 */
246 op_pixels_func put_pixels_tab[4][4];
247
248 /**
249 * Halfpel motion compensation with rounding (a+b+1)>>1.
250 * This is an array[4][4] of motion compensation functions for 4
251 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
252 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
253 * @param block destination into which the result is averaged (a+b+1)>>1
254 * @param pixels source
255 * @param line_size number of bytes in a horizontal line of block
256 * @param h height
257 */
258 op_pixels_func avg_pixels_tab[4][4];
259
260 /**
261 * Halfpel motion compensation with no rounding (a+b)>>1.
262 * this is an array[2][4] of motion compensation functions for 2
263 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
264 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
265 * @param block destination where the result is stored
266 * @param pixels source
267 * @param line_size number of bytes in a horizontal line of block
268 * @param h height
269 */
270 op_pixels_func put_no_rnd_pixels_tab[4][4];
271
272 /**
273 * Halfpel motion compensation with no rounding (a+b)>>1.
274 * this is an array[2][4] of motion compensation functions for 2
275 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
276 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
277 * @param block destination into which the result is averaged (a+b)>>1
278 * @param pixels source
279 * @param line_size number of bytes in a horizontal line of block
280 * @param h height
281 */
282 op_pixels_func avg_no_rnd_pixels_tab[4][4];
283
284 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);
285
286 /**
287 * Thirdpel motion compensation with rounding (a+b+1)>>1.
288 * this is an array[12] of motion compensation functions for the 9 thirdpe
289 * positions<br>
290 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
291 * @param block destination where the result is stored
292 * @param pixels source
293 * @param line_size number of bytes in a horizontal line of block
294 * @param h height
295 */
296 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
297 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
298
299 qpel_mc_func put_qpel_pixels_tab[2][16];
300 qpel_mc_func avg_qpel_pixels_tab[2][16];
301 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
302 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
303 qpel_mc_func put_mspel_pixels_tab[8];
304
305 /**
306 * h264 Chroma MC
307 */
308 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
309 /* This is really one func used in VC-1 decoding */
310 h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
311 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
312
313 qpel_mc_func put_h264_qpel_pixels_tab[4][16];
314 qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
315
316 qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
317 qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
318
319 h264_weight_func weight_h264_pixels_tab[10];
320 h264_biweight_func biweight_h264_pixels_tab[10];
321
322 /* AVS specific */
323 qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
324 qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
325 void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
326 void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
327 void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
328 void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
329 void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
330
331 me_cmp_func pix_abs[2][4];
332
333 /* huffyuv specific */
334 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
335 void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
336 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
337 /**
338 * subtract huffyuv's variant of median prediction
339 * note, this might read from src1[-1], src2[-1]
340 */
341 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
342 /* this might write to dst[w] */
343 void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
344 void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
345
346 void (*h264_v_loop_filter_luma)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta, int8_t *tc0);
347 void (*h264_h_loop_filter_luma)(uint8_t *pix/*align 4 */, int stride, int alpha, int beta, int8_t *tc0);
348 /* v/h_loop_filter_luma_intra: align 16 */
349 void (*h264_v_loop_filter_chroma)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta, int8_t *tc0);
350 void (*h264_h_loop_filter_chroma)(uint8_t *pix/*align 4*/, int stride, int alpha, int beta, int8_t *tc0);
351 void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
352 void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
353 // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
354 void (*h264_loop_filter_strength)(int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],
355 int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field);
356
357 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
358 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
359
360 void (*h261_loop_filter)(uint8_t *src, int stride);
361
362 void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
363 void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
364
365 void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
366 void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
367
368 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
369 void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
370 void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
371 /* no alignment needed */
372 void (*flac_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
373 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
374 void (*vector_fmul)(float *dst, const float *src, int len);
375 void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
376 /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
377 void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
378 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
379 void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len);
380 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
381 void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len);
382
383 /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
384 * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
385 void (*float_to_int16)(int16_t *dst, const float *src, long len);
386 void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
387
388 /* (I)DCT */
389 void (*fdct)(DCTELEM *block/* align 16*/);
390 void (*fdct248)(DCTELEM *block/* align 16*/);
391
392 /* IDCT really*/
393 void (*idct)(DCTELEM *block/* align 16*/);
394
395 /**
396 * block -> idct -> clip to unsigned 8 bit -> dest.
397 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
398 * @param line_size size in bytes of a horizontal line of dest
399 */
400 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
401
402 /**
403 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
404 * @param line_size size in bytes of a horizontal line of dest
405 */
406 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
407
408 /**
409 * idct input permutation.
410 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
411 * IDCT)
412 * this permutation must be performed before the idct_put/add, note, normally this can be merged
413 * with the zigzag/alternate scan<br>
414 * an example to avoid confusion:
415 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
416 * - (x -> referece dct -> reference idct -> x)
417 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
418 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
419 */
420 uint8_t idct_permutation[64];
421 int idct_permutation_type;
422 #define FF_NO_IDCT_PERM 1
423 #define FF_LIBMPEG2_IDCT_PERM 2
424 #define FF_SIMPLE_IDCT_PERM 3
425 #define FF_TRANSPOSE_IDCT_PERM 4
426 #define FF_PARTTRANS_IDCT_PERM 5
427 #define FF_SSE2_IDCT_PERM 6
428
429 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
430 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
431 #define BASIS_SHIFT 16
432 #define RECON_SHIFT 6
433
434 void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
435 #define EDGE_WIDTH 16
436
437 /* h264 functions */
438 void (*h264_idct_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
439 void (*h264_idct8_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
440 void (*h264_idct_dc_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
441 void (*h264_idct8_dc_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
442 void (*h264_dct)(DCTELEM block[4][4]);
443
444 /* snow wavelet */
445 void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
446 void (*horizontal_compose97i)(IDWTELEM *b, int width);
447 void (*inner_add_yblock)(const 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);
448
449 void (*prefetch)(void *mem, int stride, int h);
450
451 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
452
453 /* vc1 functions */
454 void (*vc1_inv_trans_8x8)(DCTELEM *b);
455 void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
456 void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
457 void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
458 void (*vc1_v_overlap)(uint8_t* src, int stride);
459 void (*vc1_h_overlap)(uint8_t* src, int stride);
460 /* put 8x8 block with bicubic interpolation and quarterpel precision
461 * last argument is actually round value instead of height
462 */
463 op_pixels_func put_vc1_mspel_pixels_tab[16];
464
465 /* intrax8 functions */
466 void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
467 void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
468 int * range, int * sum, int edges);
469
470 /* ape functions */
471 /**
472 * Add contents of the second vector to the first one.
473 * @param len length of vectors, should be multiple of 16
474 */
475 void (*add_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
476 /**
477 * Add contents of the second vector to the first one.
478 * @param len length of vectors, should be multiple of 16
479 */
480 void (*sub_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
481 /**
482 * Calculate scalar product of two vectors.
483 * @param len length of vectors, should be multiple of 16
484 * @param shift number of bits to discard from product
485 */
486 int32_t (*scalarproduct_int16)(int16_t *v1, int16_t *v2/*align 16*/, int len, int shift);
487 } DSPContext;
488
489 void dsputil_static_init(void);
490 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
491
492 int ff_check_alignment(void);
493
494 /**
495 * permute block according to permuatation.
496 * @param last last non zero element in scantable order
497 */
498 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
499
500 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
501
502 #define BYTE_VEC32(c) ((c)*0x01010101UL)
503
504 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
505 {
506 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
507 }
508
509 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
510 {
511 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
512 }
513
514 static inline int get_penalty_factor(int lambda, int lambda2, int type){
515 switch(type&0xFF){
516 default:
517 case FF_CMP_SAD:
518 return lambda>>FF_LAMBDA_SHIFT;
519 case FF_CMP_DCT:
520 return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
521 case FF_CMP_W53:
522 return (4*lambda)>>(FF_LAMBDA_SHIFT);
523 case FF_CMP_W97:
524 return (2*lambda)>>(FF_LAMBDA_SHIFT);
525 case FF_CMP_SATD:
526 case FF_CMP_DCT264:
527 return (2*lambda)>>FF_LAMBDA_SHIFT;
528 case FF_CMP_RD:
529 case FF_CMP_PSNR:
530 case FF_CMP_SSE:
531 case FF_CMP_NSSE:
532 return lambda2>>FF_LAMBDA_SHIFT;
533 case FF_CMP_BIT:
534 return 1;
535 }
536 }
537
538 /**
539 * Empty mmx state.
540 * this must be called between any dsp function and float/double code.
541 * for example sin(); dsp->idct_put(); emms_c(); cos()
542 */
543 #define emms_c()
544
545 /* should be defined by architectures supporting
546 one or more MultiMedia extension */
547 int mm_support(void);
548
549 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
550 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
551 void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
552 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
553 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
554 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
555 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
556 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
557 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
558
559 #define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v)
560
561 #if defined(HAVE_MMX)
562
563 #undef emms_c
564
565 extern int mm_flags;
566
567 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
568 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
569 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
570
571 static inline void emms(void)
572 {
573 __asm__ volatile ("emms;":::"memory");
574 }
575
576
577 #define emms_c() \
578 {\
579 if (mm_flags & FF_MM_MMX)\
580 emms();\
581 }
582
583 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
584
585 #elif defined(ARCH_ARMV4L)
586
587 extern int mm_flags;
588
589 #ifdef HAVE_NEON
590 # define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
591 # define STRIDE_ALIGN 16
592 #endif
593
594 #elif defined(ARCH_POWERPC)
595
596 extern int mm_flags;
597
598 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
599 #define STRIDE_ALIGN 16
600
601 #elif defined(HAVE_MMI)
602
603 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
604 #define STRIDE_ALIGN 16
605
606 #else
607
608 #define mm_flags 0
609 #define mm_support() 0
610
611 #endif
612
613 #ifndef DECLARE_ALIGNED_8
614 # define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
615 #endif
616
617 #ifndef STRIDE_ALIGN
618 # define STRIDE_ALIGN 8
619 #endif
620
621 /* PSNR */
622 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
623 int orig_linesize[3], int coded_linesize,
624 AVCodecContext *avctx);
625
626 /* FFT computation */
627
628 /* NOTE: soon integer code will be added, so you must use the
629 FFTSample type */
630 typedef float FFTSample;
631
632 struct MDCTContext;
633
634 typedef struct FFTComplex {
635 FFTSample re, im;
636 } FFTComplex;
637
638 typedef struct FFTContext {
639 int nbits;
640 int inverse;
641 uint16_t *revtab;
642 FFTComplex *exptab;
643 FFTComplex *exptab1; /* only used by SSE code */
644 FFTComplex *tmp_buf;
645 void (*fft_permute)(struct FFTContext *s, FFTComplex *z);
646 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
647 void (*imdct_calc)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
648 void (*imdct_half)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
649 } FFTContext;
650
651 int ff_fft_init(FFTContext *s, int nbits, int inverse);
652 void ff_fft_permute_c(FFTContext *s, FFTComplex *z);
653 void ff_fft_permute_sse(FFTContext *s, FFTComplex *z);
654 void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
655 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
656 void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
657 void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
658 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
659
660 static inline void ff_fft_permute(FFTContext *s, FFTComplex *z)
661 {
662 s->fft_permute(s, z);
663 }
664 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
665 {
666 s->fft_calc(s, z);
667 }
668 void ff_fft_end(FFTContext *s);
669
670 /* MDCT computation */
671
672 typedef struct MDCTContext {
673 int n; /* size of MDCT (i.e. number of input data * 2) */
674 int nbits; /* n = 2^nbits */
675 /* pre/post rotation tables */
676 FFTSample *tcos;
677 FFTSample *tsin;
678 FFTContext fft;
679 } MDCTContext;
680
681 static inline void ff_imdct_calc(MDCTContext *s, FFTSample *output, const FFTSample *input)
682 {
683 s->fft.imdct_calc(s, output, input);
684 }
685 static inline void ff_imdct_half(MDCTContext *s, FFTSample *output, const FFTSample *input)
686 {
687 s->fft.imdct_half(s, output, input);
688 }
689
690 /**
691 * Generate a Kaiser-Bessel Derived Window.
692 * @param window pointer to half window
693 * @param alpha determines window shape
694 * @param n size of half window
695 */
696 void ff_kbd_window_init(float *window, float alpha, int n);
697
698 /**
699 * Generate a sine window.
700 * @param window pointer to half window
701 * @param n size of half window
702 */
703 void ff_sine_window_init(float *window, int n);
704 extern float ff_sine_128 [ 128];
705 extern float ff_sine_256 [ 256];
706 extern float ff_sine_512 [ 512];
707 extern float ff_sine_1024[1024];
708 extern float ff_sine_2048[2048];
709 extern float *ff_sine_windows[5];
710
711 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
712 void ff_imdct_calc_c(MDCTContext *s, FFTSample *output, const FFTSample *input);
713 void ff_imdct_half_c(MDCTContext *s, FFTSample *output, const FFTSample *input);
714 void ff_imdct_calc_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input);
715 void ff_imdct_half_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input);
716 void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input);
717 void ff_imdct_half_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input);
718 void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output, const FFTSample *input);
719 void ff_imdct_half_sse(MDCTContext *s, FFTSample *output, const FFTSample *input);
720 void ff_mdct_calc(MDCTContext *s, FFTSample *out, const FFTSample *input);
721 void ff_mdct_end(MDCTContext *s);
722
723 #define WRAPPER8_16(name8, name16)\
724 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
725 return name8(s, dst , src , stride, h)\
726 +name8(s, dst+8 , src+8 , stride, h);\
727 }
728
729 #define WRAPPER8_16_SQ(name8, name16)\
730 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
731 int score=0;\
732 score +=name8(s, dst , src , stride, 8);\
733 score +=name8(s, dst+8 , src+8 , stride, 8);\
734 if(h==16){\
735 dst += 8*stride;\
736 src += 8*stride;\
737 score +=name8(s, dst , src , stride, 8);\
738 score +=name8(s, dst+8 , src+8 , stride, 8);\
739 }\
740 return score;\
741 }
742
743
744 static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
745 {
746 int i;
747 for(i=0; i<h; i++)
748 {
749 AV_WN16(dst , AV_RN16(src ));
750 dst+=dstStride;
751 src+=srcStride;
752 }
753 }
754
755 static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
756 {
757 int i;
758 for(i=0; i<h; i++)
759 {
760 AV_WN32(dst , AV_RN32(src ));
761 dst+=dstStride;
762 src+=srcStride;
763 }
764 }
765
766 static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
767 {
768 int i;
769 for(i=0; i<h; i++)
770 {
771 AV_WN32(dst , AV_RN32(src ));
772 AV_WN32(dst+4 , AV_RN32(src+4 ));
773 dst+=dstStride;
774 src+=srcStride;
775 }
776 }
777
778 static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
779 {
780 int i;
781 for(i=0; i<h; i++)
782 {
783 AV_WN32(dst , AV_RN32(src ));
784 AV_WN32(dst+4 , AV_RN32(src+4 ));
785 dst[8]= src[8];
786 dst+=dstStride;
787 src+=srcStride;
788 }
789 }
790
791 static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
792 {
793 int i;
794 for(i=0; i<h; i++)
795 {
796 AV_WN32(dst , AV_RN32(src ));
797 AV_WN32(dst+4 , AV_RN32(src+4 ));
798 AV_WN32(dst+8 , AV_RN32(src+8 ));
799 AV_WN32(dst+12, AV_RN32(src+12));
800 dst+=dstStride;
801 src+=srcStride;
802 }
803 }
804
805 static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
806 {
807 int i;
808 for(i=0; i<h; i++)
809 {
810 AV_WN32(dst , AV_RN32(src ));
811 AV_WN32(dst+4 , AV_RN32(src+4 ));
812 AV_WN32(dst+8 , AV_RN32(src+8 ));
813 AV_WN32(dst+12, AV_RN32(src+12));
814 dst[16]= src[16];
815 dst+=dstStride;
816 src+=srcStride;
817 }
818 }
819
820 #endif /* AVCODEC_DSPUTIL_H */