3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
6 * This file is part of FFmpeg.
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.
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.
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
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
30 #ifndef AVCODEC_DSPUTIL_H
31 #define AVCODEC_DSPUTIL_H
33 #include "libavutil/intreadwrite.h"
39 typedef short DCTELEM;
41 void ff_fdct_ifast (DCTELEM *data);
42 void ff_fdct_ifast248 (DCTELEM *data);
43 void ff_jpeg_fdct_islow_8(DCTELEM *data);
44 void ff_jpeg_fdct_islow_10(DCTELEM *data);
45 void ff_fdct248_islow_8(DCTELEM *data);
46 void ff_fdct248_islow_10(DCTELEM *data);
48 void ff_j_rev_dct (DCTELEM *data);
49 void ff_j_rev_dct4 (DCTELEM *data);
50 void ff_j_rev_dct2 (DCTELEM *data);
51 void ff_j_rev_dct1 (DCTELEM *data);
52 void ff_wmv2_idct_c(DCTELEM *data);
54 void ff_fdct_mmx(DCTELEM *block);
55 void ff_fdct_mmx2(DCTELEM *block);
56 void ff_fdct_sse2(DCTELEM *block);
58 #define H264_IDCT(depth) \
59 void ff_h264_idct8_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride);\
60 void ff_h264_idct_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride);\
61 void ff_h264_idct8_dc_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride);\
62 void ff_h264_idct_dc_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride);\
63 void ff_h264_idct_add16_ ## depth ## _c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
64 void ff_h264_idct_add16intra_ ## depth ## _c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
65 void ff_h264_idct8_add4_ ## depth ## _c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
66 void ff_h264_idct_add8_422_ ## depth ## _c(uint8_t **dest, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
67 void ff_h264_idct_add8_ ## depth ## _c(uint8_t **dest, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
68 void ff_h264_luma_dc_dequant_idct_ ## depth ## _c(DCTELEM *output, DCTELEM *input, int qmul);\
69 void ff_h264_chroma422_dc_dequant_idct_ ## depth ## _c(DCTELEM *block, int qmul);\
70 void ff_h264_chroma_dc_dequant_idct_ ## depth ## _c(DCTELEM *block, int qmul);
78 void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *output, DCTELEM *input, int qp);
79 void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
82 extern const uint8_t ff_alternate_horizontal_scan[64];
83 extern const uint8_t ff_alternate_vertical_scan[64];
84 extern const uint8_t ff_zigzag_direct[64];
85 extern const uint8_t ff_zigzag248_direct[64];
87 /* pixel operations */
88 #define MAX_NEG_CROP 1024
91 extern uint32_t ff_squareTbl[512];
92 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
94 #define PUTAVG_PIXELS(depth)\
95 void ff_put_pixels8x8_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);\
96 void ff_avg_pixels8x8_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);\
97 void ff_put_pixels16x16_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);\
98 void ff_avg_pixels16x16_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);
106 #define ff_put_pixels8x8_c ff_put_pixels8x8_8_c
107 #define ff_avg_pixels8x8_c ff_avg_pixels8x8_8_c
108 #define ff_put_pixels16x16_c ff_put_pixels16x16_8_c
109 #define ff_avg_pixels16x16_c ff_avg_pixels16x16_8_c
112 void ff_put_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride);
113 void ff_avg_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride);
114 void ff_put_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride);
115 void ff_avg_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride);
117 /* 1/2^n downscaling functions from imgconvert.c */
118 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
119 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
120 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
122 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
123 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
125 /* minimum alignment rules ;)
126 If you notice errors in the align stuff, need more alignment for some ASM code
127 for some CPU or need to use a function with less aligned data then send a mail
128 to the ffmpeg-devel mailing list, ...
130 !warning These alignments might not match reality, (missing attribute((align))
131 stuff somewhere possible).
132 I (Michael) did not check them, these are just the alignments which I think
133 could be reached easily ...
135 !future video codecs might need functions with less strict alignment
139 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
140 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
141 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
142 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
143 void clear_blocks_c(DCTELEM *blocks);
146 /* add and put pixel (decoding) */
147 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
148 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller than 4
149 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
150 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);
151 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
152 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
154 typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
156 #define DEF_OLD_QPEL(name)\
157 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
158 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
159 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
161 DEF_OLD_QPEL(qpel16_mc11_old_c)
162 DEF_OLD_QPEL(qpel16_mc31_old_c)
163 DEF_OLD_QPEL(qpel16_mc12_old_c)
164 DEF_OLD_QPEL(qpel16_mc32_old_c)
165 DEF_OLD_QPEL(qpel16_mc13_old_c)
166 DEF_OLD_QPEL(qpel16_mc33_old_c)
167 DEF_OLD_QPEL(qpel8_mc11_old_c)
168 DEF_OLD_QPEL(qpel8_mc31_old_c)
169 DEF_OLD_QPEL(qpel8_mc12_old_c)
170 DEF_OLD_QPEL(qpel8_mc32_old_c)
171 DEF_OLD_QPEL(qpel8_mc13_old_c)
172 DEF_OLD_QPEL(qpel8_mc33_old_c)
174 #define CALL_2X_PIXELS(a, b, n)\
175 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
176 b(block , pixels , line_size, h);\
177 b(block+n, pixels+n, line_size, h);\
180 /* motion estimation */
181 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller than 2
182 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
183 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))*/;
188 typedef struct ScanTable{
189 const uint8_t *scantable;
190 uint8_t permutated[64];
191 uint8_t raster_end[64];
194 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
195 void ff_init_scantable_permutation(uint8_t *idct_permutation,
196 int idct_permutation_type);
198 #define EMULATED_EDGE(depth) \
199 void ff_emulated_edge_mc_ ## depth (uint8_t *buf, const uint8_t *src, int linesize,\
200 int block_w, int block_h,\
201 int src_x, int src_y, int w, int h);
212 typedef struct DSPContext {
214 * Size of DCT coefficients.
218 /* pixel ops : interface with DCT */
219 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
220 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
221 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
222 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
223 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
224 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
225 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
226 int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
228 * Motion estimation with emulated edge values.
229 * @param buf pointer to destination buffer (unaligned)
230 * @param src pointer to pixel source (unaligned)
231 * @param linesize width (in pixels) for src/buf
232 * @param block_w number of pixels (per row) to copy to buf
233 * @param block_h nummber of pixel rows to copy to buf
234 * @param src_x offset of src to start of row - this may be negative
235 * @param src_y offset of src to top of image - this may be negative
236 * @param w width of src in pixels
237 * @param h height of src in pixels
239 void (*emulated_edge_mc)(uint8_t *buf, const uint8_t *src, int linesize,
240 int block_w, int block_h,
241 int src_x, int src_y, int w, int h);
243 * translational global motion compensation.
245 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
247 * global motion compensation.
249 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
250 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
251 void (*clear_block)(DCTELEM *block/*align 16*/);
252 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
253 int (*pix_sum)(uint8_t * pix, int line_size);
254 int (*pix_norm1)(uint8_t * pix, int line_size);
255 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
257 me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
259 me_cmp_func hadamard8_diff[6];
260 me_cmp_func dct_sad[6];
261 me_cmp_func quant_psnr[6];
269 me_cmp_func dct_max[6];
270 me_cmp_func dct264_sad[6];
272 me_cmp_func me_pre_cmp[6];
273 me_cmp_func me_cmp[6];
274 me_cmp_func me_sub_cmp[6];
275 me_cmp_func mb_cmp[6];
276 me_cmp_func ildct_cmp[6]; //only width 16 used
277 me_cmp_func frame_skip_cmp[6]; //only width 8 used
279 int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
283 * Halfpel motion compensation with rounding (a+b+1)>>1.
284 * this is an array[4][4] of motion compensation functions for 4
285 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
286 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
287 * @param block destination where the result is stored
288 * @param pixels source
289 * @param line_size number of bytes in a horizontal line of block
292 op_pixels_func put_pixels_tab[4][4];
295 * Halfpel motion compensation with rounding (a+b+1)>>1.
296 * This is an array[4][4] of motion compensation functions for 4
297 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
298 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
299 * @param block destination into which the result is averaged (a+b+1)>>1
300 * @param pixels source
301 * @param line_size number of bytes in a horizontal line of block
304 op_pixels_func avg_pixels_tab[4][4];
307 * Halfpel motion compensation with no rounding (a+b)>>1.
308 * this is an array[2][4] of motion compensation functions for 2
309 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
310 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
311 * @param block destination where the result is stored
312 * @param pixels source
313 * @param line_size number of bytes in a horizontal line of block
316 op_pixels_func put_no_rnd_pixels_tab[4][4];
319 * Halfpel motion compensation with no rounding (a+b)>>1.
320 * this is an array[2][4] of motion compensation functions for 2
321 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
322 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
323 * @param block destination into which the result is averaged (a+b)>>1
324 * @param pixels source
325 * @param line_size number of bytes in a horizontal line of block
328 op_pixels_func avg_no_rnd_pixels_tab[4][4];
330 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);
333 * Thirdpel motion compensation with rounding (a+b+1)>>1.
334 * this is an array[12] of motion compensation functions for the 9 thirdpe
336 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
337 * @param block destination where the result is stored
338 * @param pixels source
339 * @param line_size number of bytes in a horizontal line of block
342 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
343 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
345 qpel_mc_func put_qpel_pixels_tab[2][16];
346 qpel_mc_func avg_qpel_pixels_tab[2][16];
347 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
348 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
349 qpel_mc_func put_mspel_pixels_tab[8];
354 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
355 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
357 qpel_mc_func put_h264_qpel_pixels_tab[4][16];
358 qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
360 qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
361 qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
363 me_cmp_func pix_abs[2][4];
365 /* huffyuv specific */
366 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
367 void (*diff_bytes)(uint8_t *dst/*align 16*/, const uint8_t *src1/*align 16*/, const uint8_t *src2/*align 1*/,int w);
369 * subtract huffyuv's variant of median prediction
370 * note, this might read from src1[-1], src2[-1]
372 void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
373 void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
374 int (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
375 void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
376 /* this might write to dst[w] */
377 void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
378 void (*bswap16_buf)(uint16_t *dst, const uint16_t *src, int len);
380 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
381 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
383 void (*h261_loop_filter)(uint8_t *src, int stride);
385 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
386 void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
387 /* assume len is a multiple of 16, and arrays are 32-byte aligned */
388 void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
389 /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
390 void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len);
391 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
392 void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, int len);
393 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
394 void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
396 * Multiply a vector of floats by a scalar float. Source and
397 * destination vectors must overlap exactly or not at all.
398 * @param dst result vector, 16-byte aligned
399 * @param src input vector, 16-byte aligned
400 * @param mul scalar value
401 * @param len length of vector, multiple of 4
403 void (*vector_fmul_scalar)(float *dst, const float *src, float mul,
406 * Calculate the scalar product of two vectors of floats.
407 * @param v1 first vector, 16-byte aligned
408 * @param v2 second vector, 16-byte aligned
409 * @param len length of vectors, multiple of 4
411 float (*scalarproduct_float)(const float *v1, const float *v2, int len);
413 * Calculate the sum and difference of two vectors of floats.
414 * @param v1 first input vector, sum output, 16-byte aligned
415 * @param v2 second input vector, difference output, 16-byte aligned
416 * @param len length of vectors, multiple of 4
418 void (*butterflies_float)(float *av_restrict v1, float *av_restrict v2, int len);
421 * Calculate the sum and difference of two vectors of floats and interleave
422 * results into a separate output vector of floats, with each sum
423 * positioned before the corresponding difference.
425 * @param dst output vector
426 * constraints: 16-byte aligned
427 * @param src0 first input vector
428 * constraints: 32-byte aligned
429 * @param src1 second input vector
430 * constraints: 32-byte aligned
431 * @param len number of elements in the input
432 * constraints: multiple of 8
434 void (*butterflies_float_interleave)(float *dst, const float *src0,
435 const float *src1, int len);
438 void (*fdct)(DCTELEM *block/* align 16*/);
439 void (*fdct248)(DCTELEM *block/* align 16*/);
442 void (*idct)(DCTELEM *block/* align 16*/);
445 * block -> idct -> clip to unsigned 8 bit -> dest.
446 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
447 * @param line_size size in bytes of a horizontal line of dest
449 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
452 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
453 * @param line_size size in bytes of a horizontal line of dest
455 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
458 * idct input permutation.
459 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
461 * this permutation must be performed before the idct_put/add, note, normally this can be merged
462 * with the zigzag/alternate scan<br>
463 * an example to avoid confusion:
464 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
465 * - (x -> reference dct -> reference idct -> x)
466 * - (x -> reference dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
467 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
469 uint8_t idct_permutation[64];
470 int idct_permutation_type;
471 #define FF_NO_IDCT_PERM 1
472 #define FF_LIBMPEG2_IDCT_PERM 2
473 #define FF_SIMPLE_IDCT_PERM 3
474 #define FF_TRANSPOSE_IDCT_PERM 4
475 #define FF_PARTTRANS_IDCT_PERM 5
476 #define FF_SSE2_IDCT_PERM 6
478 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
479 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
480 #define BASIS_SHIFT 16
481 #define RECON_SHIFT 6
483 void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w, int h, int sides);
484 #define EDGE_WIDTH 16
486 #define EDGE_BOTTOM 2
488 void (*prefetch)(void *mem, int stride, int h);
490 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
492 /* mlp/truehd functions */
493 void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff,
494 int firorder, int iirorder,
495 unsigned int filter_shift, int32_t mask, int blocksize,
496 int32_t *sample_buffer);
499 * Calculate scalar product of two vectors.
500 * @param len length of vectors, should be multiple of 16
502 int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len);
505 * Calculate scalar product of v1 and v2,
506 * and v1[i] += v3[i] * mul
507 * @param len length of vectors, should be multiple of 16
509 int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul);
512 * Apply symmetric window in 16-bit fixed-point.
513 * @param output destination array
514 * constraints: 16-byte aligned
515 * @param input source array
516 * constraints: 16-byte aligned
517 * @param window window array
518 * constraints: 16-byte aligned, at least len/2 elements
519 * @param len full window length
520 * constraints: multiple of ? greater than zero
522 void (*apply_window_int16)(int16_t *output, const int16_t *input,
523 const int16_t *window, unsigned int len);
526 * Clip each element in an array of int32_t to a given minimum and maximum value.
527 * @param dst destination array
528 * constraints: 16-byte aligned
529 * @param src source array
530 * constraints: 16-byte aligned
531 * @param min minimum value
532 * constraints: must be in the range [-(1 << 24), 1 << 24]
533 * @param max maximum value
534 * constraints: must be in the range [-(1 << 24), 1 << 24]
535 * @param len number of elements in the array
536 * constraints: multiple of 32 greater than zero
538 void (*vector_clip_int32)(int32_t *dst, const int32_t *src, int32_t min,
539 int32_t max, unsigned int len);
541 op_fill_func fill_block_tab[2];
544 void ff_dsputil_static_init(void);
545 void ff_dsputil_init(DSPContext* p, AVCodecContext *avctx);
546 attribute_deprecated void dsputil_init(DSPContext* c, AVCodecContext *avctx);
548 int ff_check_alignment(void);
551 * Return the scalar product of two vectors.
553 * @param v1 first input vector
554 * @param v2 first input vector
555 * @param len number of elements
557 * @return sum of elementwise products
559 float ff_scalarproduct_float_c(const float *v1, const float *v2, int len);
562 * permute block according to permuatation.
563 * @param last last non zero element in scantable order
565 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
567 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
569 #define BYTE_VEC32(c) ((c)*0x01010101UL)
570 #define BYTE_VEC64(c) ((c)*0x0001000100010001UL)
572 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
574 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
577 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
579 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
582 static inline uint64_t rnd_avg64(uint64_t a, uint64_t b)
584 return (a | b) - (((a ^ b) & ~BYTE_VEC64(0x01)) >> 1);
587 static inline uint64_t no_rnd_avg64(uint64_t a, uint64_t b)
589 return (a & b) + (((a ^ b) & ~BYTE_VEC64(0x01)) >> 1);
592 static inline int get_penalty_factor(int lambda, int lambda2, int type){
596 return lambda>>FF_LAMBDA_SHIFT;
598 return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
600 return (4*lambda)>>(FF_LAMBDA_SHIFT);
602 return (2*lambda)>>(FF_LAMBDA_SHIFT);
605 return (2*lambda)>>FF_LAMBDA_SHIFT;
610 return lambda2>>FF_LAMBDA_SHIFT;
616 void ff_dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
617 void ff_dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
618 void ff_dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
619 void ff_dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
620 void ff_dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
621 void ff_dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
622 void ff_dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
623 void ff_dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
624 void ff_dsputil_init_mips(DSPContext* c, AVCodecContext *avctx);
626 void ff_dsputil_init_dwt(DSPContext *c);
627 void ff_mlp_init(DSPContext* c, AVCodecContext *avctx);
628 void ff_mlp_init_x86(DSPContext* c, AVCodecContext *avctx);
630 #if (ARCH_ARM && HAVE_NEON) || ARCH_PPC || HAVE_MMI || HAVE_MMX
631 # define STRIDE_ALIGN 16
633 # define STRIDE_ALIGN 8
636 // Some broken preprocessors need a second expansion
637 // to be forced to tokenize __VA_ARGS__
640 #define LOCAL_ALIGNED_A(a, t, v, s, o, ...) \
641 uint8_t la_##v[sizeof(t s o) + (a)]; \
642 t (*v) o = (void *)FFALIGN((uintptr_t)la_##v, a)
644 #define LOCAL_ALIGNED_D(a, t, v, s, o, ...) DECLARE_ALIGNED(a, t, v) s o
646 #define LOCAL_ALIGNED(a, t, v, ...) E(LOCAL_ALIGNED_A(a, t, v, __VA_ARGS__,,))
648 #if HAVE_LOCAL_ALIGNED_8
649 # define LOCAL_ALIGNED_8(t, v, ...) E(LOCAL_ALIGNED_D(8, t, v, __VA_ARGS__,,))
651 # define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED(8, t, v, __VA_ARGS__)
654 #if HAVE_LOCAL_ALIGNED_16
655 # define LOCAL_ALIGNED_16(t, v, ...) E(LOCAL_ALIGNED_D(16, t, v, __VA_ARGS__,,))
657 # define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED(16, t, v, __VA_ARGS__)
660 #define WRAPPER8_16(name8, name16)\
661 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
662 return name8(s, dst , src , stride, h)\
663 +name8(s, dst+8 , src+8 , stride, h);\
666 #define WRAPPER8_16_SQ(name8, name16)\
667 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
669 score +=name8(s, dst , src , stride, 8);\
670 score +=name8(s, dst+8 , src+8 , stride, 8);\
674 score +=name8(s, dst , src , stride, 8);\
675 score +=name8(s, dst+8 , src+8 , stride, 8);\
681 static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
686 AV_WN16(dst , AV_RN16(src ));
692 static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
697 AV_WN32(dst , AV_RN32(src ));
703 static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
708 AV_WN32(dst , AV_RN32(src ));
709 AV_WN32(dst+4 , AV_RN32(src+4 ));
715 static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
720 AV_WN32(dst , AV_RN32(src ));
721 AV_WN32(dst+4 , AV_RN32(src+4 ));
728 static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
733 AV_WN32(dst , AV_RN32(src ));
734 AV_WN32(dst+4 , AV_RN32(src+4 ));
735 AV_WN32(dst+8 , AV_RN32(src+8 ));
736 AV_WN32(dst+12, AV_RN32(src+12));
742 static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
747 AV_WN32(dst , AV_RN32(src ));
748 AV_WN32(dst+4 , AV_RN32(src+4 ));
749 AV_WN32(dst+8 , AV_RN32(src+8 ));
750 AV_WN32(dst+12, AV_RN32(src+12));
757 #endif /* AVCODEC_DSPUTIL_H */