3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
6 * This file is part of Libav.
8 * Libav 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 * Libav 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 Libav; 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_wmv2_idct_c(DCTELEM *data);
51 void ff_fdct_mmx(DCTELEM *block);
52 void ff_fdct_mmxext(DCTELEM *block);
53 void ff_fdct_sse2(DCTELEM *block);
55 #define H264_IDCT(depth) \
56 void ff_h264_idct8_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride);\
57 void ff_h264_idct_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride);\
58 void ff_h264_idct8_dc_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride);\
59 void ff_h264_idct_dc_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride);\
60 void ff_h264_idct_add16_ ## depth ## _c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
61 void ff_h264_idct_add16intra_ ## depth ## _c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
62 void ff_h264_idct8_add4_ ## depth ## _c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
63 void ff_h264_idct_add8_422_ ## depth ## _c(uint8_t **dest, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
64 void ff_h264_idct_add8_ ## depth ## _c(uint8_t **dest, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
65 void ff_h264_luma_dc_dequant_idct_ ## depth ## _c(DCTELEM *output, DCTELEM *input, int qmul);\
66 void ff_h264_chroma422_dc_dequant_idct_ ## depth ## _c(DCTELEM *block, int qmul);\
67 void ff_h264_chroma_dc_dequant_idct_ ## depth ## _c(DCTELEM *block, int qmul);
73 void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *output, DCTELEM *input, int qp);
74 void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
77 extern const uint8_t ff_alternate_horizontal_scan[64];
78 extern const uint8_t ff_alternate_vertical_scan[64];
79 extern const uint8_t ff_zigzag_direct[64];
80 extern const uint8_t ff_zigzag248_direct[64];
82 /* pixel operations */
83 #define MAX_NEG_CROP 1024
86 extern uint32_t ff_squareTbl[512];
87 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
89 #define PUTAVG_PIXELS(depth)\
90 void ff_put_pixels8x8_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);\
91 void ff_avg_pixels8x8_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);\
92 void ff_put_pixels16x16_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);\
93 void ff_avg_pixels16x16_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);
99 #define ff_put_pixels8x8_c ff_put_pixels8x8_8_c
100 #define ff_avg_pixels8x8_c ff_avg_pixels8x8_8_c
101 #define ff_put_pixels16x16_c ff_put_pixels16x16_8_c
102 #define ff_avg_pixels16x16_c ff_avg_pixels16x16_8_c
105 void ff_put_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride);
106 void ff_avg_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride);
107 void ff_put_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride);
108 void ff_avg_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride);
110 /* 1/2^n downscaling functions from imgconvert.c */
111 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
112 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
113 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
115 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
116 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
118 /* minimum alignment rules ;)
119 If you notice errors in the align stuff, need more alignment for some ASM code
120 for some CPU or need to use a function with less aligned data then send a mail
121 to the libav-devel mailing list, ...
123 !warning These alignments might not match reality, (missing attribute((align))
124 stuff somewhere possible).
125 I (Michael) did not check them, these are just the alignments which I think
126 could be reached easily ...
128 !future video codecs might need functions with less strict alignment
132 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
133 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
134 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
135 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
136 void clear_blocks_c(DCTELEM *blocks);
139 /* add and put pixel (decoding) */
140 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
141 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller than 4
142 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
143 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);
144 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
145 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
147 typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
149 #define DEF_OLD_QPEL(name)\
150 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
151 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
152 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
154 DEF_OLD_QPEL(qpel16_mc11_old_c)
155 DEF_OLD_QPEL(qpel16_mc31_old_c)
156 DEF_OLD_QPEL(qpel16_mc12_old_c)
157 DEF_OLD_QPEL(qpel16_mc32_old_c)
158 DEF_OLD_QPEL(qpel16_mc13_old_c)
159 DEF_OLD_QPEL(qpel16_mc33_old_c)
160 DEF_OLD_QPEL(qpel8_mc11_old_c)
161 DEF_OLD_QPEL(qpel8_mc31_old_c)
162 DEF_OLD_QPEL(qpel8_mc12_old_c)
163 DEF_OLD_QPEL(qpel8_mc32_old_c)
164 DEF_OLD_QPEL(qpel8_mc13_old_c)
165 DEF_OLD_QPEL(qpel8_mc33_old_c)
167 #define CALL_2X_PIXELS(a, b, n)\
168 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
169 b(block , pixels , line_size, h);\
170 b(block+n, pixels+n, line_size, h);\
173 /* motion estimation */
174 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller than 2
175 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
176 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))*/;
181 typedef struct ScanTable{
182 const uint8_t *scantable;
183 uint8_t permutated[64];
184 uint8_t raster_end[64];
187 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
188 void ff_init_scantable_permutation(uint8_t *idct_permutation,
189 int idct_permutation_type);
191 #define EMULATED_EDGE(depth) \
192 void ff_emulated_edge_mc_ ## depth (uint8_t *buf, const uint8_t *src, int linesize,\
193 int block_w, int block_h,\
194 int src_x, int src_y, int w, int h);
203 typedef struct DSPContext {
205 * Size of DCT coefficients.
209 /* pixel ops : interface with DCT */
210 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
211 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
212 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
213 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
214 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
215 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
216 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
217 int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
219 * Motion estimation with emulated edge values.
220 * @param buf pointer to destination buffer (unaligned)
221 * @param src pointer to pixel source (unaligned)
222 * @param linesize width (in pixels) for src/buf
223 * @param block_w number of pixels (per row) to copy to buf
224 * @param block_h nummber of pixel rows to copy to buf
225 * @param src_x offset of src to start of row - this may be negative
226 * @param src_y offset of src to top of image - this may be negative
227 * @param w width of src in pixels
228 * @param h height of src in pixels
230 void (*emulated_edge_mc)(uint8_t *buf, const uint8_t *src, int linesize,
231 int block_w, int block_h,
232 int src_x, int src_y, int w, int h);
234 * translational global motion compensation.
236 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
238 * global motion compensation.
240 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
241 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
242 void (*clear_block)(DCTELEM *block/*align 16*/);
243 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
244 int (*pix_sum)(uint8_t * pix, int line_size);
245 int (*pix_norm1)(uint8_t * pix, int line_size);
246 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
248 me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
250 me_cmp_func hadamard8_diff[6];
251 me_cmp_func dct_sad[6];
252 me_cmp_func quant_psnr[6];
260 me_cmp_func dct_max[6];
261 me_cmp_func dct264_sad[6];
263 me_cmp_func me_pre_cmp[6];
264 me_cmp_func me_cmp[6];
265 me_cmp_func me_sub_cmp[6];
266 me_cmp_func mb_cmp[6];
267 me_cmp_func ildct_cmp[6]; //only width 16 used
268 me_cmp_func frame_skip_cmp[6]; //only width 8 used
270 int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
274 * Halfpel motion compensation with rounding (a+b+1)>>1.
275 * this is an array[4][4] of motion compensation functions for 4
276 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
277 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
278 * @param block destination where the result is stored
279 * @param pixels source
280 * @param line_size number of bytes in a horizontal line of block
283 op_pixels_func put_pixels_tab[4][4];
286 * Halfpel motion compensation with rounding (a+b+1)>>1.
287 * This is an array[4][4] of motion compensation functions for 4
288 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
289 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
290 * @param block destination into which the result is averaged (a+b+1)>>1
291 * @param pixels source
292 * @param line_size number of bytes in a horizontal line of block
295 op_pixels_func avg_pixels_tab[4][4];
298 * Halfpel motion compensation with no rounding (a+b)>>1.
299 * this is an array[2][4] of motion compensation functions for 2
300 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
301 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
302 * @param block destination where the result is stored
303 * @param pixels source
304 * @param line_size number of bytes in a horizontal line of block
307 op_pixels_func put_no_rnd_pixels_tab[4][4];
310 * Halfpel motion compensation with no rounding (a+b)>>1.
311 * this is an array[2][4] of motion compensation functions for 2
312 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
313 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
314 * @param block destination into which the result is averaged (a+b)>>1
315 * @param pixels source
316 * @param line_size number of bytes in a horizontal line of block
319 op_pixels_func avg_no_rnd_pixels_tab[4][4];
321 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);
324 * Thirdpel motion compensation with rounding (a+b+1)>>1.
325 * this is an array[12] of motion compensation functions for the 9 thirdpe
327 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
328 * @param block destination where the result is stored
329 * @param pixels source
330 * @param line_size number of bytes in a horizontal line of block
333 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
334 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
336 qpel_mc_func put_qpel_pixels_tab[2][16];
337 qpel_mc_func avg_qpel_pixels_tab[2][16];
338 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
339 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
340 qpel_mc_func put_mspel_pixels_tab[8];
345 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
346 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
348 qpel_mc_func put_h264_qpel_pixels_tab[4][16];
349 qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
351 qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
352 qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
354 me_cmp_func pix_abs[2][4];
356 /* huffyuv specific */
357 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
358 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
360 * subtract huffyuv's variant of median prediction
361 * note, this might read from src1[-1], src2[-1]
363 void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
364 void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
365 int (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
366 void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
367 void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
368 void (*bswap16_buf)(uint16_t *dst, const uint16_t *src, int len);
370 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
371 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
373 void (*h261_loop_filter)(uint8_t *src, int stride);
375 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
376 void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
377 /* assume len is a multiple of 16, and arrays are 32-byte aligned */
378 void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
379 /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
380 void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len);
381 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
382 void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, int len);
383 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
384 void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
386 * Calculate the scalar product of two vectors of floats.
387 * @param v1 first vector, 16-byte aligned
388 * @param v2 second vector, 16-byte aligned
389 * @param len length of vectors, multiple of 4
391 float (*scalarproduct_float)(const float *v1, const float *v2, int len);
393 * Calculate the sum and difference of two vectors of floats.
394 * @param v1 first input vector, sum output, 16-byte aligned
395 * @param v2 second input vector, difference output, 16-byte aligned
396 * @param len length of vectors, multiple of 4
398 void (*butterflies_float)(float *restrict v1, float *restrict v2, int len);
401 * Calculate the sum and difference of two vectors of floats and interleave
402 * results into a separate output vector of floats, with each sum
403 * positioned before the corresponding difference.
405 * @param dst output vector
406 * constraints: 16-byte aligned
407 * @param src0 first input vector
408 * constraints: 32-byte aligned
409 * @param src1 second input vector
410 * constraints: 32-byte aligned
411 * @param len number of elements in the input
412 * constraints: multiple of 8
414 void (*butterflies_float_interleave)(float *dst, const float *src0,
415 const float *src1, int len);
418 void (*fdct)(DCTELEM *block/* align 16*/);
419 void (*fdct248)(DCTELEM *block/* align 16*/);
422 void (*idct)(DCTELEM *block/* align 16*/);
425 * block -> idct -> clip to unsigned 8 bit -> dest.
426 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
427 * @param line_size size in bytes of a horizontal line of dest
429 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
432 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
433 * @param line_size size in bytes of a horizontal line of dest
435 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
438 * idct input permutation.
439 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
441 * this permutation must be performed before the idct_put/add, note, normally this can be merged
442 * with the zigzag/alternate scan<br>
443 * an example to avoid confusion:
444 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
445 * - (x -> reference dct -> reference idct -> x)
446 * - (x -> reference dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
447 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
449 uint8_t idct_permutation[64];
450 int idct_permutation_type;
451 #define FF_NO_IDCT_PERM 1
452 #define FF_LIBMPEG2_IDCT_PERM 2
453 #define FF_SIMPLE_IDCT_PERM 3
454 #define FF_TRANSPOSE_IDCT_PERM 4
455 #define FF_PARTTRANS_IDCT_PERM 5
456 #define FF_SSE2_IDCT_PERM 6
458 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
459 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
460 #define BASIS_SHIFT 16
461 #define RECON_SHIFT 6
463 void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w, int h, int sides);
464 #define EDGE_WIDTH 16
466 #define EDGE_BOTTOM 2
468 void (*prefetch)(void *mem, int stride, int h);
470 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
473 * Calculate scalar product of two vectors.
474 * @param len length of vectors, should be multiple of 16
476 int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len);
479 * Calculate scalar product of v1 and v2,
480 * and v1[i] += v3[i] * mul
481 * @param len length of vectors, should be multiple of 16
483 int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul);
486 * Apply symmetric window in 16-bit fixed-point.
487 * @param output destination array
488 * constraints: 16-byte aligned
489 * @param input source array
490 * constraints: 16-byte aligned
491 * @param window window array
492 * constraints: 16-byte aligned, at least len/2 elements
493 * @param len full window length
494 * constraints: multiple of ? greater than zero
496 void (*apply_window_int16)(int16_t *output, const int16_t *input,
497 const int16_t *window, unsigned int len);
500 * Clip each element in an array of int32_t to a given minimum and maximum value.
501 * @param dst destination array
502 * constraints: 16-byte aligned
503 * @param src source array
504 * constraints: 16-byte aligned
505 * @param min minimum value
506 * constraints: must be in the range [-(1 << 24), 1 << 24]
507 * @param max maximum value
508 * constraints: must be in the range [-(1 << 24), 1 << 24]
509 * @param len number of elements in the array
510 * constraints: multiple of 32 greater than zero
512 void (*vector_clip_int32)(int32_t *dst, const int32_t *src, int32_t min,
513 int32_t max, unsigned int len);
515 op_fill_func fill_block_tab[2];
518 void ff_dsputil_static_init(void);
519 void ff_dsputil_init(DSPContext* p, AVCodecContext *avctx);
521 int ff_check_alignment(void);
524 * Return the scalar product of two vectors.
526 * @param v1 first input vector
527 * @param v2 first input vector
528 * @param len number of elements
530 * @return sum of elementwise products
532 float ff_scalarproduct_float_c(const float *v1, const float *v2, int len);
535 * permute block according to permuatation.
536 * @param last last non zero element in scantable order
538 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
540 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
542 #define BYTE_VEC32(c) ((c)*0x01010101UL)
543 #define BYTE_VEC64(c) ((c)*0x0001000100010001UL)
545 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
547 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
550 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
552 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
555 static inline uint64_t rnd_avg64(uint64_t a, uint64_t b)
557 return (a | b) - (((a ^ b) & ~BYTE_VEC64(0x01)) >> 1);
560 static inline uint64_t no_rnd_avg64(uint64_t a, uint64_t b)
562 return (a & b) + (((a ^ b) & ~BYTE_VEC64(0x01)) >> 1);
565 static inline int get_penalty_factor(int lambda, int lambda2, int type){
569 return lambda>>FF_LAMBDA_SHIFT;
571 return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
573 return (4*lambda)>>(FF_LAMBDA_SHIFT);
575 return (2*lambda)>>(FF_LAMBDA_SHIFT);
578 return (2*lambda)>>FF_LAMBDA_SHIFT;
583 return lambda2>>FF_LAMBDA_SHIFT;
589 void ff_dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
590 void ff_dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
591 void ff_dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
592 void ff_dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
593 void ff_dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
594 void ff_dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
595 void ff_dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
597 void ff_dsputil_init_dwt(DSPContext *c);
599 #if (ARCH_ARM && HAVE_NEON) || ARCH_PPC || HAVE_MMX
600 # define STRIDE_ALIGN 16
602 # define STRIDE_ALIGN 8
605 // Some broken preprocessors need a second expansion
606 // to be forced to tokenize __VA_ARGS__
609 #define LOCAL_ALIGNED_A(a, t, v, s, o, ...) \
610 uint8_t la_##v[sizeof(t s o) + (a)]; \
611 t (*v) o = (void *)FFALIGN((uintptr_t)la_##v, a)
613 #define LOCAL_ALIGNED_D(a, t, v, s, o, ...) DECLARE_ALIGNED(a, t, v) s o
615 #define LOCAL_ALIGNED(a, t, v, ...) E(LOCAL_ALIGNED_A(a, t, v, __VA_ARGS__,,))
617 #if HAVE_LOCAL_ALIGNED_8
618 # define LOCAL_ALIGNED_8(t, v, ...) E(LOCAL_ALIGNED_D(8, t, v, __VA_ARGS__,,))
620 # define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED(8, t, v, __VA_ARGS__)
623 #if HAVE_LOCAL_ALIGNED_16
624 # define LOCAL_ALIGNED_16(t, v, ...) E(LOCAL_ALIGNED_D(16, t, v, __VA_ARGS__,,))
626 # define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED(16, t, v, __VA_ARGS__)
629 #define WRAPPER8_16(name8, name16)\
630 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
631 return name8(s, dst , src , stride, h)\
632 +name8(s, dst+8 , src+8 , stride, h);\
635 #define WRAPPER8_16_SQ(name8, name16)\
636 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
638 score +=name8(s, dst , src , stride, 8);\
639 score +=name8(s, dst+8 , src+8 , stride, 8);\
643 score +=name8(s, dst , src , stride, 8);\
644 score +=name8(s, dst+8 , src+8 , stride, 8);\
650 static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
655 AV_COPY16U(dst, src);
661 static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
666 AV_COPY32U(dst, src);
672 static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
677 AV_COPY64U(dst, src);
683 static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
688 AV_COPY64U(dst, src);
695 static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
700 AV_COPY128U(dst, src);
706 static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
711 AV_COPY128U(dst, src);
718 #endif /* AVCODEC_DSPUTIL_H */