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_mmx2(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_ea_idct_put_c(uint8_t *dest
, int linesize
, DCTELEM
*block
);
108 void ff_put_rv40_qpel16_mc33_c(uint8_t *dst
, uint8_t *src
, int stride
);
109 void ff_avg_rv40_qpel16_mc33_c(uint8_t *dst
, uint8_t *src
, int stride
);
110 void ff_put_rv40_qpel8_mc33_c(uint8_t *dst
, uint8_t *src
, int stride
);
111 void ff_avg_rv40_qpel8_mc33_c(uint8_t *dst
, uint8_t *src
, int stride
);
113 /* 1/2^n downscaling functions from imgconvert.c */
114 void ff_shrink22(uint8_t *dst
, int dst_wrap
, const uint8_t *src
, int src_wrap
, int width
, int height
);
115 void ff_shrink44(uint8_t *dst
, int dst_wrap
, const uint8_t *src
, int src_wrap
, int width
, int height
);
116 void ff_shrink88(uint8_t *dst
, int dst_wrap
, const uint8_t *src
, int src_wrap
, int width
, int height
);
118 void ff_gmc_c(uint8_t *dst
, uint8_t *src
, int stride
, int h
, int ox
, int oy
,
119 int dxx
, int dxy
, int dyx
, int dyy
, int shift
, int r
, int width
, int height
);
121 /* minimum alignment rules ;)
122 If you notice errors in the align stuff, need more alignment for some ASM code
123 for some CPU or need to use a function with less aligned data then send a mail
124 to the libav-devel mailing list, ...
126 !warning These alignments might not match reality, (missing attribute((align))
127 stuff somewhere possible).
128 I (Michael) did not check them, these are just the alignments which I think
129 could be reached easily ...
131 !future video codecs might need functions with less strict alignment
135 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
136 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
137 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
138 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
139 void clear_blocks_c(DCTELEM *blocks);
142 /* add and put pixel (decoding) */
143 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
144 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller than 4
145 typedef void (*op_pixels_func
)(uint8_t *block
/*align width (8 or 16)*/, const uint8_t *pixels
/*align 1*/, int line_size
, int h
);
146 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
);
147 typedef void (*qpel_mc_func
)(uint8_t *dst
/*align width (8 or 16)*/, uint8_t *src
/*align 1*/, int stride
);
148 typedef void (*h264_chroma_mc_func
)(uint8_t *dst
/*align 8*/, uint8_t *src
/*align 1*/, int srcStride
, int h
, int x
, int y
);
150 typedef void (*op_fill_func
)(uint8_t *block
/*align width (8 or 16)*/, uint8_t value
, int line_size
, int h
);
152 #define DEF_OLD_QPEL(name)\
153 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
154 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
155 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
157 DEF_OLD_QPEL(qpel16_mc11_old_c
)
158 DEF_OLD_QPEL(qpel16_mc31_old_c
)
159 DEF_OLD_QPEL(qpel16_mc12_old_c
)
160 DEF_OLD_QPEL(qpel16_mc32_old_c
)
161 DEF_OLD_QPEL(qpel16_mc13_old_c
)
162 DEF_OLD_QPEL(qpel16_mc33_old_c
)
163 DEF_OLD_QPEL(qpel8_mc11_old_c
)
164 DEF_OLD_QPEL(qpel8_mc31_old_c
)
165 DEF_OLD_QPEL(qpel8_mc12_old_c
)
166 DEF_OLD_QPEL(qpel8_mc32_old_c
)
167 DEF_OLD_QPEL(qpel8_mc13_old_c
)
168 DEF_OLD_QPEL(qpel8_mc33_old_c
)
170 #define CALL_2X_PIXELS(a, b, n)\
171 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
172 b(block , pixels , line_size, h);\
173 b(block+n, pixels+n, line_size, h);\
176 /* motion estimation */
177 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller than 2
178 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
179 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))*/;
184 typedef struct ScanTable
{
185 const uint8_t *scantable
;
186 uint8_t permutated
[64];
187 uint8_t raster_end
[64];
190 void ff_init_scantable(uint8_t *, ScanTable
*st
, const uint8_t *src_scantable
);
191 void ff_init_scantable_permutation(uint8_t *idct_permutation
,
192 int idct_permutation_type
);
194 #define EMULATED_EDGE(depth) \
195 void ff_emulated_edge_mc_ ## depth (uint8_t *buf, const uint8_t *src, int linesize,\
196 int block_w, int block_h,\
197 int src_x, int src_y, int w, int h);
203 void ff_add_pixels_clamped_c(const DCTELEM
*block
, uint8_t *dest
, int linesize
);
204 void ff_put_pixels_clamped_c(const DCTELEM
*block
, uint8_t *dest
, int linesize
);
205 void ff_put_signed_pixels_clamped_c(const DCTELEM
*block
, uint8_t *dest
, int linesize
);
210 typedef struct DSPContext
{
212 * Size of DCT coefficients.
216 /* pixel ops : interface with DCT */
217 void (*get_pixels
)(DCTELEM
*block
/*align 16*/, const uint8_t *pixels
/*align 8*/, int line_size
);
218 void (*diff_pixels
)(DCTELEM
*block
/*align 16*/, const uint8_t *s1
/*align 8*/, const uint8_t *s2
/*align 8*/, int stride
);
219 void (*put_pixels_clamped
)(const DCTELEM
*block
/*align 16*/, uint8_t *pixels
/*align 8*/, int line_size
);
220 void (*put_signed_pixels_clamped
)(const DCTELEM
*block
/*align 16*/, uint8_t *pixels
/*align 8*/, int line_size
);
221 void (*add_pixels_clamped
)(const DCTELEM
*block
/*align 16*/, uint8_t *pixels
/*align 8*/, int line_size
);
222 void (*add_pixels8
)(uint8_t *pixels
, DCTELEM
*block
, int line_size
);
223 void (*add_pixels4
)(uint8_t *pixels
, DCTELEM
*block
, int line_size
);
224 int (*sum_abs_dctelem
)(DCTELEM
*block
/*align 16*/);
226 * Motion estimation with emulated edge values.
227 * @param buf pointer to destination buffer (unaligned)
228 * @param src pointer to pixel source (unaligned)
229 * @param linesize width (in pixels) for src/buf
230 * @param block_w number of pixels (per row) to copy to buf
231 * @param block_h nummber of pixel rows to copy to buf
232 * @param src_x offset of src to start of row - this may be negative
233 * @param src_y offset of src to top of image - this may be negative
234 * @param w width of src in pixels
235 * @param h height of src in pixels
237 void (*emulated_edge_mc
)(uint8_t *buf
, const uint8_t *src
, int linesize
,
238 int block_w
, int block_h
,
239 int src_x
, int src_y
, int w
, int h
);
241 * translational global motion compensation.
243 void (*gmc1
)(uint8_t *dst
/*align 8*/, uint8_t *src
/*align 1*/, int srcStride
, int h
, int x16
, int y16
, int rounder
);
245 * global motion compensation.
247 void (*gmc
)(uint8_t *dst
/*align 8*/, uint8_t *src
/*align 1*/, int stride
, int h
, int ox
, int oy
,
248 int dxx
, int dxy
, int dyx
, int dyy
, int shift
, int r
, int width
, int height
);
249 void (*clear_block
)(DCTELEM
*block
/*align 16*/);
250 void (*clear_blocks
)(DCTELEM
*blocks
/*align 16*/);
251 int (*pix_sum
)(uint8_t * pix
, int line_size
);
252 int (*pix_norm1
)(uint8_t * pix
, int line_size
);
253 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
255 me_cmp_func sad
[6]; /* identical to pix_absAxA except additional void * */
257 me_cmp_func hadamard8_diff
[6];
258 me_cmp_func dct_sad
[6];
259 me_cmp_func quant_psnr
[6];
267 me_cmp_func dct_max
[6];
268 me_cmp_func dct264_sad
[6];
270 me_cmp_func me_pre_cmp
[6];
271 me_cmp_func me_cmp
[6];
272 me_cmp_func me_sub_cmp
[6];
273 me_cmp_func mb_cmp
[6];
274 me_cmp_func ildct_cmp
[6]; //only width 16 used
275 me_cmp_func frame_skip_cmp
[6]; //only width 8 used
277 int (*ssd_int8_vs_int16
)(const int8_t *pix1
, const int16_t *pix2
,
281 * Halfpel motion compensation with rounding (a+b+1)>>1.
282 * this is an array[4][4] of motion compensation functions for 4
283 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
284 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
285 * @param block destination where the result is stored
286 * @param pixels source
287 * @param line_size number of bytes in a horizontal line of block
290 op_pixels_func put_pixels_tab
[4][4];
293 * Halfpel motion compensation with rounding (a+b+1)>>1.
294 * This is an array[4][4] of motion compensation functions for 4
295 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
296 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
297 * @param block destination into which the result is averaged (a+b+1)>>1
298 * @param pixels source
299 * @param line_size number of bytes in a horizontal line of block
302 op_pixels_func avg_pixels_tab
[4][4];
305 * Halfpel motion compensation with no rounding (a+b)>>1.
306 * this is an array[2][4] of motion compensation functions for 2
307 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
308 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
309 * @param block destination where the result is stored
310 * @param pixels source
311 * @param line_size number of bytes in a horizontal line of block
314 op_pixels_func put_no_rnd_pixels_tab
[4][4];
317 * Halfpel motion compensation with no rounding (a+b)>>1.
318 * this is an array[2][4] of motion compensation functions for 2
319 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
320 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
321 * @param block destination into which the result is averaged (a+b)>>1
322 * @param pixels source
323 * @param line_size number of bytes in a horizontal line of block
326 op_pixels_func avg_no_rnd_pixels_tab
[4][4];
328 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
);
331 * Thirdpel motion compensation with rounding (a+b+1)>>1.
332 * this is an array[12] of motion compensation functions for the 9 thirdpe
334 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
335 * @param block destination where the result is stored
336 * @param pixels source
337 * @param line_size number of bytes in a horizontal line of block
340 tpel_mc_func put_tpel_pixels_tab
[11]; //FIXME individual func ptr per width?
341 tpel_mc_func avg_tpel_pixels_tab
[11]; //FIXME individual func ptr per width?
343 qpel_mc_func put_qpel_pixels_tab
[2][16];
344 qpel_mc_func avg_qpel_pixels_tab
[2][16];
345 qpel_mc_func put_no_rnd_qpel_pixels_tab
[2][16];
346 qpel_mc_func avg_no_rnd_qpel_pixels_tab
[2][16];
347 qpel_mc_func put_mspel_pixels_tab
[8];
352 h264_chroma_mc_func put_h264_chroma_pixels_tab
[3];
353 h264_chroma_mc_func avg_h264_chroma_pixels_tab
[3];
355 qpel_mc_func put_h264_qpel_pixels_tab
[4][16];
356 qpel_mc_func avg_h264_qpel_pixels_tab
[4][16];
358 qpel_mc_func put_2tap_qpel_pixels_tab
[4][16];
359 qpel_mc_func avg_2tap_qpel_pixels_tab
[4][16];
361 me_cmp_func pix_abs
[2][4];
363 /* huffyuv specific */
364 void (*add_bytes
)(uint8_t *dst
/*align 16*/, uint8_t *src
/*align 16*/, int w
);
365 void (*diff_bytes
)(uint8_t *dst
/*align 16*/, uint8_t *src1
/*align 16*/, uint8_t *src2
/*align 1*/,int w
);
367 * subtract huffyuv's variant of median prediction
368 * note, this might read from src1[-1], src2[-1]
370 void (*sub_hfyu_median_prediction
)(uint8_t *dst
, const uint8_t *src1
, const uint8_t *src2
, int w
, int *left
, int *left_top
);
371 void (*add_hfyu_median_prediction
)(uint8_t *dst
, const uint8_t *top
, const uint8_t *diff
, int w
, int *left
, int *left_top
);
372 int (*add_hfyu_left_prediction
)(uint8_t *dst
, const uint8_t *src
, int w
, int left
);
373 void (*add_hfyu_left_prediction_bgr32
)(uint8_t *dst
, const uint8_t *src
, int w
, int *red
, int *green
, int *blue
, int *alpha
);
374 void (*bswap_buf
)(uint32_t *dst
, const uint32_t *src
, int w
);
375 void (*bswap16_buf
)(uint16_t *dst
, const uint16_t *src
, int len
);
377 void (*h263_v_loop_filter
)(uint8_t *src
, int stride
, int qscale
);
378 void (*h263_h_loop_filter
)(uint8_t *src
, int stride
, int qscale
);
380 void (*h261_loop_filter
)(uint8_t *src
, int stride
);
382 void (*x8_v_loop_filter
)(uint8_t *src
, int stride
, int qscale
);
383 void (*x8_h_loop_filter
)(uint8_t *src
, int stride
, int qscale
);
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 void (*ac3_downmix
)(float (*samples
)[256], float (*matrix
)[2], int out_ch
, int in_ch
, int len
);
388 /* assume len is a multiple of 16, and arrays are 32-byte aligned */
389 void (*vector_fmul_reverse
)(float *dst
, const float *src0
, const float *src1
, int len
);
390 /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
391 void (*vector_fmul_add
)(float *dst
, const float *src0
, const float *src1
, const float *src2
, int len
);
392 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
393 void (*vector_fmul_window
)(float *dst
, const float *src0
, const float *src1
, const float *win
, int len
);
394 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
395 void (*vector_clipf
)(float *dst
/* align 16 */, const float *src
/* align 16 */, float min
, float max
, int len
/* align 16 */);
397 * Multiply a vector of floats by a scalar float. Source and
398 * destination vectors must overlap exactly or not at all.
399 * @param dst result vector, 16-byte aligned
400 * @param src input vector, 16-byte aligned
401 * @param mul scalar value
402 * @param len length of vector, multiple of 4
404 void (*vector_fmul_scalar
)(float *dst
, const float *src
, float mul
,
407 * Calculate the scalar product of two vectors of floats.
408 * @param v1 first vector, 16-byte aligned
409 * @param v2 second vector, 16-byte aligned
410 * @param len length of vectors, multiple of 4
412 float (*scalarproduct_float
)(const float *v1
, const float *v2
, int len
);
414 * Calculate the sum and difference of two vectors of floats.
415 * @param v1 first input vector, sum output, 16-byte aligned
416 * @param v2 second input vector, difference output, 16-byte aligned
417 * @param len length of vectors, multiple of 4
419 void (*butterflies_float
)(float *restrict v1
, float *restrict v2
, int len
);
422 * Calculate the sum and difference of two vectors of floats and interleave
423 * results into a separate output vector of floats, with each sum
424 * positioned before the corresponding difference.
426 * @param dst output vector
427 * constraints: 16-byte aligned
428 * @param src0 first input vector
429 * constraints: 32-byte aligned
430 * @param src1 second input vector
431 * constraints: 32-byte aligned
432 * @param len number of elements in the input
433 * constraints: multiple of 8
435 void (*butterflies_float_interleave
)(float *dst
, const float *src0
,
436 const float *src1
, int len
);
439 void (*fdct
)(DCTELEM
*block
/* align 16*/);
440 void (*fdct248
)(DCTELEM
*block
/* align 16*/);
443 void (*idct
)(DCTELEM
*block
/* align 16*/);
446 * block -> idct -> clip to unsigned 8 bit -> dest.
447 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
448 * @param line_size size in bytes of a horizontal line of dest
450 void (*idct_put
)(uint8_t *dest
/*align 8*/, int line_size
, DCTELEM
*block
/*align 16*/);
453 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
454 * @param line_size size in bytes of a horizontal line of dest
456 void (*idct_add
)(uint8_t *dest
/*align 8*/, int line_size
, DCTELEM
*block
/*align 16*/);
459 * idct input permutation.
460 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
462 * this permutation must be performed before the idct_put/add, note, normally this can be merged
463 * with the zigzag/alternate scan<br>
464 * an example to avoid confusion:
465 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
466 * - (x -> reference dct -> reference idct -> x)
467 * - (x -> reference dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
468 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
470 uint8_t idct_permutation
[64];
471 int idct_permutation_type
;
472 #define FF_NO_IDCT_PERM 1
473 #define FF_LIBMPEG2_IDCT_PERM 2
474 #define FF_SIMPLE_IDCT_PERM 3
475 #define FF_TRANSPOSE_IDCT_PERM 4
476 #define FF_PARTTRANS_IDCT_PERM 5
477 #define FF_SSE2_IDCT_PERM 6
479 int (*try_8x8basis
)(int16_t rem
[64], int16_t weight
[64], int16_t basis
[64], int scale
);
480 void (*add_8x8basis
)(int16_t rem
[64], int16_t basis
[64], int scale
);
481 #define BASIS_SHIFT 16
482 #define RECON_SHIFT 6
484 void (*draw_edges
)(uint8_t *buf
, int wrap
, int width
, int height
, int w
, int h
, int sides
);
485 #define EDGE_WIDTH 16
487 #define EDGE_BOTTOM 2
489 void (*prefetch
)(void *mem
, int stride
, int h
);
491 void (*shrink
[4])(uint8_t *dst
, int dst_wrap
, const uint8_t *src
, int src_wrap
, int width
, int height
);
493 /* mlp/truehd functions */
494 void (*mlp_filter_channel
)(int32_t *state
, const int32_t *coeff
,
495 int firorder
, int iirorder
,
496 unsigned int filter_shift
, int32_t mask
, int blocksize
,
497 int32_t *sample_buffer
);
499 /* intrax8 functions */
500 void (*x8_spatial_compensation
[12])(uint8_t *src
, uint8_t *dst
, int linesize
);
501 void (*x8_setup_spatial_compensation
)(uint8_t *src
, uint8_t *dst
, int linesize
,
502 int * range
, int * sum
, int edges
);
505 * Calculate scalar product of two vectors.
506 * @param len length of vectors, should be multiple of 16
508 int32_t (*scalarproduct_int16
)(const int16_t *v1
, const int16_t *v2
/*align 16*/, int len
);
511 * Calculate scalar product of v1 and v2,
512 * and v1[i] += v3[i] * mul
513 * @param len length of vectors, should be multiple of 16
515 int32_t (*scalarproduct_and_madd_int16
)(int16_t *v1
/*align 16*/, const int16_t *v2
, const int16_t *v3
, int len
, int mul
);
518 * Apply symmetric window in 16-bit fixed-point.
519 * @param output destination array
520 * constraints: 16-byte aligned
521 * @param input source array
522 * constraints: 16-byte aligned
523 * @param window window array
524 * constraints: 16-byte aligned, at least len/2 elements
525 * @param len full window length
526 * constraints: multiple of ? greater than zero
528 void (*apply_window_int16
)(int16_t *output
, const int16_t *input
,
529 const int16_t *window
, unsigned int len
);
532 * Clip each element in an array of int32_t to a given minimum and maximum value.
533 * @param dst destination array
534 * constraints: 16-byte aligned
535 * @param src source array
536 * constraints: 16-byte aligned
537 * @param min minimum value
538 * constraints: must be in the range [-(1 << 24), 1 << 24]
539 * @param max maximum value
540 * constraints: must be in the range [-(1 << 24), 1 << 24]
541 * @param len number of elements in the array
542 * constraints: multiple of 32 greater than zero
544 void (*vector_clip_int32
)(int32_t *dst
, const int32_t *src
, int32_t min
,
545 int32_t max
, unsigned int len
);
547 op_fill_func fill_block_tab
[2];
550 void ff_dsputil_static_init(void);
551 void ff_dsputil_init(DSPContext
* p
, AVCodecContext
*avctx
);
553 int ff_check_alignment(void);
556 * permute block according to permuatation.
557 * @param last last non zero element in scantable order
559 void ff_block_permute(DCTELEM
*block
, uint8_t *permutation
, const uint8_t *scantable
, int last
);
561 void ff_set_cmp(DSPContext
* c
, me_cmp_func
*cmp
, int type
);
563 #define BYTE_VEC32(c) ((c)*0x01010101UL)
564 #define BYTE_VEC64(c) ((c)*0x0001000100010001UL)
566 static inline uint32_t rnd_avg32(uint32_t a
, uint32_t b
)
568 return (a
| b
) - (((a
^ b
) & ~BYTE_VEC32(0x01)) >> 1);
571 static inline uint32_t no_rnd_avg32(uint32_t a
, uint32_t b
)
573 return (a
& b
) + (((a
^ b
) & ~BYTE_VEC32(0x01)) >> 1);
576 static inline uint64_t rnd_avg64(uint64_t a
, uint64_t b
)
578 return (a
| b
) - (((a
^ b
) & ~BYTE_VEC64(0x01)) >> 1);
581 static inline uint64_t no_rnd_avg64(uint64_t a
, uint64_t b
)
583 return (a
& b
) + (((a
^ b
) & ~BYTE_VEC64(0x01)) >> 1);
586 static inline int get_penalty_factor(int lambda
, int lambda2
, int type
){
590 return lambda
>>FF_LAMBDA_SHIFT
;
592 return (3*lambda
)>>(FF_LAMBDA_SHIFT
+1);
594 return (4*lambda
)>>(FF_LAMBDA_SHIFT
);
596 return (2*lambda
)>>(FF_LAMBDA_SHIFT
);
599 return (2*lambda
)>>FF_LAMBDA_SHIFT
;
604 return lambda2
>>FF_LAMBDA_SHIFT
;
610 void ff_dsputil_init_alpha(DSPContext
* c
, AVCodecContext
*avctx
);
611 void ff_dsputil_init_arm(DSPContext
* c
, AVCodecContext
*avctx
);
612 void ff_dsputil_init_bfin(DSPContext
* c
, AVCodecContext
*avctx
);
613 void ff_dsputil_init_mmi(DSPContext
* c
, AVCodecContext
*avctx
);
614 void ff_dsputil_init_mmx(DSPContext
* c
, AVCodecContext
*avctx
);
615 void ff_dsputil_init_ppc(DSPContext
* c
, AVCodecContext
*avctx
);
616 void ff_dsputil_init_sh4(DSPContext
* c
, AVCodecContext
*avctx
);
617 void ff_dsputil_init_vis(DSPContext
* c
, AVCodecContext
*avctx
);
619 void ff_dsputil_init_dwt(DSPContext
*c
);
620 void ff_intrax8dsp_init(DSPContext
* c
, AVCodecContext
*avctx
);
621 void ff_mlp_init(DSPContext
* c
, AVCodecContext
*avctx
);
622 void ff_mlp_init_x86(DSPContext
* c
, AVCodecContext
*avctx
);
624 #if (ARCH_ARM && HAVE_NEON) || ARCH_PPC || HAVE_MMI || HAVE_MMX
625 # define STRIDE_ALIGN 16
627 # define STRIDE_ALIGN 8
630 #define LOCAL_ALIGNED_A(a, t, v, s, o, ...) \
631 uint8_t la_##v[sizeof(t s o) + (a)]; \
632 t (*v) o = (void *)FFALIGN((uintptr_t)la_##v, a)
634 #define LOCAL_ALIGNED_D(a, t, v, s, o, ...) DECLARE_ALIGNED(a, t, v) s o
636 #define LOCAL_ALIGNED(a, t, v, ...) LOCAL_ALIGNED_A(a, t, v, __VA_ARGS__,,)
638 #if HAVE_LOCAL_ALIGNED_8
639 # define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED_D(8, t, v, __VA_ARGS__,,)
641 # define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED(8, t, v, __VA_ARGS__)
644 #if HAVE_LOCAL_ALIGNED_16
645 # define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED_D(16, t, v, __VA_ARGS__,,)
647 # define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED(16, t, v, __VA_ARGS__)
650 #define WRAPPER8_16(name8, name16)\
651 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
652 return name8(s, dst , src , stride, h)\
653 +name8(s, dst+8 , src+8 , stride, h);\
656 #define WRAPPER8_16_SQ(name8, name16)\
657 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
659 score +=name8(s, dst , src , stride, 8);\
660 score +=name8(s, dst+8 , src+8 , stride, 8);\
664 score +=name8(s, dst , src , stride, 8);\
665 score +=name8(s, dst+8 , src+8 , stride, 8);\
671 static inline void copy_block2(uint8_t *dst
, const uint8_t *src
, int dstStride
, int srcStride
, int h
)
676 AV_WN16(dst
, AV_RN16(src
));
682 static inline void copy_block4(uint8_t *dst
, const uint8_t *src
, int dstStride
, int srcStride
, int h
)
687 AV_WN32(dst
, AV_RN32(src
));
693 static inline void copy_block8(uint8_t *dst
, const uint8_t *src
, int dstStride
, int srcStride
, int h
)
698 AV_WN32(dst
, AV_RN32(src
));
699 AV_WN32(dst
+4 , AV_RN32(src
+4 ));
705 static inline void copy_block9(uint8_t *dst
, const uint8_t *src
, int dstStride
, int srcStride
, int h
)
710 AV_WN32(dst
, AV_RN32(src
));
711 AV_WN32(dst
+4 , AV_RN32(src
+4 ));
718 static inline void copy_block16(uint8_t *dst
, const uint8_t *src
, int dstStride
, int srcStride
, int h
)
723 AV_WN32(dst
, AV_RN32(src
));
724 AV_WN32(dst
+4 , AV_RN32(src
+4 ));
725 AV_WN32(dst
+8 , AV_RN32(src
+8 ));
726 AV_WN32(dst
+12, AV_RN32(src
+12));
732 static inline void copy_block17(uint8_t *dst
, const uint8_t *src
, int dstStride
, int srcStride
, int h
)
737 AV_WN32(dst
, AV_RN32(src
));
738 AV_WN32(dst
+4 , AV_RN32(src
+4 ));
739 AV_WN32(dst
+8 , AV_RN32(src
+8 ));
740 AV_WN32(dst
+12, AV_RN32(src
+12));
747 #endif /* AVCODEC_DSPUTIL_H */