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
38 typedef short DCTELEM
;
40 typedef short IDWTELEM
;
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
);
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
);
53 void ff_fdct_mmx(DCTELEM
*block
);
54 void ff_fdct_mmx2(DCTELEM
*block
);
55 void ff_fdct_sse2(DCTELEM
*block
);
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
);
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
);
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];
77 /* pixel operations */
78 #define MAX_NEG_CROP 1024
81 extern uint32_t ff_squareTbl
[512];
82 extern uint8_t ff_cropTbl
[256 + 2 * MAX_NEG_CROP
];
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*/);
89 /* 1/2^n downscaling functions from imgconvert.c */
90 void ff_img_copy_plane(uint8_t *dst
, int dst_wrap
, const uint8_t *src
, int src_wrap
, int width
, int height
);
91 void ff_shrink22(uint8_t *dst
, int dst_wrap
, const uint8_t *src
, int src_wrap
, int width
, int height
);
92 void ff_shrink44(uint8_t *dst
, int dst_wrap
, const uint8_t *src
, int src_wrap
, int width
, int height
);
93 void ff_shrink88(uint8_t *dst
, int dst_wrap
, const uint8_t *src
, int src_wrap
, int width
, int height
);
95 void ff_gmc_c(uint8_t *dst
, uint8_t *src
, int stride
, int h
, int ox
, int oy
,
96 int dxx
, int dxy
, int dyx
, int dyy
, int shift
, int r
, int width
, int height
);
98 /* minimum alignment rules ;)
99 If you notice errors in the align stuff, need more alignment for some ASM code
100 for some CPU or need to use a function with less aligned data then send a mail
101 to the ffmpeg-devel mailing list, ...
103 !warning These alignments might not match reality, (missing attribute((align))
104 stuff somewhere possible).
105 I (Michael) did not check them, these are just the alignments which I think
106 could be reached easily ...
108 !future video codecs might need functions with less strict alignment
112 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
113 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
114 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
115 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
116 void clear_blocks_c(DCTELEM *blocks);
119 /* add and put pixel (decoding) */
120 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
121 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
122 typedef void (*op_pixels_func
)(uint8_t *block
/*align width (8 or 16)*/, const uint8_t *pixels
/*align 1*/, int line_size
, int h
);
123 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
);
124 typedef void (*qpel_mc_func
)(uint8_t *dst
/*align width (8 or 16)*/, uint8_t *src
/*align 1*/, int stride
);
125 typedef void (*h264_chroma_mc_func
)(uint8_t *dst
/*align 8*/, uint8_t *src
/*align 1*/, int srcStride
, int h
, int x
, int y
);
126 typedef void (*h264_weight_func
)(uint8_t *block
, int stride
, int log2_denom
, int weight
, int offset
);
127 typedef void (*h264_biweight_func
)(uint8_t *dst
, uint8_t *src
, int stride
, int log2_denom
, int weightd
, int weights
, int offset
);
129 #define DEF_OLD_QPEL(name)\
130 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
131 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
132 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
134 DEF_OLD_QPEL(qpel16_mc11_old_c
)
135 DEF_OLD_QPEL(qpel16_mc31_old_c
)
136 DEF_OLD_QPEL(qpel16_mc12_old_c
)
137 DEF_OLD_QPEL(qpel16_mc32_old_c
)
138 DEF_OLD_QPEL(qpel16_mc13_old_c
)
139 DEF_OLD_QPEL(qpel16_mc33_old_c
)
140 DEF_OLD_QPEL(qpel8_mc11_old_c
)
141 DEF_OLD_QPEL(qpel8_mc31_old_c
)
142 DEF_OLD_QPEL(qpel8_mc12_old_c
)
143 DEF_OLD_QPEL(qpel8_mc32_old_c
)
144 DEF_OLD_QPEL(qpel8_mc13_old_c
)
145 DEF_OLD_QPEL(qpel8_mc33_old_c
)
147 #define CALL_2X_PIXELS(a, b, n)\
148 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
149 b(block , pixels , line_size, h);\
150 b(block+n, pixels+n, line_size, h);\
153 /* motion estimation */
154 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
155 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
156 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 typedef struct slice_buffer_s slice_buffer
;
165 typedef struct ScanTable
{
166 const uint8_t *scantable
;
167 uint8_t permutated
[64];
168 uint8_t raster_end
[64];
170 /** Used by dct_quantize_altivec to find last-non-zero */
171 DECLARE_ALIGNED(16, uint8_t, inverse
[64]);
175 void ff_init_scantable(uint8_t *, ScanTable
*st
, const uint8_t *src_scantable
);
177 void ff_emulated_edge_mc(uint8_t *buf
, uint8_t *src
, int linesize
,
178 int block_w
, int block_h
,
179 int src_x
, int src_y
, int w
, int h
);
184 typedef struct DSPContext
{
185 /* pixel ops : interface with DCT */
186 void (*get_pixels
)(DCTELEM
*block
/*align 16*/, const uint8_t *pixels
/*align 8*/, int line_size
);
187 void (*diff_pixels
)(DCTELEM
*block
/*align 16*/, const uint8_t *s1
/*align 8*/, const uint8_t *s2
/*align 8*/, int stride
);
188 void (*put_pixels_clamped
)(const DCTELEM
*block
/*align 16*/, uint8_t *pixels
/*align 8*/, int line_size
);
189 void (*put_signed_pixels_clamped
)(const DCTELEM
*block
/*align 16*/, uint8_t *pixels
/*align 8*/, int line_size
);
190 void (*add_pixels_clamped
)(const DCTELEM
*block
/*align 16*/, uint8_t *pixels
/*align 8*/, int line_size
);
191 void (*add_pixels8
)(uint8_t *pixels
, DCTELEM
*block
, int line_size
);
192 void (*add_pixels4
)(uint8_t *pixels
, DCTELEM
*block
, int line_size
);
193 int (*sum_abs_dctelem
)(DCTELEM
*block
/*align 16*/);
195 * translational global motion compensation.
197 void (*gmc1
)(uint8_t *dst
/*align 8*/, uint8_t *src
/*align 1*/, int srcStride
, int h
, int x16
, int y16
, int rounder
);
199 * global motion compensation.
201 void (*gmc
)(uint8_t *dst
/*align 8*/, uint8_t *src
/*align 1*/, int stride
, int h
, int ox
, int oy
,
202 int dxx
, int dxy
, int dyx
, int dyy
, int shift
, int r
, int width
, int height
);
203 void (*clear_blocks
)(DCTELEM
*blocks
/*align 16*/);
204 int (*pix_sum
)(uint8_t * pix
, int line_size
);
205 int (*pix_norm1
)(uint8_t * pix
, int line_size
);
206 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
208 me_cmp_func sad
[5]; /* identical to pix_absAxA except additional void * */
210 me_cmp_func hadamard8_diff
[5];
211 me_cmp_func dct_sad
[5];
212 me_cmp_func quant_psnr
[5];
220 me_cmp_func dct_max
[5];
221 me_cmp_func dct264_sad
[5];
223 me_cmp_func me_pre_cmp
[5];
224 me_cmp_func me_cmp
[5];
225 me_cmp_func me_sub_cmp
[5];
226 me_cmp_func mb_cmp
[5];
227 me_cmp_func ildct_cmp
[5]; //only width 16 used
228 me_cmp_func frame_skip_cmp
[5]; //only width 8 used
230 int (*ssd_int8_vs_int16
)(const int8_t *pix1
, const int16_t *pix2
,
234 * Halfpel motion compensation with rounding (a+b+1)>>1.
235 * this is an array[4][4] of motion compensation functions for 4
236 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
237 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
238 * @param block destination where the result is stored
239 * @param pixels source
240 * @param line_size number of bytes in a horizontal line of block
243 op_pixels_func put_pixels_tab
[4][4];
246 * Halfpel motion compensation with rounding (a+b+1)>>1.
247 * This is an array[4][4] of motion compensation functions for 4
248 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
249 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
250 * @param block destination into which the result is averaged (a+b+1)>>1
251 * @param pixels source
252 * @param line_size number of bytes in a horizontal line of block
255 op_pixels_func avg_pixels_tab
[4][4];
258 * Halfpel motion compensation with no rounding (a+b)>>1.
259 * this is an array[2][4] of motion compensation functions for 2
260 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
261 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
262 * @param block destination where the result is stored
263 * @param pixels source
264 * @param line_size number of bytes in a horizontal line of block
267 op_pixels_func put_no_rnd_pixels_tab
[4][4];
270 * Halfpel motion compensation with no rounding (a+b)>>1.
271 * this is an array[2][4] of motion compensation functions for 2
272 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
273 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
274 * @param block destination into which the result is averaged (a+b)>>1
275 * @param pixels source
276 * @param line_size number of bytes in a horizontal line of block
279 op_pixels_func avg_no_rnd_pixels_tab
[4][4];
281 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
);
284 * Thirdpel motion compensation with rounding (a+b+1)>>1.
285 * this is an array[12] of motion compensation functions for the 9 thirdpe
287 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
288 * @param block destination where the result is stored
289 * @param pixels source
290 * @param line_size number of bytes in a horizontal line of block
293 tpel_mc_func put_tpel_pixels_tab
[11]; //FIXME individual func ptr per width?
294 tpel_mc_func avg_tpel_pixels_tab
[11]; //FIXME individual func ptr per width?
296 qpel_mc_func put_qpel_pixels_tab
[2][16];
297 qpel_mc_func avg_qpel_pixels_tab
[2][16];
298 qpel_mc_func put_no_rnd_qpel_pixels_tab
[2][16];
299 qpel_mc_func avg_no_rnd_qpel_pixels_tab
[2][16];
300 qpel_mc_func put_mspel_pixels_tab
[8];
305 h264_chroma_mc_func put_h264_chroma_pixels_tab
[3];
306 /* This is really one func used in VC-1 decoding */
307 h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab
[3];
308 h264_chroma_mc_func avg_h264_chroma_pixels_tab
[3];
310 qpel_mc_func put_h264_qpel_pixels_tab
[4][16];
311 qpel_mc_func avg_h264_qpel_pixels_tab
[4][16];
313 qpel_mc_func put_2tap_qpel_pixels_tab
[4][16];
314 qpel_mc_func avg_2tap_qpel_pixels_tab
[4][16];
316 h264_weight_func weight_h264_pixels_tab
[10];
317 h264_biweight_func biweight_h264_pixels_tab
[10];
320 qpel_mc_func put_cavs_qpel_pixels_tab
[2][16];
321 qpel_mc_func avg_cavs_qpel_pixels_tab
[2][16];
322 void (*cavs_filter_lv
)(uint8_t *pix
, int stride
, int alpha
, int beta
, int tc
, int bs1
, int bs2
);
323 void (*cavs_filter_lh
)(uint8_t *pix
, int stride
, int alpha
, int beta
, int tc
, int bs1
, int bs2
);
324 void (*cavs_filter_cv
)(uint8_t *pix
, int stride
, int alpha
, int beta
, int tc
, int bs1
, int bs2
);
325 void (*cavs_filter_ch
)(uint8_t *pix
, int stride
, int alpha
, int beta
, int tc
, int bs1
, int bs2
);
326 void (*cavs_idct8_add
)(uint8_t *dst
, DCTELEM
*block
, int stride
);
328 me_cmp_func pix_abs
[2][4];
330 /* huffyuv specific */
331 void (*add_bytes
)(uint8_t *dst
/*align 16*/, uint8_t *src
/*align 16*/, int w
);
332 void (*add_bytes_l2
)(uint8_t *dst
/*align 16*/, uint8_t *src1
/*align 16*/, uint8_t *src2
/*align 16*/, int w
);
333 void (*diff_bytes
)(uint8_t *dst
/*align 16*/, uint8_t *src1
/*align 16*/, uint8_t *src2
/*align 1*/,int w
);
335 * subtract huffyuv's variant of median prediction
336 * note, this might read from src1[-1], src2[-1]
338 void (*sub_hfyu_median_prediction
)(uint8_t *dst
, uint8_t *src1
, uint8_t *src2
, int w
, int *left
, int *left_top
);
339 /* this might write to dst[w] */
340 void (*add_png_paeth_prediction
)(uint8_t *dst
, uint8_t *src
, uint8_t *top
, int w
, int bpp
);
341 void (*bswap_buf
)(uint32_t *dst
, const uint32_t *src
, int w
);
343 void (*h264_v_loop_filter_luma
)(uint8_t *pix
/*align 16*/, int stride
, int alpha
, int beta
, int8_t *tc0
);
344 void (*h264_h_loop_filter_luma
)(uint8_t *pix
/*align 4 */, int stride
, int alpha
, int beta
, int8_t *tc0
);
345 /* v/h_loop_filter_luma_intra: align 16 */
346 void (*h264_v_loop_filter_chroma
)(uint8_t *pix
/*align 8*/, int stride
, int alpha
, int beta
, int8_t *tc0
);
347 void (*h264_h_loop_filter_chroma
)(uint8_t *pix
/*align 4*/, int stride
, int alpha
, int beta
, int8_t *tc0
);
348 void (*h264_v_loop_filter_chroma_intra
)(uint8_t *pix
/*align 8*/, int stride
, int alpha
, int beta
);
349 void (*h264_h_loop_filter_chroma_intra
)(uint8_t *pix
/*align 8*/, int stride
, int alpha
, int beta
);
350 // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
351 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],
352 int bidir
, int edges
, int step
, int mask_mv0
, int mask_mv1
, int field
);
354 void (*h263_v_loop_filter
)(uint8_t *src
, int stride
, int qscale
);
355 void (*h263_h_loop_filter
)(uint8_t *src
, int stride
, int qscale
);
357 void (*h261_loop_filter
)(uint8_t *src
, int stride
);
359 void (*x8_v_loop_filter
)(uint8_t *src
, int stride
, int qscale
);
360 void (*x8_h_loop_filter
)(uint8_t *src
, int stride
, int qscale
);
362 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
363 void (*vorbis_inverse_coupling
)(float *mag
, float *ang
, int blocksize
);
364 void (*ac3_downmix
)(float (*samples
)[256], float (*matrix
)[2], int out_ch
, int in_ch
, int len
);
365 /* no alignment needed */
366 void (*flac_compute_autocorr
)(const int32_t *data
, int len
, int lag
, double *autoc
);
367 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
368 void (*vector_fmul
)(float *dst
, const float *src
, int len
);
369 void (*vector_fmul_reverse
)(float *dst
, const float *src0
, const float *src1
, int len
);
370 /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
371 void (*vector_fmul_add_add
)(float *dst
, const float *src0
, const float *src1
, const float *src2
, int src3
, int len
, int step
);
372 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
373 void (*vector_fmul_window
)(float *dst
, const float *src0
, const float *src1
, const float *win
, float add_bias
, int len
);
374 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
375 void (*int32_to_float_fmul_scalar
)(float *dst
, const int *src
, float mul
, int len
);
377 /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
378 * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
379 void (*float_to_int16
)(int16_t *dst
, const float *src
, long len
);
380 void (*float_to_int16_interleave
)(int16_t *dst
, const float **src
, long len
, int channels
);
383 void (*fdct
)(DCTELEM
*block
/* align 16*/);
384 void (*fdct248
)(DCTELEM
*block
/* align 16*/);
387 void (*idct
)(DCTELEM
*block
/* align 16*/);
390 * block -> idct -> clip to unsigned 8 bit -> dest.
391 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
392 * @param line_size size in bytes of a horizontal line of dest
394 void (*idct_put
)(uint8_t *dest
/*align 8*/, int line_size
, DCTELEM
*block
/*align 16*/);
397 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
398 * @param line_size size in bytes of a horizontal line of dest
400 void (*idct_add
)(uint8_t *dest
/*align 8*/, int line_size
, DCTELEM
*block
/*align 16*/);
403 * idct input permutation.
404 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
406 * this permutation must be performed before the idct_put/add, note, normally this can be merged
407 * with the zigzag/alternate scan<br>
408 * an example to avoid confusion:
409 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
410 * - (x -> referece dct -> reference idct -> x)
411 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
412 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
414 uint8_t idct_permutation
[64];
415 int idct_permutation_type
;
416 #define FF_NO_IDCT_PERM 1
417 #define FF_LIBMPEG2_IDCT_PERM 2
418 #define FF_SIMPLE_IDCT_PERM 3
419 #define FF_TRANSPOSE_IDCT_PERM 4
420 #define FF_PARTTRANS_IDCT_PERM 5
421 #define FF_SSE2_IDCT_PERM 6
423 int (*try_8x8basis
)(int16_t rem
[64], int16_t weight
[64], int16_t basis
[64], int scale
);
424 void (*add_8x8basis
)(int16_t rem
[64], int16_t basis
[64], int scale
);
425 #define BASIS_SHIFT 16
426 #define RECON_SHIFT 6
428 void (*draw_edges
)(uint8_t *buf
, int wrap
, int width
, int height
, int w
);
429 #define EDGE_WIDTH 16
432 void (*h264_idct_add
)(uint8_t *dst
/*align 4*/, DCTELEM
*block
/*align 16*/, int stride
);
433 void (*h264_idct8_add
)(uint8_t *dst
/*align 8*/, DCTELEM
*block
/*align 16*/, int stride
);
434 void (*h264_idct_dc_add
)(uint8_t *dst
/*align 4*/, DCTELEM
*block
/*align 16*/, int stride
);
435 void (*h264_idct8_dc_add
)(uint8_t *dst
/*align 8*/, DCTELEM
*block
/*align 16*/, int stride
);
436 void (*h264_dct
)(DCTELEM block
[4][4]);
439 void (*vertical_compose97i
)(IDWTELEM
*b0
, IDWTELEM
*b1
, IDWTELEM
*b2
, IDWTELEM
*b3
, IDWTELEM
*b4
, IDWTELEM
*b5
, int width
);
440 void (*horizontal_compose97i
)(IDWTELEM
*b
, int width
);
441 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
);
443 void (*prefetch
)(void *mem
, int stride
, int h
);
445 void (*shrink
[4])(uint8_t *dst
, int dst_wrap
, const uint8_t *src
, int src_wrap
, int width
, int height
);
448 void (*vc1_inv_trans_8x8
)(DCTELEM
*b
);
449 void (*vc1_inv_trans_8x4
)(uint8_t *dest
, int line_size
, DCTELEM
*block
);
450 void (*vc1_inv_trans_4x8
)(uint8_t *dest
, int line_size
, DCTELEM
*block
);
451 void (*vc1_inv_trans_4x4
)(uint8_t *dest
, int line_size
, DCTELEM
*block
);
452 void (*vc1_v_overlap
)(uint8_t* src
, int stride
);
453 void (*vc1_h_overlap
)(uint8_t* src
, int stride
);
454 /* put 8x8 block with bicubic interpolation and quarterpel precision
455 * last argument is actually round value instead of height
457 op_pixels_func put_vc1_mspel_pixels_tab
[16];
459 /* intrax8 functions */
460 void (*x8_spatial_compensation
[12])(uint8_t *src
, uint8_t *dst
, int linesize
);
461 void (*x8_setup_spatial_compensation
)(uint8_t *src
, uint8_t *dst
, int linesize
,
462 int * range
, int * sum
, int edges
);
466 * Add contents of the second vector to the first one.
467 * @param len length of vectors, should be multiple of 16
469 void (*add_int16
)(int16_t *v1
/*align 16*/, int16_t *v2
, int len
);
471 * Add contents of the second vector to the first one.
472 * @param len length of vectors, should be multiple of 16
474 void (*sub_int16
)(int16_t *v1
/*align 16*/, int16_t *v2
, int len
);
476 * Calculate scalar product of two vectors.
477 * @param len length of vectors, should be multiple of 16
478 * @param shift number of bits to discard from product
480 int32_t (*scalarproduct_int16
)(int16_t *v1
, int16_t *v2
/*align 16*/, int len
, int shift
);
483 void dsputil_static_init(void);
484 void dsputil_init(DSPContext
* p
, AVCodecContext
*avctx
);
486 int ff_check_alignment(void);
489 * permute block according to permuatation.
490 * @param last last non zero element in scantable order
492 void ff_block_permute(DCTELEM
*block
, uint8_t *permutation
, const uint8_t *scantable
, int last
);
494 void ff_set_cmp(DSPContext
* c
, me_cmp_func
*cmp
, int type
);
496 #define BYTE_VEC32(c) ((c)*0x01010101UL)
498 static inline uint32_t rnd_avg32(uint32_t a
, uint32_t b
)
500 return (a
| b
) - (((a
^ b
) & ~BYTE_VEC32(0x01)) >> 1);
503 static inline uint32_t no_rnd_avg32(uint32_t a
, uint32_t b
)
505 return (a
& b
) + (((a
^ b
) & ~BYTE_VEC32(0x01)) >> 1);
508 static inline int get_penalty_factor(int lambda
, int lambda2
, int type
){
512 return lambda
>>FF_LAMBDA_SHIFT
;
514 return (3*lambda
)>>(FF_LAMBDA_SHIFT
+1);
516 return (4*lambda
)>>(FF_LAMBDA_SHIFT
);
518 return (2*lambda
)>>(FF_LAMBDA_SHIFT
);
521 return (2*lambda
)>>FF_LAMBDA_SHIFT
;
526 return lambda2
>>FF_LAMBDA_SHIFT
;
534 * this must be called between any dsp function and float/double code.
535 * for example sin(); dsp->idct_put(); emms_c(); cos()
539 /* should be defined by architectures supporting
540 one or more MultiMedia extension */
541 int mm_support(void);
543 void dsputil_init_alpha(DSPContext
* c
, AVCodecContext
*avctx
);
544 void dsputil_init_armv4l(DSPContext
* c
, AVCodecContext
*avctx
);
545 void dsputil_init_bfin(DSPContext
* c
, AVCodecContext
*avctx
);
546 void dsputil_init_mlib(DSPContext
* c
, AVCodecContext
*avctx
);
547 void dsputil_init_mmi(DSPContext
* c
, AVCodecContext
*avctx
);
548 void dsputil_init_mmx(DSPContext
* c
, AVCodecContext
*avctx
);
549 void dsputil_init_ppc(DSPContext
* c
, AVCodecContext
*avctx
);
550 void dsputil_init_sh4(DSPContext
* c
, AVCodecContext
*avctx
);
551 void dsputil_init_vis(DSPContext
* c
, AVCodecContext
*avctx
);
553 #define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v)
555 #if defined(HAVE_MMX)
559 #define MM_MMX 0x0001 /* standard MMX */
560 #define MM_3DNOW 0x0004 /* AMD 3DNOW */
561 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
562 #define MM_SSE 0x0008 /* SSE functions */
563 #define MM_SSE2 0x0010 /* PIV SSE2 functions */
564 #define MM_3DNOWEXT 0x0020 /* AMD 3DNowExt */
565 #define MM_SSE3 0x0040 /* Prescott SSE3 functions */
566 #define MM_SSSE3 0x0080 /* Conroe SSSE3 functions */
570 void add_pixels_clamped_mmx(const DCTELEM
*block
, uint8_t *pixels
, int line_size
);
571 void put_pixels_clamped_mmx(const DCTELEM
*block
, uint8_t *pixels
, int line_size
);
572 void put_signed_pixels_clamped_mmx(const DCTELEM
*block
, uint8_t *pixels
, int line_size
);
574 static inline void emms(void)
576 asm volatile ("emms;":::"memory");
582 if (mm_flags & MM_MMX)\
586 void dsputil_init_pix_mmx(DSPContext
* c
, AVCodecContext
*avctx
);
588 #elif defined(ARCH_ARMV4L)
590 #define MM_IWMMXT 0x0100 /* XScale IWMMXT */
595 # define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
596 # define STRIDE_ALIGN 16
599 #elif defined(ARCH_POWERPC)
601 #define MM_ALTIVEC 0x0001 /* standard AltiVec */
605 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
606 #define STRIDE_ALIGN 16
608 #elif defined(HAVE_MMI)
610 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
611 #define STRIDE_ALIGN 16
616 #define mm_support() 0
620 #ifndef DECLARE_ALIGNED_8
621 # define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
625 # define STRIDE_ALIGN 8
629 void get_psnr(uint8_t *orig_image
[3], uint8_t *coded_image
[3],
630 int orig_linesize
[3], int coded_linesize
,
631 AVCodecContext
*avctx
);
633 /* FFT computation */
635 /* NOTE: soon integer code will be added, so you must use the
637 typedef float FFTSample
;
641 typedef struct FFTComplex
{
645 typedef struct FFTContext
{
650 FFTComplex
*exptab1
; /* only used by SSE code */
652 void (*fft_permute
)(struct FFTContext
*s
, FFTComplex
*z
);
653 void (*fft_calc
)(struct FFTContext
*s
, FFTComplex
*z
);
654 void (*imdct_calc
)(struct MDCTContext
*s
, FFTSample
*output
, const FFTSample
*input
);
655 void (*imdct_half
)(struct MDCTContext
*s
, FFTSample
*output
, const FFTSample
*input
);
658 int ff_fft_init(FFTContext
*s
, int nbits
, int inverse
);
659 void ff_fft_permute_c(FFTContext
*s
, FFTComplex
*z
);
660 void ff_fft_permute_sse(FFTContext
*s
, FFTComplex
*z
);
661 void ff_fft_calc_c(FFTContext
*s
, FFTComplex
*z
);
662 void ff_fft_calc_sse(FFTContext
*s
, FFTComplex
*z
);
663 void ff_fft_calc_3dn(FFTContext
*s
, FFTComplex
*z
);
664 void ff_fft_calc_3dn2(FFTContext
*s
, FFTComplex
*z
);
665 void ff_fft_calc_altivec(FFTContext
*s
, FFTComplex
*z
);
667 static inline void ff_fft_permute(FFTContext
*s
, FFTComplex
*z
)
669 s
->fft_permute(s
, z
);
671 static inline void ff_fft_calc(FFTContext
*s
, FFTComplex
*z
)
675 void ff_fft_end(FFTContext
*s
);
677 /* MDCT computation */
679 typedef struct MDCTContext
{
680 int n
; /* size of MDCT (i.e. number of input data * 2) */
681 int nbits
; /* n = 2^nbits */
682 /* pre/post rotation tables */
688 static inline void ff_imdct_calc(MDCTContext
*s
, FFTSample
*output
, const FFTSample
*input
)
690 s
->fft
.imdct_calc(s
, output
, input
);
692 static inline void ff_imdct_half(MDCTContext
*s
, FFTSample
*output
, const FFTSample
*input
)
694 s
->fft
.imdct_half(s
, output
, input
);
698 * Generate a Kaiser-Bessel Derived Window.
699 * @param window pointer to half window
700 * @param alpha determines window shape
701 * @param n size of half window
703 void ff_kbd_window_init(float *window
, float alpha
, int n
);
706 * Generate a sine window.
707 * @param window pointer to half window
708 * @param n size of half window
710 void ff_sine_window_init(float *window
, int n
);
711 extern float ff_sine_128
[ 128];
712 extern float ff_sine_256
[ 256];
713 extern float ff_sine_512
[ 512];
714 extern float ff_sine_1024
[1024];
715 extern float ff_sine_2048
[2048];
716 extern float *ff_sine_windows
[5];
718 int ff_mdct_init(MDCTContext
*s
, int nbits
, int inverse
);
719 void ff_imdct_calc_c(MDCTContext
*s
, FFTSample
*output
, const FFTSample
*input
);
720 void ff_imdct_half_c(MDCTContext
*s
, FFTSample
*output
, const FFTSample
*input
);
721 void ff_imdct_calc_3dn(MDCTContext
*s
, FFTSample
*output
, const FFTSample
*input
);
722 void ff_imdct_half_3dn(MDCTContext
*s
, FFTSample
*output
, const FFTSample
*input
);
723 void ff_imdct_calc_3dn2(MDCTContext
*s
, FFTSample
*output
, const FFTSample
*input
);
724 void ff_imdct_half_3dn2(MDCTContext
*s
, FFTSample
*output
, const FFTSample
*input
);
725 void ff_imdct_calc_sse(MDCTContext
*s
, FFTSample
*output
, const FFTSample
*input
);
726 void ff_imdct_half_sse(MDCTContext
*s
, FFTSample
*output
, const FFTSample
*input
);
727 void ff_mdct_calc(MDCTContext
*s
, FFTSample
*out
, const FFTSample
*input
);
728 void ff_mdct_end(MDCTContext
*s
);
730 #define WRAPPER8_16(name8, name16)\
731 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
732 return name8(s, dst , src , stride, h)\
733 +name8(s, dst+8 , src+8 , stride, h);\
736 #define WRAPPER8_16_SQ(name8, name16)\
737 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
739 score +=name8(s, dst , src , stride, 8);\
740 score +=name8(s, dst+8 , src+8 , stride, 8);\
744 score +=name8(s, dst , src , stride, 8);\
745 score +=name8(s, dst+8 , src+8 , stride, 8);\
751 static inline void copy_block2(uint8_t *dst
, uint8_t *src
, int dstStride
, int srcStride
, int h
)
756 AV_WN16(dst
, AV_RN16(src
));
762 static inline void copy_block4(uint8_t *dst
, uint8_t *src
, int dstStride
, int srcStride
, int h
)
767 AV_WN32(dst
, AV_RN32(src
));
773 static inline void copy_block8(uint8_t *dst
, uint8_t *src
, int dstStride
, int srcStride
, int h
)
778 AV_WN32(dst
, AV_RN32(src
));
779 AV_WN32(dst
+4 , AV_RN32(src
+4 ));
785 static inline void copy_block9(uint8_t *dst
, uint8_t *src
, int dstStride
, int srcStride
, int h
)
790 AV_WN32(dst
, AV_RN32(src
));
791 AV_WN32(dst
+4 , AV_RN32(src
+4 ));
798 static inline void copy_block16(uint8_t *dst
, uint8_t *src
, int dstStride
, int srcStride
, int h
)
803 AV_WN32(dst
, AV_RN32(src
));
804 AV_WN32(dst
+4 , AV_RN32(src
+4 ));
805 AV_WN32(dst
+8 , AV_RN32(src
+8 ));
806 AV_WN32(dst
+12, AV_RN32(src
+12));
812 static inline void copy_block17(uint8_t *dst
, uint8_t *src
, int dstStride
, int srcStride
, int h
)
817 AV_WN32(dst
, AV_RN32(src
));
818 AV_WN32(dst
+4 , AV_RN32(src
+4 ));
819 AV_WN32(dst
+8 , AV_RN32(src
+8 ));
820 AV_WN32(dst
+12, AV_RN32(src
+12));
827 #endif /* AVCODEC_DSPUTIL_H */