1 /*
2  * DSP utils
3  * Copyright (c) 2000, 2001, 2002 Fabrice Bellard.
4  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5  *
6  * This file is part of FFmpeg.
7  *
8  * FFmpeg is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU Lesser General Public
10  * License as published by the Free Software Foundation; either
11  * version 2.1 of the License, or (at your option) any later version.
12  *
13  * FFmpeg is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16  * Lesser General Public License for more details.
17  *
18  * You should have received a copy of the GNU Lesser General Public
19  * License along with FFmpeg; if not, write to the Free Software
20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21  */
22 
23 /**
24  * @file dsputil.h
25  * DSP utils.
26  * note, many functions in here may use MMX which trashes the FPU state, it is
27  * absolutely necessary to call emms_c() between dsp & float/double code
28  */
29 
30 #ifndef FFMPEG_DSPUTIL_H
31 #define FFMPEG_DSPUTIL_H
32 
33 #include "avcodec.h"
34 
35 
36 //#define DEBUG
37 /* dct code */
38 typedef short DCTELEM;
39 typedef int DWTELEM;
40 typedef short IDWTELEM;
41 
42 void fdct_ifast (DCTELEM *data);
43 void fdct_ifast248 (DCTELEM *data);
44 void ff_jpeg_fdct_islow (DCTELEM *data);
45 void ff_fdct248_islow (DCTELEM *data);
46 
47 void j_rev_dct (DCTELEM *data);
48 void j_rev_dct4 (DCTELEM *data);
49 void j_rev_dct2 (DCTELEM *data);
50 void j_rev_dct1 (DCTELEM *data);
51 void ff_wmv2_idct_c(DCTELEM *data);
52 
53 void ff_fdct_mmx(DCTELEM *block);
54 void ff_fdct_mmx2(DCTELEM *block);
55 void ff_fdct_sse2(DCTELEM *block);
56 
57 void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
58 void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
59 void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
60 void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
61 void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
62 void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
63 
64 void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1,
65                               const float *src2, int src3, int blocksize, int step);
66 void ff_float_to_int16_c(int16_t *dst, const float *src, int len);
67 
68 /* encoding scans */
69 extern const uint8_t ff_alternate_horizontal_scan[64];
70 extern const uint8_t ff_alternate_vertical_scan[64];
71 extern const uint8_t ff_zigzag_direct[64];
72 extern const uint8_t ff_zigzag248_direct[64];
73 
74 /* pixel operations */
75 #define MAX_NEG_CROP 1024
76 
77 /* temporary */
78 extern uint32_t ff_squareTbl[512];
79 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
80 
81 /* VP3 DSP functions */
82 void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
83 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
84 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
85 
86 /* 1/2^n downscaling functions from imgconvert.c */
87 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
88 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
89 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
90 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
91 
92 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
93               int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
94 
95 /* minimum alignment rules ;)
96 If you notice errors in the align stuff, need more alignment for some ASM code
97 for some CPU or need to use a function with less aligned data then send a mail
98 to the ffmpeg-devel mailing list, ...
99 
100 !warning These alignments might not match reality, (missing attribute((align))
101 stuff somewhere possible).
102 I (Michael) did not check them, these are just the alignments which I think
103 could be reached easily ...
104 
105 !future video codecs might need functions with less strict alignment
106 */
107 
108 /*
109 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
110 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
111 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
112 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
113 void clear_blocks_c(DCTELEM *blocks);
114 */
115 
116 /* add and put pixel (decoding) */
117 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
118 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
119 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
120 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);
121 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
122 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
123 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
124 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
125 
126 #define DEF_OLD_QPEL(name)\
127 void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
128 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
129 void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
130 
131 DEF_OLD_QPEL(qpel16_mc11_old_c)
132 DEF_OLD_QPEL(qpel16_mc31_old_c)
133 DEF_OLD_QPEL(qpel16_mc12_old_c)
134 DEF_OLD_QPEL(qpel16_mc32_old_c)
135 DEF_OLD_QPEL(qpel16_mc13_old_c)
136 DEF_OLD_QPEL(qpel16_mc33_old_c)
137 DEF_OLD_QPEL(qpel8_mc11_old_c)
138 DEF_OLD_QPEL(qpel8_mc31_old_c)
139 DEF_OLD_QPEL(qpel8_mc12_old_c)
140 DEF_OLD_QPEL(qpel8_mc32_old_c)
141 DEF_OLD_QPEL(qpel8_mc13_old_c)
142 DEF_OLD_QPEL(qpel8_mc33_old_c)
143 
144 #define CALL_2X_PIXELS(a, b, n)\
145 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
146     b(block  , pixels  , line_size, h);\
147     b(block+n, pixels+n, line_size, h);\
148 }
149 
150 /* motion estimation */
151 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
152 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
153 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))*/;
154 
155 
156 // for snow slices
157 typedef struct slice_buffer_s slice_buffer;
158 
159 /**
160  * Scantable.
161  */
162 typedef struct ScanTable{
163     const uint8_t *scantable;
164     uint8_t permutated[64];
165     uint8_t raster_end[64];
166 #ifdef ARCH_POWERPC
167                 /** Used by dct_quantize_altivec to find last-non-zero */
168     DECLARE_ALIGNED(16, uint8_t, inverse[64]);
169 #endif
170 } ScanTable;
171 
172 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
173 
174 void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize,
175                          int block_w, int block_h,
176                          int src_x, int src_y, int w, int h);
177 
178 /**
179  * DSPContext.
180  */
181 typedef struct DSPContext {
182     /* pixel ops : interface with DCT */
183     void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
184     void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
185     void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
186     void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
187     void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
188     void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
189     void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
190     int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
191     /**
192      * translational global motion compensation.
193      */
194     void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
195     /**
196      * global motion compensation.
197      */
198     void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
199                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
200     void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
201     int (*pix_sum)(uint8_t * pix, int line_size);
202     int (*pix_norm1)(uint8_t * pix, int line_size);
203 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
204 
205     me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
206     me_cmp_func sse[5];
207     me_cmp_func hadamard8_diff[5];
208     me_cmp_func dct_sad[5];
209     me_cmp_func quant_psnr[5];
210     me_cmp_func bit[5];
211     me_cmp_func rd[5];
212     me_cmp_func vsad[5];
213     me_cmp_func vsse[5];
214     me_cmp_func nsse[5];
215     me_cmp_func w53[5];
216     me_cmp_func w97[5];
217     me_cmp_func dct_max[5];
218     me_cmp_func dct264_sad[5];
219 
220     me_cmp_func me_pre_cmp[5];
221     me_cmp_func me_cmp[5];
222     me_cmp_func me_sub_cmp[5];
223     me_cmp_func mb_cmp[5];
224     me_cmp_func ildct_cmp[5]; //only width 16 used
225     me_cmp_func frame_skip_cmp[5]; //only width 8 used
226 
227     int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
228                              int size);
229 
230     /**
231      * Halfpel motion compensation with rounding (a+b+1)>>1.
232      * this is an array[4][4] of motion compensation functions for 4
233      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
234      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
235      * @param block destination where the result is stored
236      * @param pixels source
237      * @param line_size number of bytes in a horizontal line of block
238      * @param h height
239      */
240     op_pixels_func put_pixels_tab[4][4];
241 
242     /**
243      * Halfpel motion compensation with rounding (a+b+1)>>1.
244      * This is an array[4][4] of motion compensation functions for 4
245      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
246      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
247      * @param block destination into which the result is averaged (a+b+1)>>1
248      * @param pixels source
249      * @param line_size number of bytes in a horizontal line of block
250      * @param h height
251      */
252     op_pixels_func avg_pixels_tab[4][4];
253 
254     /**
255      * Halfpel motion compensation with no rounding (a+b)>>1.
256      * this is an array[2][4] of motion compensation functions for 2
257      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
258      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
259      * @param block destination where the result is stored
260      * @param pixels source
261      * @param line_size number of bytes in a horizontal line of block
262      * @param h height
263      */
264     op_pixels_func put_no_rnd_pixels_tab[4][4];
265 
266     /**
267      * Halfpel motion compensation with no rounding (a+b)>>1.
268      * this is an array[2][4] of motion compensation functions for 2
269      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
270      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
271      * @param block destination into which the result is averaged (a+b)>>1
272      * @param pixels source
273      * @param line_size number of bytes in a horizontal line of block
274      * @param h height
275      */
276     op_pixels_func avg_no_rnd_pixels_tab[4][4];
277 
278     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);
279 
280     /**
281      * Thirdpel motion compensation with rounding (a+b+1)>>1.
282      * this is an array[12] of motion compensation functions for the 9 thirdpe
283      * positions<br>
284      * *pixels_tab[ xthirdpel + 4*ythirdpel ]
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
288      * @param h height
289      */
290     tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
291     tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
292 
293     qpel_mc_func put_qpel_pixels_tab[2][16];
294     qpel_mc_func avg_qpel_pixels_tab[2][16];
295     qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
296     qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
297     qpel_mc_func put_mspel_pixels_tab[8];
298 
299     /**
300      * h264 Chroma MC
301      */
302     h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
303     /* This is really one func used in VC-1 decoding */
304     h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
305     h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
306 
307     qpel_mc_func put_h264_qpel_pixels_tab[4][16];
308     qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
309 
310     qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
311     qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
312 
313     h264_weight_func weight_h264_pixels_tab[10];
314     h264_biweight_func biweight_h264_pixels_tab[10];
315 
316     /* AVS specific */
317     qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
318     qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
319     void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
320     void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
321     void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
322     void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
323     void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
324 
325     me_cmp_func pix_abs[2][4];
326 
327     /* huffyuv specific */
328     void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
329     void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
330     void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
331     /**
332      * subtract huffyuv's variant of median prediction
333      * note, this might read from src1[-1], src2[-1]
334      */
335     void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
336     /* this might write to dst[w] */
337     void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
338     void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
339 
340     void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
341     void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
342     void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
343     void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
344     void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
345     void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
346     // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
347     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],
348                                       int bidir, int edges, int step, int mask_mv0, int mask_mv1);
349 
350     void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
351     void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
352 
353     void (*h261_loop_filter)(uint8_t *src, int stride);
354 
355     void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
356     void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
357 
358     /* assume len is a multiple of 4, and arrays are 16-byte aligned */
359     void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
360     /* no alignment needed */
361     void (*flac_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
362     /* assume len is a multiple of 8, and arrays are 16-byte aligned */
363     void (*vector_fmul)(float *dst, const float *src, int len);
364     void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
365     /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
366     void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
367 
368     /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
369      * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
370     void (*float_to_int16)(int16_t *dst, const float *src, int len);
371 
372     /* (I)DCT */
373     void (*fdct)(DCTELEM *block/* align 16*/);
374     void (*fdct248)(DCTELEM *block/* align 16*/);
375 
376     /* IDCT really*/
377     void (*idct)(DCTELEM *block/* align 16*/);
378 
379     /**
380      * block -> idct -> clip to unsigned 8 bit -> dest.
381      * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
382      * @param line_size size in bytes of a horizontal line of dest
383      */
384     void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
385 
386     /**
387      * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
388      * @param line_size size in bytes of a horizontal line of dest
389      */
390     void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
391 
392     /**
393      * idct input permutation.
394      * several optimized IDCTs need a permutated input (relative to the normal order of the reference
395      * IDCT)
396      * this permutation must be performed before the idct_put/add, note, normally this can be merged
397      * with the zigzag/alternate scan<br>
398      * an example to avoid confusion:
399      * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
400      * - (x -> referece dct -> reference idct -> x)
401      * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
402      * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
403      */
404     uint8_t idct_permutation[64];
405     int idct_permutation_type;
406 #define FF_NO_IDCT_PERM 1
407 #define FF_LIBMPEG2_IDCT_PERM 2
408 #define FF_SIMPLE_IDCT_PERM 3
409 #define FF_TRANSPOSE_IDCT_PERM 4
410 #define FF_PARTTRANS_IDCT_PERM 5
411 #define FF_SSE2_IDCT_PERM 6
412 
413     int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
414     void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
415 #define BASIS_SHIFT 16
416 #define RECON_SHIFT 6
417 
418     void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
419 #define EDGE_WIDTH 16
420 
421     /* h264 functions */
422     void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
423     void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
424     void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
425     void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
426     void (*h264_dct)(DCTELEM block[4][4]);
427 
428     /* snow wavelet */
429     void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
430     void (*horizontal_compose97i)(IDWTELEM *b, int width);
431     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);
432 
433     void (*prefetch)(void *mem, int stride, int h);
434 
435     void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
436 
437     /* vc1 functions */
438     void (*vc1_inv_trans_8x8)(DCTELEM *b);
439     void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
440     void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
441     void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
442     void (*vc1_v_overlap)(uint8_t* src, int stride);
443     void (*vc1_h_overlap)(uint8_t* src, int stride);
444     /* put 8x8 block with bicubic interpolation and quarterpel precision
445      * last argument is actually round value instead of height
446      */
447     op_pixels_func put_vc1_mspel_pixels_tab[16];
448 
449     /* intrax8 functions */
450     void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
451     void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
452            int * range, int * sum,  int edges);
453 
454 } DSPContext;
455 
456 void dsputil_static_init(void);
457 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
458 
459 int ff_check_alignment(void);
460 
461 /**
462  * permute block according to permuatation.
463  * @param last last non zero element in scantable order
464  */
465 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
466 
467 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
468 
469 #define         BYTE_VEC32(c)   ((c)*0x01010101UL)
470 
rnd_avg32(uint32_t a,uint32_t b)471 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
472 {
473     return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
474 }
475 
no_rnd_avg32(uint32_t a,uint32_t b)476 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
477 {
478     return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
479 }
480 
get_penalty_factor(int lambda,int lambda2,int type)481 static inline int get_penalty_factor(int lambda, int lambda2, int type){
482     switch(type&0xFF){
483     default:
484     case FF_CMP_SAD:
485         return lambda>>FF_LAMBDA_SHIFT;
486     case FF_CMP_DCT:
487         return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
488     case FF_CMP_W53:
489         return (4*lambda)>>(FF_LAMBDA_SHIFT);
490     case FF_CMP_W97:
491         return (2*lambda)>>(FF_LAMBDA_SHIFT);
492     case FF_CMP_SATD:
493     case FF_CMP_DCT264:
494         return (2*lambda)>>FF_LAMBDA_SHIFT;
495     case FF_CMP_RD:
496     case FF_CMP_PSNR:
497     case FF_CMP_SSE:
498     case FF_CMP_NSSE:
499         return lambda2>>FF_LAMBDA_SHIFT;
500     case FF_CMP_BIT:
501         return 1;
502     }
503 }
504 
505 /**
506  * Empty mmx state.
507  * this must be called between any dsp function and float/double code.
508  * for example sin(); dsp->idct_put(); emms_c(); cos()
509  */
510 #define emms_c()
511 
512 /* should be defined by architectures supporting
513    one or more MultiMedia extension */
514 int mm_support(void);
515 
516 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
517 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
518 void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
519 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
520 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
521 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
522 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
523 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
524 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
525 
526 #define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v)
527 
528 #if defined(HAVE_MMX)
529 
530 #undef emms_c
531 
532 #define MM_MMX    0x0001 /* standard MMX */
533 #define MM_3DNOW  0x0004 /* AMD 3DNOW */
534 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
535 #define MM_SSE    0x0008 /* SSE functions */
536 #define MM_SSE2   0x0010 /* PIV SSE2 functions */
537 #define MM_3DNOWEXT  0x0020 /* AMD 3DNowExt */
538 #define MM_SSE3   0x0040 /* Prescott SSE3 functions */
539 #define MM_SSSE3  0x0080 /* Conroe SSSE3 functions */
540 
541 extern int mm_flags;
542 
543 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
544 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
545 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
546 
emms(void)547 static inline void emms(void)
548 {
549     asm volatile ("emms;":::"memory");
550 }
551 
552 
553 #define emms_c() \
554 {\
555     if (mm_flags & MM_MMX)\
556         emms();\
557 }
558 
559 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
560 
561 #elif defined(ARCH_ARMV4L)
562 
563 #define MM_IWMMXT    0x0100 /* XScale IWMMXT */
564 
565 extern int mm_flags;
566 
567 #elif defined(ARCH_POWERPC)
568 
569 #define MM_ALTIVEC    0x0001 /* standard AltiVec */
570 
571 extern int mm_flags;
572 
573 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
574 #define STRIDE_ALIGN 16
575 
576 #elif defined(HAVE_MMI)
577 
578 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
579 #define STRIDE_ALIGN 16
580 
581 #endif
582 
583 #ifndef DECLARE_ALIGNED_8
584 #   define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
585 #endif
586 
587 #ifndef STRIDE_ALIGN
588 #   define STRIDE_ALIGN 8
589 #endif
590 
591 /* PSNR */
592 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
593               int orig_linesize[3], int coded_linesize,
594               AVCodecContext *avctx);
595 
596 /* FFT computation */
597 
598 /* NOTE: soon integer code will be added, so you must use the
599    FFTSample type */
600 typedef float FFTSample;
601 
602 struct MDCTContext;
603 
604 typedef struct FFTComplex {
605     FFTSample re, im;
606 } FFTComplex;
607 
608 typedef struct FFTContext {
609     int nbits;
610     int inverse;
611     uint16_t *revtab;
612     FFTComplex *exptab;
613     FFTComplex *exptab1; /* only used by SSE code */
614     void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
615     void (*imdct_calc)(struct MDCTContext *s, FFTSample *output,
616                        const FFTSample *input, FFTSample *tmp);
617 } FFTContext;
618 
619 int ff_fft_init(FFTContext *s, int nbits, int inverse);
620 void ff_fft_permute(FFTContext *s, FFTComplex *z);
621 void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
622 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
623 void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
624 void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
625 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
626 
ff_fft_calc(FFTContext * s,FFTComplex * z)627 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
628 {
629     s->fft_calc(s, z);
630 }
631 void ff_fft_end(FFTContext *s);
632 
633 /* MDCT computation */
634 
635 typedef struct MDCTContext {
636     int n;  /* size of MDCT (i.e. number of input data * 2) */
637     int nbits; /* n = 2^nbits */
638     /* pre/post rotation tables */
639     FFTSample *tcos;
640     FFTSample *tsin;
641     FFTContext fft;
642 } MDCTContext;
643 
644 /**
645  * Generate a Kaiser-Bessel Derived Window.
646  * @param   window  pointer to half window
647  * @param   alpha   determines window shape
648  * @param   n       size of half window
649  */
650 void ff_kbd_window_init(float *window, float alpha, int n);
651 
652 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
653 void ff_imdct_calc(MDCTContext *s, FFTSample *output,
654                 const FFTSample *input, FFTSample *tmp);
655 void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output,
656                         const FFTSample *input, FFTSample *tmp);
657 void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output,
658                        const FFTSample *input, FFTSample *tmp);
659 void ff_mdct_calc(MDCTContext *s, FFTSample *out,
660                const FFTSample *input, FFTSample *tmp);
661 void ff_mdct_end(MDCTContext *s);
662 
663 #define WRAPPER8_16(name8, name16)\
664 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
665     return name8(s, dst           , src           , stride, h)\
666           +name8(s, dst+8         , src+8         , stride, h);\
667 }
668 
669 #define WRAPPER8_16_SQ(name8, name16)\
670 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
671     int score=0;\
672     score +=name8(s, dst           , src           , stride, 8);\
673     score +=name8(s, dst+8         , src+8         , stride, 8);\
674     if(h==16){\
675         dst += 8*stride;\
676         src += 8*stride;\
677         score +=name8(s, dst           , src           , stride, 8);\
678         score +=name8(s, dst+8         , src+8         , stride, 8);\
679     }\
680     return score;\
681 }
682 
683 
copy_block2(uint8_t * dst,uint8_t * src,int dstStride,int srcStride,int h)684 static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
685 {
686     int i;
687     for(i=0; i<h; i++)
688     {
689         AV_WN16(dst   , AV_RN16(src   ));
690         dst+=dstStride;
691         src+=srcStride;
692     }
693 }
694 
copy_block4(uint8_t * dst,uint8_t * src,int dstStride,int srcStride,int h)695 static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
696 {
697     int i;
698     for(i=0; i<h; i++)
699     {
700         AV_WN32(dst   , AV_RN32(src   ));
701         dst+=dstStride;
702         src+=srcStride;
703     }
704 }
705 
copy_block8(uint8_t * dst,uint8_t * src,int dstStride,int srcStride,int h)706 static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
707 {
708     int i;
709     for(i=0; i<h; i++)
710     {
711         AV_WN32(dst   , AV_RN32(src   ));
712         AV_WN32(dst+4 , AV_RN32(src+4 ));
713         dst+=dstStride;
714         src+=srcStride;
715     }
716 }
717 
copy_block9(uint8_t * dst,uint8_t * src,int dstStride,int srcStride,int h)718 static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
719 {
720     int i;
721     for(i=0; i<h; i++)
722     {
723         AV_WN32(dst   , AV_RN32(src   ));
724         AV_WN32(dst+4 , AV_RN32(src+4 ));
725         dst[8]= src[8];
726         dst+=dstStride;
727         src+=srcStride;
728     }
729 }
730 
copy_block16(uint8_t * dst,uint8_t * src,int dstStride,int srcStride,int h)731 static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
732 {
733     int i;
734     for(i=0; i<h; i++)
735     {
736         AV_WN32(dst   , AV_RN32(src   ));
737         AV_WN32(dst+4 , AV_RN32(src+4 ));
738         AV_WN32(dst+8 , AV_RN32(src+8 ));
739         AV_WN32(dst+12, AV_RN32(src+12));
740         dst+=dstStride;
741         src+=srcStride;
742     }
743 }
744 
copy_block17(uint8_t * dst,uint8_t * src,int dstStride,int srcStride,int h)745 static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
746 {
747     int i;
748     for(i=0; i<h; i++)
749     {
750         AV_WN32(dst   , AV_RN32(src   ));
751         AV_WN32(dst+4 , AV_RN32(src+4 ));
752         AV_WN32(dst+8 , AV_RN32(src+8 ));
753         AV_WN32(dst+12, AV_RN32(src+12));
754         dst[16]= src[16];
755         dst+=dstStride;
756         src+=srcStride;
757     }
758 }
759 
760 #endif /* FFMPEG_DSPUTIL_H */
761