FFmpeg: libavcodec/dsputil.h Source File

00001 /*
00002  * DSP utils
00003  * Copyright (c) 2000, 2001, 2002 Fabrice Bellard
00004  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
00005  *
00006  * This file is part of FFmpeg.
00007  *
00008  * FFmpeg is free software; you can redistribute it and/or
00009  * modify it under the terms of the GNU Lesser General Public
00010  * License as published by the Free Software Foundation; either
00011  * version 2.1 of the License, or (at your option) any later version.
00012  *
00013  * FFmpeg is distributed in the hope that it will be useful,
00014  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00015  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00016  * Lesser General Public License for more details.
00017  *
00018  * You should have received a copy of the GNU Lesser General Public
00019  * License along with FFmpeg; if not, write to the Free Software
00020  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00021  */
00022 
00030 #ifndef AVCODEC_DSPUTIL_H
00031 #define AVCODEC_DSPUTIL_H
00032 
00033 #include "libavutil/intreadwrite.h"
00034 #include "avcodec.h"
00035 
00036 
00037 //#define DEBUG
00038 /* dct code */
00039 typedef short DCTELEM;
00040 
00041 void fdct_ifast (DCTELEM *data);
00042 void fdct_ifast248 (DCTELEM *data);
00043 void ff_jpeg_fdct_islow (DCTELEM *data);
00044 void ff_fdct248_islow (DCTELEM *data);
00045 
00046 void j_rev_dct (DCTELEM *data);
00047 void j_rev_dct4 (DCTELEM *data);
00048 void j_rev_dct2 (DCTELEM *data);
00049 void j_rev_dct1 (DCTELEM *data);
00050 void ff_wmv2_idct_c(DCTELEM *data);
00051 
00052 void ff_fdct_mmx(DCTELEM *block);
00053 void ff_fdct_mmx2(DCTELEM *block);
00054 void ff_fdct_sse2(DCTELEM *block);
00055 
00056 void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
00057 void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
00058 void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
00059 void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
00060 void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
00061 void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
00062 void ff_h264_idct_add16_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
00063 void ff_h264_idct_add16intra_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
00064 void ff_h264_idct8_add4_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
00065 void ff_h264_idct_add8_c(uint8_t **dest, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
00066 
00067 void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1,
00068                              const float *win, float add_bias, int len);
00069 void ff_float_to_int16_c(int16_t *dst, const float *src, long len);
00070 void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels);
00071 
00072 /* encoding scans */
00073 extern const uint8_t ff_alternate_horizontal_scan[64];
00074 extern const uint8_t ff_alternate_vertical_scan[64];
00075 extern const uint8_t ff_zigzag_direct[64];
00076 extern const uint8_t ff_zigzag248_direct[64];
00077 
00078 /* pixel operations */
00079 #define MAX_NEG_CROP 1024
00080 
00081 /* temporary */
00082 extern uint32_t ff_squareTbl[512];
00083 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
00084 
00085 void ff_put_pixels8x8_c(uint8_t *dst, uint8_t *src, int stride);
00086 void ff_avg_pixels8x8_c(uint8_t *dst, uint8_t *src, int stride);
00087 void ff_put_pixels16x16_c(uint8_t *dst, uint8_t *src, int stride);
00088 void ff_avg_pixels16x16_c(uint8_t *dst, uint8_t *src, int stride);
00089 
00090 /* VP3 DSP functions */
00091 void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
00092 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
00093 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
00094 void ff_vp3_idct_dc_add_c(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
00095 
00096 void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
00097 void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
00098 
00099 /* Bink functions */
00100 void ff_bink_idct_c    (DCTELEM *block);
00101 void ff_bink_idct_add_c(uint8_t *dest, int linesize, DCTELEM *block);
00102 void ff_bink_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
00103 
00104 /* EA functions */
00105 void ff_ea_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
00106 
00107 /* 1/2^n downscaling functions from imgconvert.c */
00108 #if LIBAVCODEC_VERSION_MAJOR < 53
00109 
00112 attribute_deprecated
00113 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
00114 #endif
00115 
00116 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
00117 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
00118 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
00119 
00120 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
00121               int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
00122 
00123 /* minimum alignment rules ;)
00124 If you notice errors in the align stuff, need more alignment for some ASM code
00125 for some CPU or need to use a function with less aligned data then send a mail
00126 to the ffmpeg-devel mailing list, ...
00127 
00128 !warning These alignments might not match reality, (missing attribute((align))
00129 stuff somewhere possible).
00130 I (Michael) did not check them, these are just the alignments which I think
00131 could be reached easily ...
00132 
00133 !future video codecs might need functions with less strict alignment
00134 */
00135 
00136 /*
00137 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
00138 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
00139 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
00140 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
00141 void clear_blocks_c(DCTELEM *blocks);
00142 */
00143 
00144 /* add and put pixel (decoding) */
00145 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
00146 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
00147 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
00148 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);
00149 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
00150 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
00151 
00152 typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
00153 
00154 #define DEF_OLD_QPEL(name)\
00155 void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
00156 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
00157 void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
00158 
00159 DEF_OLD_QPEL(qpel16_mc11_old_c)
00160 DEF_OLD_QPEL(qpel16_mc31_old_c)
00161 DEF_OLD_QPEL(qpel16_mc12_old_c)
00162 DEF_OLD_QPEL(qpel16_mc32_old_c)
00163 DEF_OLD_QPEL(qpel16_mc13_old_c)
00164 DEF_OLD_QPEL(qpel16_mc33_old_c)
00165 DEF_OLD_QPEL(qpel8_mc11_old_c)
00166 DEF_OLD_QPEL(qpel8_mc31_old_c)
00167 DEF_OLD_QPEL(qpel8_mc12_old_c)
00168 DEF_OLD_QPEL(qpel8_mc32_old_c)
00169 DEF_OLD_QPEL(qpel8_mc13_old_c)
00170 DEF_OLD_QPEL(qpel8_mc33_old_c)
00171 
00172 #define CALL_2X_PIXELS(a, b, n)\
00173 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
00174     b(block  , pixels  , line_size, h);\
00175     b(block+n, pixels+n, line_size, h);\
00176 }
00177 
00178 /* motion estimation */
00179 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
00180 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
00181 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))*/;
00182 
00186 typedef struct ScanTable{
00187     const uint8_t *scantable;
00188     uint8_t permutated[64];
00189     uint8_t raster_end[64];
00190 #if ARCH_PPC
00191 
00192     DECLARE_ALIGNED(16, uint8_t, inverse)[64];
00193 #endif
00194 } ScanTable;
00195 
00196 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
00197 
00198 void ff_emulated_edge_mc(uint8_t *buf, const uint8_t *src, int linesize,
00199                          int block_w, int block_h,
00200                          int src_x, int src_y, int w, int h);
00201 
00205 typedef struct DSPContext {
00206     /* pixel ops : interface with DCT */
00207     void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
00208     void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
00209     void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
00210     void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
00211     void (*put_pixels_nonclamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
00212     void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
00213     void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
00214     void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
00215     int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
00219     void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
00223     void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
00224                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
00225     void (*clear_block)(DCTELEM *block/*align 16*/);
00226     void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
00227     int (*pix_sum)(uint8_t * pix, int line_size);
00228     int (*pix_norm1)(uint8_t * pix, int line_size);
00229 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
00230 
00231     me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
00232     me_cmp_func sse[6];
00233     me_cmp_func hadamard8_diff[6];
00234     me_cmp_func dct_sad[6];
00235     me_cmp_func quant_psnr[6];
00236     me_cmp_func bit[6];
00237     me_cmp_func rd[6];
00238     me_cmp_func vsad[6];
00239     me_cmp_func vsse[6];
00240     me_cmp_func nsse[6];
00241     me_cmp_func w53[6];
00242     me_cmp_func w97[6];
00243     me_cmp_func dct_max[6];
00244     me_cmp_func dct264_sad[6];
00245 
00246     me_cmp_func me_pre_cmp[6];
00247     me_cmp_func me_cmp[6];
00248     me_cmp_func me_sub_cmp[6];
00249     me_cmp_func mb_cmp[6];
00250     me_cmp_func ildct_cmp[6]; //only width 16 used
00251     me_cmp_func frame_skip_cmp[6]; //only width 8 used
00252 
00253     int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
00254                              int size);
00255 
00266     op_pixels_func put_pixels_tab[4][4];
00267 
00278     op_pixels_func avg_pixels_tab[4][4];
00279 
00290     op_pixels_func put_no_rnd_pixels_tab[4][4];
00291 
00302     op_pixels_func avg_no_rnd_pixels_tab[4][4];
00303 
00304     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);
00305 
00316     tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
00317     tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
00318 
00319     qpel_mc_func put_qpel_pixels_tab[2][16];
00320     qpel_mc_func avg_qpel_pixels_tab[2][16];
00321     qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
00322     qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
00323     qpel_mc_func put_mspel_pixels_tab[8];
00324 
00328     h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
00329     h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
00330     /* This is really one func used in VC-1 decoding */
00331     h264_chroma_mc_func put_no_rnd_vc1_chroma_pixels_tab[3];
00332     h264_chroma_mc_func avg_no_rnd_vc1_chroma_pixels_tab[3];
00333 
00334     qpel_mc_func put_h264_qpel_pixels_tab[4][16];
00335     qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
00336 
00337     qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
00338     qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
00339 
00340     me_cmp_func pix_abs[2][4];
00341 
00342     /* huffyuv specific */
00343     void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
00344     void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
00345     void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
00350     void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
00351     void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
00352     int  (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
00353     void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
00354     /* this might write to dst[w] */
00355     void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
00356     void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
00357 
00358     void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
00359     void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
00360 
00361     void (*h261_loop_filter)(uint8_t *src, int stride);
00362 
00363     void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
00364     void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
00365 
00366     void (*vp3_idct_dc_add)(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
00367     void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
00368     void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
00369 
00370     /* assume len is a multiple of 4, and arrays are 16-byte aligned */
00371     void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
00372     void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
00373     /* no alignment needed */
00374     void (*lpc_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
00375     /* assume len is a multiple of 8, and arrays are 16-byte aligned */
00376     void (*vector_fmul)(float *dst, const float *src, int len);
00377     void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
00378     /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
00379     void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len);
00380     /* assume len is a multiple of 4, and arrays are 16-byte aligned */
00381     void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len);
00382     /* assume len is a multiple of 8, and arrays are 16-byte aligned */
00383     void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len);
00384     void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
00393     void (*vector_fmul_scalar)(float *dst, const float *src, float mul,
00394                                int len);
00407     void (*vector_fmul_sv_scalar[2])(float *dst, const float *src,
00408                                      const float **sv, float mul, int len);
00419     void (*sv_fmul_scalar[2])(float *dst, const float **sv,
00420                               float mul, int len);
00427     float (*scalarproduct_float)(const float *v1, const float *v2, int len);
00434     void (*butterflies_float)(float *restrict v1, float *restrict v2, int len);
00435 
00436     /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
00437      * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
00438     void (*float_to_int16)(int16_t *dst, const float *src, long len);
00439     void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
00440 
00441     /* (I)DCT */
00442     void (*fdct)(DCTELEM *block/* align 16*/);
00443     void (*fdct248)(DCTELEM *block/* align 16*/);
00444 
00445     /* IDCT really*/
00446     void (*idct)(DCTELEM *block/* align 16*/);
00447 
00453     void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
00454 
00459     void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
00460 
00473     uint8_t idct_permutation[64];
00474     int idct_permutation_type;
00475 #define FF_NO_IDCT_PERM 1
00476 #define FF_LIBMPEG2_IDCT_PERM 2
00477 #define FF_SIMPLE_IDCT_PERM 3
00478 #define FF_TRANSPOSE_IDCT_PERM 4
00479 #define FF_PARTTRANS_IDCT_PERM 5
00480 #define FF_SSE2_IDCT_PERM 6
00481 
00482     int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
00483     void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
00484 #define BASIS_SHIFT 16
00485 #define RECON_SHIFT 6
00486 
00487     void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
00488 #define EDGE_WIDTH 16
00489 
00490     void (*prefetch)(void *mem, int stride, int h);
00491 
00492     void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
00493 
00494     /* mlp/truehd functions */
00495     void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff,
00496                                int firorder, int iirorder,
00497                                unsigned int filter_shift, int32_t mask, int blocksize,
00498                                int32_t *sample_buffer);
00499 
00500     /* vc1 functions */
00501     void (*vc1_inv_trans_8x8)(DCTELEM *b);
00502     void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
00503     void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
00504     void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
00505     void (*vc1_inv_trans_8x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
00506     void (*vc1_inv_trans_8x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
00507     void (*vc1_inv_trans_4x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
00508     void (*vc1_inv_trans_4x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
00509     void (*vc1_v_overlap)(uint8_t* src, int stride);
00510     void (*vc1_h_overlap)(uint8_t* src, int stride);
00511     void (*vc1_v_loop_filter4)(uint8_t *src, int stride, int pq);
00512     void (*vc1_h_loop_filter4)(uint8_t *src, int stride, int pq);
00513     void (*vc1_v_loop_filter8)(uint8_t *src, int stride, int pq);
00514     void (*vc1_h_loop_filter8)(uint8_t *src, int stride, int pq);
00515     void (*vc1_v_loop_filter16)(uint8_t *src, int stride, int pq);
00516     void (*vc1_h_loop_filter16)(uint8_t *src, int stride, int pq);
00517     /* put 8x8 block with bicubic interpolation and quarterpel precision
00518      * last argument is actually round value instead of height
00519      */
00520     op_pixels_func put_vc1_mspel_pixels_tab[16];
00521     op_pixels_func avg_vc1_mspel_pixels_tab[16];
00522 
00523     /* intrax8 functions */
00524     void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
00525     void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
00526            int * range, int * sum,  int edges);
00527 
00533     int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len, int shift);
00534     /* ape functions */
00540     int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul);
00541 
00542     /* rv30 functions */
00543     qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
00544     qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
00545 
00546     /* rv40 functions */
00547     qpel_mc_func put_rv40_qpel_pixels_tab[4][16];
00548     qpel_mc_func avg_rv40_qpel_pixels_tab[4][16];
00549     h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
00550     h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
00551 
00552     /* bink functions */
00553     op_fill_func fill_block_tab[2];
00554     void (*scale_block)(const uint8_t src[64]/*align 8*/, uint8_t *dst/*align 8*/, int linesize);
00555 } DSPContext;
00556 
00557 void dsputil_static_init(void);
00558 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
00559 
00560 int ff_check_alignment(void);
00561 
00566 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
00567 
00568 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
00569 
00570 #define         BYTE_VEC32(c)   ((c)*0x01010101UL)
00571 
00572 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
00573 {
00574     return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
00575 }
00576 
00577 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
00578 {
00579     return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
00580 }
00581 
00582 static inline int get_penalty_factor(int lambda, int lambda2, int type){
00583     switch(type&0xFF){
00584     default:
00585     case FF_CMP_SAD:
00586         return lambda>>FF_LAMBDA_SHIFT;
00587     case FF_CMP_DCT:
00588         return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
00589     case FF_CMP_W53:
00590         return (4*lambda)>>(FF_LAMBDA_SHIFT);
00591     case FF_CMP_W97:
00592         return (2*lambda)>>(FF_LAMBDA_SHIFT);
00593     case FF_CMP_SATD:
00594     case FF_CMP_DCT264:
00595         return (2*lambda)>>FF_LAMBDA_SHIFT;
00596     case FF_CMP_RD:
00597     case FF_CMP_PSNR:
00598     case FF_CMP_SSE:
00599     case FF_CMP_NSSE:
00600         return lambda2>>FF_LAMBDA_SHIFT;
00601     case FF_CMP_BIT:
00602         return 1;
00603     }
00604 }
00605 
00611 #define emms_c()
00612 
00613 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
00614 void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
00615 void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
00616 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
00617 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
00618 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
00619 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
00620 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
00621 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
00622 
00623 void ff_dsputil_init_dwt(DSPContext *c);
00624 void ff_rv30dsp_init(DSPContext* c, AVCodecContext *avctx);
00625 void ff_rv40dsp_init(DSPContext* c, AVCodecContext *avctx);
00626 void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx);
00627 void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx);
00628 void ff_mlp_init(DSPContext* c, AVCodecContext *avctx);
00629 void ff_mlp_init_x86(DSPContext* c, AVCodecContext *avctx);
00630 
00631 #if HAVE_MMX
00632 
00633 #undef emms_c
00634 
00635 static inline void emms(void)
00636 {
00637     __asm__ volatile ("emms;":::"memory");
00638 }
00639 
00640 #define emms_c() emms()
00641 
00642 #elif ARCH_ARM
00643 
00644 #if HAVE_NEON
00645 #   define STRIDE_ALIGN 16
00646 #endif
00647 
00648 #elif ARCH_PPC
00649 
00650 #define STRIDE_ALIGN 16
00651 
00652 #elif HAVE_MMI
00653 
00654 #define STRIDE_ALIGN 16
00655 
00656 #endif
00657 
00658 #ifndef STRIDE_ALIGN
00659 #   define STRIDE_ALIGN 8
00660 #endif
00661 
00662 #define LOCAL_ALIGNED(a, t, v, s, ...)                          \
00663     uint8_t la_##v[sizeof(t s __VA_ARGS__) + (a)];              \
00664     t (*v) __VA_ARGS__ = (void *)FFALIGN((uintptr_t)la_##v, a)
00665 
00666 #if HAVE_LOCAL_ALIGNED_8
00667 #   define LOCAL_ALIGNED_8(t, v, s, ...) DECLARE_ALIGNED(8, t, v) s __VA_ARGS__
00668 #else
00669 #   define LOCAL_ALIGNED_8(t, v, s, ...) LOCAL_ALIGNED(8, t, v, s, __VA_ARGS__)
00670 #endif
00671 
00672 #if HAVE_LOCAL_ALIGNED_16
00673 #   define LOCAL_ALIGNED_16(t, v, s, ...) DECLARE_ALIGNED(16, t, v) s __VA_ARGS__
00674 #else
00675 #   define LOCAL_ALIGNED_16(t, v, s, ...) LOCAL_ALIGNED(16, t, v, s, __VA_ARGS__)
00676 #endif
00677 
00678 /* PSNR */
00679 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
00680               int orig_linesize[3], int coded_linesize,
00681               AVCodecContext *avctx);
00682 
00683 #define WRAPPER8_16(name8, name16)\
00684 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
00685     return name8(s, dst           , src           , stride, h)\
00686           +name8(s, dst+8         , src+8         , stride, h);\
00687 }
00688 
00689 #define WRAPPER8_16_SQ(name8, name16)\
00690 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
00691     int score=0;\
00692     score +=name8(s, dst           , src           , stride, 8);\
00693     score +=name8(s, dst+8         , src+8         , stride, 8);\
00694     if(h==16){\
00695         dst += 8*stride;\
00696         src += 8*stride;\
00697         score +=name8(s, dst           , src           , stride, 8);\
00698         score +=name8(s, dst+8         , src+8         , stride, 8);\
00699     }\
00700     return score;\
00701 }
00702 
00703 
00704 static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
00705 {
00706     int i;
00707     for(i=0; i<h; i++)
00708     {
00709         AV_WN16(dst   , AV_RN16(src   ));
00710         dst+=dstStride;
00711         src+=srcStride;
00712     }
00713 }
00714 
00715 static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
00716 {
00717     int i;
00718     for(i=0; i<h; i++)
00719     {
00720         AV_WN32(dst   , AV_RN32(src   ));
00721         dst+=dstStride;
00722         src+=srcStride;
00723     }
00724 }
00725 
00726 static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
00727 {
00728     int i;
00729     for(i=0; i<h; i++)
00730     {
00731         AV_WN32(dst   , AV_RN32(src   ));
00732         AV_WN32(dst+4 , AV_RN32(src+4 ));
00733         dst+=dstStride;
00734         src+=srcStride;
00735     }
00736 }
00737 
00738 static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
00739 {
00740     int i;
00741     for(i=0; i<h; i++)
00742     {
00743         AV_WN32(dst   , AV_RN32(src   ));
00744         AV_WN32(dst+4 , AV_RN32(src+4 ));
00745         dst[8]= src[8];
00746         dst+=dstStride;
00747         src+=srcStride;
00748     }
00749 }
00750 
00751 static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
00752 {
00753     int i;
00754     for(i=0; i<h; i++)
00755     {
00756         AV_WN32(dst   , AV_RN32(src   ));
00757         AV_WN32(dst+4 , AV_RN32(src+4 ));
00758         AV_WN32(dst+8 , AV_RN32(src+8 ));
00759         AV_WN32(dst+12, AV_RN32(src+12));
00760         dst+=dstStride;
00761         src+=srcStride;
00762     }
00763 }
00764 
00765 static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
00766 {
00767     int i;
00768     for(i=0; i<h; i++)
00769     {
00770         AV_WN32(dst   , AV_RN32(src   ));
00771         AV_WN32(dst+4 , AV_RN32(src+4 ));
00772         AV_WN32(dst+8 , AV_RN32(src+8 ));
00773         AV_WN32(dst+12, AV_RN32(src+12));
00774         dst[16]= src[16];
00775         dst+=dstStride;
00776         src+=srcStride;
00777     }
00778 }
00779 
00780 #endif /* AVCODEC_DSPUTIL_H */