MythTV: 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 FFMPEG_DSPUTIL_H
00031 #define FFMPEG_DSPUTIL_H
00032 
00033 #include "avcodec.h"
00034 
00035 #include "intreadwrite.h"
00036 
00037 //#define DEBUG
00038 /* dct code */
00039 typedef short DCTELEM;
00040 typedef int DWTELEM;
00041 typedef short IDWTELEM;
00042 
00043 void fdct_ifast (DCTELEM *data);
00044 void fdct_ifast248 (DCTELEM *data);
00045 void ff_jpeg_fdct_islow (DCTELEM *data);
00046 void ff_fdct248_islow (DCTELEM *data);
00047 
00048 void j_rev_dct (DCTELEM *data);
00049 void j_rev_dct4 (DCTELEM *data);
00050 void j_rev_dct2 (DCTELEM *data);
00051 void j_rev_dct1 (DCTELEM *data);
00052 void ff_wmv2_idct_c(DCTELEM *data);
00053 
00054 void ff_fdct_mmx(DCTELEM *block);
00055 void ff_fdct_mmx2(DCTELEM *block);
00056 void ff_fdct_sse2(DCTELEM *block);
00057 
00058 void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
00059 void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
00060 void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
00061 void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
00062 void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
00063 void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
00064 
00065 void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1,
00066                               const float *src2, int src3, int blocksize, int step);
00067 void ff_float_to_int16_c(int16_t *dst, const float *src, int len);
00068 
00069 /* encoding scans */
00070 extern const uint8_t ff_alternate_horizontal_scan[64];
00071 extern const uint8_t ff_alternate_vertical_scan[64];
00072 extern const uint8_t ff_zigzag_direct[64];
00073 extern const uint8_t ff_zigzag248_direct[64];
00074 
00075 /* pixel operations */
00076 #define MAX_NEG_CROP 1024
00077 
00078 /* temporary */
00079 extern uint32_t ff_squareTbl[512];
00080 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
00081 
00082 /* VP3 DSP functions */
00083 void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
00084 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
00085 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
00086 
00087 /* 1/2^n downscaling functions from imgconvert.c */
00088 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
00089 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
00090 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
00091 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
00092 
00093 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
00094               int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
00095 
00096 /* minimum alignment rules ;)
00097 if u notice errors in the align stuff, need more alignment for some asm code for some cpu
00098 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
00099 
00100 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
00101 i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
00102 
00103 !future video codecs might need functions with less strict alignment
00104 */
00105 
00106 /*
00107 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
00108 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
00109 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
00110 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
00111 void clear_blocks_c(DCTELEM *blocks);
00112 */
00113 
00114 /* add and put pixel (decoding) */
00115 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
00116 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
00117 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
00118 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);
00119 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
00120 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
00121 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
00122 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
00123 
00124 #define DEF_OLD_QPEL(name)\
00125 void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
00126 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
00127 void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
00128 
00129 DEF_OLD_QPEL(qpel16_mc11_old_c)
00130 DEF_OLD_QPEL(qpel16_mc31_old_c)
00131 DEF_OLD_QPEL(qpel16_mc12_old_c)
00132 DEF_OLD_QPEL(qpel16_mc32_old_c)
00133 DEF_OLD_QPEL(qpel16_mc13_old_c)
00134 DEF_OLD_QPEL(qpel16_mc33_old_c)
00135 DEF_OLD_QPEL(qpel8_mc11_old_c)
00136 DEF_OLD_QPEL(qpel8_mc31_old_c)
00137 DEF_OLD_QPEL(qpel8_mc12_old_c)
00138 DEF_OLD_QPEL(qpel8_mc32_old_c)
00139 DEF_OLD_QPEL(qpel8_mc13_old_c)
00140 DEF_OLD_QPEL(qpel8_mc33_old_c)
00141 
00142 #define CALL_2X_PIXELS(a, b, n)\
00143 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
00144     b(block  , pixels  , line_size, h);\
00145     b(block+n, pixels+n, line_size, h);\
00146 }
00147 
00148 /* motion estimation */
00149 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
00150 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
00151 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))*/;
00152 
00153 
00154 // for snow slices
00155 typedef struct slice_buffer_s slice_buffer;
00156 
00160 typedef struct DSPContext {
00161     /* pixel ops : interface with DCT */
00162     void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
00163     void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
00164     void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
00165     void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
00166     void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
00167     void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
00168     void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
00169     int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
00173     void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
00177     void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
00178                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
00179     void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
00180     int (*pix_sum)(uint8_t * pix, int line_size);
00181     int (*pix_norm1)(uint8_t * pix, int line_size);
00182 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
00183 
00184     me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
00185     me_cmp_func sse[5];
00186     me_cmp_func hadamard8_diff[5];
00187     me_cmp_func dct_sad[5];
00188     me_cmp_func quant_psnr[5];
00189     me_cmp_func bit[5];
00190     me_cmp_func rd[5];
00191     me_cmp_func vsad[5];
00192     me_cmp_func vsse[5];
00193     me_cmp_func nsse[5];
00194     me_cmp_func w53[5];
00195     me_cmp_func w97[5];
00196     me_cmp_func dct_max[5];
00197     me_cmp_func dct264_sad[5];
00198 
00199     me_cmp_func me_pre_cmp[5];
00200     me_cmp_func me_cmp[5];
00201     me_cmp_func me_sub_cmp[5];
00202     me_cmp_func mb_cmp[5];
00203     me_cmp_func ildct_cmp[5]; //only width 16 used
00204     me_cmp_func frame_skip_cmp[5]; //only width 8 used
00205 
00206     int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
00207                              int size);
00208 
00219     op_pixels_func put_pixels_tab[4][4];
00220 
00231     op_pixels_func avg_pixels_tab[4][4];
00232 
00243     op_pixels_func put_no_rnd_pixels_tab[4][4];
00244 
00255     op_pixels_func avg_no_rnd_pixels_tab[4][4];
00256 
00257     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);
00258 
00269     tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
00270     tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
00271 
00272     qpel_mc_func put_qpel_pixels_tab[2][16];
00273     qpel_mc_func avg_qpel_pixels_tab[2][16];
00274     qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
00275     qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
00276     qpel_mc_func put_mspel_pixels_tab[8];
00277 
00281     h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
00282     /* This is really one func used in VC-1 decoding */
00283     h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
00284     h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
00285 
00286     qpel_mc_func put_h264_qpel_pixels_tab[4][16];
00287     qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
00288 
00289     qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
00290     qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
00291 
00292     h264_weight_func weight_h264_pixels_tab[10];
00293     h264_biweight_func biweight_h264_pixels_tab[10];
00294 
00295     /* AVS specific */
00296     qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
00297     qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
00298     void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
00299     void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
00300     void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
00301     void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
00302     void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
00303 
00304     me_cmp_func pix_abs[2][4];
00305 
00306     /* huffyuv specific */
00307     void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
00308     void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
00313     void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
00314     void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
00315 
00316     void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
00317     void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
00318     void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
00319     void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
00320     void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
00321     void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
00322     // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
00323     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],
00324                                       int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field);
00325 
00326     void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
00327     void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
00328 
00329     void (*h261_loop_filter)(uint8_t *src, int stride);
00330 
00331     void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
00332     void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
00333 
00334     /* assume len is a multiple of 4, and arrays are 16-byte aligned */
00335     void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
00336     /* no alignment needed */
00337     void (*flac_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
00338     /* assume len is a multiple of 8, and arrays are 16-byte aligned */
00339     void (*vector_fmul)(float *dst, const float *src, int len);
00340     void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
00341     /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
00342     void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
00343 
00344     /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
00345      * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
00346     void (*float_to_int16)(int16_t *dst, const float *src, int len);
00347 
00348     /* (I)DCT */
00349     void (*fdct)(DCTELEM *block/* align 16*/);
00350     void (*fdct248)(DCTELEM *block/* align 16*/);
00351 
00352     /* IDCT really*/
00353     void (*idct)(DCTELEM *block/* align 16*/);
00354 
00360     void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
00361 
00366     void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
00367 
00380     uint8_t idct_permutation[64];
00381     int idct_permutation_type;
00382 #define FF_NO_IDCT_PERM 1
00383 #define FF_LIBMPEG2_IDCT_PERM 2
00384 #define FF_SIMPLE_IDCT_PERM 3
00385 #define FF_TRANSPOSE_IDCT_PERM 4
00386 #define FF_PARTTRANS_IDCT_PERM 5
00387 
00388     int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
00389     void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
00390 #define BASIS_SHIFT 16
00391 #define RECON_SHIFT 6
00392 
00393     /* h264 functions */
00394     void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
00395     void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
00396     void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
00397     void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
00398     void (*h264_dct)(DCTELEM block[4][4]);
00399 
00400     /* snow wavelet */
00401     void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
00402     void (*horizontal_compose97i)(IDWTELEM *b, int width);
00403     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);
00404 
00405     void (*prefetch)(void *mem, int stride, int h);
00406 
00407     void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
00408 
00409     /* vc1 functions */
00410     void (*vc1_inv_trans_8x8)(DCTELEM *b);
00411     void (*vc1_inv_trans_8x4)(DCTELEM *b, int n);
00412     void (*vc1_inv_trans_4x8)(DCTELEM *b, int n);
00413     void (*vc1_inv_trans_4x4)(DCTELEM *b, int n);
00414     void (*vc1_v_overlap)(uint8_t* src, int stride);
00415     void (*vc1_h_overlap)(uint8_t* src, int stride);
00416     /* put 8x8 block with bicubic interpolation and quarterpel precision
00417      * last argument is actually round value instead of height
00418      */
00419     op_pixels_func put_vc1_mspel_pixels_tab[16];
00420 
00421     /* intrax8 functions */
00422     void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
00423     void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
00424            int * range, int * sum,  int edges);
00425 
00426 } DSPContext;
00427 
00428 void dsputil_static_init(void);
00429 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
00430 
00431 int ff_check_alignment(void);
00432 
00437 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
00438 
00439 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
00440 
00441 #define         BYTE_VEC32(c)   ((c)*0x01010101UL)
00442 
00443 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
00444 {
00445     return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
00446 }
00447 
00448 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
00449 {
00450     return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
00451 }
00452 
00453 static inline int get_penalty_factor(int lambda, int lambda2, int type){
00454     switch(type&0xFF){
00455     default:
00456     case FF_CMP_SAD:
00457         return lambda>>FF_LAMBDA_SHIFT;
00458     case FF_CMP_DCT:
00459         return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
00460     case FF_CMP_W53:
00461         return (4*lambda)>>(FF_LAMBDA_SHIFT);
00462     case FF_CMP_W97:
00463         return (2*lambda)>>(FF_LAMBDA_SHIFT);
00464     case FF_CMP_SATD:
00465     case FF_CMP_DCT264:
00466         return (2*lambda)>>FF_LAMBDA_SHIFT;
00467     case FF_CMP_RD:
00468     case FF_CMP_PSNR:
00469     case FF_CMP_SSE:
00470     case FF_CMP_NSSE:
00471         return lambda2>>FF_LAMBDA_SHIFT;
00472     case FF_CMP_BIT:
00473         return 1;
00474     }
00475 }
00476 
00482 #define emms_c()
00483 
00484 /* should be defined by architectures supporting
00485    one or more MultiMedia extension */
00486 int mm_support(void);
00487 
00488 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
00489 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
00490 void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
00491 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
00492 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
00493 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
00494 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
00495 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
00496 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
00497 
00498 #define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v)
00499 
00500 #if defined(HAVE_MMX)
00501 
00502 #undef emms_c
00503 
00504 #define MM_MMX    0x0001 /* standard MMX */
00505 #define MM_3DNOW  0x0004 /* AMD 3DNOW */
00506 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
00507 #define MM_SSE    0x0008 /* SSE functions */
00508 #define MM_SSE2   0x0010 /* PIV SSE2 functions */
00509 #define MM_3DNOWEXT  0x0020 /* AMD 3DNowExt */
00510 #define MM_SSE3   0x0040 /* Prescott SSE3 functions */
00511 #define MM_SSSE3  0x0080 /* Conroe SSSE3 functions */
00512 
00513 extern int mm_flags;
00514 
00515 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
00516 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
00517 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
00518 
00519 static inline void emms(void)
00520 {
00521     __asm __volatile ("emms;":::"memory");
00522 }
00523 
00524 
00525 #define emms_c() \
00526 {\
00527     if (mm_flags & MM_MMX)\
00528         emms();\
00529 }
00530 
00531 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
00532 
00533 #define STRIDE_ALIGN 8
00534 
00535 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
00536 
00537 #elif defined(ARCH_ARMV4L)
00538 
00539 /* This is to use 4 bytes read to the IDCT pointers for some 'zero'
00540    line optimizations */
00541 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(4, t, v)
00542 #define STRIDE_ALIGN 4
00543 
00544 #define MM_IWMMXT    0x0100 /* XScale IWMMXT */
00545 
00546 extern int mm_flags;
00547 
00548 #elif defined(HAVE_MLIB)
00549 
00550 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
00551 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
00552 #define STRIDE_ALIGN 8
00553 
00554 #elif defined(HAVE_VIS)
00555 
00556 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
00557 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
00558 #define STRIDE_ALIGN 8
00559 
00560 #elif defined(ARCH_ALPHA)
00561 
00562 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
00563 #define STRIDE_ALIGN 8
00564 
00565 #elif defined(ARCH_POWERPC)
00566 
00567 #define MM_ALTIVEC    0x0001 /* standard AltiVec */
00568 
00569 extern int mm_flags;
00570 
00571 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
00572 #define STRIDE_ALIGN 16
00573 
00574 #elif defined(HAVE_MMI)
00575 
00576 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
00577 #define STRIDE_ALIGN 16
00578 
00579 #elif defined(ARCH_SH4)
00580 
00581 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
00582 #define STRIDE_ALIGN 8
00583 
00584 #elif defined(ARCH_BFIN)
00585 
00586 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
00587 #define STRIDE_ALIGN 8
00588 
00589 #else
00590 
00591 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
00592 #define STRIDE_ALIGN 8
00593 
00594 #endif
00595 
00596 /* PSNR */
00597 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
00598               int orig_linesize[3], int coded_linesize,
00599               AVCodecContext *avctx);
00600 
00601 /* FFT computation */
00602 
00603 /* NOTE: soon integer code will be added, so you must use the
00604    FFTSample type */
00605 typedef float FFTSample;
00606 
00607 struct MDCTContext;
00608 
00609 typedef struct FFTComplex {
00610     FFTSample re, im;
00611 } FFTComplex;
00612 
00613 typedef struct FFTContext {
00614     int nbits;
00615     int inverse;
00616     uint16_t *revtab;
00617     FFTComplex *exptab;
00618     FFTComplex *exptab1; /* only used by SSE code */
00619     void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
00620     void (*imdct_calc)(struct MDCTContext *s, FFTSample *output,
00621                        const FFTSample *input, FFTSample *tmp);
00622 } FFTContext;
00623 
00624 int ff_fft_init(FFTContext *s, int nbits, int inverse);
00625 void ff_fft_permute(FFTContext *s, FFTComplex *z);
00626 void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
00627 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
00628 void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
00629 void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
00630 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
00631 
00632 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
00633 {
00634     s->fft_calc(s, z);
00635 }
00636 void ff_fft_end(FFTContext *s);
00637 
00638 /* MDCT computation */
00639 
00640 typedef struct MDCTContext {
00641     int n;  /* size of MDCT (i.e. number of input data * 2) */
00642     int nbits; /* n = 2^nbits */
00643     /* pre/post rotation tables */
00644     FFTSample *tcos;
00645     FFTSample *tsin;
00646     FFTContext fft;
00647 } MDCTContext;
00648 
00649 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
00650 void ff_imdct_calc(MDCTContext *s, FFTSample *output,
00651                 const FFTSample *input, FFTSample *tmp);
00652 void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output,
00653                         const FFTSample *input, FFTSample *tmp);
00654 void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output,
00655                        const FFTSample *input, FFTSample *tmp);
00656 void ff_mdct_calc(MDCTContext *s, FFTSample *out,
00657                const FFTSample *input, FFTSample *tmp);
00658 void ff_mdct_end(MDCTContext *s);
00659 
00660 #define WARPER8_16(name8, name16)\
00661 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
00662     return name8(s, dst           , src           , stride, h)\
00663           +name8(s, dst+8         , src+8         , stride, h);\
00664 }
00665 
00666 #define WARPER8_16_SQ(name8, name16)\
00667 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
00668     int score=0;\
00669     score +=name8(s, dst           , src           , stride, 8);\
00670     score +=name8(s, dst+8         , src+8         , stride, 8);\
00671     if(h==16){\
00672         dst += 8*stride;\
00673         src += 8*stride;\
00674         score +=name8(s, dst           , src           , stride, 8);\
00675         score +=name8(s, dst+8         , src+8         , stride, 8);\
00676     }\
00677     return score;\
00678 }
00679 
00680 
00681 static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
00682 {
00683     int i;
00684     for(i=0; i<h; i++)
00685     {
00686         AV_WN16(dst   , AV_RN16(src   ));
00687         dst+=dstStride;
00688         src+=srcStride;
00689     }
00690 }
00691 
00692 static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
00693 {
00694     int i;
00695     for(i=0; i<h; i++)
00696     {
00697         AV_WN32(dst   , AV_RN32(src   ));
00698         dst+=dstStride;
00699         src+=srcStride;
00700     }
00701 }
00702 
00703 static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
00704 {
00705     int i;
00706     for(i=0; i<h; i++)
00707     {
00708         AV_WN32(dst   , AV_RN32(src   ));
00709         AV_WN32(dst+4 , AV_RN32(src+4 ));
00710         dst+=dstStride;
00711         src+=srcStride;
00712     }
00713 }
00714 
00715 static inline void copy_block9(uint8_t *dst, 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         AV_WN32(dst+4 , AV_RN32(src+4 ));
00722         dst[8]= src[8];
00723         dst+=dstStride;
00724         src+=srcStride;
00725     }
00726 }
00727 
00728 static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
00729 {
00730     int i;
00731     for(i=0; i<h; i++)
00732     {
00733         AV_WN32(dst   , AV_RN32(src   ));
00734         AV_WN32(dst+4 , AV_RN32(src+4 ));
00735         AV_WN32(dst+8 , AV_RN32(src+8 ));
00736         AV_WN32(dst+12, AV_RN32(src+12));
00737         dst+=dstStride;
00738         src+=srcStride;
00739     }
00740 }
00741 
00742 static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
00743 {
00744     int i;
00745     for(i=0; i<h; i++)
00746     {
00747         AV_WN32(dst   , AV_RN32(src   ));
00748         AV_WN32(dst+4 , AV_RN32(src+4 ));
00749         AV_WN32(dst+8 , AV_RN32(src+8 ));
00750         AV_WN32(dst+12, AV_RN32(src+12));
00751         dst[16]= src[16];
00752         dst+=dstStride;
00753         src+=srcStride;
00754     }
00755 }
00756 
00757 #endif /* FFMPEG_DSPUTIL_H */