/* * DSP utils * Copyright (c) 2000, 2001, 2002 Fabrice Bellard * Copyright (c) 2002-2004 Michael Niedermayer * * This file is part of FFmpeg. * * FFmpeg is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public * License as published by the Free Software Foundation; either * version 2.1 of the License, or (at your option) any later version. * * FFmpeg is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * Lesser General Public License for more details. * * You should have received a copy of the GNU Lesser General Public * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ /** * @file * DSP utils. * note, many functions in here may use MMX which trashes the FPU state, it is * absolutely necessary to call emms_c() between dsp & float/double code */ #ifndef AVCODEC_DSPUTIL_H #define AVCODEC_DSPUTIL_H #include "libavutil/intreadwrite.h" #include "avcodec.h" #include "h264_idct.h" // void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len); void ff_float_to_int16_c(int16_t *dst, const float *src, long len); void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels); /* encoding scans */ extern const uint8_t ff_alternate_horizontal_scan[64]; extern const uint8_t ff_alternate_vertical_scan[64]; extern const uint8_t ff_zigzag_direct[64]; extern const uint8_t ff_zigzag248_direct[64]; /* pixel operations */ #define MAX_NEG_CROP 1024 /* temporary */ extern uint32_t ff_squareTbl[512]; extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP]; /* VP3 DSP functions */ void ff_vp3_idct_c(DCTELEM *block/* align 16*/); void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); void ff_vp3_idct_dc_add_c(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/); void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values); void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values); /* VP6 DSP functions */ void ff_vp6_filter_diag4_c(uint8_t *dst, uint8_t *src, int stride, const int16_t *h_weights, const int16_t *v_weights); /* Bink functions */ void ff_bink_idct_c (DCTELEM *block); void ff_bink_idct_add_c(uint8_t *dest, int linesize, DCTELEM *block); void ff_bink_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block); /* CAVS functions */ void ff_put_cavs_qpel8_mc00_c(uint8_t *dst, uint8_t *src, int stride); void ff_avg_cavs_qpel8_mc00_c(uint8_t *dst, uint8_t *src, int stride); void ff_put_cavs_qpel16_mc00_c(uint8_t *dst, uint8_t *src, int stride); void ff_avg_cavs_qpel16_mc00_c(uint8_t *dst, uint8_t *src, int stride); /* VC1 functions */ void ff_put_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int rnd); void ff_avg_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int rnd); /* EA functions */ void ff_ea_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block); /* 1/2^n downscaling functions from imgconvert.c */ void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy, int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); /* minimum alignment rules ;) If you notice errors in the align stuff, need more alignment for some ASM code for some CPU or need to use a function with less aligned data then send a mail to the ffmpeg-devel mailing list, ... !warning These alignments might not match reality, (missing attribute((align)) stuff somewhere possible). I (Michael) did not check them, these are just the alignments which I think could be reached easily ... !future video codecs might need functions with less strict alignment */ /* void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size); void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride); void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); void clear_blocks_c(DCTELEM *blocks); */ /* add and put pixel (decoding) */ // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h); 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); typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y); typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h); #define DEF_OLD_QPEL(name)\ void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); DEF_OLD_QPEL(qpel16_mc11_old_c) DEF_OLD_QPEL(qpel16_mc31_old_c) DEF_OLD_QPEL(qpel16_mc12_old_c) DEF_OLD_QPEL(qpel16_mc32_old_c) DEF_OLD_QPEL(qpel16_mc13_old_c) DEF_OLD_QPEL(qpel16_mc33_old_c) DEF_OLD_QPEL(qpel8_mc11_old_c) DEF_OLD_QPEL(qpel8_mc31_old_c) DEF_OLD_QPEL(qpel8_mc12_old_c) DEF_OLD_QPEL(qpel8_mc32_old_c) DEF_OLD_QPEL(qpel8_mc13_old_c) DEF_OLD_QPEL(qpel8_mc33_old_c) #define CALL_2X_PIXELS(a, b, n)\ static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ b(block , pixels , line_size, h);\ b(block+n, pixels+n, line_size, h);\ } /* motion estimation */ // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2 // although currently h<4 is not used as functions with width <8 are neither used nor implemented 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))*/; /** * Scantable. */ typedef struct ScanTable{ const uint8_t *scantable; uint8_t permutated[64]; uint8_t raster_end[64]; #if ARCH_PPC /** Used by dct_quantize_altivec to find last-non-zero */ DECLARE_ALIGNED(16, uint8_t, inverse)[64]; #endif } ScanTable; void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable); void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize, int block_w, int block_h, int src_x, int src_y, int w, int h); /** * DSPContext. */ typedef struct DSPContext { /* pixel ops : interface with DCT */ void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); void (*put_pixels_nonclamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size); void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size); void (*clear_block)(DCTELEM *block/*align 16*/); void (*clear_blocks)(DCTELEM *blocks/*align 16*/); /** * Halfpel motion compensation with rounding (a+b+1)>>1. * this is an array[4][4] of motion compensation functions for 4 * horizontal blocksizes (8,16) and the 4 halfpel positions
* *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] * @param block destination where the result is stored * @param pixels source * @param line_size number of bytes in a horizontal line of block * @param h height */ op_pixels_func put_pixels_tab[4][4]; /** * Halfpel motion compensation with rounding (a+b+1)>>1. * This is an array[4][4] of motion compensation functions for 4 * horizontal blocksizes (8,16) and the 4 halfpel positions
* *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] * @param block destination into which the result is averaged (a+b+1)>>1 * @param pixels source * @param line_size number of bytes in a horizontal line of block * @param h height */ op_pixels_func avg_pixels_tab[4][4]; /** * Halfpel motion compensation with no rounding (a+b)>>1. * this is an array[2][4] of motion compensation functions for 2 * horizontal blocksizes (8,16) and the 4 halfpel positions
* *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] * @param block destination where the result is stored * @param pixels source * @param line_size number of bytes in a horizontal line of block * @param h height */ op_pixels_func put_no_rnd_pixels_tab[4][4]; /** * Halfpel motion compensation with no rounding (a+b)>>1. * this is an array[2][4] of motion compensation functions for 2 * horizontal blocksizes (8,16) and the 4 halfpel positions
* *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] * @param block destination into which the result is averaged (a+b)>>1 * @param pixels source * @param line_size number of bytes in a horizontal line of block * @param h height */ op_pixels_func avg_no_rnd_pixels_tab[4][4]; 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); qpel_mc_func put_qpel_pixels_tab[2][16]; qpel_mc_func avg_qpel_pixels_tab[2][16]; qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16]; qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; qpel_mc_func put_mspel_pixels_tab[8]; /** * h264 Chroma MC */ h264_chroma_mc_func put_h264_chroma_pixels_tab[3]; h264_chroma_mc_func avg_h264_chroma_pixels_tab[3]; /* This is really one func used in VC-1 decoding */ h264_chroma_mc_func put_no_rnd_vc1_chroma_pixels_tab[3]; h264_chroma_mc_func avg_no_rnd_vc1_chroma_pixels_tab[3]; qpel_mc_func put_h264_qpel_pixels_tab[4][16]; qpel_mc_func avg_h264_qpel_pixels_tab[4][16]; qpel_mc_func put_2tap_qpel_pixels_tab[4][16]; qpel_mc_func avg_2tap_qpel_pixels_tab[4][16]; /* (I)DCT */ void (*fdct)(DCTELEM *block/* align 16*/); void (*fdct248)(DCTELEM *block/* align 16*/); /* IDCT really*/ void (*idct)(DCTELEM *block/* align 16*/); /** * block -> idct -> clip to unsigned 8 bit -> dest. * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) * @param line_size size in bytes of a horizontal line of dest */ void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); /** * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. * @param line_size size in bytes of a horizontal line of dest */ void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w); #define EDGE_WIDTH 32 void (*prefetch)(void *mem, int stride, int h); } DSPContext; void dsputil_static_init(void); void dsputil_init(DSPContext* p); int ff_check_alignment(void); /** * permute block according to permuatation. * @param last last non zero element in scantable order */ void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type); #define BYTE_VEC32(c) ((c)*0x01010101UL) static inline uint32_t rnd_avg32(uint32_t a, uint32_t b) { return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); } static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b) { return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); } /** * Empty mmx state. * this must be called between any dsp function and float/double code. * for example sin(); dsp->idct_put(); emms_c(); cos() */ #define emms_c() /* should be defined by architectures supporting one or more MultiMedia extension */ int mm_support(void); extern int mm_flags; void dsputil_init_arm(DSPContext* c); void dsputil_init_mmx(DSPContext* c); void dsputil_init_ppc(DSPContext* c); void ff_dsputil_init_dwt(DSPContext *c); #if HAVE_MMX #undef emms_c static inline void emms(void) { __asm__ volatile ("emms;":::"memory"); } #define emms_c() \ {\ if (mm_flags & FF_MM_MMX)\ emms();\ } #elif ARCH_ARM #if HAVE_NEON # define STRIDE_ALIGN 16 #endif #elif ARCH_PPC || ARCH_PPC64 || ARCH_CELL #define STRIDE_ALIGN 16 #endif #ifndef STRIDE_ALIGN # define STRIDE_ALIGN 8 #endif #define WRAPPER8_16(name8, name16)\ static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ return name8(s, dst , src , stride, h)\ +name8(s, dst+8 , src+8 , stride, h);\ } #define WRAPPER8_16_SQ(name8, name16)\ static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ int score=0;\ score +=name8(s, dst , src , stride, 8);\ score +=name8(s, dst+8 , src+8 , stride, 8);\ if(h==16){\ dst += 8*stride;\ src += 8*stride;\ score +=name8(s, dst , src , stride, 8);\ score +=name8(s, dst+8 , src+8 , stride, 8);\ }\ return score;\ } static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) { int i; for(i=0; i