dsputil.h

64-bits H.264 from ffmpeg 2008 version Build in VC++ 2008 no error warning, Jesse Stone, Taiwan

/*
 * DSP utils
 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard.
 * Copyright (c) 2002-2004 Michael Niedermayer <[e-mail]>
 *
 * 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 dsputil.h
 * 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 FFMPEG_DSPUTIL_H
#define FFMPEG_DSPUTIL_H

/* dct code */
typedef short DCTELEM;
typedef int DWTELEM;
typedef short IDWTELEM;
#define AV_RN16(a) (*((const uint16_t*)(a)))
#define AV_RN32(a) (*((const uint32_t*)(a)))
#define AV_RN64(a) (*((const uint64_t*)(a)))

#define AV_WN16(a, b) *((uint16_t*)(a)) = (b)
#define AV_WN32(a, b) *((uint32_t*)(a)) = (b)
#define AV_WN64(a, b) *((uint64_t*)(a)) = (b)
void fdct_ifast (DCTELEM *data);
void fdct_ifast248 (DCTELEM *data);
void ff_jpeg_fdct_islow (DCTELEM *data);
void ff_fdct248_islow (DCTELEM *data);

void j_rev_dct (DCTELEM *data);
void j_rev_dct4 (DCTELEM *data);
void j_rev_dct2 (DCTELEM *data);
void j_rev_dct1 (DCTELEM *data);
void ff_wmv2_idct_c(DCTELEM *data);

void ff_fdct_mmx(DCTELEM *block);
void ff_fdct_mmx2(DCTELEM *block);
void ff_fdct_sse2(DCTELEM *block);

void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);

void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1,
                              const float *src2, int src3, int blocksize, int step);
void ff_float_to_int16_c(int16_t *dst, const float *src, int len);

/* 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*/);

/* 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 (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);

#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))*/;


// for snow slices
typedef struct slice_buffer_s slice_buffer;

/**
 * Scantable.
 */
typedef struct ScanTable{
    const uint8_t *scantable;
    uint8_t permutated[64];
    uint8_t raster_end[64];
#ifdef ARCH_POWERPC
                /** 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 (*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);
    int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
    /**
     * translational global motion compensation.
     */
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
    /**
     * global motion compensation.
     */
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
    int (*pix_sum)(uint8_t * pix, int line_size);
    int (*pix_norm1)(uint8_t * pix, int line_size);
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4

    me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
    me_cmp_func sse[5];
    me_cmp_func hadamard8_diff[5];
    me_cmp_func dct_sad[5];
    me_cmp_func quant_psnr[5];
    me_cmp_func bit[5];
    me_cmp_func rd[5];
    me_cmp_func vsad[5];
    me_cmp_func vsse[5];
    me_cmp_func nsse[5];
    me_cmp_func w53[5];
    me_cmp_func w97[5];
    me_cmp_func dct_max[5];
    me_cmp_func dct264_sad[5];

    me_cmp_func me_pre_cmp[5];
    me_cmp_func me_cmp[5];
    me_cmp_func me_sub_cmp[5];
    me_cmp_func mb_cmp[5];
    me_cmp_func ildct_cmp[5]; //only width 16 used
    me_cmp_func frame_skip_cmp[5]; //only width 8 used

    int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
                             int size);

    /**
     * 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<br>
     * *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<br>
     * *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<br>
     * *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<br>
     * *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);

    /**
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
     * this is an array[12] of motion compensation functions for the 9 thirdpe
     * positions<br>
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
     * @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
     */
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?

    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];
    /* This is really one func used in VC-1 decoding */
    h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
    h264_chroma_mc_func avg_h264_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];

    h264_weight_func weight_h264_pixels_tab[10];
    h264_biweight_func biweight_h264_pixels_tab[10];

    /* AVS specific */
    qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
    qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
    void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
    void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
    void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
    void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
    void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);

    me_cmp_func pix_abs[2][4];

    /* huffyuv specific */
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
    void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
    /**
     * subtract huffyuv's variant of median prediction
     * note, this might read from src1[-1], src2[-1]
     */
    void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
    /* this might write to dst[w] */
    void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
    //void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);

    void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
    void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
    void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
    void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
    void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
    void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
    // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
    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],
                                      int bidir, int edges, int step, int mask_mv0, int mask_mv1);

    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);

    void (*h261_loop_filter)(uint8_t *src, int stride);

    void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
    void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);

    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
    void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
    /* no alignment needed */
    void (*flac_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
    void (*vector_fmul)(float *dst, const float *src, int len);
    void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
    /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
    void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);

    /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
     * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
    void (*float_to_int16)(int16_t *dst, const float *src, int len);

    /* (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.