dsputil.c

mediastreamer2是开源的网络传输媒体流的库

    ff_spatial_dwt(tmp, w, h, 32, type, dec_count);

    s=0;
    assert(w==h);
    for(level=0; level<dec_count; level++){
        for(ori= level ? 1 : 0; ori<4; ori++){
            int size= w>>(dec_count-level);
            int sx= (ori&1) ? size : 0;
            int stride= 32<<(dec_count-level);
            int sy= (ori&2) ? stride>>1 : 0;

            for(i=0; i<size; i++){
                for(j=0; j<size; j++){
                    int v= tmp[sx + sy + i*stride + j] * scale[type][dec_count-3][level][ori];
                    s += FFABS(v);
                }
            }
        }
    }
    assert(s>=0);
    return s>>9;
}

static int w53_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
    return w_c(v, pix1, pix2, line_size,  8, h, 1);
}

static int w97_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
    return w_c(v, pix1, pix2, line_size,  8, h, 0);
}

static int w53_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
    return w_c(v, pix1, pix2, line_size, 16, h, 1);
}

static int w97_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
    return w_c(v, pix1, pix2, line_size, 16, h, 0);
}

int w53_32_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
    return w_c(v, pix1, pix2, line_size, 32, h, 1);
}

int w97_32_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
    return w_c(v, pix1, pix2, line_size, 32, h, 0);
}
#endif

static void get_pixels_c(DCTELEM *restrict block, const uint8_t *pixels, int line_size)
{
    int i;

    /* read the pixels */
    for(i=0;i<8;i++) {
        block[0] = pixels[0];
        block[1] = pixels[1];
        block[2] = pixels[2];
        block[3] = pixels[3];
        block[4] = pixels[4];
        block[5] = pixels[5];
        block[6] = pixels[6];
        block[7] = pixels[7];
        pixels += line_size;
        block += 8;
    }
}

static void diff_pixels_c(DCTELEM *restrict block, const uint8_t *s1,
                          const uint8_t *s2, int stride){
    int i;

    /* read the pixels */
    for(i=0;i<8;i++) {
        block[0] = s1[0] - s2[0];
        block[1] = s1[1] - s2[1];
        block[2] = s1[2] - s2[2];
        block[3] = s1[3] - s2[3];
        block[4] = s1[4] - s2[4];
        block[5] = s1[5] - s2[5];
        block[6] = s1[6] - s2[6];
        block[7] = s1[7] - s2[7];
        s1 += stride;
        s2 += stride;
        block += 8;
    }
}


static void put_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels,
                                 int line_size)
{
    int i;
    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

    /* read the pixels */
    for(i=0;i<8;i++) {
        pixels[0] = cm[block[0]];
        pixels[1] = cm[block[1]];
        pixels[2] = cm[block[2]];
        pixels[3] = cm[block[3]];
        pixels[4] = cm[block[4]];
        pixels[5] = cm[block[5]];
        pixels[6] = cm[block[6]];
        pixels[7] = cm[block[7]];

        pixels += line_size;
        block += 8;
    }
}

static void put_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels,
                                 int line_size)
{
    int i;
    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

    /* read the pixels */
    for(i=0;i<4;i++) {
        pixels[0] = cm[block[0]];
        pixels[1] = cm[block[1]];
        pixels[2] = cm[block[2]];
        pixels[3] = cm[block[3]];

        pixels += line_size;
        block += 8;
    }
}

static void put_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels,
                                 int line_size)
{
    int i;
    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

    /* read the pixels */
    for(i=0;i<2;i++) {
        pixels[0] = cm[block[0]];
        pixels[1] = cm[block[1]];

        pixels += line_size;
        block += 8;
    }
}

static void put_signed_pixels_clamped_c(const DCTELEM *block,
                                        uint8_t *restrict pixels,
                                        int line_size)
{
    int i, j;

    for (i = 0; i < 8; i++) {
        for (j = 0; j < 8; j++) {
            if (*block < -128)
                *pixels = 0;
            else if (*block > 127)
                *pixels = 255;
            else
                *pixels = (uint8_t)(*block + 128);
            block++;
            pixels++;
        }
        pixels += (line_size - 8);
    }
}

static void add_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels,
                          int line_size)
{
    int i;
    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

    /* read the pixels */
    for(i=0;i<8;i++) {
        pixels[0] = cm[pixels[0] + block[0]];
        pixels[1] = cm[pixels[1] + block[1]];
        pixels[2] = cm[pixels[2] + block[2]];
        pixels[3] = cm[pixels[3] + block[3]];
        pixels[4] = cm[pixels[4] + block[4]];
        pixels[5] = cm[pixels[5] + block[5]];
        pixels[6] = cm[pixels[6] + block[6]];
        pixels[7] = cm[pixels[7] + block[7]];
        pixels += line_size;
        block += 8;
    }
}

static void add_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels,
                          int line_size)
{
    int i;
    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

    /* read the pixels */
    for(i=0;i<4;i++) {
        pixels[0] = cm[pixels[0] + block[0]];
        pixels[1] = cm[pixels[1] + block[1]];
        pixels[2] = cm[pixels[2] + block[2]];
        pixels[3] = cm[pixels[3] + block[3]];
        pixels += line_size;
        block += 8;
    }
}

static void add_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels,
                          int line_size)
{
    int i;
    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

    /* read the pixels */
    for(i=0;i<2;i++) {
        pixels[0] = cm[pixels[0] + block[0]];
        pixels[1] = cm[pixels[1] + block[1]];
        pixels += line_size;
        block += 8;
    }
}

static void add_pixels8_c(uint8_t *restrict pixels, DCTELEM *block, int line_size)
{
    int i;
    for(i=0;i<8;i++) {
        pixels[0] += block[0];
        pixels[1] += block[1];
        pixels[2] += block[2];
        pixels[3] += block[3];
        pixels[4] += block[4];
        pixels[5] += block[5];
        pixels[6] += block[6];
        pixels[7] += block[7];
        pixels += line_size;
        block += 8;
    }
}

static void add_pixels4_c(uint8_t *restrict pixels, DCTELEM *block, int line_size)
{
    int i;
    for(i=0;i<4;i++) {
        pixels[0] += block[0];
        pixels[1] += block[1];
        pixels[2] += block[2];
        pixels[3] += block[3];
        pixels += line_size;
        block += 4;
    }
}

static int sum_abs_dctelem_c(DCTELEM *block)
{
    int sum=0, i;
    for(i=0; i<64; i++)
        sum+= FFABS(block[i]);
    return sum;
}

#if 0

#define PIXOP2(OPNAME, OP) \
static void OPNAME ## _pixels(uint8_t *block, const uint8_t *pixels, int line_size, int h)\
{\
    int i;\
    for(i=0; i<h; i++){\
        OP(*((uint64_t*)block), AV_RN64(pixels));\
        pixels+=line_size;\
        block +=line_size;\
    }\
}\
\
static void OPNAME ## _no_rnd_pixels_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\
{\
    int i;\
    for(i=0; i<h; i++){\
        const uint64_t a= AV_RN64(pixels  );\
        const uint64_t b= AV_RN64(pixels+1);\
        OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\
        pixels+=line_size;\
        block +=line_size;\
    }\
}\
\
static void OPNAME ## _pixels_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\
{\
    int i;\
    for(i=0; i<h; i++){\
        const uint64_t a= AV_RN64(pixels  );\
        const uint64_t b= AV_RN64(pixels+1);\
        OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\
        pixels+=line_size;\
        block +=line_size;\
    }\
}\
\
static void OPNAME ## _no_rnd_pixels_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\
{\
    int i;\
    for(i=0; i<h; i++){\
        const uint64_t a= AV_RN64(pixels          );\
        const uint64_t b= AV_RN64(pixels+line_size);\
        OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\
        pixels+=line_size;\
        block +=line_size;\
    }\
}\
\
static void OPNAME ## _pixels_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\
{\
    int i;\
    for(i=0; i<h; i++){\
        const uint64_t a= AV_RN64(pixels          );\
        const uint64_t b= AV_RN64(pixels+line_size);\
        OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\
        pixels+=line_size;\
        block +=line_size;\
    }\
}\
\
static void OPNAME ## _pixels_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\
{\
        int i;\
        const uint64_t a= AV_RN64(pixels  );\
        const uint64_t b= AV_RN64(pixels+1);\
        uint64_t l0=  (a&0x0303030303030303ULL)\
                    + (b&0x0303030303030303ULL)\
                    + 0x0202020202020202ULL;\
        uint64_t h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\
                   + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\
        uint64_t l1,h1;\
\
        pixels+=line_size;\
        for(i=0; i<h; i+=2){\
            uint64_t a= AV_RN64(pixels  );\
            uint64_t b= AV_RN64(pixels+1);\
            l1=  (a&0x0303030303030303ULL)\
               + (b&0x0303030303030303ULL);\
            h1= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\
              + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\
            OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\
            pixels+=line_size;\
            block +=line_size;\
            a= AV_RN64(pixels  );\
            b= AV_RN64(pixels+1);\
            l0=  (a&0x0303030303030303ULL)\
               + (b&0x0303030303030303ULL)\
               + 0x0202020202020202ULL;\
            h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\
              + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\
            OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\
            pixels+=line_size;\
            block +=line_size;\