comparison ppc/h264_altivec.c @ 8535:8f3e20061aff libavcodec

offset and weights are signed, fixes some non-bitexact issues. Patch by David Conrad %lessen42 A gmail P com%

8534:b80cf351176e

void biweight_h264_WxH_altivec(uint8_t *dst, uint8_t *src, int stride, int log2_denom,
                               int weightd, int weights, int offset, int w, int h)
{
    int y, dst_aligned, src_aligned;
    vec_u8 vsrc, vdst;
    vec_u16 vtemp, vlog2_denom, vweights, vweightd, voffset, v0, v1, v2, v3;
    DECLARE_ALIGNED_16(int32_t, temp[4]);
    LOAD_ZERO;

    offset = ((offset + 1) | 1) << log2_denom;
    temp[0] = log2_denom+1;

            else
                v2 = v3;
        }

        if (w == 16 || dst_aligned) {
            v0 = vec_mladd(v0, vweightd, zero_u16v);
            v2 = vec_mladd(v2, vweights, zero_u16v);

            v0 = vec_adds(v0, voffset);
            v0 = vec_adds(v0, v2);
            v0 = vec_sra(v0, vlog2_denom);
        }
        if (w == 16 || !dst_aligned) {
            v1 = vec_mladd(v1, vweightd, zero_u16v);
            v3 = vec_mladd(v3, vweights, zero_u16v);

            v1 = vec_adds(v1, voffset);
            v1 = vec_adds(v1, v3);
            v1 = vec_sra(v1, vlog2_denom);
        }

8535:8f3e20061aff

void biweight_h264_WxH_altivec(uint8_t *dst, uint8_t *src, int stride, int log2_denom,
                               int weightd, int weights, int offset, int w, int h)
{
    int y, dst_aligned, src_aligned;
    vec_u8 vsrc, vdst;
    vec_s16 vtemp, vweights, vweightd, voffset, v0, v1, v2, v3;
    vec_u16 vlog2_denom;
    DECLARE_ALIGNED_16(int32_t, temp[4]);
    LOAD_ZERO;

    offset = ((offset + 1) | 1) << log2_denom;
    temp[0] = log2_denom+1;

            else
                v2 = v3;
        }

        if (w == 16 || dst_aligned) {
            v0 = vec_mladd(v0, vweightd, zero_s16v);
            v2 = vec_mladd(v2, vweights, zero_s16v);

            v0 = vec_adds(v0, voffset);
            v0 = vec_adds(v0, v2);
            v0 = vec_sra(v0, vlog2_denom);
        }
        if (w == 16 || !dst_aligned) {
            v1 = vec_mladd(v1, vweightd, zero_s16v);
            v3 = vec_mladd(v3, vweights, zero_s16v);

            v1 = vec_adds(v1, voffset);
            v1 = vec_adds(v1, v3);
            v1 = vec_sra(v1, vlog2_denom);
        }