static inline __m128i difference_byte_SSE2(const __m128i& sc, const __m128i& dc,
        const __m128i& sa, const __m128i& da) {
    __m128i tmp1 = _mm_mullo_epi16(sc, da);
    __m128i tmp2 = _mm_mullo_epi16(dc, sa);
    __m128i tmp = SkMin32_SSE2(tmp1, tmp2);

    __m128i ret1 = _mm_add_epi32(sc, dc);
    __m128i ret2 = _mm_slli_epi32(SkDiv255Round_SSE2(tmp), 1);
    __m128i ret = _mm_sub_epi32(ret1, ret2);

    ret = clamp_signed_byte_SSE2(ret);
    return ret;
}
示例#2
0
static inline __m128i colorburn_byte_SSE2(const __m128i& sc, const __m128i& dc,
                                          const __m128i& sa, const __m128i& da) {
    __m128i ida = _mm_sub_epi32(_mm_set1_epi32(255), da);
    __m128i isa = _mm_sub_epi32(_mm_set1_epi32(255), sa);

    // if (dc == da)
    __m128i cmp1 = _mm_cmpeq_epi32(dc, da);
    __m128i tmp1 = _mm_mullo_epi16(sa, da);
    __m128i tmp2 = _mm_mullo_epi16(sc, ida);
    __m128i tmp3 = _mm_mullo_epi16(dc, isa);
    __m128i rc1 = _mm_add_epi32(tmp1, tmp2);
    rc1 = _mm_add_epi32(rc1, tmp3);
    rc1 = clamp_div255round_SSE2(rc1);
    rc1 = _mm_and_si128(cmp1, rc1);

    // else if (0 == sc)
    __m128i cmp2 = _mm_cmpeq_epi32(sc, _mm_setzero_si128());
    __m128i rc2 = SkAlphaMulAlpha_SSE2(dc, isa);
    __m128i cmp = _mm_andnot_si128(cmp1, cmp2);
    rc2 = _mm_and_si128(cmp, rc2);

    // else
    __m128i cmp3 = _mm_or_si128(cmp1, cmp2);
    __m128i tmp4 = _mm_sub_epi32(da, dc);
    tmp4 = Multiply32_SSE2(tmp4, sa);
    tmp4 = shim_mm_div_epi32(tmp4, sc);

    __m128i tmp5 = _mm_sub_epi32(da, SkMin32_SSE2(da, tmp4));
    tmp5 = Multiply32_SSE2(sa, tmp5);
    __m128i rc3 = _mm_add_epi32(tmp5, tmp2);
    rc3 = _mm_add_epi32(rc3, tmp3);
    rc3 = clamp_div255round_SSE2(rc3);
    rc3 = _mm_andnot_si128(cmp3, rc3);

    __m128i rc = _mm_or_si128(rc1, rc2);
    rc = _mm_or_si128(rc, rc3);

    return rc;
}
示例#3
0
static inline __m128i colordodge_byte_SSE2(const __m128i& sc, const __m128i& dc,
                                           const __m128i& sa, const __m128i& da) {
    __m128i diff = _mm_sub_epi32(sa, sc);
    __m128i ida = _mm_sub_epi32(_mm_set1_epi32(255), da);
    __m128i isa = _mm_sub_epi32(_mm_set1_epi32(255), sa);

    // if (0 == dc)
    __m128i cmp1 = _mm_cmpeq_epi32(dc, _mm_setzero_si128());
    __m128i rc1 = _mm_and_si128(cmp1, SkAlphaMulAlpha_SSE2(sc, ida));

    // else if (0 == diff)
    __m128i cmp2 = _mm_cmpeq_epi32(diff, _mm_setzero_si128());
    __m128i cmp = _mm_andnot_si128(cmp1, cmp2);
    __m128i tmp1 = _mm_mullo_epi16(sa, da);
    __m128i tmp2 = _mm_mullo_epi16(sc, ida);
    __m128i tmp3 = _mm_mullo_epi16(dc, isa);
    __m128i rc2 = _mm_add_epi32(tmp1, tmp2);
    rc2 = _mm_add_epi32(rc2, tmp3);
    rc2 = clamp_div255round_SSE2(rc2);
    rc2 = _mm_and_si128(cmp, rc2);

    // else
    __m128i cmp3 = _mm_or_si128(cmp1, cmp2);
    __m128i value = _mm_mullo_epi16(dc, sa);
    diff = shim_mm_div_epi32(value, diff);

    __m128i tmp4 = SkMin32_SSE2(da, diff);
    tmp4 = Multiply32_SSE2(sa, tmp4);
    __m128i rc3 = _mm_add_epi32(tmp4, tmp2);
    rc3 = _mm_add_epi32(rc3, tmp3);
    rc3 = clamp_div255round_SSE2(rc3);
    rc3 = _mm_andnot_si128(cmp3, rc3);

    __m128i rc = _mm_or_si128(rc1, rc2);
    rc = _mm_or_si128(rc, rc3);

    return rc;
}