lerp_simd_test.cpp

Color4 Lerp(Color4 a, Color4 b, float t)
{
    // nice branchless SIMD color lerp

    __m128 one = _mm_set1_ps(1);

    __m128 t4 = _mm_set1_ps(t);
    t4 = _mm_max_ps(t4, _mm_setzero_ps());
    t4 = _mm_min_ps(t4, one);

    __m128i a8 = _mm_set1_epi32(a.rgba);
    __m128i b8 = _mm_set1_epi32(b.rgba);

    __m128 aFloat = _mm_cvtepi32_ps(
        _mm_unpacklo_epi16(
            _mm_unpacklo_epi8(
                a8,
                _mm_setzero_si128()
            ),
            _mm_setzero_si128()
        )
    );

    __m128 bFloat = _mm_cvtepi32_ps(
        _mm_unpacklo_epi16(
            _mm_unpacklo_epi8(
                b8,
                _mm_setzero_si128()
                ),
            _mm_setzero_si128()
        )
    );

    aFloat = _mm_mul_ps(aFloat, _mm_sub_ps(one, t4));
    bFloat = _mm_mul_ps(bFloat, t4);
    __m128 cFloat = _mm_add_ps(aFloat, bFloat);
    cFloat = _mm_add_ps(cFloat, _mm_set1_ps(0.5f));

    __m128i c8 = _mm_packus_epi16(
        _mm_packs_epi32(
            _mm_cvtps_epi32(cFloat),
            _mm_setzero_si128()
        ),
        _mm_setzero_si128()
    );

    return Color4(*((uint32_t*)&c8));
}