tmp

m256_compressstoreu_ps.rs

// +avx, +popcnt, +bmi2
unsafe fn mm256_compressstroeu_ps(array: *mut f32, mask: u8, src: __m256)
{
    let convert = _pext_u64(0x0706050403020100, _pdep_u64(mask as u64, 0x0101010101010101)*0xFF);
    let permute = _mm256_cvtepu8_epi32(_mm_cvtsi64_si128(convert as i64));
    let compress = _mm256_permutevar8x32_ps(src, permute);

    let imm8 = _popcnt64(mask as i64);
    let ret = match imm8 {
        0 => { _mm256_blend_ps(_mm256_loadu_ps(array), compress, 0x00) },
        1 => { _mm256_blend_ps(_mm256_loadu_ps(array), compress, 0x01) },
        2 => { _mm256_blend_ps(_mm256_loadu_ps(array), compress, 0x03) },
        3 => { _mm256_blend_ps(_mm256_loadu_ps(array), compress, 0x07) },
        4 => { _mm256_blend_ps(_mm256_loadu_ps(array), compress, 0x0f) },
        5 => { _mm256_blend_ps(_mm256_loadu_ps(array), compress, 0x1f) },
        6 => { _mm256_blend_ps(_mm256_loadu_ps(array), compress, 0x3f) },
        7 => { _mm256_blend_ps(_mm256_loadu_ps(array), compress, 0x7f) },
        8 => { _mm256_blend_ps(_mm256_loadu_ps(array), compress, 0xff) },
        _ => { _mm256_loadu_ps(array) },
    };
    _mm256_storeu_ps(array, ret);
}