simd_conv_r4()

static void fl::gfx::blur_detail::simd_conv_r4 ( const u8 * p0, const u8 * p1, const u8 * p2, const u8 * p3, const u8 * p4, const u8 * p5, const u8 * p6, const u8 * p7, const u8 * p8, u8 * out, int nbytes )

static

Definition at line 725 of file blur.cpp.hpp.

                                                                          {
    namespace fsimd = fl::simd; // ok bare using
    const auto w8w = fsimd::set1_u16_16(8), w28w = fsimd::set1_u16_16(28);
    const auto w56w = fsimd::set1_u16_16(56), w70w = fsimd::set1_u16_16(70);
    const auto w8 = fsimd::set1_u16_8(8), w28 = fsimd::set1_u16_8(28);
    const auto w56 = fsimd::set1_u16_8(56), w70 = fsimd::set1_u16_8(70);
    int i = 0;
    for (; i + 31 < nbytes; i += 32) {
        auto v0 = fsimd::load_u8_32(p0+i), v1 = fsimd::load_u8_32(p1+i), v2 = fsimd::load_u8_32(p2+i);
        auto v3 = fsimd::load_u8_32(p3+i), v4 = fsimd::load_u8_32(p4+i), v5 = fsimd::load_u8_32(p5+i);
        auto v6v = fsimd::load_u8_32(p6+i), v7 = fsimd::load_u8_32(p7+i), v8v = fsimd::load_u8_32(p8+i);
        auto s08 = fsimd::add_u16_16(fsimd::widen_lo_u8x32_to_u16(v0), fsimd::widen_lo_u8x32_to_u16(v8v));
        auto s17 = fsimd::add_u16_16(fsimd::widen_lo_u8x32_to_u16(v1), fsimd::widen_lo_u8x32_to_u16(v7));
        auto s26 = fsimd::add_u16_16(fsimd::widen_lo_u8x32_to_u16(v2), fsimd::widen_lo_u8x32_to_u16(v6v));
        auto s35 = fsimd::add_u16_16(fsimd::widen_lo_u8x32_to_u16(v3), fsimd::widen_lo_u8x32_to_u16(v5));
        auto lo = fsimd::add_u16_16(s08, fsimd::add_u16_16(fsimd::mullo_u16_16(s17, w8w),
                   fsimd::add_u16_16(fsimd::mullo_u16_16(s26, w28w),
                   fsimd::add_u16_16(fsimd::mullo_u16_16(s35, w56w), fsimd::mullo_u16_16(fsimd::widen_lo_u8x32_to_u16(v4), w70w)))));
        lo = fsimd::srli_u16_16(lo, 8);
        auto s08h = fsimd::add_u16_16(fsimd::widen_hi_u8x32_to_u16(v0), fsimd::widen_hi_u8x32_to_u16(v8v));
        auto s17h = fsimd::add_u16_16(fsimd::widen_hi_u8x32_to_u16(v1), fsimd::widen_hi_u8x32_to_u16(v7));
        auto s26h = fsimd::add_u16_16(fsimd::widen_hi_u8x32_to_u16(v2), fsimd::widen_hi_u8x32_to_u16(v6v));
        auto s35h = fsimd::add_u16_16(fsimd::widen_hi_u8x32_to_u16(v3), fsimd::widen_hi_u8x32_to_u16(v5));
        auto hi = fsimd::add_u16_16(s08h, fsimd::add_u16_16(fsimd::mullo_u16_16(s17h, w8w),
                   fsimd::add_u16_16(fsimd::mullo_u16_16(s26h, w28w),
                   fsimd::add_u16_16(fsimd::mullo_u16_16(s35h, w56w), fsimd::mullo_u16_16(fsimd::widen_hi_u8x32_to_u16(v4), w70w)))));
        hi = fsimd::srli_u16_16(hi, 8);
        fsimd::store_u8_32(out+i, fsimd::narrow_u16x16_to_u8(lo, hi));
    }
    for (; i + 15 < nbytes; i += 16) {
        auto v0 = fsimd::load_u8_16(p0+i), v1 = fsimd::load_u8_16(p1+i), v2 = fsimd::load_u8_16(p2+i);
        auto v3 = fsimd::load_u8_16(p3+i), v4 = fsimd::load_u8_16(p4+i), v5 = fsimd::load_u8_16(p5+i);
        auto v6v = fsimd::load_u8_16(p6+i), v7 = fsimd::load_u8_16(p7+i), v8v = fsimd::load_u8_16(p8+i);
        auto s08 = fsimd::add_u16_8(fsimd::widen_lo_u8_to_u16(v0), fsimd::widen_lo_u8_to_u16(v8v));
        auto s17 = fsimd::add_u16_8(fsimd::widen_lo_u8_to_u16(v1), fsimd::widen_lo_u8_to_u16(v7));
        auto s26 = fsimd::add_u16_8(fsimd::widen_lo_u8_to_u16(v2), fsimd::widen_lo_u8_to_u16(v6v));
        auto s35 = fsimd::add_u16_8(fsimd::widen_lo_u8_to_u16(v3), fsimd::widen_lo_u8_to_u16(v5));
        auto lo = fsimd::add_u16_8(s08, fsimd::add_u16_8(fsimd::mullo_u16_8(s17, w8),
                   fsimd::add_u16_8(fsimd::mullo_u16_8(s26, w28),
                   fsimd::add_u16_8(fsimd::mullo_u16_8(s35, w56), fsimd::mullo_u16_8(fsimd::widen_lo_u8_to_u16(v4), w70)))));
        lo = fsimd::srli_u16_8(lo, 8);
        auto s08h = fsimd::add_u16_8(fsimd::widen_hi_u8_to_u16(v0), fsimd::widen_hi_u8_to_u16(v8v));
        auto s17h = fsimd::add_u16_8(fsimd::widen_hi_u8_to_u16(v1), fsimd::widen_hi_u8_to_u16(v7));
        auto s26h = fsimd::add_u16_8(fsimd::widen_hi_u8_to_u16(v2), fsimd::widen_hi_u8_to_u16(v6v));
        auto s35h = fsimd::add_u16_8(fsimd::widen_hi_u8_to_u16(v3), fsimd::widen_hi_u8_to_u16(v5));
        auto hi = fsimd::add_u16_8(s08h, fsimd::add_u16_8(fsimd::mullo_u16_8(s17h, w8),
                   fsimd::add_u16_8(fsimd::mullo_u16_8(s26h, w28),
                   fsimd::add_u16_8(fsimd::mullo_u16_8(s35h, w56), fsimd::mullo_u16_8(fsimd::widen_hi_u8_to_u16(v4), w70)))));
        hi = fsimd::srli_u16_8(hi, 8);
        fsimd::store_u8_16(out+i, fsimd::narrow_u16_to_u8(lo, hi));
    }
    for (; i < nbytes; ++i) {
        u16 s08=(u16)p0[i]+(u16)p8[i], s17=(u16)p1[i]+(u16)p7[i];
        u16 s26=(u16)p2[i]+(u16)p6[i], s35=(u16)p3[i]+(u16)p5[i];
        out[i]=(u8)((s08+s17*8+s26*28+s35*56+(u16)p4[i]*70)>>8);
    }
}

References fl::FL_NO_EXCEPT.

Referenced by fl::gfx::blur_detail::simd_hconv_dispatch< 4 >::apply(), and fl::gfx::blur_detail::simd_vconv_dispatch< 4 >::apply().

Here is the caller graph for this function: