tiny.simd.sumsOfAdjQuadAbsDiff

1pub fn sumsOfAdjQuadAbsDiff(2    comptime D: type,3    comptime a_offset: usize,4    comptime b_offset: usize,5    a: D.Vector,6    b: D.Vector,7) D.repartition(wideLane(D.Lane)).Vector {8    requireByteLane(D, "sumsOfAdjQuadAbsDiff");9    if (a_offset > 1 or b_offset > 3) @compileError("quad offsets exceed a 128-bit block");10    const W = wideLane(D.Lane);11    const R = D.repartition(W);12    var result: R.Vector = @splat(0);13    inline for (0..R.lane_count) |index| {14        const a_base = a_offset * 4 + (index / 8) * 16 + (index & 7);15        const b_base = b_offset * 4 + (index / 8) * 16;16        if (a_base + 3 < D.lane_count and b_base + 3 < D.lane_count) {17            var total: W = 0;18            inline for (0..4) |offset| {19                total += absoluteDifference(W, a[a_base + offset], b[b_base + offset]);20            }21            result[index] = total;22        }23    }24    return result;25}

1pub const sumsOfAdjQuadAbsDiff = reduce.sumsOfAdjQuadAbsDiff;