simd.sum is worked out in full and checked where it lands, as any plain number; simd.sum.lane checks it at the lane's type, and simd.sum.wrap wraps it there as the chip's…
simd_sum_forms.zg
# `simd.sum` is worked out in full and checked where it lands, as any plain number; `simd.sum.lane`# checks it at the lane's type, and `simd.sum.wrap` wraps it there as the chip's own sum does, never a# check (owner, 2026-10-07: "There should be a way to do the other option too", then "yea").MAIN{let'v':simd[int32,4]=simd.fill[value1200000000];let's':int64=simd.sum['v'];let'w':int32=simd.sum.wrap['v'];let'u':simd[int32,4]=simd.fill[value-500000000];let'l':int32=simd.sum.lane['u'];std::print.useful['s'," ",'w'," ",'l',newln];}