We used to miss non-Q YMM integer vectors, and, non-Q/D XMM integer vectors. While there, change the v4i32 patterns to prefer MOVNTDQ. llvm-svn: 233668