Put the NONTEMPORAL case first

That's how it was in `store_with_flags` before this PR, so let's do that here too just to be sure we get the right thing.

Author: scottmcm

compiler/rustc_codegen_ssa/src/traits/builder.rs

@@ -293,16 +293,17 @@ pub trait BuilderMethods<'a, 'tcx>:
         debug_assert!(src.llextra.is_none(), "cannot directly copy from unsized values");
         debug_assert!(dst.llextra.is_none(), "cannot directly copy into unsized values");
         debug_assert_eq!(dst.layout.size, src.layout.size);
-        if self.sess().opts.optimize == OptLevel::No && self.is_backend_immediate(dst.layout) {
-            // If we're not optimizing, the aliasing information from `memcpy`
-            // isn't useful, so just load-store the value for smaller code.
-            let temp = self.load_operand(src);
-            temp.val.store_with_flags(self, dst, flags);
-        } else if flags.contains(MemFlags::NONTEMPORAL) {
+        if flags.contains(MemFlags::NONTEMPORAL) {
             // HACK(nox): This is inefficient but there is no nontemporal memcpy.
             let ty = self.backend_type(dst.layout);
             let val = self.load(ty, src.llval, src.align);
             self.store_with_flags(val, dst.llval, dst.align, flags);
+        } else if self.sess().opts.optimize == OptLevel::No && self.is_backend_immediate(dst.layout)
+        {
+            // If we're not optimizing, the aliasing information from `memcpy`
+            // isn't useful, so just load-store the value for smaller code.
+            let temp = self.load_operand(src);
+            temp.val.store_with_flags(self, dst, flags);
         } else if !dst.layout.is_zst() {
             let bytes = self.const_usize(dst.layout.size.bytes());
             self.memcpy(dst.llval, dst.align, src.llval, src.align, bytes, flags);