ret <4 x i32> %A
}
-Compiles into:
+On targets without SSE4.1, this compiles into:
LCPI1_0: ## <4 x i32>
.long 10
punpckldq %xmm2, %xmm0
ret
+It would be better to synthesize integer vector multiplication by constants
+using shifts and adds, pslld and paddd here. And even on targets with SSE4.1,
+simple cases such as multiplication by powers of two would be better as
+vector shifts than as multiplications.
+
//===---------------------------------------------------------------------===//
We compile this: