[llvm] [RISCV][ISel] Combine scalable vector add/sub/mul with zero/sign extension (PR #72340)

[Wang Pengcheng via llvm-commits]

================
@@ -1231,16 +1231,17 @@ define <vscale x 1 x i64> @ctlz_nxv1i64(<vscale x 1 x i64> %va) {
 ;
 ; CHECK-F-LABEL: ctlz_nxv1i64:
 ; CHECK-F:       # %bb.0:
-; CHECK-F-NEXT:    vsetvli a0, zero, e32, mf2, ta, ma
+; CHECK-F-NEXT:    li a0, 190
----------------
wangpc-pp wrote:

Thanks for explanation! But I meant that we need a `vmv.v.x` to construct the immediate. The (macro) instructions may not be reduced and we have more vtype toggles now for this `ctlz_nxv1i64` case?

https://github.com/llvm/llvm-project/pull/72340