[llvm] [VPlan] Extract reverse operation for reverse accesses (PR #146525)

[Luke Lau via llvm-commits]

================
@@ -23,6 +23,7 @@ define void @lshift_significand(i32 %n, ptr nocapture writeonly %dst) {
 ; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]]
 ; CHECK-NEXT:    [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
 ; CHECK-NEXT:    [[TMP9:%.*]] = mul nuw i64 [[TMP8]], 2
+; CHECK-NEXT:    [[VP_REVERSE:%.*]] = call <vscale x 2 x i64> @llvm.vector.reverse.nxv2i64(<vscale x 2 x i64> zeroinitializer)
----------------
lukel97 wrote:

This is a cool optimisation, I guess the LICM transform pulls the VPInstruction::Reverse out of the loop body so convertToEVLRecipes doesn't see it?

https://github.com/llvm/llvm-project/pull/146525