[llvm] [LV] Prefer fixed VFs for epilogue vectorization (PR #211814)

[David Sherwood via llvm-commits]

================
@@ -607,8 +607,8 @@ define i16 @reduce_udiv(ptr %src, i16 %x, i64 %N) #0 {
 ; VSCALEFORTUNING2-NEXT:    [[TMP1:%.*]] = call i64 @llvm.vscale.i64()
 ; VSCALEFORTUNING2-NEXT:    [[TMP5:%.*]] = shl nuw i64 [[TMP1]], 2
 ; VSCALEFORTUNING2-NEXT:    [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[TMP0]], [[TMP5]]
-; VSCALEFORTUNING2-NEXT:    [[TMP6:%.*]] = call i64 @llvm.vscale.i64()
-; VSCALEFORTUNING2-NEXT:    [[TMP9:%.*]] = shl nuw i64 [[TMP6]], 2
+; VSCALEFORTUNING2-NEXT:    [[TMP3:%.*]] = call i64 @llvm.vscale.i64()
----------------
david-arm wrote:

Changes in this file can be reverted.

https://github.com/llvm/llvm-project/pull/211814