[llvm] [LV] Honor -force-vector-interleave when generating diff checks (PR #225768)

[Florian Hahn via llvm-commits]

================
@@ -0,0 +1,43 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --filter-out-after "^vector.ph:" --version 6
+; RUN: opt -passes loop-vectorize -mattr=+sve2 -force-vector-width="vscale x 2" -force-vector-interleave=1 -S < %s | FileCheck %s
+
+target triple = "aarch64-unknown-linux-gnu"
+
+; Tests that diff checks honor `-force-vector-interleave` for the check width, not the cost model.
+
+define void @diff_check_d(ptr %src, ptr %dst, i64 %n) {
+; CHECK-LABEL: define void @diff_check_d(
+; CHECK-SAME: ptr [[SRC:%.*]], ptr [[DST:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[SRC2:%.*]] = ptrtoaddr ptr [[SRC]] to i64
+; CHECK-NEXT:    [[DST1:%.*]] = ptrtoaddr ptr [[DST]] to i64
+; CHECK-NEXT:    [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT:    [[TMP1:%.*]] = shl nuw i64 [[TMP0]], 1
+; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], [[TMP1]]
+; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], [[SCALAR_PH:label %.*]], label %[[VECTOR_MEMCHECK:.*]]
+; CHECK:       [[VECTOR_MEMCHECK]]:
+; CHECK-NEXT:    [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT:    [[TMP3:%.*]] = shl i64 [[TMP2]], 4
+; CHECK-NEXT:    [[TMP4:%.*]] = add i64 [[TMP3]], -1
+; CHECK-NEXT:    [[TMP5:%.*]] = sub i64 [[DST1]], [[SRC2]]
+; CHECK-NEXT:    [[TMP6:%.*]] = sub i64 [[TMP5]], 1
+; CHECK-NEXT:    [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP6]], [[TMP4]]
+; CHECK-NEXT:    br i1 [[DIFF_CHECK]], [[SCALAR_PH]], label %[[EXIT:.*]]
+; CHECK:       [[EXIT]]:
----------------
fhahn wrote:

can you regenerate with `--reset-variable-names`, `EXIT` here is misleading (should be `VECTOR_PH`)

https://github.com/llvm/llvm-project/pull/225768