[RLEV] Add coverage for expansion of umin EVL idiom

1 files changed, 104 insertions, 1 deletions

diff --git a/llvm/test/Transforms/IndVarSimplify/rewrite-loop-exit-value.ll b/llvm/test/Transforms/IndVarSimplify/rewrite-loop-exit-value.ll
index 0d3d242..653970c 100644
--- a/llvm/test/Transforms/IndVarSimplify/rewrite-loop-exit-value.ll
+++ b/llvm/test/Transforms/IndVarSimplify/rewrite-loop-exit-value.ll
@@ -1,5 +1,5 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
-; RUN: opt -passes='loop(indvars),instcombine' -S < %s | FileCheck %s
+; RUN: opt -passes='loop(indvars),instcombine' -replexitval=always -S < %s | FileCheck %s
 
 ;; Test that loop's exit value is rewritten to its initial
 ;; value from loop preheader
@@ -197,3 +197,106 @@ crit_edge:
   ret i16 %conv
 }
 
+define i32 @vscale_slt_with_vp_umin(ptr nocapture %A, i32 %n) mustprogress vscale_range(2,1024) {
+; CHECK-LABEL: @vscale_slt_with_vp_umin(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VSCALE:%.*]] = call i32 @llvm.vscale.i32()
+; CHECK-NEXT:    [[VF:%.*]] = shl nuw nsw i32 [[VSCALE]], 2
+; CHECK-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[N:%.*]], 0
+; CHECK-NEXT:    br i1 [[CMP4]], label [[FOR_BODY_PREHEADER:%.*]], label [[EARLY_EXIT:%.*]]
+; CHECK:       for.body.preheader:
+; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
+; CHECK:       early.exit:
+; CHECK-NEXT:    ret i32 0
+; CHECK:       for.body:
+; CHECK-NEXT:    [[I_05:%.*]] = phi i32 [ [[ADD:%.*]], [[FOR_BODY]] ], [ 0, [[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT:    [[LEFT:%.*]] = sub nsw i32 [[N]], [[I_05]]
+; CHECK-NEXT:    [[VF_CAPPED:%.*]] = call i32 @llvm.umin.i32(i32 [[VF]], i32 [[LEFT]])
+; CHECK-NEXT:    store i32 [[VF_CAPPED]], ptr [[A:%.*]], align 4
+; CHECK-NEXT:    [[ADD]] = add nuw nsw i32 [[I_05]], [[VF]]
+; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[ADD]], [[N]]
+; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END:%.*]]
+; CHECK:       for.end:
+; CHECK-NEXT:    [[TMP0:%.*]] = add nsw i32 [[N]], -1
+; CHECK-NEXT:    [[TMP1:%.*]] = udiv i32 [[TMP0]], [[VF]]
+; CHECK-NEXT:    [[TMP2:%.*]] = mul i32 [[TMP1]], [[VSCALE]]
+; CHECK-NEXT:    [[TMP3:%.*]] = shl i32 [[TMP2]], 2
+; CHECK-NEXT:    [[TMP4:%.*]] = sub i32 [[N]], [[TMP3]]
+; CHECK-NEXT:    [[UMIN:%.*]] = call i32 @llvm.umin.i32(i32 [[VF]], i32 [[TMP4]])
+; CHECK-NEXT:    ret i32 [[UMIN]]
+;
+entry:
+  %vscale = call i32 @llvm.vscale.i32()
+  %VF = shl nuw nsw i32 %vscale, 2
+  %cmp4 = icmp sgt i32 %n, 0
+  br i1 %cmp4, label %for.body, label %early.exit
+
+early.exit:
+  ret i32 0
+
+for.body:
+  %i.05 = phi i32 [ %add, %for.body ], [ 0, %entry ]
+  %arrayidx = getelementptr inbounds i32, ptr %A, i32 %i.05
+  %left = sub i32 %n, %i.05
+  %VF.capped = call i32 @llvm.umin.i32(i32 %VF, i32 %left)
+  store i32 %VF.capped, ptr %A
+
+  %add = add nsw i32 %i.05, %VF
+  %cmp = icmp slt i32 %add, %n
+  br i1 %cmp, label %for.body, label %for.end
+
+for.end:
+  ret i32 %VF.capped
+}
+
+define i32 @vscale_slt_with_vp_umin2(ptr nocapture %A, i32 %n) mustprogress vscale_range(2,1024) {
+; CHECK-LABEL: @vscale_slt_with_vp_umin2(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VSCALE:%.*]] = call i32 @llvm.vscale.i32()
+; CHECK-NEXT:    [[VF:%.*]] = shl nuw nsw i32 [[VSCALE]], 2
+; CHECK-NEXT:    [[CMP4:%.*]] = icmp slt i32 [[VF]], [[N:%.*]]
+; CHECK-NEXT:    br i1 [[CMP4]], label [[FOR_BODY_PREHEADER:%.*]], label [[EARLY_EXIT:%.*]]
+; CHECK:       for.body.preheader:
+; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
+; CHECK:       early.exit:
+; CHECK-NEXT:    ret i32 0
+; CHECK:       for.body:
+; CHECK-NEXT:    [[I_05:%.*]] = phi i32 [ [[ADD:%.*]], [[FOR_BODY]] ], [ 0, [[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT:    [[LEFT:%.*]] = sub i32 [[N]], [[I_05]]
+; CHECK-NEXT:    [[VF_CAPPED:%.*]] = call i32 @llvm.umin.i32(i32 [[VF]], i32 [[LEFT]])
+; CHECK-NEXT:    store i32 [[VF_CAPPED]], ptr [[A:%.*]], align 4
+; CHECK-NEXT:    [[ADD]] = add nuw nsw i32 [[I_05]], [[VF]]
+; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[ADD]], [[N]]
+; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END:%.*]]
+; CHECK:       for.end:
+; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[N]], -1
+; CHECK-NEXT:    [[TMP1:%.*]] = udiv i32 [[TMP0]], [[VF]]
+; CHECK-NEXT:    [[TMP2:%.*]] = mul i32 [[TMP1]], [[VSCALE]]
+; CHECK-NEXT:    [[TMP3:%.*]] = shl i32 [[TMP2]], 2
+; CHECK-NEXT:    [[TMP4:%.*]] = sub i32 [[N]], [[TMP3]]
+; CHECK-NEXT:    [[UMIN:%.*]] = call i32 @llvm.umin.i32(i32 [[VF]], i32 [[TMP4]])
+; CHECK-NEXT:    ret i32 [[UMIN]]
+;
+entry:
+  %vscale = call i32 @llvm.vscale.i32()
+  %VF = shl nuw nsw i32 %vscale, 2
+  %cmp4 = icmp sgt i32 %n, %VF
+  br i1 %cmp4, label %for.body, label %early.exit
+
+early.exit:
+  ret i32 0
+
+for.body:
+  %i.05 = phi i32 [ %add, %for.body ], [ 0, %entry ]
+  %arrayidx = getelementptr inbounds i32, ptr %A, i32 %i.05
+  %left = sub i32 %n, %i.05
+  %VF.capped = call i32 @llvm.umin.i32(i32 %VF, i32 %left)
+  store i32 %VF.capped, ptr %A
+
+  %add = add nsw i32 %i.05, %VF
+  %cmp = icmp slt i32 %add, %n
+  br i1 %cmp, label %for.body, label %for.end
+
+for.end:
+  ret i32 %VF.capped
+}