[llvm] 8f56e38 - [SCEV] Do not apply info from loop guards in AddRecs.
Florian Hahn via llvm-commits
llvm-commits at lists.llvm.org
Fri Oct 9 06:47:48 PDT 2020
Author: Florian Hahn
Date: 2020-10-09T14:47:26+01:00
New Revision: 8f56e382f782632c4b137339ab3c362fdbab62a5
URL: https://github.com/llvm/llvm-project/commit/8f56e382f782632c4b137339ab3c362fdbab62a5
DIFF: https://github.com/llvm/llvm-project/commit/8f56e382f782632c4b137339ab3c362fdbab62a5.diff
LOG: [SCEV] Do not apply info from loop guards in AddRecs.
We cannot guarantee that the replacement expression is loop-invariant in
all AddRecs in the source expression. Use a rewriter that skips
AddRecExpr for now.
Fixes PR47776.
Added:
llvm/test/Transforms/LoopStrengthReduce/X86/pr47776-do-not-apply-info-from-guards-to-addrecs.ll
Modified:
llvm/lib/Analysis/ScalarEvolution.cpp
Removed:
################################################################################
diff --git a/llvm/lib/Analysis/ScalarEvolution.cpp b/llvm/lib/Analysis/ScalarEvolution.cpp
index f569db9f6152..7e69af2e4a95 100644
--- a/llvm/lib/Analysis/ScalarEvolution.cpp
+++ b/llvm/lib/Analysis/ScalarEvolution.cpp
@@ -12705,6 +12705,27 @@ const SCEV* ScalarEvolution::computeMaxBackedgeTakenCount(const Loop *L) {
return getUMinFromMismatchedTypes(ExitCounts);
}
+/// This rewriter is similar to SCEVParameterRewriter (it replaces SCEVUnknown
+/// components following the Map (Value -> SCEV)), but skips AddRecExpr because
+/// we cannot guarantee that the replacement is loop invariant in the loop of
+/// the AddRec.
+class SCEVLoopGuardRewriter : public SCEVRewriteVisitor<SCEVLoopGuardRewriter> {
+ ValueToSCEVMapTy ⤅
+
+public:
+ SCEVLoopGuardRewriter(ScalarEvolution &SE, ValueToSCEVMapTy &M)
+ : SCEVRewriteVisitor(SE), Map(M) {}
+
+ const SCEV *visitAddRecExpr(const SCEVAddRecExpr *Expr) { return Expr; }
+
+ const SCEV *visitUnknown(const SCEVUnknown *Expr) {
+ auto I = Map.find(Expr->getValue());
+ if (I == Map.end())
+ return Expr;
+ return I->second;
+ }
+};
+
const SCEV *ScalarEvolution::applyLoopGuards(const SCEV *Expr, const Loop *L) {
auto CollectCondition = [&](ICmpInst::Predicate Predicate, const SCEV *LHS,
const SCEV *RHS, ValueToSCEVMapTy &RewriteMap) {
@@ -12787,5 +12808,6 @@ const SCEV *ScalarEvolution::applyLoopGuards(const SCEV *Expr, const Loop *L) {
if (RewriteMap.empty())
return Expr;
- return SCEVParameterRewriter::rewrite(Expr, *this, RewriteMap);
+ SCEVLoopGuardRewriter Rewriter(*this, RewriteMap);
+ return Rewriter.visit(Expr);
}
diff --git a/llvm/test/Transforms/LoopStrengthReduce/X86/pr47776-do-not-apply-info-from-guards-to-addrecs.ll b/llvm/test/Transforms/LoopStrengthReduce/X86/pr47776-do-not-apply-info-from-guards-to-addrecs.ll
new file mode 100644
index 000000000000..9e9bca6d312b
--- /dev/null
+++ b/llvm/test/Transforms/LoopStrengthReduce/X86/pr47776-do-not-apply-info-from-guards-to-addrecs.ll
@@ -0,0 +1,112 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -loop-reduce -S %s | FileCheck %s
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128-ni:1-p2:32:8:8:32-ni:2"
+target triple = "x86_64-unknown-linux-gnu"
+
+; Make sure we do not crash when applying info from loop guards to expressions in @bar.
+; Test case for PR47776.
+
+define void @bar() personality i32* ()* @zot {
+; CHECK-LABEL: @bar(
+; CHECK-NEXT: bb:
+; CHECK-NEXT: br label [[BB1:%.*]]
+; CHECK: bb1.loopexit:
+; CHECK-NEXT: br label [[BB1]]
+; CHECK: bb1:
+; CHECK-NEXT: [[TMP:%.*]] = phi i32 [ 1, [[BB:%.*]] ], [ 0, [[BB1_LOOPEXIT:%.*]] ]
+; CHECK-NEXT: br label [[BB2:%.*]]
+; CHECK: bb2:
+; CHECK-NEXT: [[TMP3:%.*]] = phi i64 [ 0, [[BB1]] ], [ [[TMP7:%.*]], [[BB5:%.*]] ]
+; CHECK-NEXT: [[TMP4:%.*]] = invoke i32 @fn()
+; CHECK-NEXT: to label [[BB5]] unwind label [[BB23_LOOPEXIT_SPLIT_LP:%.*]]
+; CHECK: bb5:
+; CHECK-NEXT: [[TMP6:%.*]] = load atomic i32, i32 addrspace(1)* undef unordered, align 8
+; CHECK-NEXT: [[TMP7]] = add nuw nsw i64 [[TMP3]], 1
+; CHECK-NEXT: [[C_0:%.*]] = icmp ult i64 [[TMP7]], 10000
+; CHECK-NEXT: br i1 [[C_0]], label [[BB2]], label [[BB8:%.*]]
+; CHECK: bb8:
+; CHECK-NEXT: [[TMP9:%.*]] = icmp ult i32 [[TMP]], [[TMP6]]
+; CHECK-NEXT: br i1 [[TMP9]], label [[BB10:%.*]], label [[BB29:%.*]]
+; CHECK: bb10:
+; CHECK-NEXT: [[TMP11:%.*]] = mul i32 [[TMP]], -1
+; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[TMP11]] to i64
+; CHECK-NEXT: [[TMP1:%.*]] = add nsw i64 [[TMP0]], 1
+; CHECK-NEXT: [[TMP2:%.*]] = sub i64 [[TMP1]], [[TMP7]]
+; CHECK-NEXT: [[TMP1:%.*]] = trunc i64 [[TMP2]] to i32
+; CHECK-NEXT: [[TMP16:%.*]] = and i32 [[TMP1]], 7
+; CHECK-NEXT: br label [[BB17:%.*]]
+; CHECK: bb17:
+; CHECK-NEXT: [[TMP18:%.*]] = phi i32 [ [[TMP21:%.*]], [[BB20:%.*]] ], [ [[TMP16]], [[BB10]] ]
+; CHECK-NEXT: [[TMP19:%.*]] = invoke i32 @fn()
+; CHECK-NEXT: to label [[BB20]] unwind label [[BB23_LOOPEXIT:%.*]]
+; CHECK: bb20:
+; CHECK-NEXT: [[TMP21]] = add i32 [[TMP18]], -1
+; CHECK-NEXT: [[TMP22:%.*]] = icmp eq i32 [[TMP21]], 0
+; CHECK-NEXT: br i1 [[TMP22]], label [[BB1_LOOPEXIT]], label [[BB17]]
+; CHECK: bb23.loopexit:
+; CHECK-NEXT: [[LPAD_LOOPEXIT:%.*]] = landingpad token
+; CHECK-NEXT: cleanup
+; CHECK-NEXT: br label [[BB23:%.*]]
+; CHECK: bb23.loopexit.split-lp:
+; CHECK-NEXT: [[LPAD_LOOPEXIT_SPLIT_LP:%.*]] = landingpad token
+; CHECK-NEXT: cleanup
+; CHECK-NEXT: br label [[BB23]]
+; CHECK: bb23:
+; CHECK-NEXT: ret void
+; CHECK: bb29:
+; CHECK-NEXT: ret void
+;
+bb:
+ br label %bb1
+
+bb1: ; preds = %bb20, %bb
+ %tmp = phi i32 [ 1, %bb ], [ 0, %bb20 ]
+ br label %bb2
+
+bb2: ; preds = %bb5, %bb1
+ %tmp3 = phi i64 [ 0, %bb1 ], [ %tmp7, %bb5 ]
+ %tmp4 = invoke i32 @fn()
+ to label %bb5 unwind label %bb23
+
+bb5: ; preds = %bb2
+ %tmp6 = load atomic i32, i32 addrspace(1)* undef unordered, align 8
+ %tmp7 = add nuw nsw i64 %tmp3, 1
+ %c.0 = icmp ult i64 %tmp7, 10000
+ br i1 %c.0, label %bb2, label %bb8
+
+bb8: ; preds = %bb5
+ %tmp9 = icmp ult i32 %tmp, %tmp6
+ br i1 %tmp9, label %bb10, label %bb29
+
+bb10: ; preds = %bb8
+ %tmp11 = mul i32 %tmp, -1
+ %tmp12 = zext i32 %tmp11 to i64
+ %tmp13 = sub i64 %tmp12, %tmp7
+ %tmp14 = add i64 %tmp13, 1
+ %tmp15 = trunc i64 %tmp14 to i32
+ %tmp16 = and i32 %tmp15, 7
+ br label %bb17
+
+bb17: ; preds = %bb20, %bb10
+ %tmp18 = phi i32 [ %tmp21, %bb20 ], [ %tmp16, %bb10 ]
+ %tmp19 = invoke i32 @fn()
+ to label %bb20 unwind label %bb23
+
+bb20: ; preds = %bb17
+ %tmp21 = add i32 %tmp18, -1
+ %tmp22 = icmp eq i32 %tmp21, 0
+ br i1 %tmp22, label %bb1, label %bb17
+
+bb23: ; preds = %bb17
+ %tmp24 = landingpad token
+ cleanup
+ ret void
+
+bb29: ; preds = %bb8
+ ret void
+}
+
+declare i32* @zot() #1
+
+declare i32 @fn()
More information about the llvm-commits
mailing list