[llvm-branch-commits] [llvm] release/19.x: [SCEV] Do not allow refinement in the rewriting of BEValue (#117152) (PR #118216)
Yingwei Zheng via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Sun Dec 1 07:27:34 PST 2024
https://github.com/dtcxzyw created https://github.com/llvm/llvm-project/pull/118216
Backport https://github.com/llvm/llvm-project/commit/f7ef0721d60f85e1f699f8d1b83d4402ae19b122
>From 6f08a0f1eb21de59f1c9cb2b8c86597d43e23b31 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng <dtcxzyw2333 at gmail.com>
Date: Sun, 1 Dec 2024 22:18:46 +0800
Subject: [PATCH] release/19.x: [SCEV] Do not allow refinement in the rewriting
of BEValue (#117152)
---
llvm/include/llvm/Analysis/ScalarEvolution.h | 6 ++
llvm/lib/Analysis/ScalarEvolution.cpp | 47 +++++++---
.../test/Analysis/ScalarEvolution/pr117133.ll | 94 +++++++++++++++++++
.../Transforms/IndVarSimplify/pr117133.ll | 44 +++++++++
.../trip-count-expansion-may-introduce-ub.ll | 10 +-
5 files changed, 184 insertions(+), 17 deletions(-)
create mode 100644 llvm/test/Analysis/ScalarEvolution/pr117133.ll
create mode 100644 llvm/test/Transforms/IndVarSimplify/pr117133.ll
diff --git a/llvm/include/llvm/Analysis/ScalarEvolution.h b/llvm/include/llvm/Analysis/ScalarEvolution.h
index d9bfca763819f1..6d64154af59731 100644
--- a/llvm/include/llvm/Analysis/ScalarEvolution.h
+++ b/llvm/include/llvm/Analysis/ScalarEvolution.h
@@ -2132,6 +2132,12 @@ class ScalarEvolution {
bool isGuaranteedToTransferExecutionTo(const Instruction *A,
const Instruction *B);
+ /// Returns true if \p Op is guaranteed not to cause immediate UB.
+ bool isGuaranteedNotToCauseUB(const SCEV *Op);
+
+ /// Returns true if \p Op is guaranteed to not be poison.
+ static bool isGuaranteedNotToBePoison(const SCEV *Op);
+
/// Return true if the SCEV corresponding to \p I is never poison. Proving
/// this is more complex than proving that just \p I is never poison, since
/// SCEV commons expressions across control flow, and you can have cases
diff --git a/llvm/lib/Analysis/ScalarEvolution.cpp b/llvm/lib/Analysis/ScalarEvolution.cpp
index 412cfe73d3e559..cd272b34aab100 100644
--- a/llvm/lib/Analysis/ScalarEvolution.cpp
+++ b/llvm/lib/Analysis/ScalarEvolution.cpp
@@ -4217,7 +4217,7 @@ bool ScalarEvolution::canReuseInstruction(
// Either the value can't be poison, or the S would also be poison if it
// is.
- if (PoisonVals.contains(V) || isGuaranteedNotToBePoison(V))
+ if (PoisonVals.contains(V) || ::isGuaranteedNotToBePoison(V))
continue;
auto *I = dyn_cast<Instruction>(V);
@@ -4320,6 +4320,8 @@ ScalarEvolution::getSequentialMinMaxExpr(SCEVTypes Kind,
}
for (unsigned i = 1, e = Ops.size(); i != e; ++i) {
+ if (!isGuaranteedNotToCauseUB(Ops[i]))
+ continue;
// We can replace %x umin_seq %y with %x umin %y if either:
// * %y being poison implies %x is also poison.
// * %x cannot be the saturating value (e.g. zero for umin).
@@ -5936,18 +5938,22 @@ const SCEV *ScalarEvolution::createAddRecFromPHI(PHINode *PN) {
// We can generalize this saying that i is the shifted value of BEValue
// by one iteration:
// PHI(f(0), f({1,+,1})) --> f({0,+,1})
- const SCEV *Shifted = SCEVShiftRewriter::rewrite(BEValue, L, *this);
- const SCEV *Start = SCEVInitRewriter::rewrite(Shifted, L, *this, false);
- if (Shifted != getCouldNotCompute() &&
- Start != getCouldNotCompute()) {
- const SCEV *StartVal = getSCEV(StartValueV);
- if (Start == StartVal) {
- // Okay, for the entire analysis of this edge we assumed the PHI
- // to be symbolic. We now need to go back and purge all of the
- // entries for the scalars that use the symbolic expression.
- forgetMemoizedResults(SymbolicName);
- insertValueToMap(PN, Shifted);
- return Shifted;
+
+ // Do not allow refinement in rewriting of BEValue.
+ if (isGuaranteedNotToCauseUB(BEValue)) {
+ const SCEV *Shifted = SCEVShiftRewriter::rewrite(BEValue, L, *this);
+ const SCEV *Start = SCEVInitRewriter::rewrite(Shifted, L, *this, false);
+ if (Shifted != getCouldNotCompute() && Start != getCouldNotCompute() &&
+ ::impliesPoison(BEValue, Start)) {
+ const SCEV *StartVal = getSCEV(StartValueV);
+ if (Start == StartVal) {
+ // Okay, for the entire analysis of this edge we assumed the PHI
+ // to be symbolic. We now need to go back and purge all of the
+ // entries for the scalars that use the symbolic expression.
+ forgetMemoizedResults(SymbolicName);
+ insertValueToMap(PN, Shifted);
+ return Shifted;
+ }
}
}
}
@@ -7319,6 +7325,21 @@ bool ScalarEvolution::isGuaranteedToTransferExecutionTo(const Instruction *A,
return false;
}
+bool ScalarEvolution::isGuaranteedNotToBePoison(const SCEV *Op) {
+ SCEVPoisonCollector PC(/* LookThroughMaybePoisonBlocking */ true);
+ visitAll(Op, PC);
+ return PC.MaybePoison.empty();
+}
+
+bool ScalarEvolution::isGuaranteedNotToCauseUB(const SCEV *Op) {
+ return !SCEVExprContains(Op, [this](const SCEV *S) {
+ auto *UDiv = dyn_cast<SCEVUDivExpr>(S);
+ // The UDiv may be UB if the divisor is poison or zero. Unless the divisor
+ // is a non-zero constant, we have to assume the UDiv may be UB.
+ return UDiv && (!isKnownNonZero(UDiv->getOperand(1)) ||
+ !isGuaranteedNotToBePoison(UDiv->getOperand(1)));
+ });
+}
bool ScalarEvolution::isSCEVExprNeverPoison(const Instruction *I) {
// Only proceed if we can prove that I does not yield poison.
diff --git a/llvm/test/Analysis/ScalarEvolution/pr117133.ll b/llvm/test/Analysis/ScalarEvolution/pr117133.ll
new file mode 100644
index 00000000000000..c6b5d9c26d44cd
--- /dev/null
+++ b/llvm/test/Analysis/ScalarEvolution/pr117133.ll
@@ -0,0 +1,94 @@
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S -disable-output "-passes=print<scalar-evolution>" < %s 2>&1 | FileCheck %s
+
+define i32 @widget() {
+; CHECK-LABEL: 'widget'
+; CHECK-NEXT: Classifying expressions for: @widget
+; CHECK-NEXT: %phi = phi i32 [ 0, %b ], [ %udiv6, %b5 ]
+; CHECK-NEXT: --> %phi U: [0,1) S: [0,1) Exits: <<Unknown>> LoopDispositions: { %b1: Variant }
+; CHECK-NEXT: %phi2 = phi i32 [ 1, %b ], [ %add, %b5 ]
+; CHECK-NEXT: --> {1,+,1}<nuw><nsw><%b1> U: [1,2) S: [1,2) Exits: <<Unknown>> LoopDispositions: { %b1: Computable }
+; CHECK-NEXT: %udiv = udiv i32 10, %phi2
+; CHECK-NEXT: --> (10 /u {1,+,1}<nuw><nsw><%b1>) U: [10,11) S: [10,11) Exits: <<Unknown>> LoopDispositions: { %b1: Computable }
+; CHECK-NEXT: %urem = urem i32 %udiv, 10
+; CHECK-NEXT: --> ((-10 * ((10 /u {1,+,1}<nuw><nsw><%b1>) /u 10))<nuw><nsw> + (10 /u {1,+,1}<nuw><nsw><%b1>)) U: [0,1) S: [0,1) Exits: <<Unknown>> LoopDispositions: { %b1: Computable }
+; CHECK-NEXT: %udiv6 = udiv i32 %phi2, 0
+; CHECK-NEXT: --> ({1,+,1}<nuw><nsw><%b1> /u 0) U: empty-set S: empty-set Exits: <<Unknown>> LoopDispositions: { %b1: Computable }
+; CHECK-NEXT: %add = add i32 %phi2, 1
+; CHECK-NEXT: --> {2,+,1}<nuw><nsw><%b1> U: [2,3) S: [2,3) Exits: <<Unknown>> LoopDispositions: { %b1: Computable }
+; CHECK-NEXT: Determining loop execution counts for: @widget
+; CHECK-NEXT: Loop %b1: <multiple exits> Unpredictable backedge-taken count.
+; CHECK-NEXT: exit count for b1: ***COULDNOTCOMPUTE***
+; CHECK-NEXT: exit count for b3: i32 0
+; CHECK-NEXT: Loop %b1: constant max backedge-taken count is i32 0
+; CHECK-NEXT: Loop %b1: symbolic max backedge-taken count is i32 0
+; CHECK-NEXT: symbolic max exit count for b1: ***COULDNOTCOMPUTE***
+; CHECK-NEXT: symbolic max exit count for b3: i32 0
+;
+b:
+ br label %b1
+
+b1: ; preds = %b5, %b
+ %phi = phi i32 [ 0, %b ], [ %udiv6, %b5 ]
+ %phi2 = phi i32 [ 1, %b ], [ %add, %b5 ]
+ %icmp = icmp eq i32 %phi, 0
+ br i1 %icmp, label %b3, label %b8
+
+b3: ; preds = %b1
+ %udiv = udiv i32 10, %phi2
+ %urem = urem i32 %udiv, 10
+ %icmp4 = icmp eq i32 %urem, 0
+ br i1 %icmp4, label %b7, label %b5
+
+b5: ; preds = %b3
+ %udiv6 = udiv i32 %phi2, 0
+ %add = add i32 %phi2, 1
+ br label %b1
+
+b7: ; preds = %b3
+ ret i32 5
+
+b8: ; preds = %b1
+ ret i32 7
+}
+
+; Don't fold %indvar2 into (zext {0,+,1}) * %a
+define i64 @test_poisonous(i64 %a, i32 %n) {
+; CHECK-LABEL: 'test_poisonous'
+; CHECK-NEXT: Classifying expressions for: @test_poisonous
+; CHECK-NEXT: %indvar1 = phi i32 [ 0, %entry ], [ %indvar1.next, %loop.body ]
+; CHECK-NEXT: --> {0,+,1}<%loop.body> U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop.body: Computable }
+; CHECK-NEXT: %indvar2 = phi i64 [ 0, %entry ], [ %mul, %loop.body ]
+; CHECK-NEXT: --> %indvar2 U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop.body: Variant }
+; CHECK-NEXT: %indvar1.next = add i32 %indvar1, 1
+; CHECK-NEXT: --> {1,+,1}<%loop.body> U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop.body: Computable }
+; CHECK-NEXT: %ext = zext i32 %indvar1.next to i64
+; CHECK-NEXT: --> (zext i32 {1,+,1}<%loop.body> to i64) U: [0,4294967296) S: [0,4294967296) Exits: <<Unknown>> LoopDispositions: { %loop.body: Computable }
+; CHECK-NEXT: %mul = mul i64 %ext, %a
+; CHECK-NEXT: --> ((zext i32 {1,+,1}<%loop.body> to i64) * %a) U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop.body: Computable }
+; CHECK-NEXT: Determining loop execution counts for: @test_poisonous
+; CHECK-NEXT: Loop %loop.body: Unpredictable backedge-taken count.
+; CHECK-NEXT: Loop %loop.body: Unpredictable constant max backedge-taken count.
+; CHECK-NEXT: Loop %loop.body: Unpredictable symbolic max backedge-taken count.
+; CHECK-NEXT: Loop %loop.body: Predicated backedge-taken count is (-1 + (1 smax (1 + (sext i32 %n to i64))<nsw>))<nsw>
+; CHECK-NEXT: Predicates:
+; CHECK-NEXT: {1,+,1}<%loop.body> Added Flags: <nssw>
+; CHECK-NEXT: Loop %loop.body: Predicated symbolic max backedge-taken count is (-1 + (1 smax (1 + (sext i32 %n to i64))<nsw>))<nsw>
+; CHECK-NEXT: Predicates:
+; CHECK-NEXT: {1,+,1}<%loop.body> Added Flags: <nssw>
+;
+entry:
+ br label %loop.body
+
+loop.body:
+ %indvar1 = phi i32 [ 0, %entry ], [ %indvar1.next, %loop.body ]
+ %indvar2 = phi i64 [ 0, %entry ], [ %mul, %loop.body ]
+ %indvar1.next = add i32 %indvar1, 1
+ %ext = zext i32 %indvar1.next to i64
+ %mul = mul i64 %ext, %a
+ %exitcond = icmp sgt i32 %indvar1.next, %n
+ br i1 %exitcond, label %loop.exit, label %loop.body
+
+loop.exit:
+ ret i64 %mul
+}
diff --git a/llvm/test/Transforms/IndVarSimplify/pr117133.ll b/llvm/test/Transforms/IndVarSimplify/pr117133.ll
new file mode 100644
index 00000000000000..f62dc3f04a0966
--- /dev/null
+++ b/llvm/test/Transforms/IndVarSimplify/pr117133.ll
@@ -0,0 +1,44 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S -passes=indvars < %s | FileCheck %s
+
+define i32 @widget() {
+; CHECK-LABEL: define i32 @widget() {
+; CHECK-NEXT: [[B:.*:]]
+; CHECK-NEXT: br label %[[B1:.*]]
+; CHECK: [[B1]]:
+; CHECK-NEXT: br i1 true, label %[[B3:.*]], label %[[B8:.*]]
+; CHECK: [[B3]]:
+; CHECK-NEXT: br i1 true, label %[[B7:.*]], label %[[B5:.*]]
+; CHECK: [[B5]]:
+; CHECK-NEXT: br label %[[B1]]
+; CHECK: [[B7]]:
+; CHECK-NEXT: ret i32 5
+; CHECK: [[B8]]:
+; CHECK-NEXT: ret i32 7
+;
+b:
+ br label %b1
+
+b1:
+ %phi = phi i32 [ 0, %b ], [ %udiv6, %b5 ]
+ %phi2 = phi i32 [ 1, %b ], [ %add, %b5 ]
+ %icmp = icmp eq i32 %phi, 0
+ br i1 %icmp, label %b3, label %b8
+
+b3:
+ %udiv = udiv i32 10, %phi2
+ %urem = urem i32 %udiv, 10
+ %icmp4 = icmp eq i32 %urem, 0
+ br i1 %icmp4, label %b7, label %b5
+
+b5:
+ %udiv6 = udiv i32 %phi2, 0
+ %add = add i32 %phi2, 1
+ br label %b1
+
+b7:
+ ret i32 5
+
+b8:
+ ret i32 7
+}
diff --git a/llvm/test/Transforms/LoopVectorize/trip-count-expansion-may-introduce-ub.ll b/llvm/test/Transforms/LoopVectorize/trip-count-expansion-may-introduce-ub.ll
index bf687a31d8d178..a5477929971338 100644
--- a/llvm/test/Transforms/LoopVectorize/trip-count-expansion-may-introduce-ub.ll
+++ b/llvm/test/Transforms/LoopVectorize/trip-count-expansion-may-introduce-ub.ll
@@ -465,9 +465,10 @@ define i64 @multi_exit_4_exit_count_with_udiv_by_value_in_latch(ptr %dst, i64 %N
; CHECK-LABEL: define i64 @multi_exit_4_exit_count_with_udiv_by_value_in_latch(
; CHECK-SAME: ptr [[DST:%.*]], i64 [[N:%.*]]) {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N]], i64 0)
; CHECK-NEXT: [[TMP0:%.*]] = udiv i64 42, [[N]]
-; CHECK-NEXT: [[UMIN:%.*]] = call i64 @llvm.umin.i64(i64 [[SMAX]], i64 [[TMP0]])
+; CHECK-NEXT: [[TMP8:%.*]] = freeze i64 [[TMP0]]
+; CHECK-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N]], i64 0)
+; CHECK-NEXT: [[UMIN:%.*]] = call i64 @llvm.umin.i64(i64 [[TMP8]], i64 [[SMAX]])
; CHECK-NEXT: [[TMP1:%.*]] = add nuw nsw i64 [[UMIN]], 1
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ule i64 [[TMP1]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
@@ -653,12 +654,13 @@ define i64 @multi_exit_4_exit_count_with_urem_by_value_in_latch(ptr %dst, i64 %N
; CHECK-LABEL: define i64 @multi_exit_4_exit_count_with_urem_by_value_in_latch(
; CHECK-SAME: ptr [[DST:%.*]], i64 [[N:%.*]]) {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N]], i64 0)
; CHECK-NEXT: [[TMP0:%.*]] = udiv i64 42, [[N]]
; CHECK-NEXT: [[TMP1:%.*]] = mul nuw i64 [[N]], [[TMP0]]
; CHECK-NEXT: [[TMP2:%.*]] = sub i64 42, [[TMP1]]
; CHECK-NEXT: [[SMAX1:%.*]] = call i64 @llvm.smax.i64(i64 [[TMP2]], i64 0)
-; CHECK-NEXT: [[UMIN:%.*]] = call i64 @llvm.umin.i64(i64 [[SMAX]], i64 [[SMAX1]])
+; CHECK-NEXT: [[TMP10:%.*]] = freeze i64 [[SMAX1]]
+; CHECK-NEXT: [[SMAX2:%.*]] = call i64 @llvm.smax.i64(i64 [[N]], i64 0)
+; CHECK-NEXT: [[UMIN:%.*]] = call i64 @llvm.umin.i64(i64 [[TMP10]], i64 [[SMAX2]])
; CHECK-NEXT: [[TMP3:%.*]] = add nuw i64 [[UMIN]], 1
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ule i64 [[TMP3]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
More information about the llvm-branch-commits
mailing list