[llvm] 75881db - [JumpThreading] Don't phi translate past loop phi (#70664)

[via llvm-commits]

Author: Nikita Popov
Date: 2023-10-31T10:20:07+01:00
New Revision: 75881dbb0fa5dcfe08518b6fb72621cbf60f45e2

URL: https://github.com/llvm/llvm-project/commit/75881dbb0fa5dcfe08518b6fb72621cbf60f45e2
DIFF: https://github.com/llvm/llvm-project/commit/75881dbb0fa5dcfe08518b6fb72621cbf60f45e2.diff

LOG: [JumpThreading] Don't phi translate past loop phi (#70664)

When evaluating comparisons in predecessors, phi operands are translated
into the predecessor. If the translation is across a backedge, this
means that the two operands of the icmp will be from two different loop
iterations, resulting in incorrect simplification.

Fix this by not performing the phi translation for phis in loop headers.

Note: This is not a complete fix. If the
jump-threading-across-loop-headers option is enabled, the LoopHeaders
variable does not get populated. Additional changes will be needed to
fix that case.

Related to https://github.com/llvm/llvm-project/issues/70651.

Added: 
    

Modified: 
    llvm/lib/Transforms/Scalar/JumpThreading.cpp
    llvm/test/Transforms/JumpThreading/pr70651.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Scalar/JumpThreading.cpp b/llvm/lib/Transforms/Scalar/JumpThreading.cpp
index f2b9d784ead8af3..7a8128c5b6c0901 100644
--- a/llvm/lib/Transforms/Scalar/JumpThreading.cpp
+++ b/llvm/lib/Transforms/Scalar/JumpThreading.cpp
@@ -761,7 +761,10 @@ bool JumpThreadingPass::computeValueKnownInPredecessorsImpl(
     PHINode *PN = dyn_cast<PHINode>(CmpLHS);
     if (!PN)
       PN = dyn_cast<PHINode>(CmpRHS);
-    if (PN && PN->getParent() == BB) {
+    // Do not perform phi translation across a loop header phi, because this
+    // may result in comparison of values from two 
diff erent loop iterations.
+    // FIXME: This check is broken if LoopHeaders is not populated.
+    if (PN && PN->getParent() == BB && !LoopHeaders.contains(BB)) {
       const DataLayout &DL = PN->getModule()->getDataLayout();
       // We can do this simplification if any comparisons fold to true or false.
       // See if any do.

diff  --git a/llvm/test/Transforms/JumpThreading/pr70651.ll b/llvm/test/Transforms/JumpThreading/pr70651.ll
index a156be541874a6a..2c6059bfdb3d566 100644
--- a/llvm/test/Transforms/JumpThreading/pr70651.ll
+++ b/llvm/test/Transforms/JumpThreading/pr70651.ll
@@ -1,7 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 3
 ; RUN: opt -S -passes=jump-threading < %s | FileCheck %s
+; RUN: opt -S -passes=jump-threading -jump-threading-across-loop-headers < %s | FileCheck %s --check-prefix=THREAD-LOOP
 
-; FIXME: This is a miscompile.
+; FIXME: This is a miscompile if -jump-threading-across-loop-headers is enabled.
 define i64 @test(i64 %v) {
 ; CHECK-LABEL: define i64 @test(
 ; CHECK-SAME: i64 [[V:%.*]]) {
@@ -12,10 +13,24 @@ define i64 @test(i64 %v) {
 ; CHECK-NEXT:    [[SUM:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[SUM_NEXT:%.*]], [[FOR_BODY]] ]
 ; CHECK-NEXT:    [[SUM_NEXT]] = add i64 [[SUM]], [[V]]
 ; CHECK-NEXT:    [[OVERFLOW:%.*]] = icmp ult i64 [[SUM_NEXT]], [[SUM]]
-; CHECK-NEXT:    br i1 [[V_NONNEG]], label [[FOR_BODY]], label [[EXIT:%.*]]
+; CHECK-NEXT:    [[CMP:%.*]] = xor i1 [[V_NONNEG]], [[OVERFLOW]]
+; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[EXIT:%.*]]
 ; CHECK:       exit:
 ; CHECK-NEXT:    ret i64 [[SUM]]
 ;
+; THREAD-LOOP-LABEL: define i64 @test(
+; THREAD-LOOP-SAME: i64 [[V:%.*]]) {
+; THREAD-LOOP-NEXT:  entry:
+; THREAD-LOOP-NEXT:    [[V_NONNEG:%.*]] = icmp sgt i64 [[V]], -1
+; THREAD-LOOP-NEXT:    br label [[FOR_BODY:%.*]]
+; THREAD-LOOP:       for.body:
+; THREAD-LOOP-NEXT:    [[SUM:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[SUM_NEXT:%.*]], [[FOR_BODY]] ]
+; THREAD-LOOP-NEXT:    [[SUM_NEXT]] = add i64 [[SUM]], [[V]]
+; THREAD-LOOP-NEXT:    [[OVERFLOW:%.*]] = icmp ult i64 [[SUM_NEXT]], [[SUM]]
+; THREAD-LOOP-NEXT:    br i1 [[V_NONNEG]], label [[FOR_BODY]], label [[EXIT:%.*]]
+; THREAD-LOOP:       exit:
+; THREAD-LOOP-NEXT:    ret i64 [[SUM]]
+;
 entry:
   %v.nonneg = icmp sgt i64 %v, -1
   br label %for.body