[llvm] 1cc5946 - [NewGVN] Use performSymbolicEvaluation instead of createExpression.

Florian Hahn via llvm-commits llvm-commits at lists.llvm.org
Sat Apr 24 10:49:47 PDT 2021


Author: Florian Hahn
Date: 2021-04-24T18:49:07+01:00
New Revision: 1cc5946cc80d75be76ac8a41e1a3697059273681

URL: https://github.com/llvm/llvm-project/commit/1cc5946cc80d75be76ac8a41e1a3697059273681
DIFF: https://github.com/llvm/llvm-project/commit/1cc5946cc80d75be76ac8a41e1a3697059273681.diff

LOG: [NewGVN] Use performSymbolicEvaluation instead of createExpression.

performSymbolicEvaluation is used to obtain the symbolic expression when
visiting instructions and this is used to determine their congruence
class.

performSymbolicEvaluation only creates expressions for certain
instructions (via createExpression). For unsupported instructions,
'unknown' expression are created.

The use of createExpression in processOutgoingEdges means we may
simplify the condition in processOutgoingEdges to a constant in the
initial round of processing, but we use Unknown(I) for the congruence
class. If an operand of I changes the expression Unknown(I) stays the
same, so there is no update of the congruence class of I. Hence it
won't get re-visited. So if an operand of I changes in a way that causes
createExpression to return different result, this update is missed.

This patch updates the code to use performSymbolicEvaluation, to be
symmetric with the congruence class updating code.

Reviewed By: asbirlea

Differential Revision: https://reviews.llvm.org/D99990

Added: 
    llvm/test/Transforms/NewGVN/compare-condition-changes.ll

Modified: 
    llvm/lib/Transforms/Scalar/NewGVN.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Scalar/NewGVN.cpp b/llvm/lib/Transforms/Scalar/NewGVN.cpp
index b29c550b1fffe..6b2932f63dda7 100644
--- a/llvm/lib/Transforms/Scalar/NewGVN.cpp
+++ b/llvm/lib/Transforms/Scalar/NewGVN.cpp
@@ -2437,8 +2437,9 @@ void NewGVN::processOutgoingEdges(Instruction *TI, BasicBlock *B) {
     Value *CondEvaluated = findConditionEquivalence(Cond);
     if (!CondEvaluated) {
       if (auto *I = dyn_cast<Instruction>(Cond)) {
-        auto Res = createExpression(I);
-        if (const auto *CE = dyn_cast<ConstantExpression>(Res.Expr)) {
+        SmallPtrSet<Value *, 4> Visited;
+        auto Res = performSymbolicEvaluation(I, Visited);
+        if (const auto *CE = dyn_cast_or_null<ConstantExpression>(Res.Expr)) {
           CondEvaluated = CE->getConstantValue();
           addAdditionalUsers(Res, I);
         } else {

diff  --git a/llvm/test/Transforms/NewGVN/compare-condition-changes.ll b/llvm/test/Transforms/NewGVN/compare-condition-changes.ll
new file mode 100644
index 0000000000000..b03b19abd84f8
--- /dev/null
+++ b/llvm/test/Transforms/NewGVN/compare-condition-changes.ll
@@ -0,0 +1,80 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -newgvn -S %s | FileCheck %s
+
+; Test cases to make sure the blocks are properly marked as executable, if the
+; state of the branch condition changes.
+
+; Test case to make sure the case where a condition cannot be simplified is
+; handled properly.
+define i1 @test1() {
+; CHECK-LABEL: @test1(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[CALL:%.*]] = tail call i1 @foo()
+; CHECK-NEXT:    br i1 [[CALL]], label [[THEN:%.*]], label [[ELSE:%.*]]
+; CHECK:       then:
+; CHECK-NEXT:    ret i1 true
+; CHECK:       else:
+; CHECK-NEXT:    ret i1 false
+;
+entry:
+  %call = tail call i1 @foo()
+  br i1 %call, label %then, label %else
+
+then:
+  ret i1 true
+
+else:
+  ret i1 false
+}
+
+declare i1 @foo()
+
+; Make sure state changes are propagated across freeze to branches.
+define void @test2(i1 %c) {
+; CHECK-LABEL: @test2(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    br label [[LOOP_HEADER:%.*]]
+; CHECK:       loop.header:
+; CHECK-NEXT:    [[P_0:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[INC:%.*]], [[LOOP_LATCH:%.*]] ]
+; CHECK-NEXT:    [[P_1:%.*]] = phi i32 [ 1, [[ENTRY]] ], [ [[P_2:%.*]], [[LOOP_LATCH]] ]
+; CHECK-NEXT:    br label [[LOOP_BB_1:%.*]]
+; CHECK:       loop.bb.1:
+; CHECK-NEXT:    [[INC]] = add nsw i32 [[P_0]], 1
+; CHECK-NEXT:    [[C_1:%.*]] = icmp slt i32 [[P_0]], 0
+; CHECK-NEXT:    [[C_1_FREEZE:%.*]] = freeze i1 [[C_1]]
+; CHECK-NEXT:    br i1 [[C_1_FREEZE]], label [[LOOP_BB_2:%.*]], label [[LOOP_LATCH]]
+; CHECK:       loop.bb.2:
+; CHECK-NEXT:    br label [[LOOP_LATCH]]
+; CHECK:       loop.latch:
+; CHECK-NEXT:    [[P_2]] = phi i32 [ 0, [[LOOP_BB_2]] ], [ [[P_1]], [[LOOP_BB_1]] ]
+; CHECK-NEXT:    [[C_2:%.*]] = icmp eq i32 [[P_2]], 123
+; CHECK-NEXT:    br i1 [[C_2]], label [[EXIT:%.*]], label [[LOOP_HEADER]]
+; CHECK:       exit:
+; CHECK-NEXT:    ret void
+;
+entry:
+  br label %loop.header
+
+loop.header:
+  %p.0 = phi i32 [ 0, %entry ], [ %p.3, %loop.latch ]
+  %p.1 = phi i32 [ 1, %entry ], [ %p.2, %loop.latch ]
+  br label %loop.bb.1
+
+loop.bb.1:
+  %inc = add nsw i32 %p.0, 1
+  %c.1 = icmp slt i32 %p.0, 0
+  %c.1.freeze = freeze i1 %c.1
+  br i1 %c.1.freeze, label %loop.bb.2, label %loop.latch
+
+loop.bb.2:
+  br label %loop.latch
+
+loop.latch:
+  %p.2 = phi i32 [ 0, %loop.bb.2 ], [ %p.1, %loop.bb.1 ]
+  %p.3 = phi i32 [ %inc, %loop.bb.2 ], [ %inc, %loop.bb.1 ]
+  %c.2 = icmp eq i32 %p.2, 123
+  br i1 %c.2, label %exit, label %loop.header
+
+exit:
+  ret void
+}


        


More information about the llvm-commits mailing list