[JumpThreading] Let SimplifyPartiallyRedundantLoad look into freeze

[aqjune]

Summary

This patch allows SimplifyPartiallyRedundantLoad work when
the branch condition was frozen.

[efriedma]

llvm/lib/Transforms/Scalar/JumpThreading.cpp

1181

Am I understanding correctly, that SimplifyPartiallyRedundantLoad is actually a legal transform for arbitrary load instructions, but we have a heuristic so we only perform it for "important" instructions?

[aqjune]

llvm/lib/Transforms/Scalar/JumpThreading.cpp

1181

Yes, that is correct. Since freeze is still there, it does not make the code more undefined.

[efriedma]

LGTM

[aqjune]

Leave pre-load.ll's diff only

[aqjune]

Since this can be tested independent from D84940, it is landed early

llvm/lib/Transforms/Scalar/JumpThreading.cpp

 //===- JumpThreading.cpp - Thread control through conditional blocks ------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
if (SwitchInst *SI = dyn_cast<SwitchInst>(BB->getTerminator()))
     if (TryToUnfoldSelect(SI, BB))
       return true;
 
   // Check for some cases that are worth simplifying.  Right now we want to look
   // for loads that are used by a switch or by the condition for the branch.  If
   // we see one, check to see if it's partially redundant.  If so, insert a PHI
   // which can then be used to thread the values.
   Value *SimplifyValue = CondInst;
+
+  if (auto *FI = dyn_cast<FreezeInst>(SimplifyValue))
+    // Look into freeze's operand
+    SimplifyValue = FI->getOperand(0);
+
   if (CmpInst *CondCmp = dyn_cast<CmpInst>(SimplifyValue))
     if (isa<Constant>(CondCmp->getOperand(1)))
       SimplifyValue = CondCmp->getOperand(0);
 
   // TODO: There are other places where load PRE would be profitable, such as
   // more complex comparisons.
   if (LoadInst *LoadI = dyn_cast<LoadInst>(SimplifyValue))
     if (SimplifyPartiallyRedundantLoad(LoadI))

llvm/test/Transforms/JumpThreading/pre-load.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -jump-threading -S < %s | FileCheck %s
+
+@x = global i32 0
+@y = global i32 0
+
+declare void @f()
+declare void @g()
+
+define i32 @pre(i1 %cond, i32 %n) {
+; CHECK-LABEL: @pre(
+; CHECK-NEXT:    br i1 [[COND:%.*]], label [[C_THREAD:%.*]], label [[C:%.*]]
+; CHECK:       C.thread:
+; CHECK-NEXT:    store i32 0, i32* @x, align 4
+; CHECK-NEXT:    br label [[YES:%.*]]
+; CHECK:       C:
+; CHECK-NEXT:    [[A_PR:%.*]] = load i32, i32* @y, align 4
+; CHECK-NEXT:    [[COND2:%.*]] = icmp eq i32 [[A_PR]], 0
+; CHECK-NEXT:    br i1 [[COND2]], label [[YES]], label [[NO:%.*]]
+; CHECK:       YES:
+; CHECK-NEXT:    [[A4:%.*]] = phi i32 [ 0, [[C_THREAD]] ], [ [[A_PR]], [[C]] ]
+; CHECK-NEXT:    call void @f()
+; CHECK-NEXT:    ret i32 [[A4]]
+; CHECK:       NO:
+; CHECK-NEXT:    call void @g()
+; CHECK-NEXT:    ret i32 1
+;
+  br i1 %cond, label %A, label %B
+A:
+  store i32 0, i32* @x
+  br label %C
+B:
+  br label %C
+C:
+  %ptr = phi i32* [@x, %A], [@y, %B]
+  %a = load i32, i32* %ptr
+  %cond2 = icmp eq i32 %a, 0
+  br i1 %cond2, label %YES, label %NO
+YES:
+  call void @f()
+  ret i32 %a
+NO:
+  call void @g()
+  ret i32 1
+}
+
+define i32 @pre_freeze(i1 %cond, i32 %n) {
+; CHECK-LABEL: @pre_freeze(
+; CHECK-NEXT:    br i1 [[COND:%.*]], label [[C_THREAD:%.*]], label [[C:%.*]]
+; CHECK:       C.thread:
+; CHECK-NEXT:    store i32 0, i32* @x, align 4
+; CHECK-NEXT:    br label [[YES:%.*]]
+; CHECK:       C:
+; CHECK-NEXT:    [[A_PR:%.*]] = load i32, i32* @y, align 4
+; CHECK-NEXT:    [[COND2:%.*]] = icmp eq i32 [[A_PR]], 0
+; CHECK-NEXT:    [[COND2_FR:%.*]] = freeze i1 [[COND2]]
+; CHECK-NEXT:    br i1 [[COND2_FR]], label [[YES]], label [[NO:%.*]]
+; CHECK:       YES:
+; CHECK-NEXT:    [[A5:%.*]] = phi i32 [ 0, [[C_THREAD]] ], [ [[A_PR]], [[C]] ]
+; CHECK-NEXT:    call void @f()
+; CHECK-NEXT:    ret i32 [[A5]]
+; CHECK:       NO:
+; CHECK-NEXT:    call void @g()
+; CHECK-NEXT:    ret i32 1
+;
+  br i1 %cond, label %A, label %B
+A:
+  store i32 0, i32* @x
+  br label %C
+B:
+  br label %C
+C:
+  %ptr = phi i32* [@x, %A], [@y, %B]
+  %a = load i32, i32* %ptr
+  %cond2 = icmp eq i32 %a, 0
+  %cond2.fr = freeze i1 %cond2
+  br i1 %cond2.fr, label %YES, label %NO
+YES:
+  call void @f()
+  ret i32 %a
+NO:
+  call void @g()
+  ret i32 1
+}