[llvm] r311565 - LowerAtomic: Don't skip optnone functions; atomic still need lowering (PR34020)

[Hans Wennborg via llvm-commits]

Author: hans
Date: Wed Aug 23 08:43:28 2017
New Revision: 311565

URL: http://llvm.org/viewvc/llvm-project?rev=311565&view=rev
Log:
LowerAtomic: Don't skip optnone functions; atomic still need lowering (PR34020)

The lowering isn't really an optimization, so optnone shouldn't make a
difference. ARM relies on the pass running when using "-mthread-model
single", because in that mode, it doesn't run AtomicExpand. See bug for
more details.

Differential Revision: https://reviews.llvm.org/D37040

Modified:
    llvm/trunk/lib/Transforms/Scalar/LowerAtomic.cpp
    llvm/trunk/test/Feature/optnone-opt.ll
    llvm/trunk/test/Transforms/LowerAtomic/atomic-swap.ll

Modified: llvm/trunk/lib/Transforms/Scalar/LowerAtomic.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LowerAtomic.cpp?rev=311565&r1=311564&r2=311565&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/LowerAtomic.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/LowerAtomic.cpp Wed Aug 23 08:43:28 2017
@@ -155,8 +155,7 @@ public:
   }
 
   bool runOnFunction(Function &F) override {
-    if (skipFunction(F))
-      return false;
+    // Don't skip optnone functions; atomics still need to be lowered.
     FunctionAnalysisManager DummyFAM;
     auto PA = Impl.run(F, DummyFAM);
     return !PA.areAllPreserved();

Modified: llvm/trunk/test/Feature/optnone-opt.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Feature/optnone-opt.ll?rev=311565&r1=311564&r2=311565&view=diff
==============================================================================
--- llvm/trunk/test/Feature/optnone-opt.ll (original)
+++ llvm/trunk/test/Feature/optnone-opt.ll Wed Aug 23 08:43:28 2017
@@ -57,7 +57,6 @@ attributes #0 = { optnone noinline }
 ; Additional IR passes that opt doesn't turn on by default.
 ; OPT-MORE-DAG: Skipping pass 'Dead Code Elimination'
 ; OPT-MORE-DAG: Skipping pass 'Dead Instruction Elimination'
-; OPT-MORE-DAG: Skipping pass 'Lower atomic intrinsics
 
 ; Loop IR passes that opt doesn't turn on by default.
 ; OPT-LOOP-DAG: Skipping pass 'Delete dead loops'

Modified: llvm/trunk/test/Transforms/LowerAtomic/atomic-swap.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LowerAtomic/atomic-swap.ll?rev=311565&r1=311564&r2=311565&view=diff
==============================================================================
--- llvm/trunk/test/Transforms/LowerAtomic/atomic-swap.ll (original)
+++ llvm/trunk/test/Transforms/LowerAtomic/atomic-swap.ll Wed Aug 23 08:43:28 2017
@@ -26,3 +26,14 @@ define i8 @swap() {
   ret i8 %j
 ; CHECK: ret i8 [[INST]]
 }
+
+
+define i8 @swap_optnone() noinline optnone {
+; CHECK-LABEL: @swap_optnone(
+  %i = alloca i8
+  %j = atomicrmw xchg i8* %i, i8 42 monotonic
+; CHECK: [[INST:%[a-z0-9]+]] = load
+; CHECK-NEXT: store
+  ret i8 %j
+; CHECK: ret i8 [[INST]]
+}