[PartialInlining] Reduce outlining overhead by removing unneeded live-out(s)

Differential Revision: http://reviews.llvm.org/D33694

llvm-svn: 304375

llvm/lib/Transforms/IPO/PartialInlining.cpp

@@ -652,12 +652,21 @@ Function *PartialInlinerImpl::unswitchFunction(Function *F) {
   // only split block when necessary:
   PHINode *FirstPhi = getFirstPHI(PreReturn);
   unsigned NumPredsFromEntries = OI->ReturnBlockPreds.size();
+  auto IsTrivialPhi = [](PHINode *PN) -> Value * {
+    Value *CommonValue = PN->getIncomingValue(0);
+    if (all_of(PN->incoming_values(),
+               [&](Value *V) { return V == CommonValue; }))
+      return CommonValue;
+    return nullptr;
+  };
+
   if (FirstPhi && FirstPhi->getNumIncomingValues() > NumPredsFromEntries + 1) {
 
     NewReturnBlock = NewReturnBlock->splitBasicBlock(
         NewReturnBlock->getFirstNonPHI()->getIterator());
     BasicBlock::iterator I = PreReturn->begin();
     Instruction *Ins = &NewReturnBlock->front();
+    SmallVector<Instruction *, 4> DeadPhis;
     while (I != PreReturn->end()) {
       PHINode *OldPhi = dyn_cast<PHINode>(I);
       if (!OldPhi)
@@ -674,8 +683,22 @@ Function *PartialInlinerImpl::unswitchFunction(Function *F) {
         RetPhi->addIncoming(OldPhi->getIncomingValueForBlock(NewE), NewE);
         OldPhi->removeIncomingValue(NewE);
       }
+
+      // After incoming values splitting, the old phi may become trivial.
+      // Keeping the trivial phi can introduce definition inside the outline
+      // region which is live-out, causing necessary overhead (load, store
+      // arg passing etc).
+      if (auto *OldPhiVal = IsTrivialPhi(OldPhi)) {
+        OldPhi->replaceAllUsesWith(OldPhiVal);
+        DeadPhis.push_back(OldPhi);
+      }
+
       ++I;
     }
+
+    for (auto *DP : DeadPhis)
+      DP->eraseFromParent();
+
     for (auto E : OI->ReturnBlockPreds) {
       BasicBlock *NewE = cast<BasicBlock>(VMap[E]);
       NewE->getTerminator()->replaceUsesOfWith(PreReturn, NewReturnBlock);

llvm/test/Transforms/CodeExtractor/PartialInlineLiveAcross.ll

+; RUN: opt -S  -partial-inliner -max-num-inline-blocks=2 -skip-partial-inlining-cost-analysis < %s  | FileCheck %s
+; RUN: opt -S -passes=partial-inliner -max-num-inline-blocks=2 -skip-partial-inlining-cost-analysis < %s  | FileCheck %s
+define i32 @test(i32 %arg) local_unnamed_addr #0 {
+bb:
+  %tmp = tail call i32 (...) @bar() #1
+  %tmp1 = icmp slt i32 %arg, 0
+  br i1 %tmp1, label %bb6, label %bb2
+
+bb2:                                              ; preds = %bb
+  tail call void (...) @foo() #1
+  tail call void (...) @foo() #1
+  tail call void (...) @foo() #1
+  tail call void (...) @foo() #1
+  tail call void (...) @foo() #1
+  %tmp3 = tail call i32 (...) @bar() #1
+  %tmp4 = icmp eq i32 %tmp3, 10
+  br i1 %tmp4, label %bb6, label %bb5
+
+bb5:                                              ; preds = %bb2
+  tail call void (...) @foo() #1
+  tail call void (...) @foo() #1
+  tail call void (...) @foo() #1
+  tail call void (...) @foo() #1
+  br label %bb6
+
+bb6:                                              ; preds = %bb5, %bb2, %bb
+  %tmp7 = phi i32 [ %tmp, %bb5 ], [ 0, %bb ], [ %tmp, %bb2 ]
+  ret i32 %tmp7
+}
+
+declare i32 @bar(...) local_unnamed_addr #1
+
+declare void @foo(...) local_unnamed_addr #1
+
+; Function Attrs: nounwind uwtable
+define i32 @dummy_caller(i32 %arg) local_unnamed_addr #0 {
+; CHECK-LABEL: @dummy_caller
+; CHECK: codeRepl.i:
+; CHECK:  call void @test.1_bb2()
+; CHECK-NOT: load
+; CHECK  br
+
+bb:
+  %tmp = tail call i32 @test(i32 %arg)
+  ret i32 %tmp
+}
+
+; CHECK-LABEL: define internal void @test.1_bb2()
+; CHECK: .exitStub:
+; CHECK-NOT:  store i32 %tmp7, i32* %tmp7.out
+; CHECK: ret
+
+
+attributes #0 = { nounwind uwtable }
+attributes #1 = { nounwind uwtable }
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"clang version 5.0.0 (trunk 303574)"}

llvm/test/Transforms/CodeExtractor/PartialInlineNoLiveOut.ll

+; RUN: opt -S -partial-inliner -max-num-inline-blocks=2 -skip-partial-inlining-cost-analysis  < %s  | FileCheck %s
+; RUN: opt -S -passes=partial-inliner -max-num-inline-blocks=2  -skip-partial-inlining-cost-analysis < %s  | FileCheck %s
+
+define i32 @test(i32 %arg) local_unnamed_addr #0 {
+bb:
+  %tmp = tail call i32 (...) @bar() #1
+  %tmp1 = icmp slt i32 %arg, 0
+  br i1 %tmp1, label %bb6, label %bb2
+
+bb2:                                              ; preds = %bb
+  tail call void (...) @foo() #1
+  tail call void (...) @foo() #1
+  tail call void (...) @foo() #1
+  tail call void (...) @foo() #1
+  tail call void (...) @foo() #1
+  %tmp3 = tail call i32 (...) @bar() #1
+  %tmp4 = icmp eq i32 %tmp3, 10
+  br i1 %tmp4, label %bb6, label %bb5
+
+bb5:                                              ; preds = %bb2
+  tail call void (...) @foo() #1
+  tail call void (...) @foo() #1
+  tail call void (...) @foo() #1
+  tail call void (...) @foo() #1
+  br label %bb6
+
+bb6:                                              ; preds = %bb5, %bb2, %bb
+  %tmp7 = phi i32 [ 1, %bb5 ], [ 0, %bb ], [ 1, %bb2 ]
+  ret i32 %tmp7
+}
+
+; Function Attrs: nounwind uwtable
+declare i32 @bar(...) local_unnamed_addr #0
+
+; Function Attrs: nounwind uwtable
+declare void @foo(...) local_unnamed_addr #0
+
+; Function Attrs: nounwind uwtable
+define i32 @dummy_caller(i32 %arg) local_unnamed_addr #0 {
+; CHECK-LABEL: @dummy_caller
+; CHECK: codeRepl.i:
+; CHECK:  call void @test.1_bb2()
+; CHECK-NOT: load
+; CHECK  br
+bb:
+  %tmp = tail call i32 @test(i32 %arg)
+  ret i32 %tmp
+}
+
+; CHECK-LABEL: define internal void @test.1_bb2()
+; CHECK: .exitStub:
+; CHECK-NOT:  store i32 %tmp7, i32* %tmp7.out
+; CHECK: ret
+
+attributes #0 = { nounwind uwtable }
+attributes #1 = { nounwind }
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"clang version 5.0.0 (trunk 303574)"}