diff options
| -rw-r--r-- | lib/CodeGen/BranchFolding.cpp | 15 | ||||
| -rw-r--r-- | lib/CodeGen/BranchFolding.h | 4 | ||||
| -rw-r--r-- | test/CodeGen/X86/block-placement.ll | 4 |
3 files changed, 16 insertions, 7 deletions
diff --git a/lib/CodeGen/BranchFolding.cpp b/lib/CodeGen/BranchFolding.cpp index bf0ea5a125..9cd4208d64 100644 --- a/lib/CodeGen/BranchFolding.cpp +++ b/lib/CodeGen/BranchFolding.cpp @@ -406,7 +406,8 @@ void BranchFolder::ReplaceTailWithBranchTo(MachineBasicBlock::iterator OldInst, /// MBB so that the part before the iterator falls into the part starting at the /// iterator. This returns the new MBB. MachineBasicBlock *BranchFolder::SplitMBBAt(MachineBasicBlock &CurMBB, - MachineBasicBlock::iterator BBI1) { + MachineBasicBlock::iterator BBI1, + const BasicBlock *BB) { if (!TII->isLegalToSplitMBBAt(CurMBB, BBI1)) return 0; @@ -414,7 +415,7 @@ MachineBasicBlock *BranchFolder::SplitMBBAt(MachineBasicBlock &CurMBB, // Create the fall-through block. MachineFunction::iterator MBBI = &CurMBB; - MachineBasicBlock *NewMBB =MF.CreateMachineBasicBlock(CurMBB.getBasicBlock()); + MachineBasicBlock *NewMBB =MF.CreateMachineBasicBlock(BB); CurMBB.getParent()->insert(++MBBI, NewMBB); // Move all the successors of this block to the specified block. @@ -647,6 +648,7 @@ void BranchFolder::RemoveBlocksWithHash(unsigned CurHash, /// CreateCommonTailOnlyBlock - None of the blocks to be tail-merged consist /// only of the common tail. Create a block that does by splitting one. bool BranchFolder::CreateCommonTailOnlyBlock(MachineBasicBlock *&PredBB, + MachineBasicBlock *SuccBB, unsigned maxCommonTailLength, unsigned &commonTailIndex) { commonTailIndex = 0; @@ -676,7 +678,12 @@ bool BranchFolder::CreateCommonTailOnlyBlock(MachineBasicBlock *&PredBB, DEBUG(dbgs() << "\nSplitting BB#" << MBB->getNumber() << ", size " << maxCommonTailLength); - MachineBasicBlock *newMBB = SplitMBBAt(*MBB, BBI); + // If the split block unconditionally falls-thru to SuccBB, it will be + // merged. In control flow terms it should then take SuccBB's name. e.g. If + // SuccBB is an inner loop, the common tail is still part of the inner loop. + const BasicBlock *BB = (SuccBB && MBB->succ_size() == 1) ? + SuccBB->getBasicBlock() : MBB->getBasicBlock(); + MachineBasicBlock *newMBB = SplitMBBAt(*MBB, BBI, BB); if (!newMBB) { DEBUG(dbgs() << "... failed!"); return false; @@ -784,7 +791,7 @@ bool BranchFolder::TryTailMergeBlocks(MachineBasicBlock *SuccBB, !SameTails[commonTailIndex].tailIsWholeBlock())) { // None of the blocks consist entirely of the common tail. // Split a block so that one does. - if (!CreateCommonTailOnlyBlock(PredBB, + if (!CreateCommonTailOnlyBlock(PredBB, SuccBB, maxCommonTailLength, commonTailIndex)) { RemoveBlocksWithHash(CurHash, SuccBB, PredBB); continue; diff --git a/lib/CodeGen/BranchFolding.h b/lib/CodeGen/BranchFolding.h index df795dfc24..26bdca93f7 100644 --- a/lib/CodeGen/BranchFolding.h +++ b/lib/CodeGen/BranchFolding.h @@ -100,13 +100,15 @@ namespace llvm { void ReplaceTailWithBranchTo(MachineBasicBlock::iterator OldInst, MachineBasicBlock *NewDest); MachineBasicBlock *SplitMBBAt(MachineBasicBlock &CurMBB, - MachineBasicBlock::iterator BBI1); + MachineBasicBlock::iterator BBI1, + const BasicBlock *BB); unsigned ComputeSameTails(unsigned CurHash, unsigned minCommonTailLength, MachineBasicBlock *SuccBB, MachineBasicBlock *PredBB); void RemoveBlocksWithHash(unsigned CurHash, MachineBasicBlock* SuccBB, MachineBasicBlock* PredBB); bool CreateCommonTailOnlyBlock(MachineBasicBlock *&PredBB, + MachineBasicBlock *SuccBB, unsigned maxCommonTailLength, unsigned &commonTailIndex); diff --git a/test/CodeGen/X86/block-placement.ll b/test/CodeGen/X86/block-placement.ll index be627e08e0..139f8b029b 100644 --- a/test/CodeGen/X86/block-placement.ll +++ b/test/CodeGen/X86/block-placement.ll @@ -1,4 +1,4 @@ -; RUN: llc -mtriple=i686-linux < %s | FileCheck %s +; RUN: llc -mtriple=i686-linux -pre-RA-sched=source < %s | FileCheck %s declare void @error(i32 %i, i32 %a, i32 %b) @@ -997,7 +997,7 @@ define void @benchmark_heapsort(i32 %n, double* nocapture %ra) { ; CHECK: %while.body ; CHECK: %land.lhs.true ; CHECK: %if.then19 -; CHECK: %if.then19 +; CHECK: %if.end20 ; CHECK: %if.then8 ; CHECK: ret |