X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FCodeGen%2FAtomicExpandLoadLinkedPass.cpp;h=5c40069fd666d62b6e3fec63dbe286b1bc643e85;hb=fe0bf8fbf9d401a67a1842da6cefbb58aa8975a3;hp=fbb6e9662d63f999579b077fe636c82f1b363631;hpb=e65c40320b1e9cda26e7ea3c7f16c6f97a0f2be6;p=oota-llvm.git diff --git a/lib/CodeGen/AtomicExpandLoadLinkedPass.cpp b/lib/CodeGen/AtomicExpandLoadLinkedPass.cpp index fbb6e9662d6..5c40069fd66 100644 --- a/lib/CodeGen/AtomicExpandLoadLinkedPass.cpp +++ b/lib/CodeGen/AtomicExpandLoadLinkedPass.cpp @@ -21,17 +21,19 @@ #include "llvm/Support/Debug.h" #include "llvm/Target/TargetLowering.h" #include "llvm/Target/TargetMachine.h" +#include "llvm/Target/TargetSubtargetInfo.h" + using namespace llvm; #define DEBUG_TYPE "arm-atomic-expand" namespace { class AtomicExpandLoadLinked : public FunctionPass { - const TargetLowering *TLI; + const TargetMachine *TM; public: static char ID; // Pass identification, replacement for typeid explicit AtomicExpandLoadLinked(const TargetMachine *TM = nullptr) - : FunctionPass(ID), TLI(TM ? TM->getTargetLowering() : nullptr) { + : FunctionPass(ID), TM(TM) { initializeAtomicExpandLoadLinkedPass(*PassRegistry::getPassRegistry()); } @@ -59,7 +61,7 @@ FunctionPass *llvm::createAtomicExpandLoadLinkedPass(const TargetMachine *TM) { } bool AtomicExpandLoadLinked::runOnFunction(Function &F) { - if (!TLI) + if (!TM || !TM->getSubtargetImpl()->enableAtomicExpandLoadLinked()) return false; SmallVector AtomicInsts; @@ -76,7 +78,8 @@ bool AtomicExpandLoadLinked::runOnFunction(Function &F) { bool MadeChange = false; for (Instruction *Inst : AtomicInsts) { - if (!TLI->shouldExpandAtomicInIR(Inst)) + if (!TM->getSubtargetImpl()->getTargetLowering()->shouldExpandAtomicInIR( + Inst)) continue; if (AtomicRMWInst *AI = dyn_cast(Inst)) @@ -98,13 +101,15 @@ bool AtomicExpandLoadLinked::expandAtomicLoad(LoadInst *LI) { // Load instructions don't actually need a leading fence, even in the // SequentiallyConsistent case. AtomicOrdering MemOpOrder = - TLI->getInsertFencesForAtomic() ? Monotonic : LI->getOrdering(); + TM->getSubtargetImpl()->getTargetLowering()->getInsertFencesForAtomic() + ? Monotonic + : LI->getOrdering(); - // The only 64-bit load guaranteed to be single-copy atomic by the ARM ARM is + // The only 64-bit load guaranteed to be single-copy atomic by the ARM is // an ldrexd (A3.5.3). IRBuilder<> Builder(LI); - Value *Val = - TLI->emitLoadLinked(Builder, LI->getPointerOperand(), MemOpOrder); + Value *Val = TM->getSubtargetImpl()->getTargetLowering()->emitLoadLinked( + Builder, LI->getPointerOperand(), MemOpOrder); insertTrailingFence(Builder, LI->getOrdering()); @@ -165,7 +170,8 @@ bool AtomicExpandLoadLinked::expandAtomicRMW(AtomicRMWInst *AI) { // Start the main loop block now that we've taken care of the preliminaries. Builder.SetInsertPoint(LoopBB); - Value *Loaded = TLI->emitLoadLinked(Builder, Addr, MemOpOrder); + Value *Loaded = TM->getSubtargetImpl()->getTargetLowering()->emitLoadLinked( + Builder, Addr, MemOpOrder); Value *NewVal; switch (AI->getOperation()) { @@ -182,7 +188,7 @@ bool AtomicExpandLoadLinked::expandAtomicRMW(AtomicRMWInst *AI) { NewVal = Builder.CreateAnd(Loaded, AI->getValOperand(), "new"); break; case AtomicRMWInst::Nand: - NewVal = Builder.CreateAnd(Loaded, Builder.CreateNot(AI->getValOperand()), + NewVal = Builder.CreateNot(Builder.CreateAnd(Loaded, AI->getValOperand()), "new"); break; case AtomicRMWInst::Or: @@ -212,7 +218,8 @@ bool AtomicExpandLoadLinked::expandAtomicRMW(AtomicRMWInst *AI) { } Value *StoreSuccess = - TLI->emitStoreConditional(Builder, NewVal, Addr, MemOpOrder); + TM->getSubtargetImpl()->getTargetLowering()->emitStoreConditional( + Builder, NewVal, Addr, MemOpOrder); Value *TryAgain = Builder.CreateICmpNE( StoreSuccess, ConstantInt::get(IntegerType::get(Ctx, 32), 0), "tryagain"); Builder.CreateCondBr(TryAgain, LoopBB, ExitBB); @@ -243,19 +250,26 @@ bool AtomicExpandLoadLinked::expandAtomicCmpXchg(AtomicCmpXchgInst *CI) { // %loaded = @load.linked(%addr) // %should_store = icmp eq %loaded, %desired // br i1 %should_store, label %cmpxchg.trystore, - // label %cmpxchg.end/%cmpxchg.barrier + // label %cmpxchg.failure // cmpxchg.trystore: // %stored = @store_conditional(%new, %addr) - // %try_again = icmp i32 ne %stored, 0 - // br i1 %try_again, label %loop, label %cmpxchg.end - // cmpxchg.barrier: + // %success = icmp eq i32 %stored, 0 + // br i1 %success, label %cmpxchg.success, label %loop/%cmpxchg.failure + // cmpxchg.success: + // fence? + // br label %cmpxchg.end + // cmpxchg.failure: // fence? // br label %cmpxchg.end // cmpxchg.end: + // %success = phi i1 [true, %cmpxchg.success], [false, %cmpxchg.failure] + // %restmp = insertvalue { iN, i1 } undef, iN %loaded, 0 + // %res = insertvalue { iN, i1 } %restmp, i1 %success, 1 // [...] BasicBlock *ExitBB = BB->splitBasicBlock(CI, "cmpxchg.end"); - auto BarrierBB = BasicBlock::Create(Ctx, "cmpxchg.barrier", F, ExitBB); - auto TryStoreBB = BasicBlock::Create(Ctx, "cmpxchg.trystore", F, BarrierBB); + auto FailureBB = BasicBlock::Create(Ctx, "cmpxchg.failure", F, ExitBB); + auto SuccessBB = BasicBlock::Create(Ctx, "cmpxchg.success", F, FailureBB); + auto TryStoreBB = BasicBlock::Create(Ctx, "cmpxchg.trystore", F, SuccessBB); auto LoopBB = BasicBlock::Create(Ctx, "cmpxchg.start", F, TryStoreBB); // This grabs the DebugLoc from CI @@ -271,75 +285,83 @@ bool AtomicExpandLoadLinked::expandAtomicCmpXchg(AtomicCmpXchgInst *CI) { // Start the main loop block now that we've taken care of the preliminaries. Builder.SetInsertPoint(LoopBB); - Value *Loaded = TLI->emitLoadLinked(Builder, Addr, MemOpOrder); + Value *Loaded = TM->getSubtargetImpl()->getTargetLowering()->emitLoadLinked( + Builder, Addr, MemOpOrder); Value *ShouldStore = Builder.CreateICmpEQ(Loaded, CI->getCompareOperand(), "should_store"); // If the the cmpxchg doesn't actually need any ordering when it fails, we can // jump straight past that fence instruction (if it exists). - BasicBlock *FailureBB = FailureOrder == Monotonic ? ExitBB : BarrierBB; Builder.CreateCondBr(ShouldStore, TryStoreBB, FailureBB); Builder.SetInsertPoint(TryStoreBB); - Value *StoreSuccess = TLI->emitStoreConditional( - Builder, CI->getNewValOperand(), Addr, MemOpOrder); - Value *TryAgain = Builder.CreateICmpNE( + Value *StoreSuccess = + TM->getSubtargetImpl()->getTargetLowering()->emitStoreConditional( + Builder, CI->getNewValOperand(), Addr, MemOpOrder); + StoreSuccess = Builder.CreateICmpEQ( StoreSuccess, ConstantInt::get(Type::getInt32Ty(Ctx), 0), "success"); - Builder.CreateCondBr(TryAgain, LoopBB, BarrierBB); + Builder.CreateCondBr(StoreSuccess, SuccessBB, + CI->isWeak() ? FailureBB : LoopBB); // Make sure later instructions don't get reordered with a fence if necessary. - Builder.SetInsertPoint(BarrierBB); + Builder.SetInsertPoint(SuccessBB); insertTrailingFence(Builder, SuccessOrder); Builder.CreateBr(ExitBB); + Builder.SetInsertPoint(FailureBB); + insertTrailingFence(Builder, FailureOrder); + Builder.CreateBr(ExitBB); + // Finally, we have control-flow based knowledge of whether the cmpxchg // succeeded or not. We expose this to later passes by converting any // subsequent "icmp eq/ne %loaded, %oldval" into a use of an appropriate PHI. // Setup the builder so we can create any PHIs we need. - Builder.SetInsertPoint(FailureBB, FailureBB->begin()); - BasicBlock *SuccessBB = FailureOrder == Monotonic ? BarrierBB : TryStoreBB; - PHINode *Success = nullptr, *Failure = nullptr; + Builder.SetInsertPoint(ExitBB, ExitBB->begin()); + PHINode *Success = Builder.CreatePHI(Type::getInt1Ty(Ctx), 2); + Success->addIncoming(ConstantInt::getTrue(Ctx), SuccessBB); + Success->addIncoming(ConstantInt::getFalse(Ctx), FailureBB); // Look for any users of the cmpxchg that are just comparing the loaded value // against the desired one, and replace them with the CFG-derived version. + SmallVector PrunedInsts; for (auto User : CI->users()) { - ICmpInst *ICmp = dyn_cast(User); - if (!ICmp) + ExtractValueInst *EV = dyn_cast(User); + if (!EV) continue; - // Because we know ICmp uses CI, we only need one operand to be the old - // value. - if (ICmp->getOperand(0) != CI->getCompareOperand() && - ICmp->getOperand(1) != CI->getCompareOperand()) - continue; + assert(EV->getNumIndices() == 1 && EV->getIndices()[0] <= 1 && + "weird extraction from { iN, i1 }"); - if (ICmp->getPredicate() == CmpInst::ICMP_EQ) { - if (!Success) { - Success = Builder.CreatePHI(Type::getInt1Ty(Ctx), 2); - Success->addIncoming(ConstantInt::getTrue(Ctx), SuccessBB); - Success->addIncoming(ConstantInt::getFalse(Ctx), LoopBB); - } - ICmp->replaceAllUsesWith(Success); - } else if (ICmp->getPredicate() == CmpInst::ICMP_NE) { - if (!Failure) { - Failure = Builder.CreatePHI(Type::getInt1Ty(Ctx), 2); - Failure->addIncoming(ConstantInt::getFalse(Ctx), SuccessBB); - Failure->addIncoming(ConstantInt::getTrue(Ctx), LoopBB); - } - ICmp->replaceAllUsesWith(Failure); - } + if (EV->getIndices()[0] == 0) + EV->replaceAllUsesWith(Loaded); + else + EV->replaceAllUsesWith(Success); + + PrunedInsts.push_back(EV); } - CI->replaceAllUsesWith(Loaded); - CI->eraseFromParent(); + // We can remove the instructions now we're no longer iterating through them. + for (auto EV : PrunedInsts) + EV->eraseFromParent(); + if (!CI->use_empty()) { + // Some use of the full struct return that we don't understand has happened, + // so we've got to reconstruct it properly. + Value *Res; + Res = Builder.CreateInsertValue(UndefValue::get(CI->getType()), Loaded, 0); + Res = Builder.CreateInsertValue(Res, Success, 1); + + CI->replaceAllUsesWith(Res); + } + + CI->eraseFromParent(); return true; } AtomicOrdering AtomicExpandLoadLinked::insertLeadingFence(IRBuilder<> &Builder, AtomicOrdering Ord) { - if (!TLI->getInsertFencesForAtomic()) + if (!TM->getSubtargetImpl()->getTargetLowering()->getInsertFencesForAtomic()) return Ord; if (Ord == Release || Ord == AcquireRelease || Ord == SequentiallyConsistent) @@ -352,7 +374,7 @@ AtomicOrdering AtomicExpandLoadLinked::insertLeadingFence(IRBuilder<> &Builder, void AtomicExpandLoadLinked::insertTrailingFence(IRBuilder<> &Builder, AtomicOrdering Ord) { - if (!TLI->getInsertFencesForAtomic()) + if (!TM->getSubtargetImpl()->getTargetLowering()->getInsertFencesForAtomic()) return; if (Ord == Acquire || Ord == AcquireRelease)