[llvm] r320584 - [PowerPC] MachineSSA pass to reduce the number of CR-logical operations
Nemanja Ivanovic via llvm-commits
llvm-commits at lists.llvm.org
Wed Dec 13 06:47:35 PST 2017
Author: nemanjai
Date: Wed Dec 13 06:47:35 2017
New Revision: 320584
URL: http://llvm.org/viewvc/llvm-project?rev=320584&view=rev
Log:
[PowerPC] MachineSSA pass to reduce the number of CR-logical operations
The initial implementation of an MI SSA pass to reduce cr-logical operations.
Currently, the only operations handled by the pass are binary operations where
both CR-inputs come from the same block and the single use is a conditional
branch (also in the same block).
Committing this off by default to allow for a period of field testing. Will
enable it by default in a follow-up patch soon.
Differential Revision: https://reviews.llvm.org/D30431
Added:
llvm/trunk/lib/Target/PowerPC/PPCMachineBasicBlockUtils.h
llvm/trunk/lib/Target/PowerPC/PPCReduceCRLogicals.cpp
Modified:
llvm/trunk/lib/Target/PowerPC/CMakeLists.txt
llvm/trunk/lib/Target/PowerPC/PPC.h
llvm/trunk/lib/Target/PowerPC/PPCTargetMachine.cpp
llvm/trunk/test/CodeGen/PowerPC/licm-remat.ll
llvm/trunk/test/CodeGen/PowerPC/select-i1-vs-i1.ll
llvm/trunk/test/CodeGen/PowerPC/tail-dup-layout.ll
Modified: llvm/trunk/lib/Target/PowerPC/CMakeLists.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PowerPC/CMakeLists.txt?rev=320584&r1=320583&r2=320584&view=diff
==============================================================================
--- llvm/trunk/lib/Target/PowerPC/CMakeLists.txt (original)
+++ llvm/trunk/lib/Target/PowerPC/CMakeLists.txt Wed Dec 13 06:47:35 2017
@@ -39,6 +39,7 @@ add_llvm_target(PowerPCCodeGen
PPCTOCRegDeps.cpp
PPCTLSDynamicCall.cpp
PPCVSXCopy.cpp
+ PPCReduceCRLogicals.cpp
PPCVSXFMAMutate.cpp
PPCVSXSwapRemoval.cpp
PPCExpandISEL.cpp
Modified: llvm/trunk/lib/Target/PowerPC/PPC.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PowerPC/PPC.h?rev=320584&r1=320583&r2=320584&view=diff
==============================================================================
--- llvm/trunk/lib/Target/PowerPC/PPC.h (original)
+++ llvm/trunk/lib/Target/PowerPC/PPC.h Wed Dec 13 06:47:35 2017
@@ -41,6 +41,7 @@ namespace llvm {
FunctionPass *createPPCVSXCopyPass();
FunctionPass *createPPCVSXFMAMutatePass();
FunctionPass *createPPCVSXSwapRemovalPass();
+ FunctionPass *createPPCReduceCRLogicalsPass();
FunctionPass *createPPCMIPeepholePass();
FunctionPass *createPPCBranchSelectionPass();
FunctionPass *createPPCBranchCoalescingPass();
Added: llvm/trunk/lib/Target/PowerPC/PPCMachineBasicBlockUtils.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PowerPC/PPCMachineBasicBlockUtils.h?rev=320584&view=auto
==============================================================================
--- llvm/trunk/lib/Target/PowerPC/PPCMachineBasicBlockUtils.h (added)
+++ llvm/trunk/lib/Target/PowerPC/PPCMachineBasicBlockUtils.h Wed Dec 13 06:47:35 2017
@@ -0,0 +1,198 @@
+//==-- PPCMachineBasicBlockUtils.h - Functions for common MBB operations ---==//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines utility functions for commonly used operations on
+// MachineBasicBlock's.
+// NOTE: Include this file after defining DEBUG_TYPE so that the debug messages
+// can be emitted for the pass that is using this.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_LIB_TARGET_PPC_MACHINE_BASIC_BLOCK_UTILS_H
+#define LLVM_LIB_TARGET_PPC_MACHINE_BASIC_BLOCK_UTILS_H
+
+#include "PPCInstrInfo.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineBranchProbabilityInfo.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+
+#ifndef DEBUG_TYPE
+#define DEBUG_TYPE "ppc-generic-mbb-utilities"
+#endif
+
+using namespace llvm;
+
+/// Given a basic block \p Successor that potentially contains PHIs, this
+/// function will look for any incoming values in the PHIs that are supposed to
+/// be coming from \p OrigMBB but whose definition is actually in \p NewMBB.
+/// Any such PHIs will be updated to reflect reality.
+static void updatePHIs(MachineBasicBlock *Successor, MachineBasicBlock *OrigMBB,
+ MachineBasicBlock *NewMBB, MachineRegisterInfo *MRI) {
+ for (auto &MI : Successor->instrs()) {
+ if (!MI.isPHI())
+ continue;
+ // This is a really ugly-looking loop, but it was pillaged directly from
+ // MachineBasicBlock::transferSuccessorsAndUpdatePHIs().
+ for (unsigned i = 2, e = MI.getNumOperands()+1; i != e; i += 2) {
+ MachineOperand &MO = MI.getOperand(i);
+ if (MO.getMBB() == OrigMBB) {
+ // Check if the instruction is actualy defined in NewMBB.
+ if (MI.getOperand(i-1).isReg()) {
+ MachineInstr *DefMI = MRI->getVRegDef(MI.getOperand(i-1).getReg());
+ if (DefMI->getParent() == NewMBB || !OrigMBB->isSuccessor(Successor)) {
+ MO.setMBB(NewMBB);
+ break;
+ }
+ }
+ }
+ }
+ }
+}
+
+/// Given a basic block \p Successor that potentially contains PHIs, this
+/// function will look for PHIs that have an incoming value from \p OrigMBB
+/// and will add the same incoming value from \p NewMBB.
+/// NOTE: This should only be used if \p NewMBB is an immediate dominator of
+/// \p OrigMBB.
+static void addIncomingValuesToPHIs(MachineBasicBlock *Successor,
+ MachineBasicBlock *OrigMBB,
+ MachineBasicBlock *NewMBB,
+ MachineRegisterInfo *MRI) {
+ assert(OrigMBB->isSuccessor(NewMBB) && "NewMBB must be a sucessor of OrigMBB");
+ for (auto &MI : Successor->instrs()) {
+ if (!MI.isPHI())
+ continue;
+ // This is a really ugly-looking loop, but it was pillaged directly from
+ // MachineBasicBlock::transferSuccessorsAndUpdatePHIs().
+ for (unsigned i = 2, e = MI.getNumOperands()+1; i != e; i += 2) {
+ MachineOperand &MO = MI.getOperand(i);
+ if (MO.getMBB() == OrigMBB) {
+ MachineInstrBuilder MIB(*MI.getParent()->getParent(), &MI);
+ MIB.addReg(MI.getOperand(i-1).getReg()).addMBB(NewMBB);
+ break;
+ }
+ }
+ }
+}
+
+struct BlockSplitInfo {
+ MachineInstr *OrigBranch;
+ MachineInstr *SplitBefore;
+ MachineInstr *SplitCond;
+ bool InvertNewBranch;
+ bool InvertOrigBranch;
+ bool BranchToFallThrough;
+ const MachineBranchProbabilityInfo *MBPI;
+ MachineInstr *MIToDelete;
+ MachineInstr *NewCond;
+ bool allInstrsInSameMBB() {
+ if (!OrigBranch || !SplitBefore || !SplitCond)
+ return false;
+ MachineBasicBlock *MBB = OrigBranch->getParent();
+ if (SplitBefore->getParent() != MBB ||
+ SplitCond->getParent() != MBB)
+ return false;
+ if (MIToDelete && MIToDelete->getParent() != MBB)
+ return false;
+ if (NewCond && NewCond->getParent() != MBB)
+ return false;
+ return true;
+ }
+};
+
+/// Splits a MachineBasicBlock to branch before \p SplitBefore. The original
+/// branch is \p OrigBranch. The target of the new branch can either be the same
+/// as the target of the original branch or the fallthrough successor of the
+/// original block as determined by \p BranchToFallThrough. The branch
+/// conditions will be inverted according to \p InvertNewBranch and
+/// \p InvertOrigBranch. If an instruction that previously fed the branch is to
+/// be deleted, it is provided in \p MIToDelete and \p NewCond will be used as
+/// the branch condition. The branch probabilities will be set if the
+/// MachineBranchProbabilityInfo isn't null.
+static bool splitMBB(BlockSplitInfo &BSI) {
+ assert(BSI.allInstrsInSameMBB() &&
+ "All instructions must be in the same block.");
+
+ MachineBasicBlock *ThisMBB = BSI.OrigBranch->getParent();
+ MachineFunction *MF = ThisMBB->getParent();
+ MachineRegisterInfo *MRI = &MF->getRegInfo();
+ assert(MRI->isSSA() && "Can only do this while the function is in SSA form.");
+ if (ThisMBB->succ_size() != 2) {
+ DEBUG(dbgs() << "Don't know how to handle blocks that don't have exactly"
+ << " two succesors.\n");
+ return false;
+ }
+
+ const PPCInstrInfo *TII = MF->getSubtarget<PPCSubtarget>().getInstrInfo();
+ unsigned OrigBROpcode = BSI.OrigBranch->getOpcode();
+ unsigned InvertedOpcode =
+ OrigBROpcode == PPC::BC ? PPC::BCn :
+ OrigBROpcode == PPC::BCn ? PPC::BC :
+ OrigBROpcode == PPC::BCLR ? PPC::BCLRn : PPC::BCLR;
+ unsigned NewBROpcode = BSI.InvertNewBranch ? InvertedOpcode : OrigBROpcode;
+ MachineBasicBlock *OrigTarget = BSI.OrigBranch->getOperand(1).getMBB();
+ MachineBasicBlock *OrigFallThrough =
+ OrigTarget == *ThisMBB->succ_begin() ? *ThisMBB->succ_rbegin() :
+ *ThisMBB->succ_begin();
+ MachineBasicBlock *NewBRTarget =
+ BSI.BranchToFallThrough ? OrigFallThrough : OrigTarget;
+ BranchProbability ProbToNewTarget =
+ !BSI.MBPI ? BranchProbability::getUnknown() :
+ BSI.MBPI->getEdgeProbability(ThisMBB, NewBRTarget);
+
+ // Create a new basic block.
+ MachineBasicBlock::iterator InsertPoint = BSI.SplitBefore;
+ const BasicBlock *LLVM_BB = ThisMBB->getBasicBlock();
+ MachineFunction::iterator It = ThisMBB->getIterator();
+ MachineBasicBlock *NewMBB = MF->CreateMachineBasicBlock(LLVM_BB);
+ MF->insert(++It, NewMBB);
+
+ // Move everything after SplitBefore into the new block.
+ NewMBB->splice(NewMBB->end(), ThisMBB, InsertPoint, ThisMBB->end());
+ NewMBB->transferSuccessors(ThisMBB);
+
+ // Add the two successors to ThisMBB. The probabilities come from the
+ // existing blocks if available.
+ ThisMBB->addSuccessor(NewBRTarget, ProbToNewTarget);
+ ThisMBB->addSuccessor(NewMBB, ProbToNewTarget.getCompl());
+
+ // Add the branches to ThisMBB.
+ BuildMI(*ThisMBB, ThisMBB->end(), BSI.SplitBefore->getDebugLoc(),
+ TII->get(NewBROpcode)).addReg(BSI.SplitCond->getOperand(0).getReg())
+ .addMBB(NewBRTarget);
+ BuildMI(*ThisMBB, ThisMBB->end(), BSI.SplitBefore->getDebugLoc(),
+ TII->get(PPC::B)).addMBB(NewMBB);
+ if (BSI.MIToDelete)
+ BSI.MIToDelete->eraseFromParent();
+
+ // Change the condition on the original branch and invert it if requested.
+ auto FirstTerminator = NewMBB->getFirstTerminator();
+ if (BSI.NewCond) {
+ assert(FirstTerminator->getOperand(0).isReg() &&
+ "Can't update condition of unconditional branch.");
+ FirstTerminator->getOperand(0).setReg(BSI.NewCond->getOperand(0).getReg());
+ }
+ if (BSI.InvertOrigBranch)
+ FirstTerminator->setDesc(TII->get(InvertedOpcode));
+
+ // If any of the PHIs in the successors of NewMBB reference values that
+ // now come from NewMBB, they need to be updated.
+ for (auto *Succ : NewMBB->successors()) {
+ updatePHIs(Succ, ThisMBB, NewMBB, MRI);
+ }
+ addIncomingValuesToPHIs(NewBRTarget, ThisMBB, NewMBB, MRI);
+
+ DEBUG(dbgs() << "After splitting, ThisMBB:\n"; ThisMBB->dump());
+ DEBUG(dbgs() << "NewMBB:\n"; NewMBB->dump());
+ DEBUG(dbgs() << "New branch-to block:\n"; NewBRTarget->dump());
+ return true;
+}
+
+
+#endif
Added: llvm/trunk/lib/Target/PowerPC/PPCReduceCRLogicals.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PowerPC/PPCReduceCRLogicals.cpp?rev=320584&view=auto
==============================================================================
--- llvm/trunk/lib/Target/PowerPC/PPCReduceCRLogicals.cpp (added)
+++ llvm/trunk/lib/Target/PowerPC/PPCReduceCRLogicals.cpp Wed Dec 13 06:47:35 2017
@@ -0,0 +1,533 @@
+//===---- PPCReduceCRLogicals.cpp - Reduce CR Bit Logical operations ------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===---------------------------------------------------------------------===//
+//
+// This pass aims to reduce the number of logical operations on bits in the CR
+// register. These instructions have a fairly high latency and only a single
+// pipeline at their disposal in modern PPC cores. Furthermore, they have a
+// tendency to occur in fairly small blocks where there's little opportunity
+// to hide the latency between the CR logical operation and its user.
+//
+//===---------------------------------------------------------------------===//
+
+#include "PPCInstrInfo.h"
+#include "PPC.h"
+#include "PPCTargetMachine.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineDominators.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/ADT/Statistic.h"
+
+using namespace llvm;
+
+#define DEBUG_TYPE "ppc-reduce-cr-ops"
+#include "PPCMachineBasicBlockUtils.h"
+
+STATISTIC(NumContainedSingleUseBinOps,
+ "Number of single-use binary CR logical ops contained in a block");
+STATISTIC(NumToSplitBlocks,
+ "Number of binary CR logical ops that can be used to split blocks");
+STATISTIC(TotalCRLogicals, "Number of CR logical ops.");
+STATISTIC(TotalNullaryCRLogicals,
+ "Number of nullary CR logical ops (CRSET/CRUNSET).");
+STATISTIC(TotalUnaryCRLogicals, "Number of unary CR logical ops.");
+STATISTIC(TotalBinaryCRLogicals, "Number of CR logical ops.");
+STATISTIC(NumBlocksSplitOnBinaryCROp,
+ "Number of blocks split on CR binary logical ops.");
+STATISTIC(NumNotSplitIdenticalOperands,
+ "Number of blocks not split due to operands being identical.");
+STATISTIC(NumNotSplitChainCopies,
+ "Number of blocks not split due to operands being chained copies.");
+STATISTIC(NumNotSplitWrongOpcode,
+ "Number of blocks not split due to the wrong opcode.");
+
+namespace llvm {
+ void initializePPCReduceCRLogicalsPass(PassRegistry&);
+}
+
+namespace {
+
+static bool isBinary(MachineInstr &MI) {
+ return MI.getNumOperands() == 3;
+}
+
+static bool isNullary(MachineInstr &MI) {
+ return MI.getNumOperands() == 1;
+}
+
+/// Given a CR logical operation \p CROp, branch opcode \p BROp as well as
+/// a flag to indicate if the first operand of \p CROp is used as the
+/// SplitBefore operand, determines whether either of the branches are to be
+/// inverted as well as whether the new target should be the original
+/// fall-through block.
+static void
+computeBranchTargetAndInversion(unsigned CROp, unsigned BROp, bool UsingDef1,
+ bool &InvertNewBranch, bool &InvertOrigBranch,
+ bool &TargetIsFallThrough) {
+ // The conditions under which each of the output operands should be [un]set
+ // can certainly be written much more concisely with just 3 if statements or
+ // ternary expressions. However, this provides a much clearer overview to the
+ // reader as to what is set for each <CROp, BROp, OpUsed> combination.
+ if (BROp == PPC::BC || BROp == PPC::BCLR) {
+ // Regular branches.
+ switch (CROp) {
+ default:
+ llvm_unreachable("Don't know how to handle this CR logical.");
+ case PPC::CROR:
+ InvertNewBranch = false;
+ InvertOrigBranch = false;
+ TargetIsFallThrough = false;
+ return;
+ case PPC::CRAND:
+ InvertNewBranch = true;
+ InvertOrigBranch = false;
+ TargetIsFallThrough = true;
+ return;
+ case PPC::CRNAND:
+ InvertNewBranch = true;
+ InvertOrigBranch = true;
+ TargetIsFallThrough = false;
+ return;
+ case PPC::CRNOR:
+ InvertNewBranch = false;
+ InvertOrigBranch = true;
+ TargetIsFallThrough = true;
+ return;
+ case PPC::CRORC:
+ InvertNewBranch = UsingDef1;
+ InvertOrigBranch = !UsingDef1;
+ TargetIsFallThrough = false;
+ return;
+ case PPC::CRANDC:
+ InvertNewBranch = !UsingDef1;
+ InvertOrigBranch = !UsingDef1;
+ TargetIsFallThrough = true;
+ return;
+ }
+ } else if (BROp == PPC::BCn || BROp == PPC::BCLRn) {
+ // Negated branches.
+ switch (CROp) {
+ default:
+ llvm_unreachable("Don't know how to handle this CR logical.");
+ case PPC::CROR:
+ InvertNewBranch = true;
+ InvertOrigBranch = false;
+ TargetIsFallThrough = true;
+ return;
+ case PPC::CRAND:
+ InvertNewBranch = false;
+ InvertOrigBranch = false;
+ TargetIsFallThrough = false;
+ return;
+ case PPC::CRNAND:
+ InvertNewBranch = false;
+ InvertOrigBranch = true;
+ TargetIsFallThrough = true;
+ return;
+ case PPC::CRNOR:
+ InvertNewBranch = true;
+ InvertOrigBranch = true;
+ TargetIsFallThrough = false;
+ return;
+ case PPC::CRORC:
+ InvertNewBranch = !UsingDef1;
+ InvertOrigBranch = !UsingDef1;
+ TargetIsFallThrough = true;
+ return;
+ case PPC::CRANDC:
+ InvertNewBranch = UsingDef1;
+ InvertOrigBranch = !UsingDef1;
+ TargetIsFallThrough = false;
+ return;
+ }
+ } else
+ llvm_unreachable("Don't know how to handle this branch.");
+}
+
+class PPCReduceCRLogicals : public MachineFunctionPass {
+
+public:
+ static char ID;
+ struct CRLogicalOpInfo {
+ MachineInstr *MI;
+ // FIXME: If chains of copies are to be handled, this should be a vector.
+ std::pair<MachineInstr*, MachineInstr*> CopyDefs;
+ std::pair<MachineInstr*, MachineInstr*> TrueDefs;
+ unsigned IsBinary : 1;
+ unsigned IsNullary : 1;
+ unsigned ContainedInBlock : 1;
+ unsigned FeedsISEL : 1;
+ unsigned FeedsBR : 1;
+ unsigned FeedsLogical : 1;
+ unsigned SingleUse : 1;
+ unsigned DefsSingleUse : 1;
+ unsigned SubregDef1;
+ unsigned SubregDef2;
+ CRLogicalOpInfo() : MI(nullptr), IsBinary(0), IsNullary(0),
+ ContainedInBlock(0), FeedsISEL(0), FeedsBR(0),
+ FeedsLogical(0), SingleUse(0), DefsSingleUse(1),
+ SubregDef1(0), SubregDef2(0) { }
+ void dump();
+ };
+
+private:
+ const PPCInstrInfo *TII;
+ MachineFunction *MF;
+ MachineRegisterInfo *MRI;
+ const MachineBranchProbabilityInfo *MBPI;
+
+ // A vector to contain all the CR logical operations
+ std::vector<CRLogicalOpInfo> AllCRLogicalOps;
+ void initialize(MachineFunction &MFParm);
+ void collectCRLogicals();
+ bool handleCROp(CRLogicalOpInfo &CRI);
+ bool splitBlockOnBinaryCROp(CRLogicalOpInfo &CRI);
+ static bool isCRLogical(MachineInstr &MI) {
+ unsigned Opc = MI.getOpcode();
+ return Opc == PPC::CRAND || Opc == PPC::CRNAND || Opc == PPC::CROR ||
+ Opc == PPC::CRXOR || Opc == PPC::CRNOR || Opc == PPC::CREQV ||
+ Opc == PPC::CRANDC || Opc == PPC::CRORC || Opc == PPC::CRSET ||
+ Opc == PPC::CRUNSET || Opc == PPC::CR6SET || Opc == PPC::CR6UNSET;
+ }
+ bool simplifyCode() {
+ bool Changed = false;
+ // Not using a range-based for loop here as the vector may grow while being
+ // operated on.
+ for (unsigned i = 0; i < AllCRLogicalOps.size(); i++)
+ Changed |= handleCROp(AllCRLogicalOps[i]);
+ return Changed;
+ }
+
+public:
+ PPCReduceCRLogicals() : MachineFunctionPass(ID) {
+ initializePPCReduceCRLogicalsPass(*PassRegistry::getPassRegistry());
+ }
+
+ MachineInstr *lookThroughCRCopy(unsigned Reg, unsigned &Subreg,
+ MachineInstr *&CpDef);
+ bool runOnMachineFunction(MachineFunction &MF) override {
+ if (skipFunction(*MF.getFunction()))
+ return false;
+
+ // If the subtarget doesn't use CR bits, there's nothing to do.
+ const PPCSubtarget &STI = MF.getSubtarget<PPCSubtarget>();
+ if (!STI.useCRBits())
+ return false;
+
+ initialize(MF);
+ collectCRLogicals();
+ return simplifyCode();
+ }
+ CRLogicalOpInfo createCRLogicalOpInfo(MachineInstr &MI);
+ void getAnalysisUsage(AnalysisUsage &AU) const override {
+ AU.addRequired<MachineBranchProbabilityInfo>();
+ AU.addRequired<MachineDominatorTree>();
+ MachineFunctionPass::getAnalysisUsage(AU);
+ }
+};
+
+void PPCReduceCRLogicals::CRLogicalOpInfo::dump() {
+ dbgs() << "CRLogicalOpMI: ";
+ MI->dump();
+ dbgs() << "IsBinary: " << IsBinary << ", FeedsISEL: " << FeedsISEL;
+ dbgs() << ", FeedsBR: " << FeedsBR << ", FeedsLogical: ";
+ dbgs() << FeedsLogical << ", SingleUse: " << SingleUse;
+ dbgs() << ", DefsSingleUse: " << DefsSingleUse;
+ dbgs() << ", SubregDef1: " << SubregDef1 << ", SubregDef2: ";
+ dbgs() << SubregDef2 << ", ContainedInBlock: " << ContainedInBlock;
+ if (!IsNullary) {
+ dbgs() << "\nDefs:\n";
+ TrueDefs.first->dump();
+ }
+ if (IsBinary)
+ TrueDefs.second->dump();
+ dbgs() << "\n";
+ if (CopyDefs.first) {
+ dbgs() << "CopyDef1: ";
+ CopyDefs.first->dump();
+ }
+ if (CopyDefs.second) {
+ dbgs() << "CopyDef2: ";
+ CopyDefs.second->dump();
+ }
+}
+
+PPCReduceCRLogicals::CRLogicalOpInfo
+PPCReduceCRLogicals::createCRLogicalOpInfo(MachineInstr &MIParam) {
+ CRLogicalOpInfo Ret;
+ Ret.MI = &MIParam;
+ // Get the defs
+ if (isNullary(MIParam)) {
+ Ret.IsNullary = 1;
+ Ret.TrueDefs = std::make_pair(nullptr, nullptr);
+ Ret.CopyDefs = std::make_pair(nullptr, nullptr);
+ } else {
+ MachineInstr *Def1 = lookThroughCRCopy(MIParam.getOperand(1).getReg(),
+ Ret.SubregDef1, Ret.CopyDefs.first);
+ Ret.DefsSingleUse &=
+ MRI->hasOneNonDBGUse(Def1->getOperand(0).getReg());
+ Ret.DefsSingleUse &=
+ MRI->hasOneNonDBGUse(Ret.CopyDefs.first->getOperand(0).getReg());
+ assert(Def1 && "Must be able to find a definition of operand 1.");
+ if (isBinary(MIParam)) {
+ Ret.IsBinary = 1;
+ MachineInstr *Def2 = lookThroughCRCopy(MIParam.getOperand(2).getReg(),
+ Ret.SubregDef2,
+ Ret.CopyDefs.second);
+ Ret.DefsSingleUse &=
+ MRI->hasOneNonDBGUse(Def2->getOperand(0).getReg());
+ Ret.DefsSingleUse &=
+ MRI->hasOneNonDBGUse(Ret.CopyDefs.second->getOperand(0).getReg());
+ assert(Def2 && "Must be able to find a definition of operand 2.");
+ Ret.TrueDefs = std::make_pair(Def1, Def2);
+ } else {
+ Ret.TrueDefs = std::make_pair(Def1, nullptr);
+ Ret.CopyDefs.second = nullptr;
+ }
+ }
+
+ Ret.ContainedInBlock = 1;
+ // Get the uses
+ for (MachineInstr &UseMI :
+ MRI->use_nodbg_instructions(MIParam.getOperand(0).getReg())) {
+ unsigned Opc = UseMI.getOpcode();
+ if (Opc == PPC::ISEL || Opc == PPC::ISEL8)
+ Ret.FeedsISEL = 1;
+ if (Opc == PPC::BC || Opc == PPC::BCn || Opc == PPC::BCLR ||
+ Opc == PPC::BCLRn)
+ Ret.FeedsBR = 1;
+ Ret.FeedsLogical = isCRLogical(UseMI);
+ if (UseMI.getParent() != MIParam.getParent())
+ Ret.ContainedInBlock = 0;
+ }
+ Ret.SingleUse = MRI->hasOneNonDBGUse(MIParam.getOperand(0).getReg()) ? 1 : 0;
+
+ // We now know whether all the uses of the CR logical are in the same block.
+ if (!Ret.IsNullary) {
+ Ret.ContainedInBlock &=
+ (MIParam.getParent() == Ret.TrueDefs.first->getParent());
+ if (Ret.IsBinary)
+ Ret.ContainedInBlock &=
+ (MIParam.getParent() == Ret.TrueDefs.second->getParent());
+ }
+ DEBUG(Ret.dump());
+ if (Ret.IsBinary && Ret.ContainedInBlock && Ret.SingleUse) {
+ NumContainedSingleUseBinOps++;
+ if (Ret.FeedsBR && Ret.DefsSingleUse)
+ NumToSplitBlocks++;
+ }
+ return Ret;
+}
+
+/// Looks trhough a COPY instruction to the actual definition of the CR-bit
+/// register and returns the instruction that defines it.
+/// FIXME: This currently handles what is by-far the most common case:
+/// an instruction that defines a CR field followed by a single copy of a bit
+/// from that field into a virtual register. If chains of copies need to be
+/// handled, this should have a loop until a non-copy instruction is found.
+MachineInstr *PPCReduceCRLogicals::lookThroughCRCopy(unsigned Reg,
+ unsigned &Subreg,
+ MachineInstr *&CpDef) {
+ Subreg = -1;
+ if (!TargetRegisterInfo::isVirtualRegister(Reg))
+ return nullptr;
+ MachineInstr *Copy = MRI->getVRegDef(Reg);
+ CpDef = Copy;
+ if (!Copy->isCopy())
+ return Copy;
+ unsigned CopySrc = Copy->getOperand(1).getReg();
+ Subreg = Copy->getOperand(1).getSubReg();
+ if (!TargetRegisterInfo::isVirtualRegister(CopySrc)) {
+ const TargetRegisterInfo *TRI = &TII->getRegisterInfo();
+ // Set the Subreg
+ if (CopySrc == PPC::CR0EQ || CopySrc == PPC::CR6EQ)
+ Subreg = PPC::sub_eq;
+ if (CopySrc == PPC::CR0LT || CopySrc == PPC::CR6LT)
+ Subreg = PPC::sub_lt;
+ if (CopySrc == PPC::CR0GT || CopySrc == PPC::CR6GT)
+ Subreg = PPC::sub_gt;
+ if (CopySrc == PPC::CR0UN || CopySrc == PPC::CR6UN)
+ Subreg = PPC::sub_un;
+ // Loop backwards and return the first MI that modifies the physical CR Reg.
+ MachineBasicBlock::iterator Me = Copy, B = Copy->getParent()->begin();
+ while (Me != B)
+ if ((--Me)->modifiesRegister(CopySrc, TRI))
+ return &*Me;
+ return nullptr;
+ }
+ return MRI->getVRegDef(CopySrc);
+}
+
+void PPCReduceCRLogicals::initialize(MachineFunction &MFParam) {
+ MF = &MFParam;
+ MRI = &MF->getRegInfo();
+ TII = MF->getSubtarget<PPCSubtarget>().getInstrInfo();
+ MBPI = &getAnalysis<MachineBranchProbabilityInfo>();
+
+ AllCRLogicalOps.clear();
+}
+
+/// Contains all the implemented transformations on CR logical operations.
+/// For example, a binary CR logical can be used to split a block on its inputs,
+/// a unary CR logical might be used to change the condition code on a
+/// comparison feeding it. A nullary CR logical might simply be removable
+/// if the user of the bit it [un]sets can be transformed.
+bool PPCReduceCRLogicals::handleCROp(CRLogicalOpInfo &CRI) {
+ // We can definitely split a block on the inputs to a binary CR operation
+ // whose defs and (single) use are within the same block.
+ bool Changed = false;
+ if (CRI.IsBinary && CRI.ContainedInBlock && CRI.SingleUse && CRI.FeedsBR &&
+ CRI.DefsSingleUse) {
+ Changed = splitBlockOnBinaryCROp(CRI);
+ if (Changed)
+ NumBlocksSplitOnBinaryCROp++;
+ }
+ return Changed;
+}
+
+/// Splits a block that contains a CR-logical operation that feeds a branch
+/// and whose operands are produced within the block.
+/// Example:
+/// %vr5<def> = CMPDI %vr2, 0; CRRC:%vr5 G8RC:%vr2
+/// %vr6<def> = COPY %vr5:sub_eq; CRBITRC:%vr6 CRRC:%vr5
+/// %vr7<def> = CMPDI %vr3, 0; CRRC:%vr7 G8RC:%vr3
+/// %vr8<def> = COPY %vr7:sub_eq; CRBITRC:%vr8 CRRC:%vr7
+/// %vr9<def> = CROR %vr6<kill>, %vr8<kill>; CRBITRC:%vr9,%vr6,%vr8
+/// BC %vr9<kill>, <BB#2>; CRBITRC:%vr9
+/// Becomes:
+/// %vr5<def> = CMPDI %vr2, 0; CRRC:%vr5 G8RC:%vr2
+/// %vr6<def> = COPY %vr5:sub_eq; CRBITRC:%vr6 CRRC:%vr5
+/// BC %vr6<kill>, <BB#2>; CRBITRC:%vr6
+///
+/// %vr7<def> = CMPDI %vr3, 0; CRRC:%vr7 G8RC:%vr3
+/// %vr8<def> = COPY %vr7:sub_eq; CRBITRC:%vr8 CRRC:%vr7
+/// BC %vr9<kill>, <BB#2>; CRBITRC:%vr9
+bool PPCReduceCRLogicals::splitBlockOnBinaryCROp(CRLogicalOpInfo &CRI) {
+ if (CRI.CopyDefs.first == CRI.CopyDefs.second) {
+ DEBUG(dbgs() << "Unable to split as the two operands are the same\n");
+ NumNotSplitIdenticalOperands++;
+ return false;
+ }
+ if (CRI.TrueDefs.first->isCopy() || CRI.TrueDefs.second->isCopy() ||
+ CRI.TrueDefs.first->isPHI() || CRI.TrueDefs.second->isPHI()) {
+ DEBUG(dbgs() << "Unable to split because one of the operands is a PHI or "
+ "chain of copies.\n");
+ NumNotSplitChainCopies++;
+ return false;
+ }
+ // Note: keep in sync with computeBranchTargetAndInversion().
+ if (CRI.MI->getOpcode() != PPC::CROR &&
+ CRI.MI->getOpcode() != PPC::CRAND &&
+ CRI.MI->getOpcode() != PPC::CRNOR &&
+ CRI.MI->getOpcode() != PPC::CRNAND &&
+ CRI.MI->getOpcode() != PPC::CRORC &&
+ CRI.MI->getOpcode() != PPC::CRANDC) {
+ DEBUG(dbgs() << "Unable to split blocks on this opcode.\n");
+ NumNotSplitWrongOpcode++;
+ return false;
+ }
+ DEBUG(dbgs() << "Splitting the following CR op:\n"; CRI.dump());
+ MachineBasicBlock::iterator Def1It = CRI.TrueDefs.first;
+ MachineBasicBlock::iterator Def2It = CRI.TrueDefs.second;
+
+ bool UsingDef1 = false;
+ MachineInstr *SplitBefore = &*Def2It;
+ for (auto E = CRI.MI->getParent()->end(); Def2It != E; ++Def2It) {
+ if (Def1It == Def2It) { // Def2 comes before Def1.
+ SplitBefore = &*Def1It;
+ UsingDef1 = true;
+ break;
+ }
+ }
+
+ DEBUG(dbgs() << "We will split the following block:\n";);
+ DEBUG(CRI.MI->getParent()->dump());
+ DEBUG(dbgs() << "Before instruction:\n"; SplitBefore->dump());
+
+ // Get the branch instruction.
+ MachineInstr *Branch =
+ MRI->use_nodbg_begin(CRI.MI->getOperand(0).getReg())->getParent();
+
+ // We want the new block to have no code in it other than the definition
+ // of the input to the CR logical and the CR logical itself. So we move
+ // those to the bottom of the block (just before the branch). Then we
+ // will split before the CR logical.
+ MachineBasicBlock *MBB = SplitBefore->getParent();
+ auto FirstTerminator = MBB->getFirstTerminator();
+ MachineBasicBlock::iterator FirstInstrToMove =
+ UsingDef1 ? CRI.TrueDefs.first : CRI.TrueDefs.second;
+ MachineBasicBlock::iterator SecondInstrToMove =
+ UsingDef1 ? CRI.CopyDefs.first : CRI.CopyDefs.second;
+
+ // The instructions that need to be moved are not guaranteed to be
+ // contiguous. Move them individually.
+ // FIXME: If one of the operands is a chain of (single use) copies, they
+ // can all be moved and we can still split.
+ MBB->splice(FirstTerminator, MBB, FirstInstrToMove);
+ if (FirstInstrToMove != SecondInstrToMove)
+ MBB->splice(FirstTerminator, MBB, SecondInstrToMove);
+ MBB->splice(FirstTerminator, MBB, CRI.MI);
+
+ unsigned Opc = CRI.MI->getOpcode();
+ bool InvertOrigBranch, InvertNewBranch, TargetIsFallThrough;
+ computeBranchTargetAndInversion(Opc, Branch->getOpcode(), UsingDef1,
+ InvertNewBranch, InvertOrigBranch,
+ TargetIsFallThrough);
+ MachineInstr *SplitCond =
+ UsingDef1 ? CRI.CopyDefs.second : CRI.CopyDefs.first;
+ DEBUG(dbgs() << "We will " << (InvertNewBranch ? "invert" : "copy"));
+ DEBUG(dbgs() << " the original branch and the target is the " <<
+ (TargetIsFallThrough ? "fallthrough block\n" : "orig. target block\n"));
+ DEBUG(dbgs() << "Original branch instruction: "; Branch->dump());
+ BlockSplitInfo BSI { Branch, SplitBefore, SplitCond, InvertNewBranch,
+ InvertOrigBranch, TargetIsFallThrough, MBPI, CRI.MI,
+ UsingDef1 ? CRI.CopyDefs.first : CRI.CopyDefs.second };
+ bool Changed = splitMBB(BSI);
+ // If we've split on a CR logical that is fed by a CR logical,
+ // recompute the source CR logical as it may be usable for splitting.
+ if (Changed) {
+ bool Input1CRlogical =
+ CRI.TrueDefs.first && isCRLogical(*CRI.TrueDefs.first);
+ bool Input2CRlogical =
+ CRI.TrueDefs.second && isCRLogical(*CRI.TrueDefs.second);
+ if (Input1CRlogical)
+ AllCRLogicalOps.push_back(createCRLogicalOpInfo(*CRI.TrueDefs.first));
+ if (Input2CRlogical)
+ AllCRLogicalOps.push_back(createCRLogicalOpInfo(*CRI.TrueDefs.second));
+ }
+ return Changed;
+}
+
+void PPCReduceCRLogicals::collectCRLogicals() {
+ for (MachineBasicBlock &MBB : *MF) {
+ for (MachineInstr &MI : MBB) {
+ if (isCRLogical(MI)) {
+ AllCRLogicalOps.push_back(createCRLogicalOpInfo(MI));
+ TotalCRLogicals++;
+ if (AllCRLogicalOps.back().IsNullary)
+ TotalNullaryCRLogicals++;
+ else if (AllCRLogicalOps.back().IsBinary)
+ TotalBinaryCRLogicals++;
+ else
+ TotalUnaryCRLogicals++;
+ }
+ }
+ }
+}
+
+} // end annonymous namespace
+
+INITIALIZE_PASS_BEGIN(PPCReduceCRLogicals, DEBUG_TYPE,
+ "PowerPC Reduce CR logical Operation", false, false)
+INITIALIZE_PASS_DEPENDENCY(MachineDominatorTree)
+INITIALIZE_PASS_END(PPCReduceCRLogicals, DEBUG_TYPE,
+ "PowerPC Reduce CR logical Operation", false, false)
+
+char PPCReduceCRLogicals::ID = 0;
+FunctionPass*
+llvm::createPPCReduceCRLogicalsPass() { return new PPCReduceCRLogicals(); }
Modified: llvm/trunk/lib/Target/PowerPC/PPCTargetMachine.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PowerPC/PPCTargetMachine.cpp?rev=320584&r1=320583&r2=320584&view=diff
==============================================================================
--- llvm/trunk/lib/Target/PowerPC/PPCTargetMachine.cpp (original)
+++ llvm/trunk/lib/Target/PowerPC/PPCTargetMachine.cpp Wed Dec 13 06:47:35 2017
@@ -88,6 +88,10 @@ EnableMachineCombinerPass("ppc-machine-c
cl::desc("Enable the machine combiner pass"),
cl::init(true), cl::Hidden);
+static cl::opt<bool>
+ ReduceCRLogical("ppc-reduce-cr-logicals",
+ cl::desc("Expand eligible cr-logical binary ops to branches"),
+ cl::init(false), cl::Hidden);
extern "C" void LLVMInitializePowerPCTarget() {
// Register the targets
RegisterTargetMachine<PPCTargetMachine> A(getThePPC32Target());
@@ -392,6 +396,9 @@ void PPCPassConfig::addMachineSSAOptimiz
if (TM->getTargetTriple().getArch() == Triple::ppc64le &&
!DisableVSXSwapRemoval)
addPass(createPPCVSXSwapRemovalPass());
+ // Reduce the number of cr-logical ops.
+ if (ReduceCRLogical && getOptLevel() != CodeGenOpt::None)
+ addPass(createPPCReduceCRLogicalsPass());
// Target-specific peephole cleanups performed after instruction
// selection.
if (!DisableMIPeephole) {
Modified: llvm/trunk/test/CodeGen/PowerPC/licm-remat.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/PowerPC/licm-remat.ll?rev=320584&r1=320583&r2=320584&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/PowerPC/licm-remat.ll (original)
+++ llvm/trunk/test/CodeGen/PowerPC/licm-remat.ll Wed Dec 13 06:47:35 2017
@@ -1,4 +1,5 @@
-; RUN: llc -verify-machineinstrs -mtriple=powerpc64le-unknown-linux-gnu < %s | FileCheck %s
+; RUN: llc -verify-machineinstrs -ppc-reduce-cr-logicals \
+; RUN: -mtriple=powerpc64le-unknown-linux-gnu < %s | FileCheck %s
; Test case is reduced from the snappy benchmark.
; Verify MachineLICM will always hoist trivially rematerializable instructions even when register pressure is high.
@@ -21,8 +22,8 @@ define linkonce_odr void @ZN6snappyDecom
; CHECK: # %bb.0: # %entry
; CHECK: addis 3, 2, _ZN6snappy8internalL8wordmaskE at toc@ha
; CHECK-DAG: addi 25, 3, _ZN6snappy8internalL8wordmaskE at toc@l
-; CHECK-DAG: addis 4, 2, _ZN6snappy8internalL10char_tableE at toc@ha
-; CHECK-DAG: addi 24, 4, _ZN6snappy8internalL10char_tableE at toc@l
+; CHECK-DAG: addis 5, 2, _ZN6snappy8internalL10char_tableE at toc@ha
+; CHECK-DAG: addi 24, 5, _ZN6snappy8internalL10char_tableE at toc@l
; CHECK: b .LBB0_2
; CHECK: .LBB0_2: # %for.cond
; CHECK-NOT: addis {{[0-9]+}}, 2, _ZN6snappy8internalL8wordmaskE at toc@ha
Modified: llvm/trunk/test/CodeGen/PowerPC/select-i1-vs-i1.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/PowerPC/select-i1-vs-i1.ll?rev=320584&r1=320583&r2=320584&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/PowerPC/select-i1-vs-i1.ll (original)
+++ llvm/trunk/test/CodeGen/PowerPC/select-i1-vs-i1.ll Wed Dec 13 06:47:35 2017
@@ -1,5 +1,6 @@
-; RUN: llc -verify-machineinstrs < %s | FileCheck %s
-; RUN: llc -verify-machineinstrs -ppc-gen-isel=false < %s | FileCheck --check-prefix=CHECK-NO-ISEL %s
+; RUN: llc -ppc-reduce-cr-logicals -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -ppc-reduce-cr-logicals -verify-machineinstrs \
+; RUN: -ppc-gen-isel=false < %s | FileCheck --check-prefix=CHECK-NO-ISEL %s
target datalayout = "E-m:e-i64:64-n32:64"
target triple = "powerpc64-unknown-linux-gnu"
@@ -475,12 +476,13 @@ entry:
ret float %cond
; CHECK-LABEL: @testfloatslt
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB1:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bc 12, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: fmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: fmr 1, 5
; CHECK: blr
}
@@ -494,12 +496,13 @@ entry:
ret float %cond
; CHECK-LABEL: @testfloatult
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB1:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bc 4, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: fmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: fmr 1, 5
; CHECK: blr
}
@@ -513,10 +516,10 @@ entry:
ret float %cond
; CHECK-LABEL: @testfloatsle
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bc 12, 2, .LBB[[BB]]
; CHECK: fmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: fmr 1, 5
@@ -532,10 +535,10 @@ entry:
ret float %cond
; CHECK-LABEL: @testfloatule
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bc 4, 2, .LBB[[BB]]
; CHECK: fmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: fmr 1, 5
@@ -570,10 +573,10 @@ entry:
ret float %cond
; CHECK-LABEL: @testfloatsge
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bc 4, 2, .LBB[[BB]]
; CHECK: fmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: fmr 1, 5
@@ -589,10 +592,10 @@ entry:
ret float %cond
; CHECK-LABEL: @testfloatuge
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bc 12, 2, .LBB[[BB]]
; CHECK: fmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: fmr 1, 5
@@ -608,12 +611,13 @@ entry:
ret float %cond
; CHECK-LABEL: @testfloatsgt
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB1:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bc 4, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: fmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: fmr 1, 5
; CHECK: blr
}
@@ -627,12 +631,13 @@ entry:
ret float %cond
; CHECK-LABEL: @testfloatugt
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB1:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bc 12, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: fmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: fmr 1, 5
; CHECK: blr
}
@@ -665,12 +670,13 @@ entry:
ret double %cond
; CHECK-LABEL: @testdoubleslt
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB1:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bc 12, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: fmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: fmr 1, 5
; CHECK: blr
}
@@ -684,12 +690,13 @@ entry:
ret double %cond
; CHECK-LABEL: @testdoubleult
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB1:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bc 4, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: fmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: fmr 1, 5
; CHECK: blr
}
@@ -703,10 +710,10 @@ entry:
ret double %cond
; CHECK-LABEL: @testdoublesle
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bc 12, 2, .LBB[[BB]]
; CHECK: fmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: fmr 1, 5
@@ -722,10 +729,10 @@ entry:
ret double %cond
; CHECK-LABEL: @testdoubleule
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bc 4, 2, .LBB[[BB]]
; CHECK: fmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: fmr 1, 5
@@ -760,10 +767,10 @@ entry:
ret double %cond
; CHECK-LABEL: @testdoublesge
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bc 4, 2, .LBB[[BB]]
; CHECK: fmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: fmr 1, 5
@@ -779,10 +786,10 @@ entry:
ret double %cond
; CHECK-LABEL: @testdoubleuge
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bc 12, 2, .LBB[[BB]]
; CHECK: fmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: fmr 1, 5
@@ -798,12 +805,13 @@ entry:
ret double %cond
; CHECK-LABEL: @testdoublesgt
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB1:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bc 4, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: fmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: fmr 1, 5
; CHECK: blr
}
@@ -817,12 +825,13 @@ entry:
ret double %cond
; CHECK-LABEL: @testdoubleugt
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB1:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bc 12, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: fmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: fmr 1, 5
; CHECK: blr
}
@@ -856,9 +865,10 @@ entry:
; CHECK-LABEL: @testv4floatslt
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bclr 12, [[REG1]], 0
+; CHECK: bclr 12, 2, 0
+; CHECK: .LBB[[BB]]:
; CHECK: vmr 2, 3
; CHECK: blr
}
@@ -872,10 +882,11 @@ entry:
ret <4 x float> %cond
; CHECK-LABEL: @testv4floatult
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bclr 12, [[REG1]], 0
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bclr 4, 2, 0
+; CHECK: .LBB[[BB]]:
; CHECK: vmr 2, 3
; CHECK: blr
}
@@ -889,10 +900,10 @@ entry:
ret <4 x float> %cond
; CHECK-LABEL: @testv4floatsle
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bclr 12, [[REG1]], 0
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bclr 4, 2, 0
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bclr 12, 2, 0
; CHECK: vmr 2, 3
; CHECK: blr
}
@@ -907,9 +918,9 @@ entry:
; CHECK-LABEL: @testv4floatule
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bclr 12, 2, 0
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bclr 12, [[REG1]], 0
+; CHECK: bclr 4, 2, 0
; CHECK: vmr 2, 3
; CHECK: blr
}
@@ -943,9 +954,9 @@ entry:
; CHECK-LABEL: @testv4floatsge
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bclr 12, 2, 0
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bclr 12, [[REG1]], 0
+; CHECK: bclr 4, 2, 0
; CHECK: vmr 2, 3
; CHECK: blr
}
@@ -960,9 +971,9 @@ entry:
; CHECK-LABEL: @testv4floatuge
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bclr 4, 2, 0
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bclr 12, [[REG1]], 0
+; CHECK: bclr 12, 2, 0
; CHECK: vmr 2, 3
; CHECK: blr
}
@@ -976,10 +987,10 @@ entry:
ret <4 x float> %cond
; CHECK-LABEL: @testv4floatsgt
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bclr 12, [[REG1]], 0
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB1:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bclr 4, 2, 0
; CHECK: vmr 2, 3
; CHECK: blr
}
@@ -993,10 +1004,11 @@ entry:
ret <4 x float> %cond
; CHECK-LABEL: @testv4floatugt
-; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
-; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bclr 12, [[REG1]], 0
+; CHECK: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB:[0-9_]+]]
+; CHECK: fcmpu {{[0-9]+}}, 1, 2
+; CHECK: bclr 12, 2, 0
+; CHECK: .LBB[[BB]]
; CHECK: vmr 2, 3
; CHECK: blr
}
@@ -1059,9 +1071,10 @@ entry:
; CHECK-LABEL: @testv2doubleslt
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bclr 12, [[REG1]], 0
+; CHECK: bc 4, 2, .LBB[[BB]]
+; CHECK: .LBB[[BB]]:
; CHECK: vmr 2, 3
; CHECK: blr
}
@@ -1076,9 +1089,10 @@ entry:
; CHECK-LABEL: @testv2doubleult
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bclr 12, [[REG1]], 0
+; CHECK: bc 12, 2, .LBB[[BB]]
+; CHECK: .LBB[[BB]]:
; CHECK: vmr 2, 3
; CHECK: blr
}
@@ -1093,9 +1107,9 @@ entry:
; CHECK-LABEL: @testv2doublesle
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bclr 4, 2, 0
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bclr 12, [[REG1]], 0
+; CHECK: bclr 12, 2, 0
; CHECK: vmr 2, 3
; CHECK: blr
}
@@ -1110,9 +1124,9 @@ entry:
; CHECK-LABEL: @testv2doubleule
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bclr 12, 2, 0
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bclr 12, [[REG1]], 0
+; CHECK: bclr 4, 2, 0
; CHECK: vmr 2, 3
; CHECK: blr
}
@@ -1146,9 +1160,9 @@ entry:
; CHECK-LABEL: @testv2doublesge
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bclr 12, 2, 0
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bclr 12, [[REG1]], 0
+; CHECK: bclr 4, 2, 0
; CHECK: vmr 2, 3
; CHECK: blr
}
@@ -1163,9 +1177,9 @@ entry:
; CHECK-LABEL: @testv2doubleuge
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bclr 4, 2, 0
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bclr 12, [[REG1]], 0
+; CHECK: bclr 12, 2, 0
; CHECK: vmr 2, 3
; CHECK: blr
}
@@ -1180,9 +1194,10 @@ entry:
; CHECK-LABEL: @testv2doublesgt
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bclr 12, [[REG1]], 0
+; CHECK: bc 12, 2, .LBB[[BB]]
+; CHECK: .LBB[[BB]]
; CHECK: vmr 2, 3
; CHECK: blr
}
@@ -1197,9 +1212,10 @@ entry:
; CHECK-LABEL: @testv2doubleugt
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bclr 12, [[REG1]], 0
+; CHECK: bc 4, 2, .LBB[[BB]]
+; CHECK: .LBB[[BB]]
; CHECK: vmr 2, 3
; CHECK: blr
}
@@ -1231,11 +1247,12 @@ entry:
; CHECK-LABEL: @testqv4doubleslt
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB1:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 12, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: qvfmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: qvfmr 1, 5
; CHECK: blr
}
@@ -1250,11 +1267,12 @@ entry:
; CHECK-LABEL: @testqv4doubleult
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB1:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 4, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: qvfmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: qvfmr 1, 5
; CHECK: blr
}
@@ -1269,9 +1287,9 @@ entry:
; CHECK-LABEL: @testqv4doublesle
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 12, 2, .LBB[[BB]]
; CHECK: qvfmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: qvfmr 1, 5
@@ -1288,9 +1306,9 @@ entry:
; CHECK-LABEL: @testqv4doubleule
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 4, 2, .LBB[[BB]]
; CHECK: qvfmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: qvfmr 1, 5
@@ -1326,9 +1344,9 @@ entry:
; CHECK-LABEL: @testqv4doublesge
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 4, 2, .LBB[[BB]]
; CHECK: qvfmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: qvfmr 1, 5
@@ -1345,9 +1363,9 @@ entry:
; CHECK-LABEL: @testqv4doubleuge
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 12, 2, .LBB[[BB]]
; CHECK: qvfmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: qvfmr 1, 5
@@ -1364,11 +1382,12 @@ entry:
; CHECK-LABEL: @testqv4doublesgt
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB1:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 4, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: qvfmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: qvfmr 1, 5
; CHECK: blr
}
@@ -1383,11 +1402,12 @@ entry:
; CHECK-LABEL: @testqv4doubleugt
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB1:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 12, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: qvfmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: qvfmr 1, 5
; CHECK: blr
}
@@ -1421,11 +1441,12 @@ entry:
; CHECK-LABEL: @testqv4floatslt
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB1:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 12, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: qvfmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: qvfmr 1, 5
; CHECK: blr
}
@@ -1440,11 +1461,12 @@ entry:
; CHECK-LABEL: @testqv4floatult
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB1:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 4, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: qvfmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: qvfmr 1, 5
; CHECK: blr
}
@@ -1459,9 +1481,9 @@ entry:
; CHECK-LABEL: @testqv4floatsle
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 12, 2, .LBB[[BB]]
; CHECK: qvfmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: qvfmr 1, 5
@@ -1478,9 +1500,9 @@ entry:
; CHECK-LABEL: @testqv4floatule
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 4, 2, .LBB[[BB]]
; CHECK: qvfmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: qvfmr 1, 5
@@ -1516,9 +1538,9 @@ entry:
; CHECK-LABEL: @testqv4floatsge
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 4, 2, .LBB[[BB]]
; CHECK: qvfmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: qvfmr 1, 5
@@ -1535,9 +1557,9 @@ entry:
; CHECK-LABEL: @testqv4floatuge
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 12, 2, .LBB[[BB]]
; CHECK: qvfmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: qvfmr 1, 5
@@ -1554,11 +1576,12 @@ entry:
; CHECK-LABEL: @testqv4floatsgt
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB1:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 4, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: qvfmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: qvfmr 1, 5
; CHECK: blr
}
@@ -1573,11 +1596,12 @@ entry:
; CHECK-LABEL: @testqv4floatugt
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB1:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 12, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: qvfmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: qvfmr 1, 5
; CHECK: blr
}
@@ -1611,11 +1635,12 @@ entry:
; CHECK-LABEL: @testqv4i1slt
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB1:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 12, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: qvfmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: qvfmr 1, 5
; CHECK: blr
}
@@ -1630,11 +1655,12 @@ entry:
; CHECK-LABEL: @testqv4i1ult
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB1:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 4, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: qvfmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: qvfmr 1, 5
; CHECK: blr
}
@@ -1649,9 +1675,9 @@ entry:
; CHECK-LABEL: @testqv4i1sle
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 12, 2, .LBB[[BB]]
; CHECK: qvfmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: qvfmr 1, 5
@@ -1668,9 +1694,9 @@ entry:
; CHECK-LABEL: @testqv4i1ule
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 4, 2, .LBB[[BB]]
; CHECK: qvfmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: qvfmr 1, 5
@@ -1706,9 +1732,9 @@ entry:
; CHECK-LABEL: @testqv4i1sge
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 4, 2, .LBB[[BB]]
; CHECK: qvfmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: qvfmr 1, 5
@@ -1725,9 +1751,9 @@ entry:
; CHECK-LABEL: @testqv4i1uge
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crorc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 12, 2, .LBB[[BB]]
; CHECK: qvfmr 5, 6
; CHECK: .LBB[[BB]]:
; CHECK: qvfmr 1, 5
@@ -1744,11 +1770,12 @@ entry:
; CHECK-LABEL: @testqv4i1sgt
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 4, 2, .LBB[[BB1:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 4, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: qvfmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: qvfmr 1, 5
; CHECK: blr
}
@@ -1763,11 +1790,12 @@ entry:
; CHECK-LABEL: @testqv4i1ugt
; CHECK-DAG: fcmpu {{[0-9]+}}, 3, 4
+; CHECK: bc 12, 2, .LBB[[BB1:[0-9_]+]]
; CHECK-DAG: fcmpu {{[0-9]+}}, 1, 2
-; CHECK: crandc [[REG1:[0-9]+]], {{[0-9]+}}, {{[0-9]+}}
-; CHECK: bc 12, [[REG1]], .LBB[[BB:[0-9_]+]]
+; CHECK: bc 12, 2, .LBB[[BB2:[0-9_]+]]
+; CHECK: .LBB[[BB1]]:
; CHECK: qvfmr 5, 6
-; CHECK: .LBB[[BB]]:
+; CHECK: .LBB[[BB2]]:
; CHECK: qvfmr 1, 5
; CHECK: blr
}
Modified: llvm/trunk/test/CodeGen/PowerPC/tail-dup-layout.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/PowerPC/tail-dup-layout.ll?rev=320584&r1=320583&r2=320584&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/PowerPC/tail-dup-layout.ll (original)
+++ llvm/trunk/test/CodeGen/PowerPC/tail-dup-layout.ll Wed Dec 13 06:47:35 2017
@@ -1,5 +1,7 @@
-; RUN: llc -O2 -o - %s | FileCheck --check-prefix=CHECK --check-prefix=CHECK-O2 %s
-; RUN: llc -O3 -o - %s | FileCheck --check-prefix=CHECK --check-prefix=CHECK-O3 %s
+; RUN: llc -O2 -ppc-reduce-cr-logicals -o - %s | FileCheck \
+; RUN: --check-prefix=CHECK --check-prefix=CHECK-O2 %s
+; RUN: llc -O3 -ppc-reduce-cr-logicals -o - %s | FileCheck \
+; RUN: --check-prefix=CHECK --check-prefix=CHECK-O3 %s
target datalayout = "e-m:e-i64:64-n32:64"
target triple = "powerpc64le-grtev4-linux-gnu"
@@ -276,8 +278,9 @@ exit:
;CHECK: add [[TAGPTRREG:[0-9]+]], 3, 4
;CHECK: .[[LATCHLABEL:[._0-9A-Za-z]+]]: # %for.latch
;CHECK: addi
-;CHECK: .[[CHECKLABEL:[._0-9A-Za-z]+]]: # %for.check
+;CHECK-O2: .[[CHECKLABEL:[._0-9A-Za-z]+]]: # %for.check
;CHECK: lwz [[TAGREG:[0-9]+]], 0([[TAGPTRREG]])
+;CHECK-O3: .[[CHECKLABEL:[._0-9A-Za-z]+]]: # %for.check
;CHECK: # %bb.{{[0-9]+}}: # %test1
;CHECK: andi. {{[0-9]+}}, [[TAGREG]], 1
;CHECK-NEXT: bc 12, 1, .[[OPT1LABEL:[._0-9A-Za-z]+]]
More information about the llvm-commits
mailing list