// This pass must be run after register allocation. After this pass is
// executed, it is illegal to construct MO_FrameIndex operands.
//
-// This pass provides an optional shrink wrapping variant of prolog/epilog
-// insertion, enabled via --shrink-wrap. See ShrinkWrapping.cpp.
-//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "pei"
#include "PrologEpilogInserter.h"
+#include "llvm/ADT/IndexedMap.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/SetVector.h"
+#include "llvm/ADT/SmallSet.h"
+#include "llvm/ADT/Statistic.h"
#include "llvm/CodeGen/MachineDominators.h"
-#include "llvm/CodeGen/MachineLoopInfo.h"
-#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineInstr.h"
+#include "llvm/CodeGen/MachineLoopInfo.h"
+#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/RegisterScavenging.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetRegisterInfo.h"
-#include "llvm/Target/TargetFrameInfo.h"
-#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/CodeGen/StackProtector.h"
+#include "llvm/IR/DiagnosticInfo.h"
+#include "llvm/IR/InlineAsm.h"
+#include "llvm/IR/LLVMContext.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
-#include "llvm/ADT/IndexedMap.h"
-#include "llvm/ADT/SmallSet.h"
-#include "llvm/ADT/STLExtras.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetFrameLowering.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetRegisterInfo.h"
+#include "llvm/Target/TargetSubtargetInfo.h"
#include <climits>
using namespace llvm;
-// FIXME: For testing purposes only. Remove once the pre-allocation pass
-// is done.
-extern cl::opt<bool> EnableLocalStackAlloc;
+#define DEBUG_TYPE "pei"
char PEI::ID = 0;
+char &llvm::PrologEpilogCodeInserterID = PEI::ID;
+
+static cl::opt<unsigned>
+WarnStackSize("warn-stack-size", cl::Hidden, cl::init((unsigned)-1),
+ cl::desc("Warn for stack size bigger than the given"
+ " number"));
+
+INITIALIZE_PASS_BEGIN(PEI, "prologepilog",
+ "Prologue/Epilogue Insertion", false, false)
+INITIALIZE_PASS_DEPENDENCY(MachineLoopInfo)
+INITIALIZE_PASS_DEPENDENCY(MachineDominatorTree)
+INITIALIZE_PASS_DEPENDENCY(StackProtector)
+INITIALIZE_PASS_DEPENDENCY(TargetPassConfig)
+INITIALIZE_PASS_END(PEI, "prologepilog",
+ "Prologue/Epilogue Insertion & Frame Finalization",
+ false, false)
+
+STATISTIC(NumScavengedRegs, "Number of frame index regs scavenged");
+STATISTIC(NumBytesStackSpace,
+ "Number of bytes used for stack in all functions");
-INITIALIZE_PASS(PEI, "prologepilog",
- "Prologue/Epilogue Insertion", false, false);
+void PEI::getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.setPreservesCFG();
+ AU.addPreserved<MachineLoopInfo>();
+ AU.addPreserved<MachineDominatorTree>();
+ AU.addRequired<StackProtector>();
+ AU.addRequired<TargetPassConfig>();
+ MachineFunctionPass::getAnalysisUsage(AU);
+}
-/// createPrologEpilogCodeInserter - This function returns a pass that inserts
-/// prolog and epilog code, and eliminates abstract frame references.
-///
-FunctionPass *llvm::createPrologEpilogCodeInserter() { return new PEI(); }
+bool PEI::isReturnBlock(MachineBasicBlock* MBB) {
+ return (MBB && !MBB->empty() && MBB->back().isReturn());
+}
+
+/// Compute the set of return blocks
+void PEI::calculateSets(MachineFunction &Fn) {
+ // Sets used to compute spill, restore placement sets.
+ const std::vector<CalleeSavedInfo> &CSI =
+ Fn.getFrameInfo()->getCalleeSavedInfo();
+
+ // If no CSRs used, we are done.
+ if (CSI.empty())
+ return;
+
+ // Save refs to entry and return blocks.
+ EntryBlock = Fn.begin();
+ for (MachineFunction::iterator MBB = Fn.begin(), E = Fn.end();
+ MBB != E; ++MBB)
+ if (isReturnBlock(MBB))
+ ReturnBlocks.push_back(MBB);
+
+ return;
+}
+
+/// StackObjSet - A set of stack object indexes
+typedef SmallSetVector<int, 8> StackObjSet;
/// runOnMachineFunction - Insert prolog/epilog code and replace abstract
/// frame indexes with appropriate references.
///
bool PEI::runOnMachineFunction(MachineFunction &Fn) {
const Function* F = Fn.getFunction();
- const TargetRegisterInfo *TRI = Fn.getTarget().getRegisterInfo();
- RS = TRI->requiresRegisterScavenging(Fn) ? new RegScavenger() : NULL;
+ const TargetRegisterInfo *TRI = Fn.getSubtarget().getRegisterInfo();
+ const TargetFrameLowering *TFI = Fn.getSubtarget().getFrameLowering();
+
+ assert(!Fn.getRegInfo().getNumVirtRegs() && "Regalloc must assign all vregs");
+
+ RS = TRI->requiresRegisterScavenging(Fn) ? new RegScavenger() : nullptr;
FrameIndexVirtualScavenging = TRI->requiresFrameIndexScavenging(Fn);
- FrameConstantRegMap.clear();
// Calculate the MaxCallFrameSize and AdjustsStack variables for the
// function's frame information. Also eliminates call frame pseudo
// Allow the target machine to make some adjustments to the function
// e.g. UsedPhysRegs before calculateCalleeSavedRegisters.
- TRI->processFunctionBeforeCalleeSavedScan(Fn, RS);
+ TFI->processFunctionBeforeCalleeSavedScan(Fn, RS);
// Scan the function for modified callee saved registers and insert spill code
// for any callee saved registers that are modified.
calculateCalleeSavedRegisters(Fn);
// Determine placement of CSR spill/restore code:
- // - With shrink wrapping, place spills and restores to tightly
- // enclose regions in the Machine CFG of the function where
- // they are used.
- // - Without shink wrapping (default), place all spills in the
- // entry block, all restores in return blocks.
- placeCSRSpillsAndRestores(Fn);
+ // place all spills in the entry block, all restores in return blocks.
+ calculateSets(Fn);
// Add the code to save and restore the callee saved registers
- if (!F->hasFnAttr(Attribute::Naked))
+ if (!F->hasFnAttribute(Attribute::Naked))
insertCSRSpillsAndRestores(Fn);
// Allow the target machine to make final modifications to the function
// before the frame layout is finalized.
- TRI->processFunctionBeforeFrameFinalized(Fn);
+ TFI->processFunctionBeforeFrameFinalized(Fn, RS);
// Calculate actual frame offsets for all abstract stack objects...
calculateFrameObjectOffsets(Fn);
// called functions. Because of this, calculateCalleeSavedRegisters()
// must be called before this function in order to set the AdjustsStack
// and MaxCallFrameSize variables.
- if (!F->hasFnAttr(Attribute::Naked))
+ if (!F->hasFnAttribute(Attribute::Naked))
insertPrologEpilogCode(Fn);
// Replace all MO_FrameIndex operands with physical register references
replaceFrameIndices(Fn);
// If register scavenging is needed, as we've enabled doing it as a
- // post-pass, scavenge the virtual registers that frame index elimiation
+ // post-pass, scavenge the virtual registers that frame index elimination
// inserted.
if (TRI->requiresRegisterScavenging(Fn) && FrameIndexVirtualScavenging)
scavengeFrameVirtualRegs(Fn);
- delete RS;
- clearAllSets();
- return true;
-}
+ // Clear any vregs created by virtual scavenging.
+ Fn.getRegInfo().clearVirtRegs();
-#if 0
-void PEI::getAnalysisUsage(AnalysisUsage &AU) const {
- AU.setPreservesCFG();
- if (ShrinkWrapping || ShrinkWrapFunc != "") {
- AU.addRequired<MachineLoopInfo>();
- AU.addRequired<MachineDominatorTree>();
+ // Warn on stack size when we exceeds the given limit.
+ MachineFrameInfo *MFI = Fn.getFrameInfo();
+ uint64_t StackSize = MFI->getStackSize();
+ if (WarnStackSize.getNumOccurrences() > 0 && WarnStackSize < StackSize) {
+ DiagnosticInfoStackSize DiagStackSize(*F, StackSize);
+ F->getContext().diagnose(DiagStackSize);
}
- AU.addPreserved<MachineLoopInfo>();
- AU.addPreserved<MachineDominatorTree>();
- MachineFunctionPass::getAnalysisUsage(AU);
+
+ delete RS;
+ ReturnBlocks.clear();
+ return true;
}
-#endif
/// calculateCallsInformation - Calculate the MaxCallFrameSize and AdjustsStack
/// variables for the function's frame information and eliminate call frame
/// pseudo instructions.
void PEI::calculateCallsInformation(MachineFunction &Fn) {
- const TargetRegisterInfo *RegInfo = Fn.getTarget().getRegisterInfo();
+ const TargetInstrInfo &TII = *Fn.getSubtarget().getInstrInfo();
+ const TargetFrameLowering *TFI = Fn.getSubtarget().getFrameLowering();
MachineFrameInfo *MFI = Fn.getFrameInfo();
unsigned MaxCallFrameSize = 0;
bool AdjustsStack = MFI->adjustsStack();
// Get the function call frame set-up and tear-down instruction opcode
- int FrameSetupOpcode = RegInfo->getCallFrameSetupOpcode();
- int FrameDestroyOpcode = RegInfo->getCallFrameDestroyOpcode();
+ int FrameSetupOpcode = TII.getCallFrameSetupOpcode();
+ int FrameDestroyOpcode = TII.getCallFrameDestroyOpcode();
// Early exit for targets which have no call frame setup/destroy pseudo
// instructions.
FrameSDOps.push_back(I);
} else if (I->isInlineAsm()) {
// Some inline asm's need a stack frame, as indicated by operand 1.
- if (I->getOperand(1).getImm())
+ unsigned ExtraInfo = I->getOperand(InlineAsm::MIOp_ExtraInfo).getImm();
+ if (ExtraInfo & InlineAsm::Extra_IsAlignStack)
AdjustsStack = true;
}
// the target doesn't indicate otherwise, remove the call frame pseudos
// here. The sub/add sp instruction pairs are still inserted, but we don't
// need to track the SP adjustment for frame index elimination.
- if (RegInfo->canSimplifyCallFramePseudos(Fn))
- RegInfo->eliminateCallFramePseudoInstr(Fn, *I->getParent(), I);
+ if (TFI->canSimplifyCallFramePseudos(Fn))
+ TFI->eliminateCallFramePseudoInstr(Fn, *I->getParent(), I);
}
}
/// calculateCalleeSavedRegisters - Scan the function for modified callee saved
/// registers.
-void PEI::calculateCalleeSavedRegisters(MachineFunction &Fn) {
- const TargetRegisterInfo *RegInfo = Fn.getTarget().getRegisterInfo();
- const TargetFrameInfo *TFI = Fn.getTarget().getFrameInfo();
- MachineFrameInfo *MFI = Fn.getFrameInfo();
+void PEI::calculateCalleeSavedRegisters(MachineFunction &F) {
+ const TargetRegisterInfo *RegInfo = F.getSubtarget().getRegisterInfo();
+ const TargetFrameLowering *TFI = F.getSubtarget().getFrameLowering();
+ MachineFrameInfo *MFI = F.getFrameInfo();
// Get the callee saved register list...
- const unsigned *CSRegs = RegInfo->getCalleeSavedRegs(&Fn);
+ const MCPhysReg *CSRegs = RegInfo->getCalleeSavedRegs(&F);
// These are used to keep track the callee-save area. Initialize them.
MinCSFrameIndex = INT_MAX;
MaxCSFrameIndex = 0;
// Early exit for targets which have no callee saved registers.
- if (CSRegs == 0 || CSRegs[0] == 0)
+ if (!CSRegs || CSRegs[0] == 0)
return;
// In Naked functions we aren't going to save any registers.
- if (Fn.getFunction()->hasFnAttr(Attribute::Naked))
+ if (F.getFunction()->hasFnAttribute(Attribute::Naked))
return;
std::vector<CalleeSavedInfo> CSI;
for (unsigned i = 0; CSRegs[i]; ++i) {
unsigned Reg = CSRegs[i];
- if (Fn.getRegInfo().isPhysRegUsed(Reg)) {
+ // Functions which call __builtin_unwind_init get all their registers saved.
+ if (F.getRegInfo().isPhysRegUsed(Reg) || F.getMMI().callsUnwindInit()) {
// If the reg is modified, save it!
CSI.push_back(CalleeSavedInfo(Reg));
- } else {
- for (const unsigned *AliasSet = RegInfo->getAliasSet(Reg);
- *AliasSet; ++AliasSet) { // Check alias registers too.
- if (Fn.getRegInfo().isPhysRegUsed(*AliasSet)) {
- CSI.push_back(CalleeSavedInfo(Reg));
- break;
- }
- }
}
}
- if (CSI.empty())
- return; // Early exit if no callee saved registers are modified!
+ if (!TFI->assignCalleeSavedSpillSlots(F, RegInfo, CSI)) {
+ // If target doesn't implement this, use generic code.
- unsigned NumFixedSpillSlots;
- const TargetFrameInfo::SpillSlot *FixedSpillSlots =
- TFI->getCalleeSavedSpillSlots(NumFixedSpillSlots);
+ if (CSI.empty())
+ return; // Early exit if no callee saved registers are modified!
- // Now that we know which registers need to be saved and restored, allocate
- // stack slots for them.
- for (std::vector<CalleeSavedInfo>::iterator
- I = CSI.begin(), E = CSI.end(); I != E; ++I) {
- unsigned Reg = I->getReg();
- const TargetRegisterClass *RC = RegInfo->getMinimalPhysRegClass(Reg);
+ unsigned NumFixedSpillSlots;
+ const TargetFrameLowering::SpillSlot *FixedSpillSlots =
+ TFI->getCalleeSavedSpillSlots(NumFixedSpillSlots);
- int FrameIdx;
- if (RegInfo->hasReservedSpillSlot(Fn, Reg, FrameIdx)) {
- I->setFrameIdx(FrameIdx);
- continue;
- }
+ // Now that we know which registers need to be saved and restored, allocate
+ // stack slots for them.
+ for (std::vector<CalleeSavedInfo>::iterator I = CSI.begin(), E = CSI.end();
+ I != E; ++I) {
+ unsigned Reg = I->getReg();
+ const TargetRegisterClass *RC = RegInfo->getMinimalPhysRegClass(Reg);
- // Check to see if this physreg must be spilled to a particular stack slot
- // on this target.
- const TargetFrameInfo::SpillSlot *FixedSlot = FixedSpillSlots;
- while (FixedSlot != FixedSpillSlots+NumFixedSpillSlots &&
- FixedSlot->Reg != Reg)
- ++FixedSlot;
-
- if (FixedSlot == FixedSpillSlots + NumFixedSpillSlots) {
- // Nope, just spill it anywhere convenient.
- unsigned Align = RC->getAlignment();
- unsigned StackAlign = TFI->getStackAlignment();
-
- // We may not be able to satisfy the desired alignment specification of
- // the TargetRegisterClass if the stack alignment is smaller. Use the
- // min.
- Align = std::min(Align, StackAlign);
- FrameIdx = MFI->CreateStackObject(RC->getSize(), Align, true);
- if ((unsigned)FrameIdx < MinCSFrameIndex) MinCSFrameIndex = FrameIdx;
- if ((unsigned)FrameIdx > MaxCSFrameIndex) MaxCSFrameIndex = FrameIdx;
- } else {
- // Spill it to the stack where we must.
- FrameIdx = MFI->CreateFixedObject(RC->getSize(), FixedSlot->Offset, true);
- }
+ int FrameIdx;
+ if (RegInfo->hasReservedSpillSlot(F, Reg, FrameIdx)) {
+ I->setFrameIdx(FrameIdx);
+ continue;
+ }
+
+ // Check to see if this physreg must be spilled to a particular stack slot
+ // on this target.
+ const TargetFrameLowering::SpillSlot *FixedSlot = FixedSpillSlots;
+ while (FixedSlot != FixedSpillSlots + NumFixedSpillSlots &&
+ FixedSlot->Reg != Reg)
+ ++FixedSlot;
+
+ if (FixedSlot == FixedSpillSlots + NumFixedSpillSlots) {
+ // Nope, just spill it anywhere convenient.
+ unsigned Align = RC->getAlignment();
+ unsigned StackAlign = TFI->getStackAlignment();
+
+ // We may not be able to satisfy the desired alignment specification of
+ // the TargetRegisterClass if the stack alignment is smaller. Use the
+ // min.
+ Align = std::min(Align, StackAlign);
+ FrameIdx = MFI->CreateStackObject(RC->getSize(), Align, true);
+ if ((unsigned)FrameIdx < MinCSFrameIndex) MinCSFrameIndex = FrameIdx;
+ if ((unsigned)FrameIdx > MaxCSFrameIndex) MaxCSFrameIndex = FrameIdx;
+ } else {
+ // Spill it to the stack where we must.
+ FrameIdx =
+ MFI->CreateFixedSpillStackObject(RC->getSize(), FixedSlot->Offset);
+ }
- I->setFrameIdx(FrameIdx);
+ I->setFrameIdx(FrameIdx);
+ }
}
MFI->setCalleeSavedInfo(CSI);
}
/// insertCSRSpillsAndRestores - Insert spill and restore code for
-/// callee saved registers used in the function, handling shrink wrapping.
+/// callee saved registers used in the function.
///
void PEI::insertCSRSpillsAndRestores(MachineFunction &Fn) {
// Get callee saved register information.
if (CSI.empty())
return;
- const TargetInstrInfo &TII = *Fn.getTarget().getInstrInfo();
- const TargetRegisterInfo *TRI = Fn.getTarget().getRegisterInfo();
+ const TargetInstrInfo &TII = *Fn.getSubtarget().getInstrInfo();
+ const TargetFrameLowering *TFI = Fn.getSubtarget().getFrameLowering();
+ const TargetRegisterInfo *TRI = Fn.getSubtarget().getRegisterInfo();
MachineBasicBlock::iterator I;
- if (! ShrinkWrapThisFunction) {
- // Spill using target interface.
- I = EntryBlock->begin();
- if (!TII.spillCalleeSavedRegisters(*EntryBlock, I, CSI, TRI)) {
- for (unsigned i = 0, e = CSI.size(); i != e; ++i) {
- // Add the callee-saved register as live-in.
- // It's killed at the spill.
- EntryBlock->addLiveIn(CSI[i].getReg());
-
- // Insert the spill to the stack frame.
- unsigned Reg = CSI[i].getReg();
- const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg);
- TII.storeRegToStackSlot(*EntryBlock, I, Reg, true,
- CSI[i].getFrameIdx(), RC, TRI);
- }
- }
-
- // Restore using target interface.
- for (unsigned ri = 0, re = ReturnBlocks.size(); ri != re; ++ri) {
- MachineBasicBlock* MBB = ReturnBlocks[ri];
- I = MBB->end(); --I;
-
- // Skip over all terminator instructions, which are part of the return
- // sequence.
- MachineBasicBlock::iterator I2 = I;
- while (I2 != MBB->begin() && (--I2)->getDesc().isTerminator())
- I = I2;
-
- bool AtStart = I == MBB->begin();
- MachineBasicBlock::iterator BeforeI = I;
- if (!AtStart)
- --BeforeI;
-
- // Restore all registers immediately before the return and any
- // terminators that preceed it.
- if (!TII.restoreCalleeSavedRegisters(*MBB, I, CSI, TRI)) {
- for (unsigned i = 0, e = CSI.size(); i != e; ++i) {
- unsigned Reg = CSI[i].getReg();
- const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg);
- TII.loadRegFromStackSlot(*MBB, I, Reg,
- CSI[i].getFrameIdx(),
- RC, TRI);
- assert(I != MBB->begin() &&
- "loadRegFromStackSlot didn't insert any code!");
- // Insert in reverse order. loadRegFromStackSlot can insert
- // multiple instructions.
- if (AtStart)
- I = MBB->begin();
- else {
- I = BeforeI;
- ++I;
- }
- }
- }
- }
- return;
- }
-
- // Insert spills.
- std::vector<CalleeSavedInfo> blockCSI;
- for (CSRegBlockMap::iterator BI = CSRSave.begin(),
- BE = CSRSave.end(); BI != BE; ++BI) {
- MachineBasicBlock* MBB = BI->first;
- CSRegSet save = BI->second;
-
- if (save.empty())
- continue;
-
- blockCSI.clear();
- for (CSRegSet::iterator RI = save.begin(),
- RE = save.end(); RI != RE; ++RI) {
- blockCSI.push_back(CSI[*RI]);
- }
- assert(blockCSI.size() > 0 &&
- "Could not collect callee saved register info");
-
- I = MBB->begin();
-
- // When shrink wrapping, use stack slot stores/loads.
- for (unsigned i = 0, e = blockCSI.size(); i != e; ++i) {
+ // Spill using target interface.
+ I = EntryBlock->begin();
+ if (!TFI->spillCalleeSavedRegisters(*EntryBlock, I, CSI, TRI)) {
+ for (unsigned i = 0, e = CSI.size(); i != e; ++i) {
// Add the callee-saved register as live-in.
// It's killed at the spill.
- MBB->addLiveIn(blockCSI[i].getReg());
+ EntryBlock->addLiveIn(CSI[i].getReg());
// Insert the spill to the stack frame.
- unsigned Reg = blockCSI[i].getReg();
+ unsigned Reg = CSI[i].getReg();
const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg);
- TII.storeRegToStackSlot(*MBB, I, Reg,
- true,
- blockCSI[i].getFrameIdx(),
+ TII.storeRegToStackSlot(*EntryBlock, I, Reg, true, CSI[i].getFrameIdx(),
RC, TRI);
}
}
- for (CSRegBlockMap::iterator BI = CSRRestore.begin(),
- BE = CSRRestore.end(); BI != BE; ++BI) {
- MachineBasicBlock* MBB = BI->first;
- CSRegSet restore = BI->second;
+ // Restore using target interface.
+ for (unsigned ri = 0, re = ReturnBlocks.size(); ri != re; ++ri) {
+ MachineBasicBlock *MBB = ReturnBlocks[ri];
+ I = MBB->end();
+ --I;
- if (restore.empty())
- continue;
-
- blockCSI.clear();
- for (CSRegSet::iterator RI = restore.begin(),
- RE = restore.end(); RI != RE; ++RI) {
- blockCSI.push_back(CSI[*RI]);
- }
- assert(blockCSI.size() > 0 &&
- "Could not find callee saved register info");
-
- // If MBB is empty and needs restores, insert at the _beginning_.
- if (MBB->empty()) {
- I = MBB->begin();
- } else {
- I = MBB->end();
- --I;
-
- // Skip over all terminator instructions, which are part of the
- // return sequence.
- if (! I->getDesc().isTerminator()) {
- ++I;
- } else {
- MachineBasicBlock::iterator I2 = I;
- while (I2 != MBB->begin() && (--I2)->getDesc().isTerminator())
- I = I2;
- }
- }
+ // Skip over all terminator instructions, which are part of the return
+ // sequence.
+ MachineBasicBlock::iterator I2 = I;
+ while (I2 != MBB->begin() && (--I2)->isTerminator())
+ I = I2;
bool AtStart = I == MBB->begin();
MachineBasicBlock::iterator BeforeI = I;
--BeforeI;
// Restore all registers immediately before the return and any
- // terminators that preceed it.
- for (unsigned i = 0, e = blockCSI.size(); i != e; ++i) {
- unsigned Reg = blockCSI[i].getReg();
- const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg);
- TII.loadRegFromStackSlot(*MBB, I, Reg,
- blockCSI[i].getFrameIdx(),
- RC, TRI);
- assert(I != MBB->begin() &&
- "loadRegFromStackSlot didn't insert any code!");
- // Insert in reverse order. loadRegFromStackSlot can insert
- // multiple instructions.
- if (AtStart)
- I = MBB->begin();
- else {
- I = BeforeI;
- ++I;
+ // terminators that precede it.
+ if (!TFI->restoreCalleeSavedRegisters(*MBB, I, CSI, TRI)) {
+ for (unsigned i = 0, e = CSI.size(); i != e; ++i) {
+ unsigned Reg = CSI[i].getReg();
+ const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg);
+ TII.loadRegFromStackSlot(*MBB, I, Reg, CSI[i].getFrameIdx(), RC, TRI);
+ assert(I != MBB->begin() &&
+ "loadRegFromStackSlot didn't insert any code!");
+ // Insert in reverse order. loadRegFromStackSlot can insert
+ // multiple instructions.
+ if (AtStart)
+ I = MBB->begin();
+ else {
+ I = BeforeI;
+ ++I;
+ }
}
}
}
}
}
+/// AssignProtectedObjSet - Helper function to assign large stack objects (i.e.,
+/// those required to be close to the Stack Protector) to stack offsets.
+static void
+AssignProtectedObjSet(const StackObjSet &UnassignedObjs,
+ SmallSet<int, 16> &ProtectedObjs,
+ MachineFrameInfo *MFI, bool StackGrowsDown,
+ int64_t &Offset, unsigned &MaxAlign) {
+
+ for (StackObjSet::const_iterator I = UnassignedObjs.begin(),
+ E = UnassignedObjs.end(); I != E; ++I) {
+ int i = *I;
+ AdjustStackOffset(MFI, i, StackGrowsDown, Offset, MaxAlign);
+ ProtectedObjs.insert(i);
+ }
+}
+
/// calculateFrameObjectOffsets - Calculate actual frame offsets for all of the
/// abstract stack objects.
///
void PEI::calculateFrameObjectOffsets(MachineFunction &Fn) {
- const TargetFrameInfo &TFI = *Fn.getTarget().getFrameInfo();
+ const TargetFrameLowering &TFI = *Fn.getSubtarget().getFrameLowering();
+ StackProtector *SP = &getAnalysis<StackProtector>();
bool StackGrowsDown =
- TFI.getStackGrowthDirection() == TargetFrameInfo::StackGrowsDown;
+ TFI.getStackGrowthDirection() == TargetFrameLowering::StackGrowsDown;
// Loop over all of the stack objects, assigning sequential addresses...
MachineFrameInfo *MFI = Fn.getFrameInfo();
unsigned Align = MFI->getObjectAlignment(i);
// Adjust to alignment boundary
- Offset = (Offset+Align-1)/Align*Align;
+ Offset = RoundUpToAlignment(Offset, Align);
MFI->setObjectOffset(i, -Offset); // Set the computed offset
}
for (int i = MaxCSFI; i >= MinCSFI ; --i) {
unsigned Align = MFI->getObjectAlignment(i);
// Adjust to alignment boundary
- Offset = (Offset+Align-1)/Align*Align;
+ Offset = RoundUpToAlignment(Offset, Align);
MFI->setObjectOffset(i, Offset);
Offset += MFI->getObjectSize(i);
unsigned MaxAlign = MFI->getMaxAlignment();
// Make sure the special register scavenging spill slot is closest to the
- // frame pointer if a frame pointer is required.
- const TargetRegisterInfo *RegInfo = Fn.getTarget().getRegisterInfo();
- if (RS && RegInfo->hasFP(Fn) && !RegInfo->needsStackRealignment(Fn)) {
- int SFI = RS->getScavengingFrameIndex();
- if (SFI >= 0)
- AdjustStackOffset(MFI, SFI, StackGrowsDown, Offset, MaxAlign);
+ // incoming stack pointer if a frame pointer is required and is closer
+ // to the incoming rather than the final stack pointer.
+ const TargetRegisterInfo *RegInfo = Fn.getSubtarget().getRegisterInfo();
+ bool EarlyScavengingSlots = (TFI.hasFP(Fn) &&
+ TFI.isFPCloseToIncomingSP() &&
+ RegInfo->useFPForScavengingIndex(Fn) &&
+ !RegInfo->needsStackRealignment(Fn));
+ if (RS && EarlyScavengingSlots) {
+ SmallVector<int, 2> SFIs;
+ RS->getScavengingFrameIndices(SFIs);
+ for (SmallVectorImpl<int>::iterator I = SFIs.begin(),
+ IE = SFIs.end(); I != IE; ++I)
+ AdjustStackOffset(MFI, *I, StackGrowsDown, Offset, MaxAlign);
}
- // Store the offset of the start of the local allocation block. This
- // will be used later when resolving frame base virtual register pseudos.
- MFI->setLocalFrameBaseOffset(Offset);
- if (EnableLocalStackAlloc) {
- // Allocate the local block
- Offset += MFI->getLocalFrameSize();
+ // FIXME: Once this is working, then enable flag will change to a target
+ // check for whether the frame is large enough to want to use virtual
+ // frame index registers. Functions which don't want/need this optimization
+ // will continue to use the existing code path.
+ if (MFI->getUseLocalStackAllocationBlock()) {
+ unsigned Align = MFI->getLocalFrameMaxAlign();
+
+ // Adjust to alignment boundary.
+ Offset = RoundUpToAlignment(Offset, Align);
+
+ DEBUG(dbgs() << "Local frame base offset: " << Offset << "\n");
// Resolve offsets for objects in the local block.
for (unsigned i = 0, e = MFI->getLocalFrameObjectCount(); i != e; ++i) {
std::pair<int, int64_t> Entry = MFI->getLocalFrameObjectMap(i);
- int64_t FIOffset = MFI->getLocalFrameBaseOffset() + Entry.second;
-
- AdjustStackOffset(MFI, Entry.first, StackGrowsDown, FIOffset, MaxAlign);
+ int64_t FIOffset = (StackGrowsDown ? -Offset : Offset) + Entry.second;
+ DEBUG(dbgs() << "alloc FI(" << Entry.first << ") at SP[" <<
+ FIOffset << "]\n");
+ MFI->setObjectOffset(Entry.first, FIOffset);
}
+ // Allocate the local block
+ Offset += MFI->getLocalFrameSize();
+
+ MaxAlign = std::max(Align, MaxAlign);
}
- // FIXME: Allocate locals. Once the block allocation pass is turned on,
- // this simplifies to just the second loop, since all of the large objects
- // will have already been handled. The second loop can also simplify a
- // bit, as the conditionals inside aren't all necessary.
// Make sure that the stack protector comes before the local variables on the
// stack.
- SmallSet<int, 16> LargeStackObjs;
+ SmallSet<int, 16> ProtectedObjs;
if (MFI->getStackProtectorIndex() >= 0) {
+ StackObjSet LargeArrayObjs;
+ StackObjSet SmallArrayObjs;
+ StackObjSet AddrOfObjs;
+
AdjustStackOffset(MFI, MFI->getStackProtectorIndex(), StackGrowsDown,
Offset, MaxAlign);
// Assign large stack objects first.
for (unsigned i = 0, e = MFI->getObjectIndexEnd(); i != e; ++i) {
- if (MFI->isObjectPreAllocated(i))
+ if (MFI->isObjectPreAllocated(i) &&
+ MFI->getUseLocalStackAllocationBlock())
continue;
if (i >= MinCSFrameIndex && i <= MaxCSFrameIndex)
continue;
- if (RS && (int)i == RS->getScavengingFrameIndex())
+ if (RS && RS->isScavengingFrameIndex((int)i))
continue;
if (MFI->isDeadObjectIndex(i))
continue;
if (MFI->getStackProtectorIndex() == (int)i)
continue;
- if (!MFI->MayNeedStackProtector(i))
- continue;
- AdjustStackOffset(MFI, i, StackGrowsDown, Offset, MaxAlign);
- LargeStackObjs.insert(i);
+ switch (SP->getSSPLayout(MFI->getObjectAllocation(i))) {
+ case StackProtector::SSPLK_None:
+ continue;
+ case StackProtector::SSPLK_SmallArray:
+ SmallArrayObjs.insert(i);
+ continue;
+ case StackProtector::SSPLK_AddrOf:
+ AddrOfObjs.insert(i);
+ continue;
+ case StackProtector::SSPLK_LargeArray:
+ LargeArrayObjs.insert(i);
+ continue;
+ }
+ llvm_unreachable("Unexpected SSPLayoutKind.");
}
+
+ AssignProtectedObjSet(LargeArrayObjs, ProtectedObjs, MFI, StackGrowsDown,
+ Offset, MaxAlign);
+ AssignProtectedObjSet(SmallArrayObjs, ProtectedObjs, MFI, StackGrowsDown,
+ Offset, MaxAlign);
+ AssignProtectedObjSet(AddrOfObjs, ProtectedObjs, MFI, StackGrowsDown,
+ Offset, MaxAlign);
}
// Then assign frame offsets to stack objects that are not used to spill
// callee saved registers.
for (unsigned i = 0, e = MFI->getObjectIndexEnd(); i != e; ++i) {
- if (MFI->isObjectPreAllocated(i))
+ if (MFI->isObjectPreAllocated(i) &&
+ MFI->getUseLocalStackAllocationBlock())
continue;
if (i >= MinCSFrameIndex && i <= MaxCSFrameIndex)
continue;
- if (RS && (int)i == RS->getScavengingFrameIndex())
+ if (RS && RS->isScavengingFrameIndex((int)i))
continue;
if (MFI->isDeadObjectIndex(i))
continue;
if (MFI->getStackProtectorIndex() == (int)i)
continue;
- if (LargeStackObjs.count(i))
+ if (ProtectedObjs.count(i))
continue;
AdjustStackOffset(MFI, i, StackGrowsDown, Offset, MaxAlign);
// Make sure the special register scavenging spill slot is closest to the
// stack pointer.
- if (RS && (!RegInfo->hasFP(Fn) || RegInfo->needsStackRealignment(Fn))) {
- int SFI = RS->getScavengingFrameIndex();
- if (SFI >= 0)
- AdjustStackOffset(MFI, SFI, StackGrowsDown, Offset, MaxAlign);
+ if (RS && !EarlyScavengingSlots) {
+ SmallVector<int, 2> SFIs;
+ RS->getScavengingFrameIndices(SFIs);
+ for (SmallVectorImpl<int>::iterator I = SFIs.begin(),
+ IE = SFIs.end(); I != IE; ++I)
+ AdjustStackOffset(MFI, *I, StackGrowsDown, Offset, MaxAlign);
}
- if (!RegInfo->targetHandlesStackFrameRounding()) {
+ if (!TFI.targetHandlesStackFrameRounding()) {
// If we have reserved argument space for call sites in the function
// immediately on entry to the current function, count it as part of the
// overall stack size.
- if (MFI->adjustsStack() && RegInfo->hasReservedCallFrame(Fn))
+ if (MFI->adjustsStack() && TFI.hasReservedCallFrame(Fn))
Offset += MFI->getMaxCallFrameSize();
// Round up the size to a multiple of the alignment. If the function has
// If the frame pointer is eliminated, all frame offsets will be relative to
// SP not FP. Align to MaxAlign so this works.
StackAlign = std::max(StackAlign, MaxAlign);
- unsigned AlignMask = StackAlign - 1;
- Offset = (Offset + AlignMask) & ~uint64_t(AlignMask);
+ Offset = RoundUpToAlignment(Offset, StackAlign);
}
// Update frame info to pretend that this is part of the stack...
- MFI->setStackSize(Offset - LocalAreaOffset);
+ int64_t StackSize = Offset - LocalAreaOffset;
+ MFI->setStackSize(StackSize);
+ NumBytesStackSpace += StackSize;
}
/// insertPrologEpilogCode - Scan the function for modified callee saved
/// prolog and epilog code to the function.
///
void PEI::insertPrologEpilogCode(MachineFunction &Fn) {
- const TargetRegisterInfo *TRI = Fn.getTarget().getRegisterInfo();
+ const TargetFrameLowering &TFI = *Fn.getSubtarget().getFrameLowering();
// Add prologue to the function...
- TRI->emitPrologue(Fn);
+ TFI.emitPrologue(Fn);
// Add epilogue to restore the callee-save registers in each exiting block
for (MachineFunction::iterator I = Fn.begin(), E = Fn.end(); I != E; ++I) {
// If last instruction is a return instruction, add an epilogue
- if (!I->empty() && I->back().getDesc().isReturn())
- TRI->emitEpilogue(Fn, *I);
+ if (!I->empty() && I->back().isReturn())
+ TFI.emitEpilogue(Fn, *I);
}
+
+ // Emit additional code that is required to support segmented stacks, if
+ // we've been asked for it. This, when linked with a runtime with support
+ // for segmented stacks (libgcc is one), will result in allocating stack
+ // space in small chunks instead of one large contiguous block.
+ if (Fn.shouldSplitStack())
+ TFI.adjustForSegmentedStacks(Fn);
+
+ // Emit additional code that is required to explicitly handle the stack in
+ // HiPE native code (if needed) when loaded in the Erlang/OTP runtime. The
+ // approach is rather similar to that of Segmented Stacks, but it uses a
+ // different conditional check and another BIF for allocating more stack
+ // space.
+ if (Fn.getFunction()->getCallingConv() == CallingConv::HiPE)
+ TFI.adjustForHiPEPrologue(Fn);
}
/// replaceFrameIndices - Replace all MO_FrameIndex operands with physical
/// register references and actual offsets.
///
void PEI::replaceFrameIndices(MachineFunction &Fn) {
- if (!Fn.getFrameInfo()->hasStackObjects()) return; // Nothing to do?
+ const TargetFrameLowering &TFI = *Fn.getSubtarget().getFrameLowering();
+ if (!TFI.needsFrameIndexResolution(Fn)) return;
- const TargetMachine &TM = Fn.getTarget();
- assert(TM.getRegisterInfo() && "TM::getRegisterInfo() must be implemented!");
- const TargetRegisterInfo &TRI = *TM.getRegisterInfo();
- const TargetFrameInfo *TFI = TM.getFrameInfo();
- bool StackGrowsDown =
- TFI->getStackGrowthDirection() == TargetFrameInfo::StackGrowsDown;
- int FrameSetupOpcode = TRI.getCallFrameSetupOpcode();
- int FrameDestroyOpcode = TRI.getCallFrameDestroyOpcode();
+ // Store SPAdj at exit of a basic block.
+ SmallVector<int, 8> SPState;
+ SPState.resize(Fn.getNumBlockIDs());
+ SmallPtrSet<MachineBasicBlock*, 8> Reachable;
- for (MachineFunction::iterator BB = Fn.begin(),
- E = Fn.end(); BB != E; ++BB) {
-#ifndef NDEBUG
- int SPAdjCount = 0; // frame setup / destroy count.
-#endif
- int SPAdj = 0; // SP offset due to call frame setup / destroy.
- if (RS && !FrameIndexVirtualScavenging) RS->enterBasicBlock(BB);
+ // Iterate over the reachable blocks in DFS order.
+ for (auto DFI = df_ext_begin(&Fn, Reachable), DFE = df_ext_end(&Fn, Reachable);
+ DFI != DFE; ++DFI) {
+ int SPAdj = 0;
+ // Check the exit state of the DFS stack predecessor.
+ if (DFI.getPathLength() >= 2) {
+ MachineBasicBlock *StackPred = DFI.getPath(DFI.getPathLength() - 2);
+ assert(Reachable.count(StackPred) &&
+ "DFS stack predecessor is already visited.\n");
+ SPAdj = SPState[StackPred->getNumber()];
+ }
+ MachineBasicBlock *BB = *DFI;
+ replaceFrameIndices(BB, Fn, SPAdj);
+ SPState[BB->getNumber()] = SPAdj;
+ }
- for (MachineBasicBlock::iterator I = BB->begin(); I != BB->end(); ) {
+ // Handle the unreachable blocks.
+ for (MachineFunction::iterator BB = Fn.begin(), E = Fn.end(); BB != E; ++BB) {
+ if (Reachable.count(BB))
+ // Already handled in DFS traversal.
+ continue;
+ int SPAdj = 0;
+ replaceFrameIndices(BB, Fn, SPAdj);
+ }
+}
- if (I->getOpcode() == FrameSetupOpcode ||
- I->getOpcode() == FrameDestroyOpcode) {
-#ifndef NDEBUG
- // Track whether we see even pairs of them
- SPAdjCount += I->getOpcode() == FrameSetupOpcode ? 1 : -1;
-#endif
- // Remember how much SP has been adjusted to create the call
- // frame.
- int Size = I->getOperand(0).getImm();
-
- if ((!StackGrowsDown && I->getOpcode() == FrameSetupOpcode) ||
- (StackGrowsDown && I->getOpcode() == FrameDestroyOpcode))
- Size = -Size;
-
- SPAdj += Size;
-
- MachineBasicBlock::iterator PrevI = BB->end();
- if (I != BB->begin()) PrevI = prior(I);
- TRI.eliminateCallFramePseudoInstr(Fn, *BB, I);
-
- // Visit the instructions created by eliminateCallFramePseudoInstr().
- if (PrevI == BB->end())
- I = BB->begin(); // The replaced instr was the first in the block.
- else
- I = llvm::next(PrevI);
+void PEI::replaceFrameIndices(MachineBasicBlock *BB, MachineFunction &Fn,
+ int &SPAdj) {
+ assert(Fn.getSubtarget().getRegisterInfo() &&
+ "getRegisterInfo() must be implemented!");
+ const TargetInstrInfo &TII = *Fn.getSubtarget().getInstrInfo();
+ const TargetRegisterInfo &TRI = *Fn.getSubtarget().getRegisterInfo();
+ const TargetFrameLowering *TFI = Fn.getSubtarget().getFrameLowering();
+ int FrameSetupOpcode = TII.getCallFrameSetupOpcode();
+ int FrameDestroyOpcode = TII.getCallFrameDestroyOpcode();
+
+ if (RS && !FrameIndexVirtualScavenging) RS->enterBasicBlock(BB);
+
+ bool InsideCallSequence = false;
+
+ for (MachineBasicBlock::iterator I = BB->begin(); I != BB->end(); ) {
+
+ if (I->getOpcode() == FrameSetupOpcode ||
+ I->getOpcode() == FrameDestroyOpcode) {
+ InsideCallSequence = (I->getOpcode() == FrameSetupOpcode);
+ SPAdj += TII.getSPAdjust(I);
+
+ MachineBasicBlock::iterator PrevI = BB->end();
+ if (I != BB->begin()) PrevI = std::prev(I);
+ TFI->eliminateCallFramePseudoInstr(Fn, *BB, I);
+
+ // Visit the instructions created by eliminateCallFramePseudoInstr().
+ if (PrevI == BB->end())
+ I = BB->begin(); // The replaced instr was the first in the block.
+ else
+ I = std::next(PrevI);
+ continue;
+ }
+
+ MachineInstr *MI = I;
+ bool DoIncr = true;
+ for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
+ if (!MI->getOperand(i).isFI())
+ continue;
+
+ // Frame indicies in debug values are encoded in a target independent
+ // way with simply the frame index and offset rather than any
+ // target-specific addressing mode.
+ if (MI->isDebugValue()) {
+ assert(i == 0 && "Frame indicies can only appear as the first "
+ "operand of a DBG_VALUE machine instruction");
+ unsigned Reg;
+ MachineOperand &Offset = MI->getOperand(1);
+ Offset.setImm(Offset.getImm() +
+ TFI->getFrameIndexReference(
+ Fn, MI->getOperand(0).getIndex(), Reg));
+ MI->getOperand(0).ChangeToRegister(Reg, false /*isDef*/);
continue;
}
- MachineInstr *MI = I;
- bool DoIncr = true;
- for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i)
- if (MI->getOperand(i).isFI()) {
- // Some instructions (e.g. inline asm instructions) can have
- // multiple frame indices and/or cause eliminateFrameIndex
- // to insert more than one instruction. We need the register
- // scavenger to go through all of these instructions so that
- // it can update its register information. We keep the
- // iterator at the point before insertion so that we can
- // revisit them in full.
- bool AtBeginning = (I == BB->begin());
- if (!AtBeginning) --I;
-
- // If this instruction has a FrameIndex operand, we need to
- // use that target machine register info object to eliminate
- // it.
- TargetRegisterInfo::FrameIndexValue Value;
- unsigned VReg =
- TRI.eliminateFrameIndex(MI, SPAdj, &Value,
- FrameIndexVirtualScavenging ? NULL : RS);
- if (VReg) {
- assert (FrameIndexVirtualScavenging &&
- "Not scavenging, but virtual returned from "
- "eliminateFrameIndex()!");
- FrameConstantRegMap[VReg] = FrameConstantEntry(Value, SPAdj);
- }
-
- // Reset the iterator if we were at the beginning of the BB.
- if (AtBeginning) {
- I = BB->begin();
- DoIncr = false;
- }
-
- MI = 0;
- break;
- }
+ // TODO: This code should be commoned with the code for
+ // PATCHPOINT. There's no good reason for the difference in
+ // implementation other than historical accident. The only
+ // remaining difference is the unconditional use of the stack
+ // pointer as the base register.
+ if (MI->getOpcode() == TargetOpcode::STATEPOINT) {
+ assert((!MI->isDebugValue() || i == 0) &&
+ "Frame indicies can only appear as the first operand of a "
+ "DBG_VALUE machine instruction");
+ unsigned Reg;
+ MachineOperand &Offset = MI->getOperand(i + 1);
+ const unsigned refOffset =
+ TFI->getFrameIndexReferenceFromSP(Fn, MI->getOperand(i).getIndex(),
+ Reg);
+
+ Offset.setImm(Offset.getImm() + refOffset);
+ MI->getOperand(i).ChangeToRegister(Reg, false /*isDef*/);
+ continue;
+ }
+
+ // Frame allocations are target independent. Simply swap the index with
+ // the offset.
+ if (MI->getOpcode() == TargetOpcode::FRAME_ALLOC) {
+ assert(TFI->hasFP(Fn) && "frame alloc requires FP");
+ MachineOperand &FI = MI->getOperand(i);
+ unsigned Reg;
+ int FrameOffset = TFI->getFrameIndexReference(Fn, FI.getIndex(), Reg);
+ FI.ChangeToImmediate(FrameOffset);
+ continue;
+ }
- if (DoIncr && I != BB->end()) ++I;
+ // Some instructions (e.g. inline asm instructions) can have
+ // multiple frame indices and/or cause eliminateFrameIndex
+ // to insert more than one instruction. We need the register
+ // scavenger to go through all of these instructions so that
+ // it can update its register information. We keep the
+ // iterator at the point before insertion so that we can
+ // revisit them in full.
+ bool AtBeginning = (I == BB->begin());
+ if (!AtBeginning) --I;
+
+ // If this instruction has a FrameIndex operand, we need to
+ // use that target machine register info object to eliminate
+ // it.
+ TRI.eliminateFrameIndex(MI, SPAdj, i,
+ FrameIndexVirtualScavenging ? nullptr : RS);
+
+ // Reset the iterator if we were at the beginning of the BB.
+ if (AtBeginning) {
+ I = BB->begin();
+ DoIncr = false;
+ }
- // Update register states.
- if (RS && !FrameIndexVirtualScavenging && MI) RS->forward(MI);
+ MI = nullptr;
+ break;
}
- // If we have evenly matched pairs of frame setup / destroy instructions,
- // make sure the adjustments come out to zero. If we don't have matched
- // pairs, we can't be sure the missing bit isn't in another basic block
- // due to a custom inserter playing tricks, so just asserting SPAdj==0
- // isn't sufficient. See tMOVCC on Thumb1, for example.
- assert((SPAdjCount || SPAdj == 0) &&
- "Unbalanced call frame setup / destroy pairs?");
- }
-}
+ // If we are looking at a call sequence, we need to keep track of
+ // the SP adjustment made by each instruction in the sequence.
+ // This includes both the frame setup/destroy pseudos (handled above),
+ // as well as other instructions that have side effects w.r.t the SP.
+ // Note that this must come after eliminateFrameIndex, because
+ // if I itself referred to a frame index, we shouldn't count its own
+ // adjustment.
+ if (MI && InsideCallSequence)
+ SPAdj += TII.getSPAdjust(MI);
-/// findLastUseReg - find the killing use of the specified register within
-/// the instruciton range. Return the operand number of the kill in Operand.
-static MachineBasicBlock::iterator
-findLastUseReg(MachineBasicBlock::iterator I, MachineBasicBlock::iterator ME,
- unsigned Reg) {
- // Scan forward to find the last use of this virtual register
- for (++I; I != ME; ++I) {
- MachineInstr *MI = I;
- bool isDefInsn = false;
- bool isKillInsn = false;
- for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i)
- if (MI->getOperand(i).isReg()) {
- unsigned OpReg = MI->getOperand(i).getReg();
- if (OpReg == 0 || !TargetRegisterInfo::isVirtualRegister(OpReg))
- continue;
- assert (OpReg == Reg
- && "overlapping use of scavenged index register!");
- // If this is the killing use, we have a candidate.
- if (MI->getOperand(i).isKill())
- isKillInsn = true;
- else if (MI->getOperand(i).isDef())
- isDefInsn = true;
- }
- if (isKillInsn && !isDefInsn)
- return I;
+ if (DoIncr && I != BB->end()) ++I;
+
+ // Update register states.
+ if (RS && !FrameIndexVirtualScavenging && MI) RS->forward(MI);
}
- // If we hit the end of the basic block, there was no kill of
- // the virtual register, which is wrong.
- assert (0 && "scavenged index register never killed!");
- return ME;
}
/// scavengeFrameVirtualRegs - Replace all frame index virtual registers
/// with physical registers. Use the register scavenger to find an
/// appropriate register to use.
-void PEI::scavengeFrameVirtualRegs(MachineFunction &Fn) {
+///
+/// FIXME: Iterating over the instruction stream is unnecessary. We can simply
+/// iterate over the vreg use list, which at this point only contains machine
+/// operands for which eliminateFrameIndex need a new scratch reg.
+void
+PEI::scavengeFrameVirtualRegs(MachineFunction &Fn) {
// Run through the instructions and find any virtual registers.
for (MachineFunction::iterator BB = Fn.begin(),
E = Fn.end(); BB != E; ++BB) {
RS->enterBasicBlock(BB);
- // FIXME: The logic flow in this function is still too convoluted.
- // It needs a cleanup refactoring. Do that in preparation for tracking
- // more than one scratch register value and using ranges to find
- // available scratch registers.
- unsigned CurrentVirtReg = 0;
- unsigned CurrentScratchReg = 0;
- bool havePrevValue = false;
- TargetRegisterInfo::FrameIndexValue PrevValue(0,0);
- TargetRegisterInfo::FrameIndexValue Value(0,0);
- MachineInstr *PrevLastUseMI = NULL;
- unsigned PrevLastUseOp = 0;
- bool trackingCurrentValue = false;
int SPAdj = 0;
// The instruction stream may change in the loop, so check BB->end()
// directly.
for (MachineBasicBlock::iterator I = BB->begin(); I != BB->end(); ) {
+ // We might end up here again with a NULL iterator if we scavenged a
+ // register for which we inserted spill code for definition by what was
+ // originally the first instruction in BB.
+ if (I == MachineBasicBlock::iterator(nullptr))
+ I = BB->begin();
+
MachineInstr *MI = I;
- bool isDefInsn = false;
- bool isKillInsn = false;
- bool clobbersScratchReg = false;
- bool DoIncr = true;
+ MachineBasicBlock::iterator J = std::next(I);
+ MachineBasicBlock::iterator P =
+ I == BB->begin() ? MachineBasicBlock::iterator(nullptr)
+ : std::prev(I);
+
+ // RS should process this instruction before we might scavenge at this
+ // location. This is because we might be replacing a virtual register
+ // defined by this instruction, and if so, registers killed by this
+ // instruction are available, and defined registers are not.
+ RS->forward(I);
+
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
if (MI->getOperand(i).isReg()) {
MachineOperand &MO = MI->getOperand(i);
unsigned Reg = MO.getReg();
if (Reg == 0)
continue;
- if (!TargetRegisterInfo::isVirtualRegister(Reg)) {
- // If we have a previous scratch reg, check and see if anything
- // here kills whatever value is in there.
- if (Reg == CurrentScratchReg) {
- if (MO.isUse()) {
- // Two-address operands implicitly kill
- if (MO.isKill() || MI->isRegTiedToDefOperand(i))
- clobbersScratchReg = true;
- } else {
- assert (MO.isDef());
- clobbersScratchReg = true;
- }
- }
+ if (!TargetRegisterInfo::isVirtualRegister(Reg))
continue;
- }
- // If this is a def, remember that this insn defines the value.
- // This lets us properly consider insns which re-use the scratch
- // register, such as r2 = sub r2, #imm, in the middle of the
- // scratch range.
- if (MO.isDef())
- isDefInsn = true;
-
- // Have we already allocated a scratch register for this virtual?
- if (Reg != CurrentVirtReg) {
- // When we first encounter a new virtual register, it
- // must be a definition.
- assert(MI->getOperand(i).isDef() &&
- "frame index virtual missing def!");
- // We can't have nested virtual register live ranges because
- // there's only a guarantee of one scavenged register at a time.
- assert (CurrentVirtReg == 0 &&
- "overlapping frame index virtual registers!");
-
- // If the target gave us information about what's in the register,
- // we can use that to re-use scratch regs.
- DenseMap<unsigned, FrameConstantEntry>::iterator Entry =
- FrameConstantRegMap.find(Reg);
- trackingCurrentValue = Entry != FrameConstantRegMap.end();
- if (trackingCurrentValue) {
- SPAdj = (*Entry).second.second;
- Value = (*Entry).second.first;
- } else {
- SPAdj = 0;
- Value.first = 0;
- Value.second = 0;
- }
-
- // If the scratch register from the last allocation is still
- // available, see if the value matches. If it does, just re-use it.
- if (trackingCurrentValue && havePrevValue && PrevValue == Value) {
- // FIXME: This assumes that the instructions in the live range
- // for the virtual register are exclusively for the purpose
- // of populating the value in the register. That's reasonable
- // for these frame index registers, but it's still a very, very
- // strong assumption. rdar://7322732. Better would be to
- // explicitly check each instruction in the range for references
- // to the virtual register. Only delete those insns that
- // touch the virtual register.
-
- // Find the last use of the new virtual register. Remove all
- // instruction between here and there, and update the current
- // instruction to reference the last use insn instead.
- MachineBasicBlock::iterator LastUseMI =
- findLastUseReg(I, BB->end(), Reg);
-
- // Remove all instructions up 'til the last use, since they're
- // just calculating the value we already have.
- BB->erase(I, LastUseMI);
- I = LastUseMI;
-
- // Extend the live range of the scratch register
- PrevLastUseMI->getOperand(PrevLastUseOp).setIsKill(false);
- RS->setUsed(CurrentScratchReg);
- CurrentVirtReg = Reg;
-
- // We deleted the instruction we were scanning the operands of.
- // Jump back to the instruction iterator loop. Don't increment
- // past this instruction since we updated the iterator already.
- DoIncr = false;
- break;
- }
-
- // Scavenge a new scratch register
- CurrentVirtReg = Reg;
- const TargetRegisterClass *RC = Fn.getRegInfo().getRegClass(Reg);
- CurrentScratchReg = RS->scavengeRegister(RC, I, SPAdj);
- PrevValue = Value;
- }
- // replace this reference to the virtual register with the
+
+ // When we first encounter a new virtual register, it
+ // must be a definition.
+ assert(MI->getOperand(i).isDef() &&
+ "frame index virtual missing def!");
+ // Scavenge a new scratch register
+ const TargetRegisterClass *RC = Fn.getRegInfo().getRegClass(Reg);
+ unsigned ScratchReg = RS->scavengeRegister(RC, J, SPAdj);
+
+ ++NumScavengedRegs;
+
+ // Replace this reference to the virtual register with the
// scratch register.
- assert (CurrentScratchReg && "Missing scratch register!");
- MI->getOperand(i).setReg(CurrentScratchReg);
-
- if (MI->getOperand(i).isKill()) {
- isKillInsn = true;
- PrevLastUseOp = i;
- PrevLastUseMI = MI;
- }
+ assert (ScratchReg && "Missing scratch register!");
+ MachineRegisterInfo &MRI = Fn.getRegInfo();
+ Fn.getRegInfo().replaceRegWith(Reg, ScratchReg);
+
+ // Make sure MRI now accounts this register as used.
+ MRI.setPhysRegUsed(ScratchReg);
+
+ // Because this instruction was processed by the RS before this
+ // register was allocated, make sure that the RS now records the
+ // register as being used.
+ RS->setRegUsed(ScratchReg);
}
}
- // If this is the last use of the scratch, stop tracking it. The
- // last use will be a kill operand in an instruction that does
- // not also define the scratch register.
- if (isKillInsn && !isDefInsn) {
- CurrentVirtReg = 0;
- havePrevValue = trackingCurrentValue;
- }
- // Similarly, notice if instruction clobbered the value in the
- // register we're tracking for possible later reuse. This is noted
- // above, but enforced here since the value is still live while we
- // process the rest of the operands of the instruction.
- if (clobbersScratchReg) {
- havePrevValue = false;
- CurrentScratchReg = 0;
- }
- if (DoIncr) {
- RS->forward(I);
+
+ // If the scavenger needed to use one of its spill slots, the
+ // spill code will have been inserted in between I and J. This is a
+ // problem because we need the spill code before I: Move I to just
+ // prior to J.
+ if (I != std::prev(J)) {
+ BB->splice(J, BB, I);
+
+ // Before we move I, we need to prepare the RS to visit I again.
+ // Specifically, RS will assert if it sees uses of registers that
+ // it believes are undefined. Because we have already processed
+ // register kills in I, when it visits I again, it will believe that
+ // those registers are undefined. To avoid this situation, unprocess
+ // the instruction I.
+ assert(RS->getCurrentPosition() == I &&
+ "The register scavenger has an unexpected position");
+ I = P;
+ RS->unprocess(P);
+ } else
++I;
- }
}
}
}