// 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.
+//
//===----------------------------------------------------------------------===//
-#include "llvm/CodeGen/Passes.h"
-#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "PrologEpilogInserter.h"
+#include "llvm/CodeGen/MachineDominators.h"
+#include "llvm/CodeGen/MachineLoopInfo.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/Support/Compiler.h"
#include "llvm/ADT/STLExtras.h"
#include <climits>
+
using namespace llvm;
-namespace {
- struct VISIBILITY_HIDDEN PEI : public MachineFunctionPass {
- static char ID;
- PEI() : MachineFunctionPass(&ID) {}
-
- const char *getPassName() const {
- return "Prolog/Epilog Insertion & Frame Finalization";
- }
-
- virtual void getAnalysisUsage(AnalysisUsage &AU) const {
- AU.addPreservedID(MachineLoopInfoID);
- AU.addPreservedID(MachineDominatorsID);
- MachineFunctionPass::getAnalysisUsage(AU);
- }
-
- /// runOnMachineFunction - Insert prolog/epilog code and replace abstract
- /// frame indexes with appropriate references.
- ///
- bool runOnMachineFunction(MachineFunction &Fn) {
- const TargetRegisterInfo *TRI = Fn.getTarget().getRegisterInfo();
- RS = TRI->requiresRegisterScavenging(Fn) ? new RegScavenger() : NULL;
-
- // Get MachineModuleInfo so that we can track the construction of the
- // frame.
- if (MachineModuleInfo *MMI = getAnalysisToUpdate<MachineModuleInfo>())
- Fn.getFrameInfo()->setMachineModuleInfo(MMI);
-
- // Allow the target machine to make some adjustments to the function
- // e.g. UsedPhysRegs before calculateCalleeSavedRegisters.
- TRI->processFunctionBeforeCalleeSavedScan(Fn, RS);
-
- // Scan the function for modified callee saved registers and insert spill
- // code for any callee saved registers that are modified. Also calculate
- // the MaxCallFrameSize and HasCalls variables for the function's frame
- // information and eliminates call frame pseudo instructions.
- calculateCalleeSavedRegisters(Fn);
-
- // Add the code to save and restore the callee saved registers
- saveCalleeSavedRegisters(Fn);
-
- // Allow the target machine to make final modifications to the function
- // before the frame layout is finalized.
- TRI->processFunctionBeforeFrameFinalized(Fn);
-
- // Calculate actual frame offsets for all of the abstract stack objects...
- calculateFrameObjectOffsets(Fn);
-
- // Add prolog and epilog code to the function. This function is required
- // to align the stack frame as necessary for any stack variables or
- // called functions. Because of this, calculateCalleeSavedRegisters
- // must be called before this function in order to set the HasCalls
- // and MaxCallFrameSize variables.
- insertPrologEpilogCode(Fn);
-
- // Replace all MO_FrameIndex operands with physical register references
- // and actual offsets.
- //
- replaceFrameIndices(Fn);
-
- delete RS;
- return true;
- }
-
- private:
- RegScavenger *RS;
-
- // MinCSFrameIndex, MaxCSFrameIndex - Keeps the range of callee saved
- // stack frame indexes.
- unsigned MinCSFrameIndex, MaxCSFrameIndex;
-
- void calculateCalleeSavedRegisters(MachineFunction &Fn);
- void saveCalleeSavedRegisters(MachineFunction &Fn);
- void calculateFrameObjectOffsets(MachineFunction &Fn);
- void replaceFrameIndices(MachineFunction &Fn);
- void insertPrologEpilogCode(MachineFunction &Fn);
- };
- char PEI::ID = 0;
-}
+char PEI::ID = 0;
+static RegisterPass<PEI>
+X("prologepilog", "Prologue/Epilogue Insertion");
/// createPrologEpilogCodeInserter - This function returns a pass that inserts
/// prolog and epilog code, and eliminates abstract frame references.
///
FunctionPass *llvm::createPrologEpilogCodeInserter() { return new PEI(); }
+/// runOnMachineFunction - Insert prolog/epilog code and replace abstract
+/// frame indexes with appropriate references.
+///
+bool PEI::runOnMachineFunction(MachineFunction &Fn) {
+ const TargetRegisterInfo *TRI = Fn.getTarget().getRegisterInfo();
+ RS = TRI->requiresRegisterScavenging(Fn) ? new RegScavenger() : NULL;
+
+ // Get MachineModuleInfo so that we can track the construction of the
+ // frame.
+ if (MachineModuleInfo *MMI = getAnalysisIfAvailable<MachineModuleInfo>())
+ Fn.getFrameInfo()->setMachineModuleInfo(MMI);
+
+ // Allow the target machine to make some adjustments to the function
+ // e.g. UsedPhysRegs before calculateCalleeSavedRegisters.
+ TRI->processFunctionBeforeCalleeSavedScan(Fn, RS);
+
+ // Scan the function for modified callee saved registers and insert spill
+ // code for any callee saved registers that are modified. Also calculate
+ // the MaxCallFrameSize and HasCalls variables for the function's frame
+ // information and eliminates call frame pseudo instructions.
+ 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 shrink wrapping
+ // - default (no shrink wrapping), place all spills in the
+ // entry block, all restores in return blocks.
+ placeCSRSpillsAndRestores(Fn);
+
+ // Add the code to save and restore the callee saved registers
+ insertCSRSpillsAndRestores(Fn);
+
+ // Allow the target machine to make final modifications to the function
+ // before the frame layout is finalized.
+ TRI->processFunctionBeforeFrameFinalized(Fn);
+
+ // Calculate actual frame offsets for all abstract stack objects...
+ calculateFrameObjectOffsets(Fn);
+
+ // Add prolog and epilog code to the function. This function is required
+ // to align the stack frame as necessary for any stack variables or
+ // called functions. Because of this, calculateCalleeSavedRegisters
+ // must be called before this function in order to set the HasCalls
+ // and MaxCallFrameSize variables.
+ insertPrologEpilogCode(Fn);
+
+ // Replace all MO_FrameIndex operands with physical register references
+ // and actual offsets.
+ //
+ replaceFrameIndices(Fn);
+
+ delete RS;
+ clearAllSets();
+ return true;
+}
+
+#if 0
+void PEI::getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.setPreservesCFG();
+ if (ShrinkWrapping || ShrinkWrapFunc != "") {
+ AU.addRequired<MachineLoopInfo>();
+ AU.addRequired<MachineDominatorTree>();
+ }
+ AU.addPreserved<MachineLoopInfo>();
+ AU.addPreserved<MachineDominatorTree>();
+ MachineFunctionPass::getAnalysisUsage(AU);
+}
+#endif
/// calculateCalleeSavedRegisters - Scan the function for modified callee saved
/// registers. Also calculate the MaxCallFrameSize and HasCalls variables for
unsigned Align = RC->getAlignment();
unsigned StackAlign = TFI->getStackAlignment();
// We may not be able to sastify the desired alignment specification of
- // the TargetRegisterClass if the stack alignment is smaller. Use the min.
+ // the TargetRegisterClass if the stack alignment is smaller.
+ // Use the min.
Align = std::min(Align, StackAlign);
FrameIdx = FFI->CreateStackObject(RC->getSize(), Align);
if ((unsigned)FrameIdx < MinCSFrameIndex) MinCSFrameIndex = FrameIdx;
FFI->setCalleeSavedInfo(CSI);
}
-/// saveCalleeSavedRegisters - Insert spill code for any callee saved registers
-/// that are modified in the function.
+/// insertCSRSpillsAndRestores - Insert spill and restore code for
+/// callee saved registers used in the function, handling shrink wrapping.
///
-void PEI::saveCalleeSavedRegisters(MachineFunction &Fn) {
+void PEI::insertCSRSpillsAndRestores(MachineFunction &Fn) {
// Get callee saved register information.
MachineFrameInfo *FFI = Fn.getFrameInfo();
const std::vector<CalleeSavedInfo> &CSI = FFI->getCalleeSavedInfo();
-
+
// Early exit if no callee saved registers are modified!
if (CSI.empty())
return;
const TargetInstrInfo &TII = *Fn.getTarget().getInstrInfo();
-
- // Now that we have a stack slot for each register to be saved, insert spill
- // code into the entry block.
- MachineBasicBlock *MBB = Fn.begin();
- MachineBasicBlock::iterator I = MBB->begin();
-
- if (!TII.spillCalleeSavedRegisters(*MBB, I, CSI)) {
- 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(CSI[i].getReg());
-
- // Insert the spill to the stack frame.
- TII.storeRegToStackSlot(*MBB, I, CSI[i].getReg(), true,
- CSI[i].getFrameIdx(), CSI[i].getRegClass());
+ MachineBasicBlock::iterator I;
+
+ if (! ShrinkWrapThisFunction) {
+ // Spill using target interface.
+ I = EntryBlock->begin();
+ if (!TII.spillCalleeSavedRegisters(*EntryBlock, I, CSI)) {
+ 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.
+ TII.storeRegToStackSlot(*EntryBlock, I, CSI[i].getReg(), true,
+ CSI[i].getFrameIdx(), CSI[i].getRegClass());
+ }
}
- }
- // Add code to restore the callee-save registers in each exiting block.
- for (MachineFunction::iterator FI = Fn.begin(), E = Fn.end(); FI != E; ++FI)
- // If last instruction is a return instruction, add an epilogue.
- if (!FI->empty() && FI->back().getDesc().isReturn()) {
- MBB = FI;
+ // 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 "return" instructions, which are part of the return
+ // Skip over all terminator instructions, which are part of the return
// sequence.
MachineBasicBlock::iterator I2 = I;
- while (I2 != MBB->begin() && (--I2)->getDesc().isReturn())
+ 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.
+
+ // Restore all registers immediately before the return and any
+ // terminators that preceed it.
if (!TII.restoreCalleeSavedRegisters(*MBB, I, CSI)) {
for (unsigned i = 0, e = CSI.size(); i != e; ++i) {
TII.loadRegFromStackSlot(*MBB, I, CSI[i].getReg(),
- CSI[i].getFrameIdx(),
- CSI[i].getRegClass());
+ CSI[i].getFrameIdx(),
+ CSI[i].getRegClass());
assert(I != MBB->begin() &&
"loadRegFromStackSlot didn't insert any code!");
- // Insert in reverse order. loadRegFromStackSlot can insert multiple
- // instructions.
+ // Insert in reverse order. loadRegFromStackSlot can insert
+ // multiple instructions.
if (AtStart)
I = MBB->begin();
else {
}
}
}
+ 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) {
+ // Add the callee-saved register as live-in.
+ // It's killed at the spill.
+ MBB->addLiveIn(blockCSI[i].getReg());
+
+ // Insert the spill to the stack frame.
+ TII.storeRegToStackSlot(*MBB, I, blockCSI[i].getReg(),
+ true,
+ blockCSI[i].getFrameIdx(),
+ blockCSI[i].getRegClass());
+ }
+ }
+
+ for (CSRegBlockMap::iterator BI = CSRRestore.begin(),
+ BE = CSRRestore.end(); BI != BE; ++BI) {
+ MachineBasicBlock* MBB = BI->first;
+ CSRegSet restore = BI->second;
+
+ 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;
+ }
+ }
+
+ 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.
+ for (unsigned i = 0, e = blockCSI.size(); i != e; ++i) {
+ TII.loadRegFromStackSlot(*MBB, I, blockCSI[i].getReg(),
+ blockCSI[i].getFrameIdx(),
+ blockCSI[i].getRegClass());
+ 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;
+ }
+ }
+ }
}
+/// AdjustStackOffset - Helper function used to adjust the stack frame offset.
+static inline void
+AdjustStackOffset(MachineFrameInfo *FFI, int FrameIdx,
+ bool StackGrowsDown, int64_t &Offset,
+ unsigned &MaxAlign) {
+ // If stack grows down, we need to add size of find the lowest address of the
+ // object.
+ if (StackGrowsDown)
+ Offset += FFI->getObjectSize(FrameIdx);
+
+ unsigned Align = FFI->getObjectAlignment(FrameIdx);
+
+ // If the alignment of this object is greater than that of the stack, then
+ // increase the stack alignment to match.
+ MaxAlign = std::max(MaxAlign, Align);
+
+ // Adjust to alignment boundary.
+ Offset = (Offset + Align - 1) / Align * Align;
+
+ if (StackGrowsDown) {
+ FFI->setObjectOffset(FrameIdx, -Offset); // Set the computed offset
+ } else {
+ FFI->setObjectOffset(FrameIdx, Offset);
+ Offset += FFI->getObjectSize(FrameIdx);
+ }
+}
/// calculateFrameObjectOffsets - Calculate actual frame offsets for all of the
/// abstract stack objects.
// If there are fixed sized objects that are preallocated in the local area,
// non-fixed objects can't be allocated right at the start of local area.
- // We currently don't support filling in holes in between fixed sized objects,
- // so we adjust 'Offset' to point to the end of last fixed sized
+ // We currently don't support filling in holes in between fixed sized
+ // objects, so we adjust 'Offset' to point to the end of last fixed sized
// preallocated object.
for (int i = FFI->getObjectIndexBegin(); i != 0; ++i) {
int64_t FixedOff;
Offset += FFI->getObjectSize(i);
unsigned Align = FFI->getObjectAlignment(i);
- // If the alignment of this object is greater than that of the stack, then
- // increase the stack alignment to match.
+ // If the alignment of this object is greater than that of the stack,
+ // then increase the stack alignment to match.
MaxAlign = std::max(MaxAlign, Align);
// Adjust to alignment boundary
Offset = (Offset+Align-1)/Align*Align;
int MaxCSFI = MaxCSFrameIndex, MinCSFI = MinCSFrameIndex;
for (int i = MaxCSFI; i >= MinCSFI ; --i) {
unsigned Align = FFI->getObjectAlignment(i);
- // If the alignment of this object is greater than that of the stack, then
- // increase the stack alignment to match.
+ // If the alignment of this object is greater than that of the stack,
+ // then increase the stack alignment to match.
MaxAlign = std::max(MaxAlign, Align);
// Adjust to alignment boundary
Offset = (Offset+Align-1)/Align*Align;
const TargetRegisterInfo *RegInfo = Fn.getTarget().getRegisterInfo();
if (RS && RegInfo->hasFP(Fn)) {
int SFI = RS->getScavengingFrameIndex();
- if (SFI >= 0) {
- // If stack grows down, we need to add size of the lowest
- // address of the object.
- if (StackGrowsDown)
- Offset += FFI->getObjectSize(SFI);
-
- unsigned Align = FFI->getObjectAlignment(SFI);
- // Adjust to alignment boundary
- Offset = (Offset+Align-1)/Align*Align;
-
- if (StackGrowsDown) {
- FFI->setObjectOffset(SFI, -Offset); // Set the computed offset
- } else {
- FFI->setObjectOffset(SFI, Offset);
- Offset += FFI->getObjectSize(SFI);
- }
- }
+ if (SFI >= 0)
+ AdjustStackOffset(FFI, SFI, StackGrowsDown, Offset, MaxAlign);
}
+ // Make sure that the stack protector comes before the local variables on the
+ // stack.
+ if (FFI->getStackProtectorIndex() >= 0)
+ AdjustStackOffset(FFI, FFI->getStackProtectorIndex(), StackGrowsDown,
+ Offset, MaxAlign);
+
// Then assign frame offsets to stack objects that are not used to spill
// callee saved registers.
for (unsigned i = 0, e = FFI->getObjectIndexEnd(); i != e; ++i) {
continue;
if (FFI->isDeadObjectIndex(i))
continue;
+ if (FFI->getStackProtectorIndex() == (int)i)
+ continue;
- // If stack grows down, we need to add size of find the lowest
- // address of the object.
- if (StackGrowsDown)
- Offset += FFI->getObjectSize(i);
-
- unsigned Align = FFI->getObjectAlignment(i);
- // If the alignment of this object is greater than that of the stack, then
- // increase the stack alignment to match.
- MaxAlign = std::max(MaxAlign, Align);
- // Adjust to alignment boundary
- Offset = (Offset+Align-1)/Align*Align;
-
- if (StackGrowsDown) {
- FFI->setObjectOffset(i, -Offset); // Set the computed offset
- } else {
- FFI->setObjectOffset(i, Offset);
- Offset += FFI->getObjectSize(i);
- }
+ AdjustStackOffset(FFI, i, StackGrowsDown, Offset, MaxAlign);
}
// Make sure the special register scavenging spill slot is closest to the
// stack pointer.
if (RS && !RegInfo->hasFP(Fn)) {
int SFI = RS->getScavengingFrameIndex();
- if (SFI >= 0) {
- // If stack grows down, we need to add size of find the lowest
- // address of the object.
- if (StackGrowsDown)
- Offset += FFI->getObjectSize(SFI);
-
- unsigned Align = FFI->getObjectAlignment(SFI);
- // If the alignment of this object is greater than that of the
- // stack, then increase the stack alignment to match.
- MaxAlign = std::max(MaxAlign, Align);
- // Adjust to alignment boundary
- Offset = (Offset+Align-1)/Align*Align;
-
- if (StackGrowsDown) {
- FFI->setObjectOffset(SFI, -Offset); // Set the computed offset
- } else {
- FFI->setObjectOffset(SFI, Offset);
- Offset += FFI->getObjectSize(SFI);
- }
- }
+ if (SFI >= 0)
+ AdjustStackOffset(FFI, SFI, StackGrowsDown, Offset, MaxAlign);
}
// Round up the size to a multiple of the alignment, but only if there are
// offsets will be relative to SP not FP; round up the stack size so this
// works.
if (!RegInfo->targetHandlesStackFrameRounding() &&
- (FFI->hasCalls() || FFI->hasVarSizedObjects() ||
+ (FFI->hasCalls() || FFI->hasVarSizedObjects() ||
(RegInfo->needsStackRealignment(Fn) &&
FFI->getObjectIndexEnd() != 0))) {
// If we have reserved argument space for call sites in the function
int FrameSetupOpcode = TRI.getCallFrameSetupOpcode();
int FrameDestroyOpcode = TRI.getCallFrameDestroyOpcode();
- for (MachineFunction::iterator BB = Fn.begin(), E = Fn.end(); BB != E; ++BB) {
+ for (MachineFunction::iterator BB = Fn.begin(),
+ E = Fn.end(); BB != E; ++BB) {
int SPAdj = 0; // SP offset due to call frame setup / destroy.
if (RS) RS->enterBasicBlock(BB);
- for (MachineBasicBlock::iterator I = BB->begin(); I != BB->end(); ) {
- MachineInstr *MI = I;
+ for (MachineBasicBlock::iterator I = BB->begin(); I != BB->end(); ) {
if (I->getOpcode() == TargetInstrInfo::DECLARE) {
// Ignore it.
++I;
SPAdj += Size;
- MachineBasicBlock::iterator PrevI = prior(I);
+ MachineBasicBlock::iterator PrevI = BB->end();
+ if (I != BB->begin()) PrevI = prior(I);
TRI.eliminateCallFramePseudoInstr(Fn, *BB, I);
// Visit the instructions created by eliminateCallFramePseudoInstr().
- I = next(PrevI);
+ if (PrevI == BB->end())
+ I = BB->begin(); // The replaced instr was the first in the block.
+ else
+ I = next(PrevI);
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
// If this instruction has a FrameIndex operand, we need to
// use that target machine register info object to eliminate
// it.
+
TRI.eliminateFrameIndex(MI, SPAdj, RS);
// Reset the iterator if we were at the beginning of the BB.
break;
}
- if (DoIncr) ++I;
+ if (DoIncr && I != BB->end()) ++I;
// Update register states.
if (RS && MI) RS->forward(MI);
assert(SPAdj == 0 && "Unbalanced call frame setup / destroy pairs?");
}
}
+