//
//===----------------------------------------------------------------------===//
+#define DEBUG_TYPE "pei"
#include "PrologEpilogInserter.h"
#include "llvm/CodeGen/MachineDominators.h"
#include "llvm/CodeGen/MachineLoopInfo.h"
#include "llvm/Target/TargetInstrInfo.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 <climits>
using namespace llvm;
+// FIXME: For testing purposes only. Remove once the pre-allocation pass
+// is done.
+extern cl::opt<bool> EnableLocalStackAlloc;
+
char PEI::ID = 0;
-static RegisterPass<PEI>
-X("prologepilog", "Prologue/Epilogue Insertion");
+INITIALIZE_PASS(PEI, "prologepilog",
+ "Prologue/Epilogue Insertion", false, false);
/// createPrologEpilogCodeInserter - This function returns a pass that inserts
/// prolog and epilog code, and eliminates abstract frame references.
FrameIndexVirtualScavenging = TRI->requiresFrameIndexScavenging(Fn);
FrameConstantRegMap.clear();
- // Calculate the MaxCallFrameSize and HasCalls variables for the function's
- // frame information. Also eliminates call frame pseudo instructions.
+ // Calculate the MaxCallFrameSize and AdjustsStack variables for the
+ // function's frame information. Also eliminates call frame pseudo
+ // instructions.
calculateCallsInformation(Fn);
// Allow the target machine to make some adjustments to the function
calculateCalleeSavedRegisters(Fn);
// Determine placement of CSR spill/restore code:
- // - with shrink wrapping, place spills and restores to tightly
+ // - 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
+ // they are used.
+ // - Without shink wrapping (default), place all spills in the
// entry block, all restores in return blocks.
placeCSRSpillsAndRestores(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
+ // 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))
insertPrologEpilogCode(Fn);
}
#endif
-/// calculateCallsInformation - Calculate the MaxCallFrameSize and HasCalls
+/// calculateCallsInformation - Calculate the MaxCallFrameSize and AdjustsStack
/// variables for the function's frame information and eliminate call frame
/// pseudo instructions.
void PEI::calculateCallsInformation(MachineFunction &Fn) {
MachineFrameInfo *MFI = Fn.getFrameInfo();
unsigned MaxCallFrameSize = 0;
- bool HasCalls = MFI->hasCalls();
+ bool AdjustsStack = MFI->adjustsStack();
// Get the function call frame set-up and tear-down instruction opcode
int FrameSetupOpcode = RegInfo->getCallFrameSetupOpcode();
" instructions should have a single immediate argument!");
unsigned Size = I->getOperand(0).getImm();
if (Size > MaxCallFrameSize) MaxCallFrameSize = Size;
- HasCalls = true;
+ AdjustsStack = true;
FrameSDOps.push_back(I);
} else if (I->isInlineAsm()) {
- // An InlineAsm might be a call; assume it is to get the stack frame
- // aligned correctly for calls.
- HasCalls = true;
+ // Some inline asm's need a stack frame, as indicated by operand 1.
+ if (I->getOperand(1).getImm())
+ AdjustsStack = true;
}
- MFI->setHasCalls(HasCalls);
+ MFI->setAdjustsStack(AdjustsStack);
MFI->setMaxCallFrameSize(MaxCallFrameSize);
for (std::vector<MachineBasicBlock::iterator>::iterator
if (Fn.getFunction()->hasFnAttr(Attribute::Naked))
return;
- // Figure out which *callee saved* registers are modified by the current
- // function, thus needing to be saved and restored in the prolog/epilog.
- const TargetRegisterClass * const *CSRegClasses =
- RegInfo->getCalleeSavedRegClasses(&Fn);
-
std::vector<CalleeSavedInfo> CSI;
for (unsigned i = 0; CSRegs[i]; ++i) {
unsigned Reg = CSRegs[i];
if (Fn.getRegInfo().isPhysRegUsed(Reg)) {
// If the reg is modified, save it!
- CSI.push_back(CalleeSavedInfo(Reg, CSRegClasses[i]));
+ 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, CSRegClasses[i]));
+ CSI.push_back(CalleeSavedInfo(Reg));
break;
}
}
for (std::vector<CalleeSavedInfo>::iterator
I = CSI.begin(), E = CSI.end(); I != E; ++I) {
unsigned Reg = I->getReg();
- const TargetRegisterClass *RC = I->getRegClass();
+ const TargetRegisterClass *RC = RegInfo->getMinimalPhysRegClass(Reg);
int FrameIdx;
if (RegInfo->hasReservedSpillSlot(Fn, Reg, FrameIdx)) {
if ((unsigned)FrameIdx > MaxCSFrameIndex) MaxCSFrameIndex = FrameIdx;
} else {
// Spill it to the stack where we must.
- FrameIdx = MFI->CreateFixedObject(RC->getSize(), FixedSlot->Offset,
- true, false);
+ FrameIdx = MFI->CreateFixedObject(RC->getSize(), FixedSlot->Offset, true);
}
I->setFrameIdx(FrameIdx);
if (! ShrinkWrapThisFunction) {
// Spill using target interface.
I = EntryBlock->begin();
- if (!TII.spillCalleeSavedRegisters(*EntryBlock, I, CSI)) {
+ 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.
- TII.storeRegToStackSlot(*EntryBlock, I, CSI[i].getReg(), true,
- CSI[i].getFrameIdx(), CSI[i].getRegClass(),TRI);
+ unsigned Reg = CSI[i].getReg();
+ const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg);
+ TII.storeRegToStackSlot(*EntryBlock, I, Reg, true,
+ CSI[i].getFrameIdx(), RC, TRI);
}
}
// Restore all registers immediately before the return and any
// terminators that preceed it.
- if (!TII.restoreCalleeSavedRegisters(*MBB, I, CSI)) {
+ if (!TII.restoreCalleeSavedRegisters(*MBB, I, CSI, TRI)) {
for (unsigned i = 0, e = CSI.size(); i != e; ++i) {
- TII.loadRegFromStackSlot(*MBB, I, CSI[i].getReg(),
+ unsigned Reg = CSI[i].getReg();
+ const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg);
+ TII.loadRegFromStackSlot(*MBB, I, Reg,
CSI[i].getFrameIdx(),
- CSI[i].getRegClass(), TRI);
+ RC, TRI);
assert(I != MBB->begin() &&
"loadRegFromStackSlot didn't insert any code!");
// Insert in reverse order. loadRegFromStackSlot can insert
MBB->addLiveIn(blockCSI[i].getReg());
// Insert the spill to the stack frame.
- TII.storeRegToStackSlot(*MBB, I, blockCSI[i].getReg(),
+ unsigned Reg = blockCSI[i].getReg();
+ const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg);
+ TII.storeRegToStackSlot(*MBB, I, Reg,
true,
blockCSI[i].getFrameIdx(),
- blockCSI[i].getRegClass(), TRI);
+ RC, TRI);
}
}
// 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(),
+ unsigned Reg = blockCSI[i].getReg();
+ const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg);
+ TII.loadRegFromStackSlot(*MBB, I, Reg,
blockCSI[i].getFrameIdx(),
- blockCSI[i].getRegClass(), TRI);
+ RC, TRI);
assert(I != MBB->begin() &&
"loadRegFromStackSlot didn't insert any code!");
// Insert in reverse order. loadRegFromStackSlot can insert
Offset = (Offset + Align - 1) / Align * Align;
if (StackGrowsDown) {
+ DEBUG(dbgs() << "alloc FI(" << FrameIdx << ") at SP[" << -Offset << "]\n");
MFI->setObjectOffset(FrameIdx, -Offset); // Set the computed offset
} else {
+ DEBUG(dbgs() << "alloc FI(" << FrameIdx << ") at SP[" << Offset << "]\n");
MFI->setObjectOffset(FrameIdx, Offset);
Offset += MFI->getObjectSize(FrameIdx);
}
AdjustStackOffset(MFI, SFI, StackGrowsDown, Offset, MaxAlign);
}
+ // 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 (EnableLocalStackAlloc) {
+ unsigned Align = MFI->getLocalFrameMaxAlign();
+
+ // Adjust to alignment boundary.
+ Offset = (Offset + Align - 1) / Align * Align;
+
+ // 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);
+
+ DEBUG(dbgs() << "Local frame base offset: " << Offset << "\n");
+
+ // Allocate the local block
+ Offset += MFI->getLocalFrameSize();
+
+ // 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);
+ }
+ }
+
// Make sure that the stack protector comes before the local variables on the
// stack.
- if (MFI->getStackProtectorIndex() >= 0)
+ SmallSet<int, 16> LargeStackObjs;
+ if (MFI->getStackProtectorIndex() >= 0) {
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))
+ continue;
+ if (i >= MinCSFrameIndex && i <= MaxCSFrameIndex)
+ continue;
+ if (RS && (int)i == RS->getScavengingFrameIndex())
+ 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);
+ }
+ }
+
// 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))
+ continue;
if (i >= MinCSFrameIndex && i <= MaxCSFrameIndex)
continue;
if (RS && (int)i == RS->getScavengingFrameIndex())
continue;
if (MFI->getStackProtectorIndex() == (int)i)
continue;
+ if (LargeStackObjs.count(i))
+ continue;
AdjustStackOffset(MFI, i, StackGrowsDown, Offset, MaxAlign);
}
// 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->hasCalls() && RegInfo->hasReservedCallFrame(Fn))
+ if (MFI->adjustsStack() && RegInfo->hasReservedCallFrame(Fn))
Offset += MFI->getMaxCallFrameSize();
// Round up the size to a multiple of the alignment. If the function has
// otherwise, for leaf functions, align to the TransientStackAlignment
// value.
unsigned StackAlign;
- if (MFI->hasCalls() || MFI->hasVarSizedObjects() ||
+ if (MFI->adjustsStack() || MFI->hasVarSizedObjects() ||
(RegInfo->needsStackRealignment(Fn) && MFI->getObjectIndexEnd() != 0))
StackAlign = TFI.getStackAlignment();
else
StackAlign = TFI.getTransientStackAlignment();
- // If the frame pointer is eliminated, all frame offsets will be relative
- // to SP not FP; align to MaxAlign so this works.
+
+ // 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);
MFI->setStackSize(Offset - LocalAreaOffset);
}
-
/// insertPrologEpilogCode - Scan the function for modified callee saved
/// registers, insert spill code for these callee saved registers, then add
/// prolog and epilog code to the function.
}
}
-
/// replaceFrameIndices - Replace all MO_FrameIndex operands with physical
/// register references and actual offsets.
///
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);
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 (RS && !FrameIndexVirtualScavenging && MI) RS->forward(MI);
}
- assert(SPAdj == 0 && "Unbalanced call frame setup / destroy pairs?");
+ // 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?");
}
}
// Scavenge a new scratch register
CurrentVirtReg = Reg;
const TargetRegisterClass *RC = Fn.getRegInfo().getRegClass(Reg);
- CurrentScratchReg = RS->FindUnusedReg(RC);
- if (CurrentScratchReg == 0)
- // No register is "free". Scavenge a register.
- CurrentScratchReg = RS->scavengeRegister(RC, I, SPAdj);
-
+ CurrentScratchReg = RS->scavengeRegister(RC, I, SPAdj);
PrevValue = Value;
}
// replace this reference to the virtual register with the
projects / oota-llvm.git / blobdiff