aboutsummaryrefslogtreecommitdiff
path: root/llvm/lib/Target/X86/X86FrameLowering.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/lib/Target/X86/X86FrameLowering.cpp')
-rw-r--r--llvm/lib/Target/X86/X86FrameLowering.cpp121
1 files changed, 78 insertions, 43 deletions
diff --git a/llvm/lib/Target/X86/X86FrameLowering.cpp b/llvm/lib/Target/X86/X86FrameLowering.cpp
index a15db03..50c56c9 100644
--- a/llvm/lib/Target/X86/X86FrameLowering.cpp
+++ b/llvm/lib/Target/X86/X86FrameLowering.cpp
@@ -223,6 +223,8 @@ flagsNeedToBePreservedBeforeTheTerminators(const MachineBasicBlock &MBB) {
return false;
}
+constexpr int64_t MaxSPChunk = (1LL << 31) - 1;
+
/// emitSPUpdate - Emit a series of instructions to increment / decrement the
/// stack pointer by a constant value.
void X86FrameLowering::emitSPUpdate(MachineBasicBlock &MBB,
@@ -242,7 +244,7 @@ void X86FrameLowering::emitSPUpdate(MachineBasicBlock &MBB,
return;
}
- uint64_t Chunk = (1LL << 31) - 1;
+ uint64_t Chunk = MaxSPChunk;
MachineFunction &MF = *MBB.getParent();
const X86Subtarget &STI = MF.getSubtarget<X86Subtarget>();
@@ -391,12 +393,15 @@ MachineInstrBuilder X86FrameLowering::BuildStackAdjustment(
return MI;
}
-int X86FrameLowering::mergeSPUpdates(MachineBasicBlock &MBB,
- MachineBasicBlock::iterator &MBBI,
- bool doMergeWithPrevious) const {
+template <typename FoundT, typename CalcT>
+int64_t X86FrameLowering::mergeSPUpdates(MachineBasicBlock &MBB,
+ MachineBasicBlock::iterator &MBBI,
+ FoundT FoundStackAdjust,
+ CalcT CalcNewOffset,
+ bool doMergeWithPrevious) const {
if ((doMergeWithPrevious && MBBI == MBB.begin()) ||
(!doMergeWithPrevious && MBBI == MBB.end()))
- return 0;
+ return CalcNewOffset(0);
MachineBasicBlock::iterator PI = doMergeWithPrevious ? std::prev(MBBI) : MBBI;
@@ -415,27 +420,38 @@ int X86FrameLowering::mergeSPUpdates(MachineBasicBlock &MBB,
if (doMergeWithPrevious && PI != MBB.begin() && PI->isCFIInstruction())
PI = std::prev(PI);
- unsigned Opc = PI->getOpcode();
- int Offset = 0;
-
- if ((Opc == X86::ADD64ri32 || Opc == X86::ADD32ri) &&
- PI->getOperand(0).getReg() == StackPtr) {
- assert(PI->getOperand(1).getReg() == StackPtr);
- Offset = PI->getOperand(2).getImm();
- } else if ((Opc == X86::LEA32r || Opc == X86::LEA64_32r) &&
- PI->getOperand(0).getReg() == StackPtr &&
- PI->getOperand(1).getReg() == StackPtr &&
- PI->getOperand(2).getImm() == 1 &&
- PI->getOperand(3).getReg() == X86::NoRegister &&
- PI->getOperand(5).getReg() == X86::NoRegister) {
- // For LEAs we have: def = lea SP, FI, noreg, Offset, noreg.
- Offset = PI->getOperand(4).getImm();
- } else if ((Opc == X86::SUB64ri32 || Opc == X86::SUB32ri) &&
- PI->getOperand(0).getReg() == StackPtr) {
- assert(PI->getOperand(1).getReg() == StackPtr);
- Offset = -PI->getOperand(2).getImm();
- } else
- return 0;
+ int64_t Offset = 0;
+ for (;;) {
+ unsigned Opc = PI->getOpcode();
+
+ if ((Opc == X86::ADD64ri32 || Opc == X86::ADD32ri) &&
+ PI->getOperand(0).getReg() == StackPtr) {
+ assert(PI->getOperand(1).getReg() == StackPtr);
+ Offset = PI->getOperand(2).getImm();
+ } else if ((Opc == X86::LEA32r || Opc == X86::LEA64_32r) &&
+ PI->getOperand(0).getReg() == StackPtr &&
+ PI->getOperand(1).getReg() == StackPtr &&
+ PI->getOperand(2).getImm() == 1 &&
+ PI->getOperand(3).getReg() == X86::NoRegister &&
+ PI->getOperand(5).getReg() == X86::NoRegister) {
+ // For LEAs we have: def = lea SP, FI, noreg, Offset, noreg.
+ Offset = PI->getOperand(4).getImm();
+ } else if ((Opc == X86::SUB64ri32 || Opc == X86::SUB32ri) &&
+ PI->getOperand(0).getReg() == StackPtr) {
+ assert(PI->getOperand(1).getReg() == StackPtr);
+ Offset = -PI->getOperand(2).getImm();
+ } else
+ return CalcNewOffset(0);
+
+ FoundStackAdjust(PI, Offset);
+ if (std::abs((int64_t)CalcNewOffset(Offset)) < MaxSPChunk)
+ break;
+
+ if (doMergeWithPrevious ? (PI == MBB.begin()) : (PI == MBB.end()))
+ return CalcNewOffset(0);
+
+ PI = doMergeWithPrevious ? std::prev(PI) : std::next(PI);
+ }
PI = MBB.erase(PI);
if (PI != MBB.end() && PI->isCFIInstruction()) {
@@ -448,7 +464,16 @@ int X86FrameLowering::mergeSPUpdates(MachineBasicBlock &MBB,
if (!doMergeWithPrevious)
MBBI = skipDebugInstructionsForward(PI, MBB.end());
- return Offset;
+ return CalcNewOffset(Offset);
+}
+
+int64_t X86FrameLowering::mergeSPAdd(MachineBasicBlock &MBB,
+ MachineBasicBlock::iterator &MBBI,
+ int64_t AddOffset,
+ bool doMergeWithPrevious) const {
+ return mergeSPUpdates(
+ MBB, MBBI, [AddOffset](int64_t Offset) { return AddOffset + Offset; },
+ doMergeWithPrevious);
}
void X86FrameLowering::BuildCFI(MachineBasicBlock &MBB,
@@ -1975,8 +2000,10 @@ void X86FrameLowering::emitPrologue(MachineFunction &MF,
// If there is an SUB32ri of ESP immediately before this instruction, merge
// the two. This can be the case when tail call elimination is enabled and
- // the callee has more arguments then the caller.
- NumBytes -= mergeSPUpdates(MBB, MBBI, true);
+ // the callee has more arguments than the caller.
+ NumBytes = mergeSPUpdates(
+ MBB, MBBI, [NumBytes](int64_t Offset) { return NumBytes - Offset; },
+ true);
// Adjust stack pointer: ESP -= numbytes.
@@ -2457,7 +2484,7 @@ void X86FrameLowering::emitEpilogue(MachineFunction &MF,
if (HasFP) {
if (X86FI->hasSwiftAsyncContext()) {
// Discard the context.
- int Offset = 16 + mergeSPUpdates(MBB, MBBI, true);
+ int64_t Offset = mergeSPAdd(MBB, MBBI, 16, true);
emitSPUpdate(MBB, MBBI, DL, Offset, /*InEpilogue*/ true);
}
// Pop EBP.
@@ -2531,7 +2558,7 @@ void X86FrameLowering::emitEpilogue(MachineFunction &MF,
// If there is an ADD32ri or SUB32ri of ESP immediately before this
// instruction, merge the two instructions.
if (NumBytes || MFI.hasVarSizedObjects())
- NumBytes += mergeSPUpdates(MBB, MBBI, true);
+ NumBytes = mergeSPAdd(MBB, MBBI, NumBytes, true);
// If dynamic alloca is used, then reset esp to point to the last callee-saved
// slot before popping them off! Same applies for the case, when stack was
@@ -2612,11 +2639,11 @@ void X86FrameLowering::emitEpilogue(MachineFunction &MF,
if (Terminator == MBB.end() || !isTailCallOpcode(Terminator->getOpcode())) {
// Add the return addr area delta back since we are not tail calling.
- int Offset = -1 * X86FI->getTCReturnAddrDelta();
+ int64_t Offset = -1 * X86FI->getTCReturnAddrDelta();
assert(Offset >= 0 && "TCDelta should never be positive");
if (Offset) {
// Check for possible merge with preceding ADD instruction.
- Offset += mergeSPUpdates(MBB, Terminator, true);
+ Offset = mergeSPAdd(MBB, Terminator, Offset, true);
emitSPUpdate(MBB, Terminator, DL, Offset, /*InEpilogue=*/true);
}
}
@@ -3814,13 +3841,24 @@ MachineBasicBlock::iterator X86FrameLowering::eliminateCallFramePseudoInstr(
// Add Amount to SP to destroy a frame, or subtract to setup.
int64_t StackAdjustment = isDestroy ? Amount : -Amount;
+ int64_t CfaAdjustment = StackAdjustment;
if (StackAdjustment) {
// Merge with any previous or following adjustment instruction. Note: the
// instructions merged with here do not have CFI, so their stack
- // adjustments do not feed into CfaAdjustment.
- StackAdjustment += mergeSPUpdates(MBB, InsertPos, true);
- StackAdjustment += mergeSPUpdates(MBB, InsertPos, false);
+ // adjustments do not feed into CfaAdjustment
+
+ auto CalcCfaAdjust = [&CfaAdjustment](MachineBasicBlock::iterator PI,
+ int64_t Offset) {
+ CfaAdjustment += Offset;
+ };
+ auto CalcNewOffset = [&StackAdjustment](int64_t Offset) {
+ return StackAdjustment + Offset;
+ };
+ StackAdjustment =
+ mergeSPUpdates(MBB, InsertPos, CalcCfaAdjust, CalcNewOffset, true);
+ StackAdjustment =
+ mergeSPUpdates(MBB, InsertPos, CalcCfaAdjust, CalcNewOffset, false);
if (StackAdjustment) {
if (!(F.hasMinSize() &&
@@ -3830,7 +3868,7 @@ MachineBasicBlock::iterator X86FrameLowering::eliminateCallFramePseudoInstr(
}
}
- if (DwarfCFI && !hasFP(MF)) {
+ if (DwarfCFI && !hasFP(MF) && CfaAdjustment) {
// If we don't have FP, but need to generate unwind information,
// we need to set the correct CFA offset after the stack adjustment.
// How much we adjust the CFA offset depends on whether we're emitting
@@ -3838,14 +3876,11 @@ MachineBasicBlock::iterator X86FrameLowering::eliminateCallFramePseudoInstr(
// offset to be correct at each call site, while for debugging we want
// it to be more precise.
- int64_t CfaAdjustment = -StackAdjustment;
// TODO: When not using precise CFA, we also need to adjust for the
// InternalAmt here.
- if (CfaAdjustment) {
- BuildCFI(
- MBB, InsertPos, DL,
- MCCFIInstruction::createAdjustCfaOffset(nullptr, CfaAdjustment));
- }
+ BuildCFI(
+ MBB, InsertPos, DL,
+ MCCFIInstruction::createAdjustCfaOffset(nullptr, -CfaAdjustment));
}
return I;
generated by cgit v1.1 at 2025-09-28 03:12:16 +0000