int AMDGPUFrameLowering::getFrameIndexOffset(const MachineFunction &MF,
int FI) const {
const MachineFrameInfo *MFI = MF.getFrameInfo();
- unsigned Offset = 0;
+ // Start the offset at 2 so we don't overwrite work group information.
+ // XXX: We should only do this when the shader actually uses this
+ // information.
+ unsigned OffsetBytes = 2 * (getStackWidth(MF) * 4);
int UpperBound = FI == -1 ? MFI->getNumObjects() : FI;
for (int i = MFI->getObjectIndexBegin(); i < UpperBound; ++i) {
- const AllocaInst *Alloca = MFI->getObjectAllocation(i);
- unsigned ArrayElements;
- const Type *AllocaType = Alloca->getAllocatedType();
- const Type *ElementType;
-
- if (AllocaType->isArrayTy()) {
- ArrayElements = AllocaType->getArrayNumElements();
- ElementType = AllocaType->getArrayElementType();
- } else {
- ArrayElements = 1;
- ElementType = AllocaType;
- }
+ OffsetBytes = RoundUpToAlignment(OffsetBytes, MFI->getObjectAlignment(i));
+ OffsetBytes += MFI->getObjectSize(i);
+ // Each register holds 4 bytes, so we must always align the offset to at
+ // least 4 bytes, so that 2 frame objects won't share the same register.
+ OffsetBytes = RoundUpToAlignment(OffsetBytes, 4);
+ }
- unsigned VectorElements;
- if (ElementType->isVectorTy()) {
- VectorElements = ElementType->getVectorNumElements();
- } else {
- VectorElements = 1;
- }
+ if (FI != -1)
+ OffsetBytes = RoundUpToAlignment(OffsetBytes, MFI->getObjectAlignment(FI));
- Offset += (VectorElements / getStackWidth(MF)) * ArrayElements;
- }
- return Offset;
+ return OffsetBytes / (getStackWidth(MF) * 4);
}
const TargetFrameLowering::SpillSlot *
AMDGPUFrameLowering::getCalleeSavedSpillSlots(unsigned &NumEntries) const {
NumEntries = 0;
- return 0;
-}
-void
-AMDGPUFrameLowering::emitPrologue(MachineFunction &MF) const {
+ return nullptr;
}
+void AMDGPUFrameLowering::emitPrologue(MachineFunction &MF,
+ MachineBasicBlock &MBB) const {}
void
AMDGPUFrameLowering::emitEpilogue(MachineFunction &MF,
MachineBasicBlock &MBB) const {