Home | History | Annotate | Download | only in NVPTX 
      1 //===-- NVPTXPrologEpilogPass.cpp - NVPTX prolog/epilog inserter ----------===//
      2 //
      3 //                     The LLVM Compiler Infrastructure
      4 //
      5 // This file is distributed under the University of Illinois Open Source
      6 // License. See LICENSE.TXT for details.
      7 //
      8 //===----------------------------------------------------------------------===//
      9 //
     10 // This file is a copy of the generic LLVM PrologEpilogInserter pass, modified
     11 // to remove unneeded functionality and to handle virtual registers. Most code
     12 // here is a copy of PrologEpilogInserter.cpp.
     13 //
     14 //===----------------------------------------------------------------------===//
     15 
     16 #include "NVPTX.h"
     17 #include "llvm/CodeGen/MachineFrameInfo.h"
     18 #include "llvm/CodeGen/MachineFunction.h"
     19 #include "llvm/CodeGen/MachineFunctionPass.h"
     20 #include "llvm/Pass.h"
     21 #include "llvm/Support/Debug.h"
     22 #include "llvm/Support/raw_ostream.h"
     23 #include "llvm/Target/TargetFrameLowering.h"
     24 #include "llvm/Target/TargetRegisterInfo.h"
     25 #include "llvm/Target/TargetSubtargetInfo.h"
     26 
     27 using namespace llvm;
     28 
     29 #define DEBUG_TYPE "nvptx-prolog-epilog"
     30 
     31 namespace {
     32 class NVPTXPrologEpilogPass : public MachineFunctionPass {
     33 public:
     34   static char ID;
     35   NVPTXPrologEpilogPass() : MachineFunctionPass(ID) {}
     36 
     37   bool runOnMachineFunction(MachineFunction &MF) override;
     38 
     39 private:
     40   void calculateFrameObjectOffsets(MachineFunction &Fn);
     41 };
     42 }
     43 
     44 MachineFunctionPass *llvm::createNVPTXPrologEpilogPass() {
     45   return new NVPTXPrologEpilogPass();
     46 }
     47 
     48 char NVPTXPrologEpilogPass::ID = 0;
     49 
     50 bool NVPTXPrologEpilogPass::runOnMachineFunction(MachineFunction &MF) {
     51   const TargetSubtargetInfo &STI = MF.getSubtarget();
     52   const TargetFrameLowering &TFI = *STI.getFrameLowering();
     53   const TargetRegisterInfo &TRI = *STI.getRegisterInfo();
     54   bool Modified = false;
     55 
     56   calculateFrameObjectOffsets(MF);
     57 
     58   for (MachineFunction::iterator BB = MF.begin(), E = MF.end(); BB != E; ++BB) {
     59     for (MachineBasicBlock::iterator I = BB->begin(); I != BB->end(); ++I) {
     60       MachineInstr *MI = I;
     61       for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
     62         if (!MI->getOperand(i).isFI())
     63           continue;
     64         TRI.eliminateFrameIndex(MI, 0, i, nullptr);
     65         Modified = true;
     66       }
     67     }
     68   }
     69 
     70   // Add function prolog/epilog
     71   TFI.emitPrologue(MF, MF.front());
     72 
     73   for (MachineFunction::iterator I = MF.begin(), E = MF.end(); I != E; ++I) {
     74     // If last instruction is a return instruction, add an epilogue
     75     if (I->isReturnBlock())
     76       TFI.emitEpilogue(MF, *I);
     77   }
     78 
     79   return Modified;
     80 }
     81 
     82 /// AdjustStackOffset - Helper function used to adjust the stack frame offset.
     83 static inline void
     84 AdjustStackOffset(MachineFrameInfo *MFI, int FrameIdx,
     85                   bool StackGrowsDown, int64_t &Offset,
     86                   unsigned &MaxAlign) {
     87   // If the stack grows down, add the object size to find the lowest address.
     88   if (StackGrowsDown)
     89     Offset += MFI->getObjectSize(FrameIdx);
     90 
     91   unsigned Align = MFI->getObjectAlignment(FrameIdx);
     92 
     93   // If the alignment of this object is greater than that of the stack, then
     94   // increase the stack alignment to match.
     95   MaxAlign = std::max(MaxAlign, Align);
     96 
     97   // Adjust to alignment boundary.
     98   Offset = (Offset + Align - 1) / Align * Align;
     99 
    100   if (StackGrowsDown) {
    101     DEBUG(dbgs() << "alloc FI(" << FrameIdx << ") at SP[" << -Offset << "]\n");
    102     MFI->setObjectOffset(FrameIdx, -Offset); // Set the computed offset
    103   } else {
    104     DEBUG(dbgs() << "alloc FI(" << FrameIdx << ") at SP[" << Offset << "]\n");
    105     MFI->setObjectOffset(FrameIdx, Offset);
    106     Offset += MFI->getObjectSize(FrameIdx);
    107   }
    108 }
    109 
    110 void
    111 NVPTXPrologEpilogPass::calculateFrameObjectOffsets(MachineFunction &Fn) {
    112   const TargetFrameLowering &TFI = *Fn.getSubtarget().getFrameLowering();
    113   const TargetRegisterInfo *RegInfo = Fn.getSubtarget().getRegisterInfo();
    114 
    115   bool StackGrowsDown =
    116     TFI.getStackGrowthDirection() == TargetFrameLowering::StackGrowsDown;
    117 
    118   // Loop over all of the stack objects, assigning sequential addresses...
    119   MachineFrameInfo *MFI = Fn.getFrameInfo();
    120 
    121   // Start at the beginning of the local area.
    122   // The Offset is the distance from the stack top in the direction
    123   // of stack growth -- so it's always nonnegative.
    124   int LocalAreaOffset = TFI.getOffsetOfLocalArea();
    125   if (StackGrowsDown)
    126     LocalAreaOffset = -LocalAreaOffset;
    127   assert(LocalAreaOffset >= 0
    128          && "Local area offset should be in direction of stack growth");
    129   int64_t Offset = LocalAreaOffset;
    130 
    131   // If there are fixed sized objects that are preallocated in the local area,
    132   // non-fixed objects can't be allocated right at the start of local area.
    133   // We currently don't support filling in holes in between fixed sized
    134   // objects, so we adjust 'Offset' to point to the end of last fixed sized
    135   // preallocated object.
    136   for (int i = MFI->getObjectIndexBegin(); i != 0; ++i) {
    137     int64_t FixedOff;
    138     if (StackGrowsDown) {
    139       // The maximum distance from the stack pointer is at lower address of
    140       // the object -- which is given by offset. For down growing stack
    141       // the offset is negative, so we negate the offset to get the distance.
    142       FixedOff = -MFI->getObjectOffset(i);
    143     } else {
    144       // The maximum distance from the start pointer is at the upper
    145       // address of the object.
    146       FixedOff = MFI->getObjectOffset(i) + MFI->getObjectSize(i);
    147     }
    148     if (FixedOff > Offset) Offset = FixedOff;
    149   }
    150 
    151   // NOTE: We do not have a call stack
    152 
    153   unsigned MaxAlign = MFI->getMaxAlignment();
    154 
    155   // No scavenger
    156 
    157   // FIXME: Once this is working, then enable flag will change to a target
    158   // check for whether the frame is large enough to want to use virtual
    159   // frame index registers. Functions which don't want/need this optimization
    160   // will continue to use the existing code path.
    161   if (MFI->getUseLocalStackAllocationBlock()) {
    162     unsigned Align = MFI->getLocalFrameMaxAlign();
    163 
    164     // Adjust to alignment boundary.
    165     Offset = (Offset + Align - 1) / Align * Align;
    166 
    167     DEBUG(dbgs() << "Local frame base offset: " << Offset << "\n");
    168 
    169     // Resolve offsets for objects in the local block.
    170     for (unsigned i = 0, e = MFI->getLocalFrameObjectCount(); i != e; ++i) {
    171       std::pair<int, int64_t> Entry = MFI->getLocalFrameObjectMap(i);
    172       int64_t FIOffset = (StackGrowsDown ? -Offset : Offset) + Entry.second;
    173       DEBUG(dbgs() << "alloc FI(" << Entry.first << ") at SP[" <<
    174             FIOffset << "]\n");
    175       MFI->setObjectOffset(Entry.first, FIOffset);
    176     }
    177     // Allocate the local block
    178     Offset += MFI->getLocalFrameSize();
    179 
    180     MaxAlign = std::max(Align, MaxAlign);
    181   }
    182 
    183   // No stack protector
    184 
    185   // Then assign frame offsets to stack objects that are not used to spill
    186   // callee saved registers.
    187   for (unsigned i = 0, e = MFI->getObjectIndexEnd(); i != e; ++i) {
    188     if (MFI->isObjectPreAllocated(i) &&
    189         MFI->getUseLocalStackAllocationBlock())
    190       continue;
    191     if (MFI->isDeadObjectIndex(i))
    192       continue;
    193 
    194     AdjustStackOffset(MFI, i, StackGrowsDown, Offset, MaxAlign);
    195   }
    196 
    197   // No scavenger
    198 
    199   if (!TFI.targetHandlesStackFrameRounding()) {
    200     // If we have reserved argument space for call sites in the function
    201     // immediately on entry to the current function, count it as part of the
    202     // overall stack size.
    203     if (MFI->adjustsStack() && TFI.hasReservedCallFrame(Fn))
    204       Offset += MFI->getMaxCallFrameSize();
    205 
    206     // Round up the size to a multiple of the alignment.  If the function has
    207     // any calls or alloca's, align to the target's StackAlignment value to
    208     // ensure that the callee's frame or the alloca data is suitably aligned;
    209     // otherwise, for leaf functions, align to the TransientStackAlignment
    210     // value.
    211     unsigned StackAlign;
    212     if (MFI->adjustsStack() || MFI->hasVarSizedObjects() ||
    213         (RegInfo->needsStackRealignment(Fn) && MFI->getObjectIndexEnd() != 0))
    214       StackAlign = TFI.getStackAlignment();
    215     else
    216       StackAlign = TFI.getTransientStackAlignment();
    217 
    218     // If the frame pointer is eliminated, all frame offsets will be relative to
    219     // SP not FP. Align to MaxAlign so this works.
    220     StackAlign = std::max(StackAlign, MaxAlign);
    221     unsigned AlignMask = StackAlign - 1;
    222     Offset = (Offset + AlignMask) & ~uint64_t(AlignMask);
    223   }
    224 
    225   // Update frame info to pretend that this is part of the stack...
    226   int64_t StackSize = Offset - LocalAreaOffset;
    227   MFI->setStackSize(StackSize);
    228 }
    229