lib/Target/NVPTX/NVPTXPrologEpilogPass.cpp

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