lib/Target/NVPTX/NVPTXPrologEpilogPass.cpp

   1 //===-- NVPTXPrologEpilogPass.cpp - NVPTX prolog/epilog inserter ----------===//
   2 //
   3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
   4 // See https://llvm.org/LICENSE.txt for license information.
   5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
   6 //
   7 //===----------------------------------------------------------------------===//
   8 //
   9 // This file is a copy of the generic LLVM PrologEpilogInserter pass, modified
  10 // to remove unneeded functionality and to handle virtual registers. Most code
  11 // here is a copy of PrologEpilogInserter.cpp.
  12 //
  13 //===----------------------------------------------------------------------===//
  14
  15 #include "NVPTX.h"
  16 #include "llvm/CodeGen/MachineFrameInfo.h"
  17 #include "llvm/CodeGen/MachineFunction.h"
  18 #include "llvm/CodeGen/MachineFunctionPass.h"
  19 #include "llvm/CodeGen/TargetFrameLowering.h"
  20 #include "llvm/CodeGen/TargetRegisterInfo.h"
  21 #include "llvm/CodeGen/TargetSubtargetInfo.h"
  22 #include "llvm/IR/DebugInfoMetadata.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
  64         // Frame indices in debug values are encoded in a target independent
  65         // way with simply the frame index and offset rather than any
  66         // target-specific addressing mode.
  67         if (MI.isDebugValue()) {
  68           assert(i == 0 && "Frame indices can only appear as the first "
  69                            "operand of a DBG_VALUE machine instruction");
  70           unsigned Reg;
  71           int64_t Offset =
  72               TFI.getFrameIndexReference(MF, MI.getOperand(0).getIndex(), Reg);
  73           MI.getOperand(0).ChangeToRegister(Reg, /*isDef=*/false);
  74           MI.getOperand(0).setIsDebug();
  75           auto *DIExpr = DIExpression::prepend(
  76               MI.getDebugExpression(), DIExpression::ApplyOffset, Offset);
  77           MI.getOperand(3).setMetadata(DIExpr);
  78           continue;
  79         }
  80
  81         TRI.eliminateFrameIndex(MI, 0, i, nullptr);
  82         Modified = true;
  83       }
  84     }
  85   }
  86
  87   // Add function prolog/epilog
  88   TFI.emitPrologue(MF, MF.front());
  89
  90   for (MachineFunction::iterator I = MF.begin(), E = MF.end(); I != E; ++I) {
  91     // If last instruction is a return instruction, add an epilogue
  92     if (I->isReturnBlock())
  93       TFI.emitEpilogue(MF, *I);
  94   }
  95
  96   return Modified;
  97 }
  98
  99 /// AdjustStackOffset - Helper function used to adjust the stack frame offset.
 100 static inline void
 101 AdjustStackOffset(MachineFrameInfo &MFI, int FrameIdx,
 102                   bool StackGrowsDown, int64_t &Offset,
 103                   unsigned &MaxAlign) {
 104   // If the stack grows down, add the object size to find the lowest address.
 105   if (StackGrowsDown)
 106     Offset += MFI.getObjectSize(FrameIdx);
 107
 108   unsigned Align = MFI.getObjectAlignment(FrameIdx);
 109
 110   // If the alignment of this object is greater than that of the stack, then
 111   // increase the stack alignment to match.
 112   MaxAlign = std::max(MaxAlign, Align);
 113
 114   // Adjust to alignment boundary.
 115   Offset = (Offset + Align - 1) / Align * Align;
 116
 117   if (StackGrowsDown) {
 118     LLVM_DEBUG(dbgs() << "alloc FI(" << FrameIdx << ") at SP[" << -Offset
 119                       << "]\n");
 120     MFI.setObjectOffset(FrameIdx, -Offset); // Set the computed offset
 121   } else {
 122     LLVM_DEBUG(dbgs() << "alloc FI(" << FrameIdx << ") at SP[" << Offset
 123                       << "]\n");
 124     MFI.setObjectOffset(FrameIdx, Offset);
 125     Offset += MFI.getObjectSize(FrameIdx);
 126   }
 127 }
 128
 129 void
 130 NVPTXPrologEpilogPass::calculateFrameObjectOffsets(MachineFunction &Fn) {
 131   const TargetFrameLowering &TFI = *Fn.getSubtarget().getFrameLowering();
 132   const TargetRegisterInfo *RegInfo = Fn.getSubtarget().getRegisterInfo();
 133
 134   bool StackGrowsDown =
 135     TFI.getStackGrowthDirection() == TargetFrameLowering::StackGrowsDown;
 136
 137   // Loop over all of the stack objects, assigning sequential addresses...
 138   MachineFrameInfo &MFI = Fn.getFrameInfo();
 139
 140   // Start at the beginning of the local area.
 141   // The Offset is the distance from the stack top in the direction
 142   // of stack growth -- so it's always nonnegative.
 143   int LocalAreaOffset = TFI.getOffsetOfLocalArea();
 144   if (StackGrowsDown)
 145     LocalAreaOffset = -LocalAreaOffset;
 146   assert(LocalAreaOffset >= 0
 147          && "Local area offset should be in direction of stack growth");
 148   int64_t Offset = LocalAreaOffset;
 149
 150   // If there are fixed sized objects that are preallocated in the local area,
 151   // non-fixed objects can't be allocated right at the start of local area.
 152   // We currently don't support filling in holes in between fixed sized
 153   // objects, so we adjust 'Offset' to point to the end of last fixed sized
 154   // preallocated object.
 155   for (int i = MFI.getObjectIndexBegin(); i != 0; ++i) {
 156     int64_t FixedOff;
 157     if (StackGrowsDown) {
 158       // The maximum distance from the stack pointer is at lower address of
 159       // the object -- which is given by offset. For down growing stack
 160       // the offset is negative, so we negate the offset to get the distance.
 161       FixedOff = -MFI.getObjectOffset(i);
 162     } else {
 163       // The maximum distance from the start pointer is at the upper
 164       // address of the object.
 165       FixedOff = MFI.getObjectOffset(i) + MFI.getObjectSize(i);
 166     }
 167     if (FixedOff > Offset) Offset = FixedOff;
 168   }
 169
 170   // NOTE: We do not have a call stack
 171
 172   unsigned MaxAlign = MFI.getMaxAlignment();
 173
 174   // No scavenger
 175
 176   // FIXME: Once this is working, then enable flag will change to a target
 177   // check for whether the frame is large enough to want to use virtual
 178   // frame index registers. Functions which don't want/need this optimization
 179   // will continue to use the existing code path.
 180   if (MFI.getUseLocalStackAllocationBlock()) {
 181     unsigned Align = MFI.getLocalFrameMaxAlign().value();
 182
 183     // Adjust to alignment boundary.
 184     Offset = (Offset + Align - 1) / Align * Align;
 185
 186     LLVM_DEBUG(dbgs() << "Local frame base offset: " << Offset << "\n");
 187
 188     // Resolve offsets for objects in the local block.
 189     for (unsigned i = 0, e = MFI.getLocalFrameObjectCount(); i != e; ++i) {
 190       std::pair<int, int64_t> Entry = MFI.getLocalFrameObjectMap(i);
 191       int64_t FIOffset = (StackGrowsDown ? -Offset : Offset) + Entry.second;
 192       LLVM_DEBUG(dbgs() << "alloc FI(" << Entry.first << ") at SP[" << FIOffset
 193                         << "]\n");
 194       MFI.setObjectOffset(Entry.first, FIOffset);
 195     }
 196     // Allocate the local block
 197     Offset += MFI.getLocalFrameSize();
 198
 199     MaxAlign = std::max(Align, MaxAlign);
 200   }
 201
 202   // No stack protector
 203
 204   // Then assign frame offsets to stack objects that are not used to spill
 205   // callee saved registers.
 206   for (unsigned i = 0, e = MFI.getObjectIndexEnd(); i != e; ++i) {
 207     if (MFI.isObjectPreAllocated(i) &&
 208         MFI.getUseLocalStackAllocationBlock())
 209       continue;
 210     if (MFI.isDeadObjectIndex(i))
 211       continue;
 212
 213     AdjustStackOffset(MFI, i, StackGrowsDown, Offset, MaxAlign);
 214   }
 215
 216   // No scavenger
 217
 218   if (!TFI.targetHandlesStackFrameRounding()) {
 219     // If we have reserved argument space for call sites in the function
 220     // immediately on entry to the current function, count it as part of the
 221     // overall stack size.
 222     if (MFI.adjustsStack() && TFI.hasReservedCallFrame(Fn))
 223       Offset += MFI.getMaxCallFrameSize();
 224
 225     // Round up the size to a multiple of the alignment.  If the function has
 226     // any calls or alloca's, align to the target's StackAlignment value to
 227     // ensure that the callee's frame or the alloca data is suitably aligned;
 228     // otherwise, for leaf functions, align to the TransientStackAlignment
 229     // value.
 230     unsigned StackAlign;
 231     if (MFI.adjustsStack() || MFI.hasVarSizedObjects() ||
 232         (RegInfo->needsStackRealignment(Fn) && MFI.getObjectIndexEnd() != 0))
 233       StackAlign = TFI.getStackAlignment();
 234     else
 235       StackAlign = TFI.getTransientStackAlignment();
 236
 237     // If the frame pointer is eliminated, all frame offsets will be relative to
 238     // SP not FP. Align to MaxAlign so this works.
 239     StackAlign = std::max(StackAlign, MaxAlign);
 240     unsigned AlignMask = StackAlign - 1;
 241     Offset = (Offset + AlignMask) & ~uint64_t(AlignMask);
 242   }
 243
 244   // Update frame info to pretend that this is part of the stack...
 245   int64_t StackSize = Offset - LocalAreaOffset;
 246   MFI.setStackSize(StackSize);
 247 }