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
27using namespace llvm;
28
29#define DEBUG_TYPE "nvptx-prolog-epilog"
30
31namespace {
32class NVPTXPrologEpilog {
33public:
34 bool run(MachineFunction &MF);
35
36private:
37 void calculateFrameObjectOffsets(MachineFunction &Fn);
38};
39} // end anonymous namespace
40
41static bool replaceFrameIndexDebugInstr(MachineFunction &MF, MachineInstr &MI,
42 unsigned OpIdx) {
43 const TargetFrameLowering *TFI = MF.getSubtarget().getFrameLowering();
44 const TargetRegisterInfo &TRI = *MF.getSubtarget().getRegisterInfo();
45 if (MI.isDebugValue()) {
46
47 MachineOperand &Op = MI.getOperand(i: OpIdx);
48 assert(MI.isDebugOperand(&Op) &&
49 "Frame indices can only appear as a debug operand in a DBG_VALUE*"
50 " machine instruction");
51 Register Reg;
52 unsigned FrameIdx = Op.getIndex();
53
54 StackOffset Offset = TFI->getFrameIndexReference(MF, FI: FrameIdx, FrameReg&: Reg);
55 Op.ChangeToRegister(Reg, isDef: false /*isDef*/);
56
57 const DIExpression *DIExpr = MI.getDebugExpression();
58 if (MI.isNonListDebugValue()) {
59 DIExpr = TRI.prependOffsetExpression(Expr: MI.getDebugExpression(),
60 PrependFlags: DIExpression::ApplyOffset, Offset);
61 } else {
62 // The debug operand at DebugOpIndex was a frame index at offset
63 // `Offset`; now the operand has been replaced with the frame
64 // register, we must add Offset with `register x, plus Offset`.
65 unsigned DebugOpIndex = MI.getDebugOperandIndex(Op: &Op);
66 SmallVector<uint64_t, 3> Ops;
67 TRI.getOffsetOpcodes(Offset, Ops);
68 DIExpr = DIExpression::appendOpsToArg(Expr: DIExpr, Ops, ArgNo: DebugOpIndex);
69 }
70 MI.getDebugExpressionOp().setMetadata(DIExpr);
71 return true;
72 }
73 return false;
74}
75
76bool NVPTXPrologEpilog::run(MachineFunction &MF) {
77 const TargetSubtargetInfo &STI = MF.getSubtarget();
78 const TargetFrameLowering &TFI = *STI.getFrameLowering();
79 const TargetRegisterInfo &TRI = *STI.getRegisterInfo();
80 bool Modified = false;
81
82 calculateFrameObjectOffsets(Fn&: MF);
83
84 for (MachineBasicBlock &BB : MF) {
85 for (MachineBasicBlock::iterator I = BB.end(); I != BB.begin();) {
86 MachineInstr &MI = *std::prev(x: I);
87
88 bool RemovedMI = false;
89 for (const auto &[Idx, Op] : enumerate(First: MI.operands())) {
90 if (!Op.isFI())
91 continue;
92
93 if (replaceFrameIndexDebugInstr(MF, MI, OpIdx: Idx))
94 continue;
95
96 // Eliminate this FrameIndex operand.
97 RemovedMI = TRI.eliminateFrameIndex(MI, SPAdj: 0, FIOperandNum: Idx, RS: nullptr);
98 Modified = true;
99 if (RemovedMI)
100 break;
101 }
102
103 if (!RemovedMI)
104 --I;
105 }
106 }
107
108 // Add function prolog/epilog
109 TFI.emitPrologue(MF, MBB&: MF.front());
110
111 for (MachineFunction::iterator I = MF.begin(), E = MF.end(); I != E; ++I) {
112 // If last instruction is a return instruction, add an epilogue
113 if (I->isReturnBlock())
114 TFI.emitEpilogue(MF, MBB&: *I);
115 }
116
117 return Modified;
118}
119
120/// AdjustStackOffset - Helper function used to adjust the stack frame offset.
121static inline void AdjustStackOffset(MachineFrameInfo &MFI, int FrameIdx,
122 bool StackGrowsDown, int64_t &Offset,
123 Align &MaxAlign) {
124 // If the stack grows down, add the object size to find the lowest address.
125 if (StackGrowsDown)
126 Offset += MFI.getObjectSize(ObjectIdx: FrameIdx);
127
128 Align Alignment = MFI.getObjectAlign(ObjectIdx: FrameIdx);
129
130 // If the alignment of this object is greater than that of the stack, then
131 // increase the stack alignment to match.
132 MaxAlign = std::max(a: MaxAlign, b: Alignment);
133
134 // Adjust to alignment boundary.
135 Offset = alignTo(Size: Offset, A: Alignment);
136
137 if (StackGrowsDown) {
138 LLVM_DEBUG(dbgs() << "alloc FI(" << FrameIdx << ") at SP[" << -Offset
139 << "]\n");
140 MFI.setObjectOffset(ObjectIdx: FrameIdx, SPOffset: -Offset); // Set the computed offset
141 } else {
142 LLVM_DEBUG(dbgs() << "alloc FI(" << FrameIdx << ") at SP[" << Offset
143 << "]\n");
144 MFI.setObjectOffset(ObjectIdx: FrameIdx, SPOffset: Offset);
145 Offset += MFI.getObjectSize(ObjectIdx: FrameIdx);
146 }
147}
148
149void NVPTXPrologEpilog::calculateFrameObjectOffsets(MachineFunction &Fn) {
150 const TargetFrameLowering &TFI = *Fn.getSubtarget().getFrameLowering();
151 const TargetRegisterInfo *RegInfo = Fn.getSubtarget().getRegisterInfo();
152
153 bool StackGrowsDown =
154 TFI.getStackGrowthDirection() == TargetFrameLowering::StackGrowsDown;
155
156 // Loop over all of the stack objects, assigning sequential addresses...
157 MachineFrameInfo &MFI = Fn.getFrameInfo();
158
159 // Start at the beginning of the local area.
160 // The Offset is the distance from the stack top in the direction
161 // of stack growth -- so it's always nonnegative.
162 int LocalAreaOffset = TFI.getOffsetOfLocalArea();
163 if (StackGrowsDown)
164 LocalAreaOffset = -LocalAreaOffset;
165 assert(LocalAreaOffset >= 0
166 && "Local area offset should be in direction of stack growth");
167 int64_t Offset = LocalAreaOffset;
168
169 // If there are fixed sized objects that are preallocated in the local area,
170 // non-fixed objects can't be allocated right at the start of local area.
171 // We currently don't support filling in holes in between fixed sized
172 // objects, so we adjust 'Offset' to point to the end of last fixed sized
173 // preallocated object.
174 for (int i = MFI.getObjectIndexBegin(); i != 0; ++i) {
175 int64_t FixedOff;
176 if (StackGrowsDown) {
177 // The maximum distance from the stack pointer is at lower address of
178 // the object -- which is given by offset. For down growing stack
179 // the offset is negative, so we negate the offset to get the distance.
180 FixedOff = -MFI.getObjectOffset(ObjectIdx: i);
181 } else {
182 // The maximum distance from the start pointer is at the upper
183 // address of the object.
184 FixedOff = MFI.getObjectOffset(ObjectIdx: i) + MFI.getObjectSize(ObjectIdx: i);
185 }
186 if (FixedOff > Offset) Offset = FixedOff;
187 }
188
189 // NOTE: We do not have a call stack
190
191 Align MaxAlign = MFI.getMaxAlign();
192
193 // No scavenger
194
195 // FIXME: Once this is working, then enable flag will change to a target
196 // check for whether the frame is large enough to want to use virtual
197 // frame index registers. Functions which don't want/need this optimization
198 // will continue to use the existing code path.
199 if (MFI.getUseLocalStackAllocationBlock()) {
200 Align Alignment = MFI.getLocalFrameMaxAlign();
201
202 // Adjust to alignment boundary.
203 Offset = alignTo(Size: Offset, A: Alignment);
204
205 LLVM_DEBUG(dbgs() << "Local frame base offset: " << Offset << "\n");
206
207 // Resolve offsets for objects in the local block.
208 for (unsigned i = 0, e = MFI.getLocalFrameObjectCount(); i != e; ++i) {
209 std::pair<int, int64_t> Entry = MFI.getLocalFrameObjectMap(i);
210 int64_t FIOffset = (StackGrowsDown ? -Offset : Offset) + Entry.second;
211 LLVM_DEBUG(dbgs() << "alloc FI(" << Entry.first << ") at SP[" << FIOffset
212 << "]\n");
213 MFI.setObjectOffset(ObjectIdx: Entry.first, SPOffset: FIOffset);
214 }
215 // Allocate the local block
216 Offset += MFI.getLocalFrameSize();
217
218 MaxAlign = std::max(a: Alignment, b: MaxAlign);
219 }
220
221 // No stack protector
222
223 // Then assign frame offsets to stack objects that are not used to spill
224 // callee saved registers.
225 for (unsigned i = 0, e = MFI.getObjectIndexEnd(); i != e; ++i) {
226 if (MFI.isObjectPreAllocated(ObjectIdx: i) &&
227 MFI.getUseLocalStackAllocationBlock())
228 continue;
229 if (MFI.isDeadObjectIndex(ObjectIdx: i))
230 continue;
231
232 AdjustStackOffset(MFI, FrameIdx: i, StackGrowsDown, Offset, MaxAlign);
233 }
234
235 // No scavenger
236
237 if (!TFI.targetHandlesStackFrameRounding()) {
238 // If we have reserved argument space for call sites in the function
239 // immediately on entry to the current function, count it as part of the
240 // overall stack size.
241 if (MFI.adjustsStack() && TFI.hasReservedCallFrame(MF: Fn))
242 Offset += MFI.getMaxCallFrameSize();
243
244 // Round up the size to a multiple of the alignment. If the function has
245 // any calls or alloca's, align to the target's StackAlignment value to
246 // ensure that the callee's frame or the alloca data is suitably aligned;
247 // otherwise, for leaf functions, align to the TransientStackAlignment
248 // value.
249 Align StackAlign;
250 if (MFI.adjustsStack() || MFI.hasVarSizedObjects() ||
251 (RegInfo->hasStackRealignment(MF: Fn) && MFI.getObjectIndexEnd() != 0))
252 StackAlign = TFI.getStackAlign();
253 else
254 StackAlign = TFI.getTransientStackAlign();
255
256 // If the frame pointer is eliminated, all frame offsets will be relative to
257 // SP not FP. Align to MaxAlign so this works.
258 Offset = alignTo(Size: Offset, A: std::max(a: StackAlign, b: MaxAlign));
259 }
260
261 // Update frame info to pretend that this is part of the stack...
262 int64_t StackSize = Offset - LocalAreaOffset;
263 MFI.setStackSize(StackSize);
264}
265
266namespace {
267class NVPTXPrologEpilogLegacyPass : public MachineFunctionPass {
268public:
269 static char ID;
270 NVPTXPrologEpilogLegacyPass() : MachineFunctionPass(ID) {}
271
272 bool runOnMachineFunction(MachineFunction &MF) override {
273 return NVPTXPrologEpilog().run(MF);
274 }
275
276 StringRef getPassName() const override { return "NVPTX Prolog Epilog Pass"; }
277
278 void getAnalysisUsage(AnalysisUsage &AU) const override {
279 AU.setPreservesCFG();
280 MachineFunctionPass::getAnalysisUsage(AU);
281 }
282};
283} // end anonymous namespace
284
285char NVPTXPrologEpilogLegacyPass::ID = 0;
286
287INITIALIZE_PASS(NVPTXPrologEpilogLegacyPass, DEBUG_TYPE,
288 "NVPTX Prologue/Epilogue Insertion", false, false)
289
290MachineFunctionPass *llvm::createNVPTXPrologEpilogLegacyPass() {
291 return new NVPTXPrologEpilogLegacyPass();
292}
293
294PreservedAnalyses
295NVPTXPrologEpilogPass::run(MachineFunction &MF,
296 MachineFunctionAnalysisManager &MFAM) {
297 if (!NVPTXPrologEpilog().run(MF))
298 return PreservedAnalyses::all();
299 return getMachineFunctionPassPreservedAnalyses().preserveSet<CFGAnalyses>();
300}
301