1 //===-- AArch64StackTaggingPreRA.cpp --- Stack Tagging for AArch64 -----===//
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
7 //===----------------------------------------------------------------------===//
10 #include "AArch64InstrInfo.h"
11 #include "AArch64MachineFunctionInfo.h"
12 #include "llvm/ADT/SetVector.h"
13 #include "llvm/ADT/Statistic.h"
14 #include "llvm/CodeGen/MachineFrameInfo.h"
15 #include "llvm/CodeGen/MachineFunction.h"
16 #include "llvm/CodeGen/MachineFunctionPass.h"
17 #include "llvm/CodeGen/MachineInstrBuilder.h"
18 #include "llvm/CodeGen/MachineRegisterInfo.h"
19 #include "llvm/CodeGen/MachineTraceMetrics.h"
20 #include "llvm/CodeGen/Passes.h"
21 #include "llvm/CodeGen/TargetInstrInfo.h"
22 #include "llvm/CodeGen/TargetRegisterInfo.h"
23 #include "llvm/CodeGen/TargetSubtargetInfo.h"
24 #include "llvm/Support/CommandLine.h"
25 #include "llvm/Support/Debug.h"
26 #include "llvm/Support/raw_ostream.h"
30 #define DEBUG_TYPE "aarch64-stack-tagging-pre-ra"
32 enum UncheckedLdStMode
{ UncheckedNever
, UncheckedSafe
, UncheckedAlways
};
34 cl::opt
<UncheckedLdStMode
> ClUncheckedLdSt(
35 "stack-tagging-unchecked-ld-st", cl::Hidden
,
36 cl::init(UncheckedSafe
),
38 "Unconditionally apply unchecked-ld-st optimization (even for large "
39 "stack frames, or in the presence of variable sized allocas)."),
41 clEnumValN(UncheckedNever
, "never", "never apply unchecked-ld-st"),
43 UncheckedSafe
, "safe",
44 "apply unchecked-ld-st when the target is definitely within range"),
45 clEnumValN(UncheckedAlways
, "always", "always apply unchecked-ld-st")));
48 ClFirstSlot("stack-tagging-first-slot-opt", cl::Hidden
, cl::init(true),
49 cl::desc("Apply first slot optimization for stack tagging "
50 "(eliminate ADDG Rt, Rn, 0, 0)."));
54 class AArch64StackTaggingPreRA
: public MachineFunctionPass
{
56 AArch64FunctionInfo
*AFI
;
57 MachineFrameInfo
*MFI
;
58 MachineRegisterInfo
*MRI
;
59 const AArch64RegisterInfo
*TRI
;
60 const AArch64InstrInfo
*TII
;
62 SmallVector
<MachineInstr
*, 16> ReTags
;
66 AArch64StackTaggingPreRA() : MachineFunctionPass(ID
) {
67 initializeAArch64StackTaggingPreRAPass(*PassRegistry::getPassRegistry());
70 bool mayUseUncheckedLoadStore();
71 void uncheckUsesOf(unsigned TaggedReg
, int FI
);
72 void uncheckLoadsAndStores();
73 std::optional
<int> findFirstSlotCandidate();
75 bool runOnMachineFunction(MachineFunction
&Func
) override
;
76 StringRef
getPassName() const override
{
77 return "AArch64 Stack Tagging PreRA";
80 void getAnalysisUsage(AnalysisUsage
&AU
) const override
{
82 MachineFunctionPass::getAnalysisUsage(AU
);
85 } // end anonymous namespace
87 char AArch64StackTaggingPreRA::ID
= 0;
89 INITIALIZE_PASS_BEGIN(AArch64StackTaggingPreRA
, "aarch64-stack-tagging-pre-ra",
90 "AArch64 Stack Tagging PreRA Pass", false, false)
91 INITIALIZE_PASS_END(AArch64StackTaggingPreRA
, "aarch64-stack-tagging-pre-ra",
92 "AArch64 Stack Tagging PreRA Pass", false, false)
94 FunctionPass
*llvm::createAArch64StackTaggingPreRAPass() {
95 return new AArch64StackTaggingPreRA();
98 static bool isUncheckedLoadOrStoreOpcode(unsigned Opcode
) {
100 case AArch64::LDRBBui
:
101 case AArch64::LDRHHui
:
102 case AArch64::LDRWui
:
103 case AArch64::LDRXui
:
105 case AArch64::LDRBui
:
106 case AArch64::LDRHui
:
107 case AArch64::LDRSui
:
108 case AArch64::LDRDui
:
109 case AArch64::LDRQui
:
111 case AArch64::LDRSHWui
:
112 case AArch64::LDRSHXui
:
114 case AArch64::LDRSBWui
:
115 case AArch64::LDRSBXui
:
117 case AArch64::LDRSWui
:
119 case AArch64::STRBBui
:
120 case AArch64::STRHHui
:
121 case AArch64::STRWui
:
122 case AArch64::STRXui
:
124 case AArch64::STRBui
:
125 case AArch64::STRHui
:
126 case AArch64::STRSui
:
127 case AArch64::STRDui
:
128 case AArch64::STRQui
:
136 case AArch64::LDPSWi
:
149 bool AArch64StackTaggingPreRA::mayUseUncheckedLoadStore() {
150 if (ClUncheckedLdSt
== UncheckedNever
)
152 else if (ClUncheckedLdSt
== UncheckedAlways
)
155 // This estimate can be improved if we had harder guarantees about stack frame
156 // layout. With LocalStackAllocation we can estimate SP offset to any
157 // preallocated slot. AArch64FrameLowering::orderFrameObjects could put tagged
158 // objects ahead of non-tagged ones, but that's not always desirable.
160 // Underestimating SP offset here may require the use of LDG to materialize
161 // the tagged address of the stack slot, along with a scratch register
162 // allocation (post-regalloc!).
164 // For now we do the safe thing here and require that the entire stack frame
165 // is within range of the shortest of the unchecked instructions.
166 unsigned FrameSize
= 0;
167 for (unsigned i
= 0, e
= MFI
->getObjectIndexEnd(); i
!= e
; ++i
)
168 FrameSize
+= MFI
->getObjectSize(i
);
169 bool EntireFrameReachableFromSP
= FrameSize
< 0xf00;
170 return !MFI
->hasVarSizedObjects() && EntireFrameReachableFromSP
;
173 void AArch64StackTaggingPreRA::uncheckUsesOf(unsigned TaggedReg
, int FI
) {
174 for (MachineInstr
&UseI
:
175 llvm::make_early_inc_range(MRI
->use_instructions(TaggedReg
))) {
176 if (isUncheckedLoadOrStoreOpcode(UseI
.getOpcode())) {
177 // FI operand is always the one before the immediate offset.
178 unsigned OpIdx
= TII
->getLoadStoreImmIdx(UseI
.getOpcode()) - 1;
179 if (UseI
.getOperand(OpIdx
).isReg() &&
180 UseI
.getOperand(OpIdx
).getReg() == TaggedReg
) {
181 UseI
.getOperand(OpIdx
).ChangeToFrameIndex(FI
);
182 UseI
.getOperand(OpIdx
).setTargetFlags(AArch64II::MO_TAGGED
);
184 } else if (UseI
.isCopy() && UseI
.getOperand(0).getReg().isVirtual()) {
185 uncheckUsesOf(UseI
.getOperand(0).getReg(), FI
);
190 void AArch64StackTaggingPreRA::uncheckLoadsAndStores() {
191 for (auto *I
: ReTags
) {
192 Register TaggedReg
= I
->getOperand(0).getReg();
193 int FI
= I
->getOperand(1).getIndex();
194 uncheckUsesOf(TaggedReg
, FI
);
202 SlotWithTag(int FI
, int Tag
) : FI(FI
), Tag(Tag
) {}
203 explicit SlotWithTag(const MachineInstr
&MI
)
204 : FI(MI
.getOperand(1).getIndex()), Tag(MI
.getOperand(4).getImm()) {}
205 bool operator==(const SlotWithTag
&Other
) const {
206 return FI
== Other
.FI
&& Tag
== Other
.Tag
;
212 template <> struct DenseMapInfo
<SlotWithTag
> {
213 static inline SlotWithTag
getEmptyKey() { return {-2, -2}; }
214 static inline SlotWithTag
getTombstoneKey() { return {-3, -3}; }
215 static unsigned getHashValue(const SlotWithTag
&V
) {
216 return hash_combine(DenseMapInfo
<int>::getHashValue(V
.FI
),
217 DenseMapInfo
<int>::getHashValue(V
.Tag
));
219 static bool isEqual(const SlotWithTag
&A
, const SlotWithTag
&B
) {
225 static bool isSlotPreAllocated(MachineFrameInfo
*MFI
, int FI
) {
226 return MFI
->getUseLocalStackAllocationBlock() &&
227 MFI
->isObjectPreAllocated(FI
);
230 // Pin one of the tagged slots to offset 0 from the tagged base pointer.
231 // This would make its address available in a virtual register (IRG's def), as
232 // opposed to requiring an ADDG instruction to materialize. This effectively
233 // eliminates a vreg (by replacing it with direct uses of IRG, which is usually
234 // live almost everywhere anyway), and therefore needs to happen before
236 std::optional
<int> AArch64StackTaggingPreRA::findFirstSlotCandidate() {
237 // Find the best (FI, Tag) pair to pin to offset 0.
238 // Looking at the possible uses of a tagged address, the advantage of pinning
240 // - COPY to physical register.
241 // Does not matter, this would trade a MOV instruction for an ADDG.
242 // - ST*G matter, but those mostly appear near the function prologue where all
243 // the tagged addresses need to be materialized anyway; also, counting ST*G
244 // uses would overweight large allocas that require more than one ST*G
246 // - Load/Store instructions in the address operand do not require a tagged
247 // pointer, so they also do not benefit. These operands have already been
248 // eliminated (see uncheckLoadsAndStores) so all remaining load/store
249 // instructions count.
250 // - Any other instruction may benefit from being pinned to offset 0.
251 LLVM_DEBUG(dbgs() << "AArch64StackTaggingPreRA::findFirstSlotCandidate\n");
255 DenseMap
<SlotWithTag
, int> RetagScore
;
256 SlotWithTag MaxScoreST
{-1, -1};
258 for (auto *I
: ReTags
) {
260 if (isSlotPreAllocated(MFI
, ST
.FI
))
263 Register RetagReg
= I
->getOperand(0).getReg();
264 if (!RetagReg
.isVirtual())
268 SmallVector
<Register
, 8> WorkList
;
269 WorkList
.push_back(RetagReg
);
271 while (!WorkList
.empty()) {
272 Register UseReg
= WorkList
.pop_back_val();
273 for (auto &UseI
: MRI
->use_instructions(UseReg
)) {
274 unsigned Opcode
= UseI
.getOpcode();
275 if (Opcode
== AArch64::STGi
|| Opcode
== AArch64::ST2Gi
||
276 Opcode
== AArch64::STZGi
|| Opcode
== AArch64::STZ2Gi
||
277 Opcode
== AArch64::STGPi
|| Opcode
== AArch64::STGloop
||
278 Opcode
== AArch64::STZGloop
|| Opcode
== AArch64::STGloop_wback
||
279 Opcode
== AArch64::STZGloop_wback
)
282 Register DstReg
= UseI
.getOperand(0).getReg();
283 if (DstReg
.isVirtual())
284 WorkList
.push_back(DstReg
);
287 LLVM_DEBUG(dbgs() << "[" << ST
.FI
<< ":" << ST
.Tag
<< "] use of %"
288 << Register::virtReg2Index(UseReg
) << " in " << UseI
294 int TotalScore
= RetagScore
[ST
] += Score
;
295 if (TotalScore
> MaxScore
||
296 (TotalScore
== MaxScore
&& ST
.FI
> MaxScoreST
.FI
)) {
297 MaxScore
= TotalScore
;
302 if (MaxScoreST
.FI
< 0)
305 // If FI's tag is already 0, we are done.
306 if (MaxScoreST
.Tag
== 0)
307 return MaxScoreST
.FI
;
309 // Otherwise, find a random victim pair (FI, Tag) where Tag == 0.
310 SlotWithTag SwapST
{-1, -1};
311 for (auto *I
: ReTags
) {
319 // Swap tags between the victim and the highest scoring pair.
320 // If SwapWith is still (-1, -1), that's fine, too - we'll simply take tag for
321 // the highest score slot without changing anything else.
322 for (auto *&I
: ReTags
) {
324 MachineOperand
&TagOp
= I
->getOperand(4);
325 if (ST
== MaxScoreST
) {
327 } else if (ST
== SwapST
) {
328 TagOp
.setImm(MaxScoreST
.Tag
);
331 return MaxScoreST
.FI
;
334 bool AArch64StackTaggingPreRA::runOnMachineFunction(MachineFunction
&Func
) {
336 MRI
= &MF
->getRegInfo();
337 AFI
= MF
->getInfo
<AArch64FunctionInfo
>();
338 TII
= static_cast<const AArch64InstrInfo
*>(MF
->getSubtarget().getInstrInfo());
339 TRI
= static_cast<const AArch64RegisterInfo
*>(
340 MF
->getSubtarget().getRegisterInfo());
341 MFI
= &MF
->getFrameInfo();
344 assert(MRI
->isSSA());
346 LLVM_DEBUG(dbgs() << "********** AArch64 Stack Tagging PreRA **********\n"
347 << "********** Function: " << MF
->getName() << '\n');
349 SmallSetVector
<int, 8> TaggedSlots
;
350 for (auto &BB
: *MF
) {
352 if (I
.getOpcode() == AArch64::TAGPstack
) {
353 ReTags
.push_back(&I
);
354 int FI
= I
.getOperand(1).getIndex();
355 TaggedSlots
.insert(FI
);
356 // There should be no offsets in TAGP yet.
357 assert(I
.getOperand(2).getImm() == 0);
362 // Take over from SSP. It does nothing for tagged slots, and should not really
363 // have been enabled in the first place.
364 for (int FI
: TaggedSlots
)
365 MFI
->setObjectSSPLayout(FI
, MachineFrameInfo::SSPLK_None
);
370 if (mayUseUncheckedLoadStore())
371 uncheckLoadsAndStores();
373 // Find a slot that is used with zero tag offset, like ADDG #fi, 0.
374 // If the base tagged pointer is set up to the address of this slot,
375 // the ADDG instruction can be eliminated.
376 std::optional
<int> BaseSlot
= findFirstSlotCandidate();
378 AFI
->setTaggedBasePointerIndex(*BaseSlot
);
380 for (auto *I
: ReTags
) {
381 int FI
= I
->getOperand(1).getIndex();
382 int Tag
= I
->getOperand(4).getImm();
383 Register Base
= I
->getOperand(3).getReg();
384 if (Tag
== 0 && FI
== BaseSlot
) {
385 BuildMI(*I
->getParent(), I
, {}, TII
->get(AArch64::COPY
),
386 I
->getOperand(0).getReg())
388 I
->eraseFromParent();