Skip to content

Commit c261b5f

Browse files
MacDueMacDue
committed
Fixups
Change-Id: Ifa920748f994e4160978da7d67ceb0c66093f729
1 parent 8d71ee3 commit c261b5f

File tree

3 files changed

+75
-87
lines changed

3 files changed

+75
-87
lines changed

‎llvm/lib/Target/AArch64/AArch64FrameLowering.cpp‎

Lines changed: 65 additions & 76 deletions
Original file line numberDiff line numberDiff line change
@@ -330,13 +330,36 @@ static int64_t getArgumentStackToRestore(MachineFunction &MF,
330330

331331
static bool produceCompactUnwindFrame(MachineFunction &MF);
332332
static bool needsWinCFI(const MachineFunction &MF);
333-
static StackOffset getZPRStackSize(const MachineFunction &MF);
334-
static StackOffset getPPRStackSize(const MachineFunction &MF);
335-
static StackOffset getSVEStackSize(const MachineFunction &MF);
336333
static Register findScratchNonCalleeSaveRegister(MachineBasicBlock *MBB,
337334
bool HasCall = false);
338335
static bool requiresSaveVG(const MachineFunction &MF);
339-
static bool hasSVEStackSize(const MachineFunction &MF);
336+
337+
static unsigned getStackHazardSize(const MachineFunction &MF) {
338+
return MF.getSubtarget<AArch64Subtarget>().getStreamingHazardSize();
339+
}
340+
341+
/// Returns the size of the entire ZPR stackframe (calleesaves + spills).
342+
static StackOffset getZPRStackSize(const MachineFunction &MF) {
343+
const AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>();
344+
return StackOffset::getScalable(AFI->getStackSizeZPR());
345+
}
346+
347+
/// Returns the size of the entire PPR stackframe (calleesaves + spills).
348+
static StackOffset getPPRStackSize(const MachineFunction &MF) {
349+
const AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>();
350+
return StackOffset::getScalable(AFI->getStackSizePPR());
351+
}
352+
353+
/// Returns the size of the entire SVE stackframe (PPRs + ZPRs).
354+
static StackOffset getSVEStackSize(const MachineFunction &MF) {
355+
return getZPRStackSize(MF) + getPPRStackSize(MF);
356+
}
357+
358+
/// Returns true if PPRs are spilled as ZPRs.
359+
static bool arePPRsSpilledAsZPR(const MachineFunction &MF) {
360+
return MF.getSubtarget().getRegisterInfo()->getSpillSize(
361+
AArch64::PPRRegClass) == 16;
362+
}
340363

341364
/// Returns true if a homogeneous prolog or epilog code can be emitted
342365
/// for the size optimization. If possible, a frame helper call is injected.
@@ -353,8 +376,10 @@ bool AArch64FrameLowering::homogeneousPrologEpilog(
353376
// TODO: Window is supported yet.
354377
if (needsWinCFI(MF))
355378
return false;
379+
356380
// TODO: SVE is not supported yet.
357-
if (hasSVEStackSize(MF))
381+
auto *AFI = MF.getInfo<AArch64FunctionInfo>();
382+
if (AFI->hasSVEStackSize())
358383
return false;
359384

360385
// Bail on stack adjustment needed on return for simplicity.
@@ -365,7 +390,6 @@ bool AArch64FrameLowering::homogeneousPrologEpilog(
365390
if (Exit && getArgumentStackToRestore(MF, *Exit))
366391
return false;
367392

368-
auto *AFI = MF.getInfo<AArch64FunctionInfo>();
369393
if (AFI->hasSwiftAsyncContext() || AFI->hasStreamingModeChanges())
370394
return false;
371395

@@ -454,38 +478,6 @@ static unsigned getFixedObjectSize(const MachineFunction &MF,
454478
}
455479
}
456480

457-
static unsigned getStackHazardSize(const MachineFunction &MF) {
458-
return MF.getSubtarget<AArch64Subtarget>().getStreamingHazardSize();
459-
}
460-
461-
/// Returns the size of the entire ZPR stackframe (calleesaves + spills).
462-
static StackOffset getZPRStackSize(const MachineFunction &MF) {
463-
const AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>();
464-
return StackOffset::getScalable(AFI->getStackSizeZPR());
465-
}
466-
467-
/// Returns the size of the entire PPR stackframe (calleesaves + spills).
468-
static StackOffset getPPRStackSize(const MachineFunction &MF) {
469-
const AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>();
470-
return StackOffset::getScalable(AFI->getStackSizePPR());
471-
}
472-
473-
/// Returns the size of the entire SVE stackframe (PPRs + ZPRs).
474-
static StackOffset getSVEStackSize(const MachineFunction &MF) {
475-
return getZPRStackSize(MF) + getPPRStackSize(MF);
476-
}
477-
478-
static bool hasSVEStackSize(const MachineFunction &MF) {
479-
const AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>();
480-
return AFI->getStackSizeZPR() > 0 || AFI->getStackSizePPR() > 0;
481-
}
482-
483-
/// Returns true if PPRs are spilled as ZPRs.
484-
static bool arePPRsSpilledAsZPR(const MachineFunction &MF) {
485-
return MF.getSubtarget().getRegisterInfo()->getSpillSize(
486-
AArch64::PPRRegClass) == 16;
487-
}
488-
489481
bool AArch64FrameLowering::canUseRedZone(const MachineFunction &MF) const {
490482
if (!EnableRedZone)
491483
return false;
@@ -511,7 +503,7 @@ bool AArch64FrameLowering::canUseRedZone(const MachineFunction &MF) const {
511503
!Subtarget.hasSVE();
512504

513505
return !(MFI.hasCalls() || hasFP(MF) || NumBytes > RedZoneSize ||
514-
hasSVEStackSize(MF) || LowerQRegCopyThroughMem);
506+
AFI->hasSVEStackSize() || LowerQRegCopyThroughMem);
515507
}
516508

517509
/// hasFPImpl - Return true if the specified function should have a dedicated
@@ -1190,7 +1182,7 @@ bool AArch64FrameLowering::shouldCombineCSRLocalStackBump(
11901182

11911183
// When there is an SVE area on the stack, always allocate the
11921184
// callee-saves and spills/locals separately.
1193-
if (hasSVEStackSize(MF))
1185+
if (AFI->hasSVEStackSize())
11941186
return false;
11951187

11961188
return true;
@@ -1634,8 +1626,8 @@ static bool isTargetWindows(const MachineFunction &MF) {
16341626
return MF.getSubtarget<AArch64Subtarget>().isTargetWindows();
16351627
}
16361628

1637-
// Convenience function to determine whether I is an SVE callee save.
1638-
static bool IsZPRCalleeSave(MachineBasicBlock::iterator I) {
1629+
// Convenience function to determine whether I is part of the ZPR callee saves.
1630+
static bool isPartOfZPRCalleeSaves(MachineBasicBlock::iterator I) {
16391631
switch (I->getOpcode()) {
16401632
default:
16411633
return false;
@@ -1655,8 +1647,8 @@ static bool IsZPRCalleeSave(MachineBasicBlock::iterator I) {
16551647
}
16561648
}
16571649

1658-
// Convenience function to determine whether I is an SVE predicate callee save.
1659-
static bool IsPPRCalleeSave(MachineBasicBlock::iterator I) {
1650+
// Convenience function to determine whether I is part of the PPR callee saves.
1651+
static bool isPartOfPPRCalleeSaves(MachineBasicBlock::iterator I) {
16601652
switch (I->getOpcode()) {
16611653
default:
16621654
return false;
@@ -1667,8 +1659,9 @@ static bool IsPPRCalleeSave(MachineBasicBlock::iterator I) {
16671659
}
16681660
}
16691661

1670-
static bool IsSVECalleeSave(MachineBasicBlock::iterator I) {
1671-
return IsZPRCalleeSave(I) || IsPPRCalleeSave(I);
1662+
// Convenience function to determine whether I is part of the SVE callee saves.
1663+
static bool isPartOfSVECalleeSaves(MachineBasicBlock::iterator I) {
1664+
return isPartOfZPRCalleeSaves(I) || isPartOfPPRCalleeSaves(I);
16721665
}
16731666

16741667
static void emitShadowCallStackPrologue(const TargetInstrInfo &TII,
@@ -1912,7 +1905,7 @@ void AArch64FrameLowering::emitPrologue(MachineFunction &MF,
19121905
IsFunclet ? getWinEHFuncletFrameSize(MF) : MFI.getStackSize();
19131906
if (!AFI->hasStackFrame() && !windowsRequiresStackProbe(MF, NumBytes)) {
19141907
assert(!HasFP && "unexpected function without stack frame but with FP");
1915-
assert(!hasSVEStackSize(MF) &&
1908+
assert(!AFI->hasSVEStackSize() &&
19161909
"unexpected function without stack frame but with SVE objects");
19171910
// All of the stack allocation is for locals.
19181911
AFI->setLocalStackSize(NumBytes);
@@ -1986,14 +1979,14 @@ void AArch64FrameLowering::emitPrologue(MachineFunction &MF,
19861979
NumBytes -= FixedObject;
19871980

19881981
// Now allocate space for the GPR callee saves.
1989-
while (MBBI != End && IsSVECalleeSave(MBBI))
1982+
while (MBBI != End && isPartOfSVECalleeSaves(MBBI))
19901983
++MBBI;
19911984
MBBI = convertCalleeSaveRestoreToSPPrePostIncDec(
19921985
MBB, MBBI, DL, TII, -AFI->getCalleeSavedStackSize(), NeedsWinCFI,
19931986
&HasWinCFI, EmitAsyncCFI);
19941987
NumBytes -= AFI->getCalleeSavedStackSize();
19951988
} else if (CombineSPBump) {
1996-
assert(!hasSVEStackSize(MF) && "Cannot combine SP bump with SVE");
1989+
assert(!AFI->hasSVEStackSize() && "Cannot combine SP bump with SVE");
19971990
emitFrameOffset(MBB, MBBI, DL, AArch64::SP, AArch64::SP,
19981991
StackOffset::getFixed(-NumBytes), TII,
19991992
MachineInstr::FrameSetup, false, NeedsWinCFI, &HasWinCFI,
@@ -2014,7 +2007,7 @@ void AArch64FrameLowering::emitPrologue(MachineFunction &MF,
20142007
// and pre-inc if we decided to combine the callee-save and local stack
20152008
// pointer bump above.
20162009
while (MBBI != End && MBBI->getFlag(MachineInstr::FrameSetup) &&
2017-
!IsSVECalleeSave(MBBI)) {
2010+
!isPartOfSVECalleeSaves(MBBI)) {
20182011
if (CombineSPBump &&
20192012
// Only fix-up frame-setup load/store instructions.
20202013
(!requiresSaveVG(MF) || !isVGInstruction(MBBI)))
@@ -2278,8 +2271,9 @@ void AArch64FrameLowering::emitPrologue(MachineFunction &MF,
22782271
<< PPRCalleeSavesSize.getScalable() << "\n");
22792272

22802273
PPRCalleeSavesBegin = MBBI;
2281-
assert(IsPPRCalleeSave(PPRCalleeSavesBegin) && "Unexpected instruction");
2282-
while (IsPPRCalleeSave(MBBI) && MBBI != MBB.getFirstTerminator())
2274+
assert(isPartOfPPRCalleeSaves(PPRCalleeSavesBegin) &&
2275+
"Unexpected instruction");
2276+
while (isPartOfPPRCalleeSaves(MBBI) && MBBI != MBB.getFirstTerminator())
22832277
++MBBI;
22842278
PPRCalleeSavesEnd = MBBI;
22852279
}
@@ -2288,8 +2282,9 @@ void AArch64FrameLowering::emitPrologue(MachineFunction &MF,
22882282
LLVM_DEBUG(dbgs() << "ZPRCalleeSavedStackSize = "
22892283
<< ZPRCalleeSavesSize.getScalable() << "\n");
22902284
ZPRCalleeSavesBegin = MBBI;
2291-
assert(IsZPRCalleeSave(ZPRCalleeSavesBegin) && "Unexpected instruction");
2292-
while (IsZPRCalleeSave(MBBI) && MBBI != MBB.getFirstTerminator())
2285+
assert(isPartOfZPRCalleeSaves(ZPRCalleeSavesBegin) &&
2286+
"Unexpected instruction");
2287+
while (isPartOfZPRCalleeSaves(MBBI) && MBBI != MBB.getFirstTerminator())
22932288
++MBBI;
22942289
ZPRCalleeSavesEnd = MBBI;
22952290
}
@@ -2523,7 +2518,7 @@ void AArch64FrameLowering::emitEpilogue(MachineFunction &MF,
25232518
while (LastPopI != Begin) {
25242519
--LastPopI;
25252520
if (!LastPopI->getFlag(MachineInstr::FrameDestroy) ||
2526-
(!FPAfterSVECalleeSaves && IsSVECalleeSave(LastPopI))) {
2521+
(!FPAfterSVECalleeSaves && isPartOfSVECalleeSaves(LastPopI))) {
25272522
++LastPopI;
25282523
break;
25292524
} else if (CombineSPBump)
@@ -2608,11 +2603,12 @@ void AArch64FrameLowering::emitEpilogue(MachineFunction &MF,
26082603

26092604
RestoreBegin = std::prev(RestoreEnd);
26102605
while (RestoreBegin != MBB.begin() &&
2611-
IsSVECalleeSave(std::prev(RestoreBegin)))
2606+
isPartOfSVECalleeSaves(std::prev(RestoreBegin)))
26122607
--RestoreBegin;
26132608

2614-
assert(IsSVECalleeSave(RestoreBegin) &&
2615-
IsSVECalleeSave(std::prev(RestoreEnd)) && "Unexpected instruction");
2609+
assert(isPartOfSVECalleeSaves(RestoreBegin) &&
2610+
isPartOfSVECalleeSaves(std::prev(RestoreEnd)) &&
2611+
"Unexpected instruction");
26162612

26172613
StackOffset CalleeSavedSizeAsOffset =
26182614
StackOffset::getScalable(SVECalleeSavedSize);
@@ -4315,14 +4311,14 @@ determineSVEStackObjectOffsets(MachineFunction &MF, bool AssignOffsets,
43154311
bool SplitSVEObjects = false) {
43164312
MachineFrameInfo &MFI = MF.getFrameInfo();
43174313

4318-
int64_t ZPRStack = 0;
4319-
int64_t PPRStack = 0;
4314+
SVEStackSizes SVEStack{};
43204315

4321-
auto [ZPROffset, PPROffset] = [&] {
4322-
if (SplitSVEObjects)
4323-
return std::tie(ZPRStack, PPRStack);
4324-
return std::tie(ZPRStack, ZPRStack);
4325-
}();
4316+
// With SplitSVEObjects we maintain separate stack offsets for predicates
4317+
// (PPRs) and SVE vectors (ZPRs). When SplitSVEObjects is disabled predicates
4318+
// are included in the SVE vector area.
4319+
int64_t &ZPROffset = SVEStack.ZPRStackSize;
4320+
int64_t &PPROffset =
4321+
SplitSVEObjects ? SVEStack.PPRStackSize : SVEStack.ZPRStackSize;
43264322

43274323
#ifndef NDEBUG
43284324
// First process all fixed stack objects.
@@ -4404,14 +4400,7 @@ determineSVEStackObjectOffsets(MachineFunction &MF, bool AssignOffsets,
44044400

44054401
PPROffset = alignTo(PPROffset, Align(16U));
44064402
ZPROffset = alignTo(ZPROffset, Align(16U));
4407-
4408-
if (&ZPROffset != &PPROffset) {
4409-
// SplitSVEObjects (PPRs and ZPRs allocated to separate areas).
4410-
return SVEStackSizes{ZPROffset, PPROffset};
4411-
}
4412-
// When SplitSVEObjects is disabled just attribute all the stack to ZPRs.
4413-
// Determining the split is not necessary.
4414-
return SVEStackSizes{ZPROffset, 0};
4403+
return SVEStack;
44154404
}
44164405

44174406
SVEStackSizes
@@ -4736,8 +4725,7 @@ void AArch64FrameLowering::processFunctionBeforeFrameFinalized(
47364725
"Upwards growing stack unsupported");
47374726

47384727
auto [ZPRStackSize, PPRStackSize] = assignSVEStackObjectOffsets(MF);
4739-
AFI->setStackSizeZPR(ZPRStackSize);
4740-
AFI->setStackSizePPR(PPRStackSize);
4728+
AFI->setStackSizeSVE(ZPRStackSize, PPRStackSize);
47414729

47424730
// If this function isn't doing Win64-style C++ EH, we don't need to do
47434731
// anything.
@@ -5269,7 +5257,8 @@ StackOffset AArch64FrameLowering::getFrameIndexReferencePreferSP(
52695257
}
52705258

52715259
// Go to common code if we cannot provide sp + offset.
5272-
if (MFI.hasVarSizedObjects() || hasSVEStackSize(MF) ||
5260+
if (MFI.hasVarSizedObjects() ||
5261+
MF.getInfo<AArch64FunctionInfo>()->hasSVEStackSize() ||
52735262
MF.getSubtarget().getRegisterInfo()->hasStackRealignment(MF))
52745263
return getFrameIndexReference(MF, FI, FrameReg);
52755264

‎llvm/lib/Target/AArch64/AArch64MachineFunctionInfo.h‎

Lines changed: 7 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -297,21 +297,21 @@ class AArch64FunctionInfo final : public MachineFunctionInfo {
297297
TailCallReservedStack = bytes;
298298
}
299299

300-
void setStackSizeZPR(uint64_t S) {
300+
void setStackSizeSVE(uint64_t ZPR, uint64_t PPR = 0) {
301+
StackSizeZPR = ZPR;
302+
StackSizePPR = PPR;
301303
HasCalculatedStackSizeSVE = true;
302-
StackSizeZPR = S;
303-
}
304-
305-
void setStackSizePPR(uint64_t S) {
306-
HasCalculatedStackSizeSVE = true;
307-
StackSizePPR = S;
308304
}
309305

310306
uint64_t getStackSizeZPR() const { return StackSizeZPR; }
311307
uint64_t getStackSizePPR() const { return StackSizePPR; }
312308

313309
bool hasCalculatedStackSizeSVE() const { return HasCalculatedStackSizeSVE; }
314310

311+
bool hasSVEStackSize() const {
312+
return getStackSizeZPR() > 0 || getStackSizePPR() > 0;
313+
}
314+
315315
bool hasStackFrame() const { return HasStackFrame; }
316316
void setHasStackFrame(bool s) { HasStackFrame = s; }
317317

‎llvm/lib/Target/AArch64/AArch64RegisterInfo.cpp‎

Lines changed: 3 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -644,8 +644,7 @@ bool AArch64RegisterInfo::hasBasePointer(const MachineFunction &MF) const {
644644
if (ST.hasSVE() || ST.isStreaming()) {
645645
// Frames that have variable sized objects and scalable SVE objects,
646646
// should always use a basepointer.
647-
if (!AFI->hasCalculatedStackSizeSVE() || AFI->getStackSizeZPR() ||
648-
AFI->getStackSizePPR())
647+
if (!AFI->hasCalculatedStackSizeSVE() || AFI->hasSVEStackSize())
649648
return true;
650649
}
651650

@@ -785,8 +784,8 @@ AArch64RegisterInfo::useFPForScavengingIndex(const MachineFunction &MF) const {
785784
assert((!MF.getSubtarget<AArch64Subtarget>().hasSVE() ||
786785
AFI->hasCalculatedStackSizeSVE()) &&
787786
"Expected SVE area to be calculated by this point");
788-
return TFI.hasFP(MF) && !hasStackRealignment(MF) && !AFI->getStackSizeZPR() &&
789-
!AFI->getStackSizePPR() && !AFI->hasStackHazardSlotIndex();
787+
return TFI.hasFP(MF) && !hasStackRealignment(MF) && !AFI->hasSVEStackSize() &&
788+
!AFI->hasStackHazardSlotIndex();
790789
}
791790

792791
bool AArch64RegisterInfo::requiresFrameIndexScavenging(

0 commit comments

Comments
 (0)