Skip to content

Commit 4e231e7

Browse files
MacDueMacDue
committed
Fixups
Change-Id: Ifa920748f994e4160978da7d67ceb0c66093f729
1 parent eba3bca commit 4e231e7

File tree

3 files changed

+72
-85
lines changed

3 files changed

+72
-85
lines changed

‎llvm/lib/Target/AArch64/AArch64FrameLowering.cpp‎

Lines changed: 62 additions & 74 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
// Conservatively, returns true if the function is likely to have SVE vectors
342365
// on the stack. This function is safe to be called before callee-saves or
@@ -496,38 +519,6 @@ static unsigned getFixedObjectSize(const MachineFunction &MF,
496519
}
497520
}
498521

499-
static unsigned getStackHazardSize(const MachineFunction &MF) {
500-
return MF.getSubtarget<AArch64Subtarget>().getStreamingHazardSize();
501-
}
502-
503-
/// Returns the size of the entire ZPR stackframe (calleesaves + spills).
504-
static StackOffset getZPRStackSize(const MachineFunction &MF) {
505-
const AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>();
506-
return StackOffset::getScalable(AFI->getStackSizeZPR());
507-
}
508-
509-
/// Returns the size of the entire PPR stackframe (calleesaves + spills).
510-
static StackOffset getPPRStackSize(const MachineFunction &MF) {
511-
const AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>();
512-
return StackOffset::getScalable(AFI->getStackSizePPR());
513-
}
514-
515-
/// Returns the size of the entire SVE stackframe (PPRs + ZPRs).
516-
static StackOffset getSVEStackSize(const MachineFunction &MF) {
517-
return getZPRStackSize(MF) + getPPRStackSize(MF);
518-
}
519-
520-
static bool hasSVEStackSize(const MachineFunction &MF) {
521-
const AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>();
522-
return AFI->getStackSizeZPR() > 0 || AFI->getStackSizePPR() > 0;
523-
}
524-
525-
/// Returns true if PPRs are spilled as ZPRs.
526-
static bool arePPRsSpilledAsZPR(const MachineFunction &MF) {
527-
return MF.getSubtarget().getRegisterInfo()->getSpillSize(
528-
AArch64::PPRRegClass) == 16;
529-
}
530-
531522
bool AArch64FrameLowering::canUseRedZone(const MachineFunction &MF) const {
532523
if (!EnableRedZone)
533524
return false;
@@ -553,7 +544,7 @@ bool AArch64FrameLowering::canUseRedZone(const MachineFunction &MF) const {
553544
!Subtarget.hasSVE();
554545

555546
return !(MFI.hasCalls() || hasFP(MF) || NumBytes > RedZoneSize ||
556-
hasSVEStackSize(MF) || LowerQRegCopyThroughMem);
547+
AFI->hasSVEStackSize() || LowerQRegCopyThroughMem);
557548
}
558549

559550
/// hasFPImpl - Return true if the specified function should have a dedicated
@@ -1253,7 +1244,7 @@ bool AArch64FrameLowering::shouldCombineCSRLocalStackBump(
12531244

12541245
// When there is an SVE area on the stack, always allocate the
12551246
// callee-saves and spills/locals separately.
1256-
if (hasSVEStackSize(MF))
1247+
if (AFI->hasSVEStackSize())
12571248
return false;
12581249

12591250
return true;
@@ -1697,8 +1688,8 @@ static bool isTargetWindows(const MachineFunction &MF) {
16971688
return MF.getSubtarget<AArch64Subtarget>().isTargetWindows();
16981689
}
16991690

1700-
// Convenience function to determine whether I is an SVE callee save.
1701-
static bool IsZPRCalleeSave(MachineBasicBlock::iterator I) {
1691+
// Convenience function to determine whether I is part of the ZPR callee saves.
1692+
static bool isPartOfZPRCalleeSaves(MachineBasicBlock::iterator I) {
17021693
switch (I->getOpcode()) {
17031694
default:
17041695
return false;
@@ -1718,8 +1709,8 @@ static bool IsZPRCalleeSave(MachineBasicBlock::iterator I) {
17181709
}
17191710
}
17201711

1721-
// Convenience function to determine whether I is an SVE predicate callee save.
1722-
static bool IsPPRCalleeSave(MachineBasicBlock::iterator I) {
1712+
// Convenience function to determine whether I is part of the PPR callee saves.
1713+
static bool isPartOfPPRCalleeSaves(MachineBasicBlock::iterator I) {
17231714
switch (I->getOpcode()) {
17241715
default:
17251716
return false;
@@ -1730,8 +1721,9 @@ static bool IsPPRCalleeSave(MachineBasicBlock::iterator I) {
17301721
}
17311722
}
17321723

1733-
static bool IsSVECalleeSave(MachineBasicBlock::iterator I) {
1734-
return IsZPRCalleeSave(I) || IsPPRCalleeSave(I);
1724+
// Convenience function to determine whether I is part of the SVE callee saves.
1725+
static bool isPartOfSVECalleeSaves(MachineBasicBlock::iterator I) {
1726+
return isPartOfZPRCalleeSaves(I) || isPartOfPPRCalleeSaves(I);
17351727
}
17361728

17371729
static void emitShadowCallStackPrologue(const TargetInstrInfo &TII,
@@ -1975,7 +1967,7 @@ void AArch64FrameLowering::emitPrologue(MachineFunction &MF,
19751967
IsFunclet ? getWinEHFuncletFrameSize(MF) : MFI.getStackSize();
19761968
if (!AFI->hasStackFrame() && !windowsRequiresStackProbe(MF, NumBytes)) {
19771969
assert(!HasFP && "unexpected function without stack frame but with FP");
1978-
assert(!hasSVEStackSize(MF) &&
1970+
assert(!AFI->hasSVEStackSize() &&
19791971
"unexpected function without stack frame but with SVE objects");
19801972
// All of the stack allocation is for locals.
19811973
AFI->setLocalStackSize(NumBytes);
@@ -2049,14 +2041,14 @@ void AArch64FrameLowering::emitPrologue(MachineFunction &MF,
20492041
NumBytes -= FixedObject;
20502042

20512043
// Now allocate space for the GPR callee saves.
2052-
while (MBBI != End && IsSVECalleeSave(MBBI))
2044+
while (MBBI != End && isPartOfSVECalleeSaves(MBBI))
20532045
++MBBI;
20542046
MBBI = convertCalleeSaveRestoreToSPPrePostIncDec(
20552047
MBB, MBBI, DL, TII, -AFI->getCalleeSavedStackSize(), NeedsWinCFI,
20562048
&HasWinCFI, EmitAsyncCFI);
20572049
NumBytes -= AFI->getCalleeSavedStackSize();
20582050
} else if (CombineSPBump) {
2059-
assert(!hasSVEStackSize(MF) && "Cannot combine SP bump with SVE");
2051+
assert(!AFI->hasSVEStackSize() && "Cannot combine SP bump with SVE");
20602052
emitFrameOffset(MBB, MBBI, DL, AArch64::SP, AArch64::SP,
20612053
StackOffset::getFixed(-NumBytes), TII,
20622054
MachineInstr::FrameSetup, false, NeedsWinCFI, &HasWinCFI,
@@ -2077,7 +2069,7 @@ void AArch64FrameLowering::emitPrologue(MachineFunction &MF,
20772069
// and pre-inc if we decided to combine the callee-save and local stack
20782070
// pointer bump above.
20792071
while (MBBI != End && MBBI->getFlag(MachineInstr::FrameSetup) &&
2080-
!IsSVECalleeSave(MBBI)) {
2072+
!isPartOfSVECalleeSaves(MBBI)) {
20812073
if (CombineSPBump &&
20822074
// Only fix-up frame-setup load/store instructions.
20832075
(!requiresSaveVG(MF) || !isVGInstruction(MBBI)))
@@ -2341,8 +2333,9 @@ void AArch64FrameLowering::emitPrologue(MachineFunction &MF,
23412333
<< PPRCalleeSavesSize.getScalable() << "\n");
23422334

23432335
PPRCalleeSavesBegin = MBBI;
2344-
assert(IsPPRCalleeSave(PPRCalleeSavesBegin) && "Unexpected instruction");
2345-
while (IsPPRCalleeSave(MBBI) && MBBI != MBB.getFirstTerminator())
2336+
assert(isPartOfPPRCalleeSaves(PPRCalleeSavesBegin) &&
2337+
"Unexpected instruction");
2338+
while (isPartOfPPRCalleeSaves(MBBI) && MBBI != MBB.getFirstTerminator())
23462339
++MBBI;
23472340
PPRCalleeSavesEnd = MBBI;
23482341
}
@@ -2351,8 +2344,9 @@ void AArch64FrameLowering::emitPrologue(MachineFunction &MF,
23512344
LLVM_DEBUG(dbgs() << "ZPRCalleeSavedStackSize = "
23522345
<< ZPRCalleeSavesSize.getScalable() << "\n");
23532346
ZPRCalleeSavesBegin = MBBI;
2354-
assert(IsZPRCalleeSave(ZPRCalleeSavesBegin) && "Unexpected instruction");
2355-
while (IsZPRCalleeSave(MBBI) && MBBI != MBB.getFirstTerminator())
2347+
assert(isPartOfZPRCalleeSaves(ZPRCalleeSavesBegin) &&
2348+
"Unexpected instruction");
2349+
while (isPartOfZPRCalleeSaves(MBBI) && MBBI != MBB.getFirstTerminator())
23562350
++MBBI;
23572351
ZPRCalleeSavesEnd = MBBI;
23582352
}
@@ -2586,7 +2580,7 @@ void AArch64FrameLowering::emitEpilogue(MachineFunction &MF,
25862580
while (LastPopI != Begin) {
25872581
--LastPopI;
25882582
if (!LastPopI->getFlag(MachineInstr::FrameDestroy) ||
2589-
(!FPAfterSVECalleeSaves && IsSVECalleeSave(LastPopI))) {
2583+
(!FPAfterSVECalleeSaves && isPartOfSVECalleeSaves(LastPopI))) {
25902584
++LastPopI;
25912585
break;
25922586
} else if (CombineSPBump)
@@ -2671,11 +2665,12 @@ void AArch64FrameLowering::emitEpilogue(MachineFunction &MF,
26712665

26722666
RestoreBegin = std::prev(RestoreEnd);
26732667
while (RestoreBegin != MBB.begin() &&
2674-
IsSVECalleeSave(std::prev(RestoreBegin)))
2668+
isPartOfSVECalleeSaves(std::prev(RestoreBegin)))
26752669
--RestoreBegin;
26762670

2677-
assert(IsSVECalleeSave(RestoreBegin) &&
2678-
IsSVECalleeSave(std::prev(RestoreEnd)) && "Unexpected instruction");
2671+
assert(isPartOfSVECalleeSaves(RestoreBegin) &&
2672+
isPartOfSVECalleeSaves(std::prev(RestoreEnd)) &&
2673+
"Unexpected instruction");
26792674

26802675
StackOffset CalleeSavedSizeAsOffset =
26812676
StackOffset::getScalable(SVECalleeSavedSize);
@@ -4384,14 +4379,14 @@ determineSVEStackObjectOffsets(MachineFunction &MF, bool AssignOffsets,
43844379
bool SplitSVEObjects = false) {
43854380
MachineFrameInfo &MFI = MF.getFrameInfo();
43864381

4387-
int64_t ZPRStack = 0;
4388-
int64_t PPRStack = 0;
4382+
SVEStackSizes SVEStack{};
43894383

4390-
auto [ZPROffset, PPROffset] = [&] {
4391-
if (SplitSVEObjects)
4392-
return std::tie(ZPRStack, PPRStack);
4393-
return std::tie(ZPRStack, ZPRStack);
4394-
}();
4384+
// With SplitSVEObjects we maintain separate stack offsets for predicates
4385+
// (PPRs) and SVE vectors (ZPRs). When SplitSVEObjects is disabled predicates
4386+
// are included in the SVE vector area.
4387+
int64_t &ZPROffset = SVEStack.ZPRStackSize;
4388+
int64_t &PPROffset =
4389+
SplitSVEObjects ? SVEStack.PPRStackSize : SVEStack.ZPRStackSize;
43954390

43964391
#ifndef NDEBUG
43974392
// First process all fixed stack objects.
@@ -4473,14 +4468,7 @@ determineSVEStackObjectOffsets(MachineFunction &MF, bool AssignOffsets,
44734468

44744469
PPROffset = alignTo(PPROffset, Align(16U));
44754470
ZPROffset = alignTo(ZPROffset, Align(16U));
4476-
4477-
if (&ZPROffset != &PPROffset) {
4478-
// SplitSVEObjects (PPRs and ZPRs allocated to separate areas).
4479-
return SVEStackSizes{ZPROffset, PPROffset};
4480-
}
4481-
// When SplitSVEObjects is disabled just attribute all the stack to ZPRs.
4482-
// Determining the split is not necessary.
4483-
return SVEStackSizes{ZPROffset, 0};
4471+
return SVEStack;
44844472
}
44854473

44864474
SVEStackSizes
@@ -4805,8 +4793,7 @@ void AArch64FrameLowering::processFunctionBeforeFrameFinalized(
48054793
"Upwards growing stack unsupported");
48064794

48074795
auto [ZPRStackSize, PPRStackSize] = assignSVEStackObjectOffsets(MF);
4808-
AFI->setStackSizeZPR(ZPRStackSize);
4809-
AFI->setStackSizePPR(PPRStackSize);
4796+
AFI->setStackSizeSVE(ZPRStackSize, PPRStackSize);
48104797

48114798
// If this function isn't doing Win64-style C++ EH, we don't need to do
48124799
// anything.
@@ -5355,7 +5342,8 @@ StackOffset AArch64FrameLowering::getFrameIndexReferencePreferSP(
53555342
}
53565343

53575344
// Go to common code if we cannot provide sp + offset.
5358-
if (MFI.hasVarSizedObjects() || hasSVEStackSize(MF) ||
5345+
if (MFI.hasVarSizedObjects() ||
5346+
MF.getInfo<AArch64FunctionInfo>()->hasSVEStackSize() ||
53595347
MF.getSubtarget().getRegisterInfo()->hasStackRealignment(MF))
53605348
return getFrameIndexReference(MF, FI, FrameReg);
53615349

‎llvm/lib/Target/AArch64/AArch64MachineFunctionInfo.h‎

Lines changed: 7 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -298,14 +298,10 @@ class AArch64FunctionInfo final : public MachineFunctionInfo {
298298
TailCallReservedStack = bytes;
299299
}
300300

301-
void setStackSizeZPR(uint64_t S) {
301+
void setStackSizeSVE(uint64_t ZPR, uint64_t PPR = 0) {
302+
StackSizeZPR = ZPR;
303+
StackSizePPR = PPR;
302304
HasCalculatedStackSizeSVE = true;
303-
StackSizeZPR = S;
304-
}
305-
306-
void setStackSizePPR(uint64_t S) {
307-
HasCalculatedStackSizeSVE = true;
308-
StackSizePPR = S;
309305
}
310306

311307
uint64_t getStackSizeZPR() const {
@@ -319,6 +315,10 @@ class AArch64FunctionInfo final : public MachineFunctionInfo {
319315

320316
bool hasCalculatedStackSizeSVE() const { return HasCalculatedStackSizeSVE; }
321317

318+
bool hasSVEStackSize() const {
319+
return getStackSizeZPR() > 0 || getStackSizePPR() > 0;
320+
}
321+
322322
bool hasStackFrame() const { return HasStackFrame; }
323323
void setHasStackFrame(bool s) { HasStackFrame = s; }
324324

‎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)