mirrors/llvm-project
0
0
Fork 0
mirror of https://github.com/llvm/llvm-project.git synced 2025-04-28 08:56:07 +00:00
Code Issues Projects Releases Wiki Activity

[MC] Aligned bundling: remove special handling for RelaxAll

Browse Source 
When both aligned bundling and RelaxAll are enabled, bundle padding is
directly written into fragments (https://reviews.llvm.org/D8072).
(The original motivation was memory usage, which has been achieved from
different angles with recent assembler improvement).

The code presents challenges with the work to replace fragment
representation (e.g. #94950 #95077). This patch removes the special
handling. RelaxAll still works but the behavior seems slightly different
as revealed by 2 changed tests. However, most `-mc-relax-all` tests are
unchanged.

RelaxAll used to be the default for clang -O0. This mode has significant
code size drawbacks and newer Clang doesn't use it (#90013).

---

flushPendingLabels: The FOffset parameter can be removed: pending labels
will be assigned to the incoming fragment at offset 0.

Pull Request: https://github.com/llvm/llvm-project/pull/95188
This commit is contained in:
Fangrui Song 2024-06-14 10:01:36 -07:00 committed by GitHub
parent 72b841d016
commit b1932b8483
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
10 changed files with 13 additions and 141 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
llvm/include/llvm/MC/MCELFStreamer.h
View File

@ -39,7 +39,6 @@ public:
/// state management
void reset() override {
SeenIdent = false;
BundleGroups.clear();
MCObjectStreamer::reset();
}
@ -142,14 +141,7 @@ private:
void finalizeCGProfileEntry(const MCSymbolRefExpr *&S, uint64_t Offset);
void finalizeCGProfile();
/// Merge the content of the fragment \p EF into the fragment \p DF.
void mergeFragment(MCDataFragment *, MCDataFragment *);
bool SeenIdent = false;
/// BundleGroups - The stack of fragments holding the bundle-locked
/// instructions.
SmallVector<MCDataFragment *, 4> BundleGroups;
};
MCELFStreamer *createARMELFStreamer(MCContext &Context,

3
llvm/include/llvm/MC/MCSection.h
View File

@ -227,8 +227,7 @@ public:
void addPendingLabel(MCSymbol* label, unsigned Subsection = 0);
/// Associate all pending labels in a subsection with a fragment.
void flushPendingLabels(MCFragment *F, uint64_t FOffset = 0,
unsigned Subsection = 0);
void flushPendingLabels(MCFragment *F, unsigned Subsection);
/// Associate all pending labels with empty data fragments. One fragment
/// will be created for each subsection as necessary.

3
llvm/include/llvm/MC/MCWasmStreamer.h
View File

@ -73,9 +73,6 @@ private:
void fixSymbolsInTLSFixups(const MCExpr *expr);
/// Merge the content of the fragment \p EF into the fragment \p DF.
void mergeFragment(MCDataFragment *, MCDataFragment *);
bool SeenIdent;
};

8
llvm/lib/MC/MCAssembler.cpp
View File

@ -420,12 +420,6 @@ void MCAsmLayout::layoutBundle(MCFragment *F) {
// The fragment's offset will point to after the padding, and its computed
// size won't include the padding.
//
// When the -mc-relax-all flag is used, we optimize bundling by writting the
// padding directly into fragments when the instructions are emitted inside
// the streamer. When the fragment is larger than the bundle size, we need to
// ensure that it's bundle aligned. This means that if we end up with
// multiple fragments, we must emit bundle padding between fragments.
//
// ".align N" is an example of a directive that introduces multiple
// fragments. We could add a special case to handle ".align N" by emitting
// within-fragment padding (which would produce less padding when N is less
@ -436,7 +430,7 @@ void MCAsmLayout::layoutBundle(MCFragment *F) {
MCEncodedFragment *EF = cast<MCEncodedFragment>(F);
uint64_t FSize = Assembler.computeFragmentSize(*this, *EF);
if (!Assembler.getRelaxAll() && FSize > Assembler.getBundleAlignSize())
if (FSize > Assembler.getBundleAlignSize())
report_fatal_error("Fragment can't be larger than a bundle size");
uint64_t RequiredBundlePadding =

89
llvm/lib/MC/MCELFStreamer.cpp
View File

@ -50,44 +50,6 @@ bool MCELFStreamer::isBundleLocked() const {
return getCurrentSectionOnly()->isBundleLocked();
}
void MCELFStreamer::mergeFragment(MCDataFragment *DF,
MCDataFragment *EF) {
MCAssembler &Assembler = getAssembler();
if (Assembler.isBundlingEnabled() && Assembler.getRelaxAll()) {
uint64_t FSize = EF->getContents().size();
if (FSize > Assembler.getBundleAlignSize())
report_fatal_error("Fragment can't be larger than a bundle size");
uint64_t RequiredBundlePadding = computeBundlePadding(
Assembler, EF, DF->getContents().size(), FSize);
if (RequiredBundlePadding > UINT8_MAX)
report_fatal_error("Padding cannot exceed 255 bytes");
if (RequiredBundlePadding > 0) {
SmallString<256> Code;
raw_svector_ostream VecOS(Code);
EF->setBundlePadding(static_cast<uint8_t>(RequiredBundlePadding));
Assembler.writeFragmentPadding(VecOS, *EF, FSize);
DF->getContents().append(Code.begin(), Code.end());
}
}
flushPendingLabels(DF, DF->getContents().size());
for (unsigned i = 0, e = EF->getFixups().size(); i != e; ++i) {
EF->getFixups()[i].setOffset(EF->getFixups()[i].getOffset() +
DF->getContents().size());
DF->getFixups().push_back(EF->getFixups()[i]);
}
if (DF->getSubtargetInfo() == nullptr && EF->getSubtargetInfo())
DF->setHasInstructions(*EF->getSubtargetInfo());
DF->getContents().append(EF->getContents().begin(), EF->getContents().end());
}
void MCELFStreamer::initSections(bool NoExecStack, const MCSubtargetInfo &STI) {
MCContext &Ctx = getContext();
switchSection(Ctx.getObjectFileInfo()->getTextSection());
@ -575,24 +537,12 @@ void MCELFStreamer::emitInstToData(const MCInst &Inst,
if (Assembler.isBundlingEnabled()) {
MCSection &Sec = *getCurrentSectionOnly();
if (Assembler.getRelaxAll() && isBundleLocked()) {
// If the -mc-relax-all flag is used and we are bundle-locked, we re-use
// the current bundle group.
DF = BundleGroups.back();
CheckBundleSubtargets(DF->getSubtargetInfo(), &STI);
}
else if (Assembler.getRelaxAll() && !isBundleLocked())
// When not in a bundle-locked group and the -mc-relax-all flag is used,
// we create a new temporary fragment which will be later merged into
// the current fragment.
DF = getContext().allocFragment<MCDataFragment>();
else if (isBundleLocked() && !Sec.isBundleGroupBeforeFirstInst()) {
if (isBundleLocked() && !Sec.isBundleGroupBeforeFirstInst()) {
// If we are bundle-locked, we re-use the current fragment.
// The bundle-locking directive ensures this is a new data fragment.
DF = cast<MCDataFragment>(getCurrentFragment());
CheckBundleSubtargets(DF->getSubtargetInfo(), &STI);
}
else if (!isBundleLocked() && Fixups.size() == 0) {
} else if (!isBundleLocked() && Fixups.size() == 0) {
// Optimize memory usage by emitting the instruction to a
// MCCompactEncodedInstFragment when not in a bundle-locked group and
// there are no fixups registered.
@ -632,13 +582,6 @@ void MCELFStreamer::emitInstToData(const MCInst &Inst,
getAssembler().getBackend().RelaxFixupKind)
DF->setLinkerRelaxable();
DF->getContents().append(Code.begin(), Code.end());
if (Assembler.isBundlingEnabled() && Assembler.getRelaxAll()) {
if (!isBundleLocked()) {
mergeFragment(getOrCreateDataFragment(&STI), DF);
delete DF;
}
}
}
void MCELFStreamer::emitBundleAlignMode(Align Alignment) {
@ -660,12 +603,6 @@ void MCELFStreamer::emitBundleLock(bool AlignToEnd) {
if (!isBundleLocked())
Sec.setBundleGroupBeforeFirstInst(true);
if (getAssembler().getRelaxAll() && !isBundleLocked()) {
// TODO: drop the lock state and set directly in the fragment
MCDataFragment *DF = getContext().allocFragment<MCDataFragment>();
BundleGroups.push_back(DF);
}
Sec.setBundleLockState(AlignToEnd ? MCSection::BundleLockedAlignToEnd
: MCSection::BundleLocked);
}
@ -680,27 +617,7 @@ void MCELFStreamer::emitBundleUnlock() {
else if (Sec.isBundleGroupBeforeFirstInst())
report_fatal_error("Empty bundle-locked group is forbidden");
// When the -mc-relax-all flag is used, we emit instructions to fragments
// stored on a stack. When the bundle unlock is emitted, we pop a fragment
// from the stack a merge it to the one below.
if (getAssembler().getRelaxAll()) {
assert(!BundleGroups.empty() && "There are no bundle groups");
MCDataFragment *DF = BundleGroups.back();
// FIXME: Use BundleGroups to track the lock state instead.
Sec.setBundleLockState(MCSection::NotBundleLocked);
// FIXME: Use more separate fragments for nested groups.
if (!isBundleLocked()) {
mergeFragment(getOrCreateDataFragment(DF->getSubtargetInfo()), DF);
BundleGroups.pop_back();
delete DF;
}
if (Sec.getBundleLockState() != MCSection::BundleLockedAlignToEnd)
getOrCreateDataFragment()->setAlignToBundleEnd(false);
} else
Sec.setBundleLockState(MCSection::NotBundleLocked);
Sec.setBundleLockState(MCSection::NotBundleLocked);
}
void MCELFStreamer::finishImpl() {

10
llvm/lib/MC/MCObjectStreamer.cpp
View File

@ -81,7 +81,7 @@ void MCObjectStreamer::flushPendingLabels(MCFragment *F, uint64_t FOffset) {
}
// Associate the labels with F.
CurSection->flushPendingLabels(F, FOffset, CurSubsectionIdx);
CurSection->flushPendingLabels(F, CurSubsectionIdx);
}
void MCObjectStreamer::flushPendingLabels() {
@ -215,7 +215,7 @@ static bool canReuseDataFragment(const MCDataFragment &F,
// When bundling is enabled, we don't want to add data to a fragment that
// already has instructions (see MCELFStreamer::emitInstToData for details)
if (Assembler.isBundlingEnabled())
return Assembler.getRelaxAll();
return false;
// If the subtarget is changed mid fragment we start a new fragment to record
// the new STI.
return !STI || F.getSubtargetInfo() == STI;
@ -292,8 +292,7 @@ void MCObjectStreamer::emitLabel(MCSymbol *Symbol, SMLoc Loc) {
// Otherwise queue the label and set its fragment pointer when we emit the
// next fragment.
auto *F = dyn_cast_or_null<MCDataFragment>(getCurrentFragment());
if (F && !(getAssembler().isBundlingEnabled() &&
getAssembler().getRelaxAll())) {
if (F) {
Symbol->setFragment(F);
Symbol->setOffset(F->getContents().size());
} else {
@ -465,9 +464,6 @@ void MCObjectStreamer::emitInstructionImpl(const MCInst &Inst,
void MCObjectStreamer::emitInstToFragment(const MCInst &Inst,
const MCSubtargetInfo &STI) {
if (getAssembler().getRelaxAll() && getAssembler().isBundlingEnabled())
llvm_unreachable("All instructions should have already been relaxed");
// Always create a new, separate fragment here, because its size can change
// during relaxation.
MCRelaxableFragment *IF =

7
llvm/lib/MC/MCSection.cpp
View File

@ -82,15 +82,14 @@ void MCSection::addPendingLabel(MCSymbol *label, unsigned Subsection) {
PendingLabels.push_back(PendingLabel(label, Subsection));
}
void MCSection::flushPendingLabels(MCFragment *F, uint64_t FOffset,
unsigned Subsection) {
void MCSection::flushPendingLabels(MCFragment *F, unsigned Subsection) {
// Set the fragment and fragment offset for all pending symbols in the
// specified Subsection, and remove those symbols from the pending list.
for (auto It = PendingLabels.begin(); It != PendingLabels.end(); ++It) {
PendingLabel& Label = *It;
if (Label.Subsection == Subsection) {
Label.Sym->setFragment(F);
Label.Sym->setOffset(FOffset);
assert(Label.Sym->getOffset() == 0);
PendingLabels.erase(It--);
}
}
@ -105,7 +104,7 @@ void MCSection::flushPendingLabels() {
MCFragment *F = new MCDataFragment();
addFragment(*F);
F->setParent(this);
flushPendingLabels(F, 0, Label.Subsection);
flushPendingLabels(F, Label.Subsection);
}
}

13
llvm/lib/MC/MCWasmStreamer.cpp
View File

@ -39,19 +39,6 @@ using namespace llvm;
MCWasmStreamer::~MCWasmStreamer() = default; // anchor.
void MCWasmStreamer::mergeFragment(MCDataFragment *DF, MCDataFragment *EF) {
flushPendingLabels(DF, DF->getContents().size());
for (unsigned I = 0, E = EF->getFixups().size(); I != E; ++I) {
EF->getFixups()[I].setOffset(EF->getFixups()[I].getOffset() +
DF->getContents().size());
DF->getFixups().push_back(EF->getFixups()[I]);
}
if (DF->getSubtargetInfo() == nullptr && EF->getSubtargetInfo())
DF->setHasInstructions(*EF->getSubtargetInfo());
DF->getContents().append(EF->getContents().begin(), EF->getContents().end());
}
void MCWasmStreamer::emitLabel(MCSymbol *S, SMLoc Loc) {
auto *Symbol = cast<MCSymbolWasm>(S);
MCObjectStreamer::emitLabel(Symbol, Loc);

6
llvm/test/MC/X86/AlignedBundling/misaligned-bundle-group.s
View File

@ -3,7 +3,7 @@
# RUN: | FileCheck -check-prefix=CHECK -check-prefix=CHECK-OPT %s
# RUN: llvm-mc -filetype=obj -triple i686-pc-linux-gnu -mcpu=pentiumpro -mc-relax-all %s -o - \
# RUN: | llvm-objdump -d --no-show-raw-insn - \
# RUN: | FileCheck -check-prefix=CHECK -check-prefix=CHECK-RELAX %s
# RUN: | FileCheck --check-prefixes=CHECK,CHECK-OPT %s
.text
foo:
@ -13,11 +13,7 @@ foo:
.bundle_lock align_to_end
# CHECK: 1: nopw %cs:(%eax,%eax)
# CHECK: 10: nopw %cs:(%eax,%eax)
# CHECK-RELAX: 1a: nop
# CHECK-RELAX: 20: nopw %cs:(%eax,%eax)
# CHECK-RELAX: 2a: nopw %cs:(%eax,%eax)
# CHECK-OPT: 1b: calll 0x1c
# CHECK-RELAX: 3b: calll 0x3c
calll bar # 5 bytes
.bundle_unlock
ret # 1 byte

7
llvm/test/MC/X86/AlignedBundling/misaligned-bundle.s
View File

@ -3,7 +3,7 @@
# RUN: | FileCheck -check-prefix=CHECK -check-prefix=CHECK-OPT %s
# RUN: llvm-mc -filetype=obj -triple i686-pc-linux-gnu -mcpu=pentiumpro -mc-relax-all %s -o - \
# RUN: | llvm-objdump --no-print-imm-hex -d --no-show-raw-insn - \
# RUN: | FileCheck -check-prefix=CHECK -check-prefix=CHECK-RELAX %s
# RUN: | FileCheck --check-prefixes=CHECK,CHECK-OPT %s
.text
foo:
@ -11,17 +11,12 @@ foo:
push %ebp # 1 byte
.align 16
# CHECK: 1: nopw %cs:(%eax,%eax)
# CHECK-RELAX: 10: nopw %cs:(%eax,%eax)
# CHECK-RELAX: 1a: nop
# CHECK-OPT: 10: movl $1, (%esp)
# CHECK-RELAX: 20: movl $1, (%esp)
movl $0x1, (%esp) # 7 bytes
movl $0x1, (%esp) # 7 bytes
# CHECK-OPT: 1e: nop
movl $0x2, 0x1(%esp) # 8 bytes
movl $0x2, 0x1(%esp) # 8 bytes
# CHECK-RELAX: 3e: nop
# CHECK-RELAX: 40: movl $2, 1(%esp)
movl $0x2, 0x1(%esp) # 8 bytes
movl $0x2, (%esp) # 7 bytes
# CHECK-OPT: 3f: nop