[llvm] MC: Store fragment content and fixups out-of-line (PR #146307)
Fangrui Song via llvm-commits
llvm-commits at lists.llvm.org
Sun Jun 29 21:27:15 PDT 2025
https://github.com/MaskRay created https://github.com/llvm/llvm-project/pull/146307
Moved `Contents` and `Fixups` SmallVector storage to MCSection, enabling
trivial destructors for all fragment subclasses and eliminating the need
for MCFragment::destroy in ~MCSection.
For appending content to the current section, use
getContentsForAppending. During assembler relaxation, prefer
setContents/setFixups, which may involve copying and reduce the benefits
of https://reviews.llvm.org/D145791.
Moving only Contents out-of-line caused a slight performance regression
(Alexis Engelke's 2024 prototype). By also moving Fragments out-of-line,
fragment destructors become trivial, resulting in
minor instructions:u increase and [modest max-rss decrease](https://llvm-compile-time-tracker.com/compare.php?from=e47d4010d34119c2b4a28e7609fde35449a8b437&to=983887637aaf87fc1ca233a8848a5637fac8f524&stat=max-rss&linkStats=on)
for the "stage1-ReleaseLTO-g (link only)" benchmark.
Now using a plain SmallVector in MCSection for storage, with potential
for future allocator optimizations.
Co-authored-by: Alexis Engelke <engelke at in.tum.de>
>From 2930a87e589e61747900ebdc36bb4fe33eed9da9 Mon Sep 17 00:00:00 2001
From: Fangrui Song <i at maskray.me>
Date: Sun, 29 Jun 2025 21:27:00 -0700
Subject: [PATCH] =?UTF-8?q?[=F0=9D=98=80=F0=9D=97=BD=F0=9D=97=BF]=20initia?=
=?UTF-8?q?l=20version?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Created using spr 1.3.5-bogner
---
llvm/include/llvm/MC/MCFragment.h | 121 +++++++----------
llvm/include/llvm/MC/MCSection.h | 5 +
llvm/lib/MC/MCAssembler.cpp | 58 ++++----
llvm/lib/MC/MCCodeView.cpp | 12 +-
llvm/lib/MC/MCELFStreamer.cpp | 10 +-
llvm/lib/MC/MCFragment.cpp | 127 +++++++++++-------
llvm/lib/MC/MCObjectStreamer.cpp | 22 ++-
llvm/lib/MC/MCSection.cpp | 6 +-
llvm/lib/MC/MachObjectWriter.cpp | 5 +-
llvm/lib/MC/WinCOFFObjectWriter.cpp | 6 +-
.../MCTargetDesc/LoongArchAsmBackend.cpp | 27 ++--
.../RISCV/MCTargetDesc/RISCVAsmBackend.cpp | 26 ++--
.../Target/X86/MCTargetDesc/X86AsmBackend.cpp | 2 +-
13 files changed, 227 insertions(+), 200 deletions(-)
diff --git a/llvm/include/llvm/MC/MCFragment.h b/llvm/include/llvm/MC/MCFragment.h
index 2affa85c744ba..789f9a0e602b2 100644
--- a/llvm/include/llvm/MC/MCFragment.h
+++ b/llvm/include/llvm/MC/MCFragment.h
@@ -30,6 +30,9 @@ class MCSection;
class MCSubtargetInfo;
class MCSymbol;
+// Represents a contiguous segment of code or data within a section. Its size is
+// determined by MCAssembler::layout. All subclasses (except
+// MCRelaxableFragment, which stores a MCInst) must have trivial destructors.
class MCFragment {
friend class MCAssembler;
friend class MCObjectStreamer;
@@ -86,12 +89,6 @@ class MCFragment {
MCFragment(const MCFragment &) = delete;
MCFragment &operator=(const MCFragment &) = delete;
- /// Destroys the current fragment.
- ///
- /// This must be used instead of delete as MCFragment is non-virtual.
- /// This method will dispatch to the appropriate subclass.
- LLVM_ABI void destroy();
-
MCFragment *getNext() const { return Next; }
FragmentType getKind() const { return Kind; }
@@ -113,10 +110,7 @@ class MCFragment {
/// Interface implemented by fragments that contain encoded instructions and/or
/// data.
-///
class MCEncodedFragment : public MCFragment {
- uint8_t BundlePadding = 0;
-
protected:
MCEncodedFragment(MCFragment::FragmentType FType, bool HasInstructions)
: MCFragment(FType, HasInstructions) {}
@@ -125,6 +119,13 @@ class MCEncodedFragment : public MCFragment {
/// It must be non-null for instructions.
const MCSubtargetInfo *STI = nullptr;
+private:
+ uint32_t ContentStart = 0;
+ uint32_t ContentSize = 0;
+ uint32_t FixupStart = 0;
+ uint32_t FixupSize = 0;
+ uint8_t BundlePadding = 0;
+
public:
static bool classof(const MCFragment *F) {
MCFragment::FragmentType Kind = F->getKind();
@@ -136,6 +137,7 @@ class MCEncodedFragment : public MCFragment {
case MCFragment::FT_Dwarf:
case MCFragment::FT_DwarfFrame:
case MCFragment::FT_PseudoProbe:
+ case MCFragment::FT_CVInlineLines:
return true;
}
}
@@ -165,48 +167,33 @@ class MCEncodedFragment : public MCFragment {
HasInstructions = true;
this->STI = &STI;
}
-};
-/// Interface implemented by fragments that contain encoded instructions and/or
-/// data and also have fixups registered.
-///
-template <unsigned ContentsSize, unsigned FixupsSize>
-class MCEncodedFragmentWithFixups : public MCEncodedFragment {
- SmallVector<char, ContentsSize> Contents;
-
- /// The list of fixups in this fragment.
- SmallVector<MCFixup, FixupsSize> Fixups;
-
-protected:
- MCEncodedFragmentWithFixups(MCFragment::FragmentType FType,
- bool HasInstructions)
- : MCEncodedFragment(FType, HasInstructions) {}
-
-public:
- SmallVectorImpl<char> &getContents() { return Contents; }
- const SmallVectorImpl<char> &getContents() const { return Contents; }
-
- void appendContents(ArrayRef<char> C) { Contents.append(C.begin(), C.end()); }
- void appendContents(size_t Num, char Elt) { Contents.append(Num, Elt); }
- void setContents(ArrayRef<char> C) { Contents.assign(C.begin(), C.end()); }
-
- void addFixup(MCFixup Fixup) { Fixups.push_back(Fixup); }
- SmallVectorImpl<MCFixup> &getFixups() { return Fixups; }
- const SmallVectorImpl<MCFixup> &getFixups() const { return Fixups; }
-
- static bool classof(const MCFragment *F) {
- MCFragment::FragmentType Kind = F->getKind();
- return Kind == MCFragment::FT_Relaxable || Kind == MCFragment::FT_Data ||
- Kind == MCFragment::FT_CVDefRange || Kind == MCFragment::FT_Dwarf ||
- Kind == MCFragment::FT_DwarfFrame;
- }
+ // Content-related functions manage parent's storage using ContentStart and
+ // ContentSize.
+ void clearContents() { ContentSize = 0; }
+ SmallVectorImpl<char> &getContentsForAppending();
+ void doneAppending();
+ void appendContents(ArrayRef<char> Contents);
+ void appendContents(size_t Num, char Elt);
+ void setContents(ArrayRef<char> Contents);
+ MutableArrayRef<char> getContents();
+ ArrayRef<char> getContents() const;
+
+ // Fixup-related functions manage parent's storage using FixupStart and
+ // FixupSize.
+ void clearFixups() { FixupSize = 0; }
+ void addFixup(MCFixup Fixup);
+ void appendFixups(ArrayRef<MCFixup> Fixups);
+ void setFixups(ArrayRef<MCFixup> Fixups);
+ MutableArrayRef<MCFixup> getFixups();
+ ArrayRef<MCFixup> getFixups() const;
};
/// Fragment for data and encoded instructions.
///
-class MCDataFragment : public MCEncodedFragmentWithFixups<32, 4> {
+class MCDataFragment : public MCEncodedFragment {
public:
- MCDataFragment() : MCEncodedFragmentWithFixups<32, 4>(FT_Data, false) {}
+ MCDataFragment() : MCEncodedFragment(FT_Data, false) {}
static bool classof(const MCFragment *F) {
return F->getKind() == MCFragment::FT_Data;
@@ -219,13 +206,13 @@ class MCDataFragment : public MCEncodedFragmentWithFixups<32, 4> {
/// A relaxable fragment holds on to its MCInst, since it may need to be
/// relaxed during the assembler layout and relaxation stage.
///
-class MCRelaxableFragment : public MCEncodedFragmentWithFixups<8, 1> {
+class MCRelaxableFragment : public MCEncodedFragment {
/// The instruction this is a fragment for.
MCInst Inst;
public:
MCRelaxableFragment(const MCInst &Inst, const MCSubtargetInfo &STI)
- : MCEncodedFragmentWithFixups(FT_Relaxable, true), Inst(Inst) {
+ : MCEncodedFragment(FT_Relaxable, true), Inst(Inst) {
this->STI = &STI;
}
@@ -374,7 +361,7 @@ class MCOrgFragment : public MCFragment {
}
};
-class MCLEBFragment final : public MCEncodedFragmentWithFixups<8, 0> {
+class MCLEBFragment final : public MCEncodedFragment {
/// True if this is a sleb128, false if uleb128.
bool IsSigned;
@@ -383,24 +370,19 @@ class MCLEBFragment final : public MCEncodedFragmentWithFixups<8, 0> {
public:
MCLEBFragment(const MCExpr &Value, bool IsSigned)
- : MCEncodedFragmentWithFixups<8, 0>(FT_LEB, false), IsSigned(IsSigned),
- Value(&Value) {
- getContents().push_back(0);
- }
+ : MCEncodedFragment(FT_LEB, false), IsSigned(IsSigned), Value(&Value) {}
const MCExpr &getValue() const { return *Value; }
void setValue(const MCExpr *Expr) { Value = Expr; }
bool isSigned() const { return IsSigned; }
- /// @}
-
static bool classof(const MCFragment *F) {
return F->getKind() == MCFragment::FT_LEB;
}
};
-class MCDwarfLineAddrFragment : public MCEncodedFragmentWithFixups<8, 1> {
+class MCDwarfLineAddrFragment : public MCEncodedFragment {
/// The value of the difference between the two line numbers
/// between two .loc dwarf directives.
int64_t LineDelta;
@@ -411,8 +393,8 @@ class MCDwarfLineAddrFragment : public MCEncodedFragmentWithFixups<8, 1> {
public:
MCDwarfLineAddrFragment(int64_t LineDelta, const MCExpr &AddrDelta)
- : MCEncodedFragmentWithFixups<8, 1>(FT_Dwarf, false),
- LineDelta(LineDelta), AddrDelta(&AddrDelta) {}
+ : MCEncodedFragment(FT_Dwarf, false), LineDelta(LineDelta),
+ AddrDelta(&AddrDelta) {}
int64_t getLineDelta() const { return LineDelta; }
@@ -423,15 +405,14 @@ class MCDwarfLineAddrFragment : public MCEncodedFragmentWithFixups<8, 1> {
}
};
-class MCDwarfCallFrameFragment : public MCEncodedFragmentWithFixups<8, 1> {
+class MCDwarfCallFrameFragment : public MCEncodedFragment {
/// The expression for the difference of the two symbols that
/// make up the address delta between two .cfi_* dwarf directives.
const MCExpr *AddrDelta;
public:
MCDwarfCallFrameFragment(const MCExpr &AddrDelta)
- : MCEncodedFragmentWithFixups<8, 1>(FT_DwarfFrame, false),
- AddrDelta(&AddrDelta) {}
+ : MCEncodedFragment(FT_DwarfFrame, false), AddrDelta(&AddrDelta) {}
const MCExpr &getAddrDelta() const { return *AddrDelta; }
void setAddrDelta(const MCExpr *E) { AddrDelta = E; }
@@ -459,13 +440,12 @@ class MCSymbolIdFragment : public MCFragment {
/// Fragment representing the binary annotations produced by the
/// .cv_inline_linetable directive.
-class MCCVInlineLineTableFragment : public MCFragment {
+class MCCVInlineLineTableFragment : public MCEncodedFragment {
unsigned SiteFuncId;
unsigned StartFileId;
unsigned StartLineNum;
const MCSymbol *FnStartSym;
const MCSymbol *FnEndSym;
- SmallString<8> Contents;
/// CodeViewContext has the real knowledge about this format, so let it access
/// our members.
@@ -475,23 +455,20 @@ class MCCVInlineLineTableFragment : public MCFragment {
MCCVInlineLineTableFragment(unsigned SiteFuncId, unsigned StartFileId,
unsigned StartLineNum, const MCSymbol *FnStartSym,
const MCSymbol *FnEndSym)
- : MCFragment(FT_CVInlineLines, false), SiteFuncId(SiteFuncId),
+ : MCEncodedFragment(FT_CVInlineLines, false), SiteFuncId(SiteFuncId),
StartFileId(StartFileId), StartLineNum(StartLineNum),
FnStartSym(FnStartSym), FnEndSym(FnEndSym) {}
const MCSymbol *getFnStartSym() const { return FnStartSym; }
const MCSymbol *getFnEndSym() const { return FnEndSym; }
- SmallString<8> &getContents() { return Contents; }
- const SmallString<8> &getContents() const { return Contents; }
-
static bool classof(const MCFragment *F) {
return F->getKind() == MCFragment::FT_CVInlineLines;
}
};
/// Fragment representing the .cv_def_range directive.
-class MCCVDefRangeFragment : public MCEncodedFragmentWithFixups<32, 4> {
+class MCCVDefRangeFragment : public MCEncodedFragment {
ArrayRef<std::pair<const MCSymbol *, const MCSymbol *>> Ranges;
StringRef FixedSizePortion;
@@ -503,8 +480,9 @@ class MCCVDefRangeFragment : public MCEncodedFragmentWithFixups<32, 4> {
MCCVDefRangeFragment(
ArrayRef<std::pair<const MCSymbol *, const MCSymbol *>> Ranges,
StringRef FixedSizePortion)
- : MCEncodedFragmentWithFixups<32, 4>(FT_CVDefRange, false),
- Ranges(Ranges), FixedSizePortion(FixedSizePortion) {}
+ : MCEncodedFragment(FT_CVDefRange, false),
+ Ranges(Ranges.begin(), Ranges.end()),
+ FixedSizePortion(FixedSizePortion) {}
ArrayRef<std::pair<const MCSymbol *, const MCSymbol *>> getRanges() const {
return Ranges;
@@ -556,15 +534,14 @@ class MCBoundaryAlignFragment : public MCFragment {
}
};
-class MCPseudoProbeAddrFragment : public MCEncodedFragmentWithFixups<8, 1> {
+class MCPseudoProbeAddrFragment : public MCEncodedFragment {
/// The expression for the difference of the two symbols that
/// make up the address delta between two .pseudoprobe directives.
const MCExpr *AddrDelta;
public:
MCPseudoProbeAddrFragment(const MCExpr *AddrDelta)
- : MCEncodedFragmentWithFixups<8, 1>(FT_PseudoProbe, false),
- AddrDelta(AddrDelta) {}
+ : MCEncodedFragment(FT_PseudoProbe, false), AddrDelta(AddrDelta) {}
const MCExpr &getAddrDelta() const { return *AddrDelta; }
diff --git a/llvm/include/llvm/MC/MCSection.h b/llvm/include/llvm/MC/MCSection.h
index a4e391dfb03ea..fb59c74b7c484 100644
--- a/llvm/include/llvm/MC/MCSection.h
+++ b/llvm/include/llvm/MC/MCSection.h
@@ -39,6 +39,7 @@ class LLVM_ABI MCSection {
public:
friend MCAssembler;
friend MCObjectStreamer;
+ friend class MCEncodedFragment;
static constexpr unsigned NonUniqueID = ~0U;
enum SectionVariant {
@@ -116,6 +117,10 @@ class LLVM_ABI MCSection {
// subsections.
SmallVector<std::pair<unsigned, FragList>, 1> Subsections;
+ // Content and fixup storage for fragments
+ SmallVector<char, 0> ContentStorage;
+ SmallVector<MCFixup, 0> FixupStorage;
+
protected:
// TODO Make Name private when possible.
StringRef Name;
diff --git a/llvm/lib/MC/MCAssembler.cpp b/llvm/lib/MC/MCAssembler.cpp
index bd2242af23f7c..f3ad5b6d71b80 100644
--- a/llvm/lib/MC/MCAssembler.cpp
+++ b/llvm/lib/MC/MCAssembler.cpp
@@ -607,12 +607,14 @@ static void writeFragment(raw_ostream &OS, const MCAssembler &Asm,
case MCFragment::FT_Data:
++stats::EmittedDataFragments;
- OS << cast<MCDataFragment>(F).getContents();
+ OS << StringRef(cast<MCDataFragment>(F).getContents().data(),
+ cast<MCDataFragment>(F).getContents().size());
break;
case MCFragment::FT_Relaxable:
++stats::EmittedRelaxableFragments;
- OS << cast<MCRelaxableFragment>(F).getContents();
+ OS << StringRef(cast<MCRelaxableFragment>(F).getContents().data(),
+ cast<MCRelaxableFragment>(F).getContents().size());
break;
case MCFragment::FT_Fill: {
@@ -692,7 +694,7 @@ static void writeFragment(raw_ostream &OS, const MCAssembler &Asm,
case MCFragment::FT_LEB: {
const MCLEBFragment &LF = cast<MCLEBFragment>(F);
- OS << LF.getContents();
+ OS << StringRef(LF.getContents().data(), LF.getContents().size());
break;
}
@@ -722,27 +724,27 @@ static void writeFragment(raw_ostream &OS, const MCAssembler &Asm,
case MCFragment::FT_Dwarf: {
const MCDwarfLineAddrFragment &OF = cast<MCDwarfLineAddrFragment>(F);
- OS << OF.getContents();
+ OS << StringRef(OF.getContents().data(), OF.getContents().size());
break;
}
case MCFragment::FT_DwarfFrame: {
const MCDwarfCallFrameFragment &CF = cast<MCDwarfCallFrameFragment>(F);
- OS << CF.getContents();
+ OS << StringRef(CF.getContents().data(), CF.getContents().size());
break;
}
case MCFragment::FT_CVInlineLines: {
const auto &OF = cast<MCCVInlineLineTableFragment>(F);
- OS << OF.getContents();
+ OS << StringRef(OF.getContents().data(), OF.getContents().size());
break;
}
case MCFragment::FT_CVDefRange: {
const auto &DRF = cast<MCCVDefRangeFragment>(F);
- OS << DRF.getContents();
+ OS << StringRef(DRF.getContents().data(), DRF.getContents().size());
break;
}
case MCFragment::FT_PseudoProbe: {
const MCPseudoProbeAddrFragment &PF = cast<MCPseudoProbeAddrFragment>(F);
- OS << PF.getContents();
+ OS << StringRef(PF.getContents().data(), PF.getContents().size());
break;
}
}
@@ -994,10 +996,11 @@ bool MCAssembler::relaxInstruction(MCRelaxableFragment &F) {
// Encode the new instruction.
F.setInst(Relaxed);
- F.getFixups().clear();
- F.getContents().clear();
- getEmitter().encodeInstruction(Relaxed, F.getContents(), F.getFixups(),
- *F.getSubtargetInfo());
+ SmallVector<char, 16> Data;
+ SmallVector<MCFixup, 1> Fixups;
+ getEmitter().encodeInstruction(Relaxed, Data, Fixups, *F.getSubtargetInfo());
+ F.setContents(Data);
+ F.setFixups(Fixups);
return true;
}
@@ -1005,8 +1008,7 @@ bool MCAssembler::relaxLEB(MCLEBFragment &LF) {
const unsigned OldSize = static_cast<unsigned>(LF.getContents().size());
unsigned PadTo = OldSize;
int64_t Value;
- SmallVectorImpl<char> &Data = LF.getContents();
- LF.getFixups().clear();
+ LF.clearFixups();
// Use evaluateKnownAbsolute for Mach-O as a hack: .subsections_via_symbols
// requires that .uleb128 A-B is foldable where A and B reside in different
// fragments. This is used by __gcc_except_table.
@@ -1027,7 +1029,7 @@ bool MCAssembler::relaxLEB(MCLEBFragment &LF) {
if (UseZeroPad)
Value = 0;
}
- Data.clear();
+ SmallVector<char, 16> Data;
raw_svector_ostream OSE(Data);
// The compiler can generate EH table assembly that is impossible to assemble
// without either adding padding to an LEB fragment or adding extra padding
@@ -1037,7 +1039,8 @@ bool MCAssembler::relaxLEB(MCLEBFragment &LF) {
encodeSLEB128(Value, OSE, PadTo);
else
encodeULEB128(Value, OSE, PadTo);
- return OldSize != LF.getContents().size();
+ LF.setContents(Data);
+ return OldSize != Data.size();
}
/// Check if the branch crosses the boundary.
@@ -1107,19 +1110,19 @@ bool MCAssembler::relaxDwarfLineAddr(MCDwarfLineAddrFragment &DF) {
return WasRelaxed;
MCContext &Context = getContext();
- uint64_t OldSize = DF.getContents().size();
+ auto OldSize = DF.getContents().size();
int64_t AddrDelta;
bool Abs = DF.getAddrDelta().evaluateKnownAbsolute(AddrDelta, *this);
assert(Abs && "We created a line delta with an invalid expression");
(void)Abs;
int64_t LineDelta;
LineDelta = DF.getLineDelta();
- SmallVectorImpl<char> &Data = DF.getContents();
- Data.clear();
- DF.getFixups().clear();
+ SmallVector<char, 8> Data;
MCDwarfLineAddr::encode(Context, getDWARFLinetableParams(), LineDelta,
AddrDelta, Data);
+ DF.setContents(Data);
+ DF.clearFixups();
return OldSize != Data.size();
}
@@ -1138,12 +1141,11 @@ bool MCAssembler::relaxDwarfCallFrameFragment(MCDwarfCallFrameFragment &DF) {
return false;
}
- SmallVectorImpl<char> &Data = DF.getContents();
- uint64_t OldSize = Data.size();
- Data.clear();
- DF.getFixups().clear();
-
+ auto OldSize = DF.getContents().size();
+ SmallVector<char, 8> Data;
MCDwarfFrameEmitter::encodeAdvanceLoc(Context, Value, Data);
+ DF.setContents(Data);
+ DF.clearFixups();
return OldSize != Data.size();
}
@@ -1173,13 +1175,13 @@ bool MCAssembler::relaxPseudoProbeAddr(MCPseudoProbeAddrFragment &PF) {
bool Abs = PF.getAddrDelta().evaluateKnownAbsolute(AddrDelta, *this);
assert(Abs && "We created a pseudo probe with an invalid expression");
(void)Abs;
- SmallVectorImpl<char> &Data = PF.getContents();
- Data.clear();
+ SmallVector<char, 8> Data;
raw_svector_ostream OSE(Data);
- PF.getFixups().clear();
// AddrDelta is a signed integer
encodeSLEB128(AddrDelta, OSE, OldSize);
+ PF.setContents(Data);
+ PF.clearFixups();
return OldSize != Data.size();
}
diff --git a/llvm/lib/MC/MCCodeView.cpp b/llvm/lib/MC/MCCodeView.cpp
index e05374783d2b4..7a50e5c046147 100644
--- a/llvm/lib/MC/MCCodeView.cpp
+++ b/llvm/lib/MC/MCCodeView.cpp
@@ -512,7 +512,7 @@ void CodeViewContext::encodeInlineLineTable(const MCAssembler &Asm,
MCCVFunctionInfo *SiteInfo = getCVFunctionInfo(Frag.SiteFuncId);
- SmallVectorImpl<char> &Buffer = Frag.getContents();
+ SmallVector<char, 0> Buffer;
Buffer.clear(); // Clear old contents if we went through relaxation.
for (const MCCVLoc &Loc : Locs) {
// Exit early if our line table would produce an oversized InlineSiteSym
@@ -606,15 +606,14 @@ void CodeViewContext::encodeInlineLineTable(const MCAssembler &Asm,
compressAnnotation(BinaryAnnotationsOpCode::ChangeCodeLength, Buffer);
compressAnnotation(std::min(EndSymLength, LocAfterLength), Buffer);
+ Frag.setContents(Buffer);
}
void CodeViewContext::encodeDefRange(const MCAssembler &Asm,
MCCVDefRangeFragment &Frag) {
MCContext &Ctx = Asm.getContext();
- SmallVectorImpl<char> &Contents = Frag.getContents();
- Contents.clear();
- SmallVectorImpl<MCFixup> &Fixups = Frag.getFixups();
- Fixups.clear();
+ SmallVector<char, 0> Contents;
+ SmallVector<MCFixup, 0> Fixups;
raw_svector_ostream OS(Contents);
// Compute all the sizes up front.
@@ -694,4 +693,7 @@ void CodeViewContext::encodeDefRange(const MCAssembler &Asm,
GapStartOffset += GapSize + RangeSize;
}
}
+
+ Frag.setContents(Contents);
+ Frag.setFixups(Fixups);
}
diff --git a/llvm/lib/MC/MCELFStreamer.cpp b/llvm/lib/MC/MCELFStreamer.cpp
index 5a514f12ce549..41be72ebeb9ab 100644
--- a/llvm/lib/MC/MCELFStreamer.cpp
+++ b/llvm/lib/MC/MCELFStreamer.cpp
@@ -449,11 +449,13 @@ void MCELFStreamer::emitInstToData(const MCInst &Inst,
// Emit instruction directly into data fragment.
size_t FixupStartIndex = DF->getFixups().size();
size_t CodeOffset = DF->getContents().size();
- Assembler.getEmitter().encodeInstruction(Inst, DF->getContents(),
- DF->getFixups(), STI);
+ SmallVector<MCFixup, 1> Fixups;
+ Assembler.getEmitter().encodeInstruction(Inst, DF->getContentsForAppending(),
+ Fixups, STI);
+ DF->doneAppending();
+ DF->appendFixups(Fixups);
- auto Fixups = MutableArrayRef(DF->getFixups()).slice(FixupStartIndex);
- for (auto &Fixup : Fixups) {
+ for (auto &Fixup : MutableArrayRef(DF->getFixups()).slice(FixupStartIndex)) {
Fixup.setOffset(Fixup.getOffset() + CodeOffset);
if (Fixup.isLinkerRelaxable()) {
DF->setLinkerRelaxable();
diff --git a/llvm/lib/MC/MCFragment.cpp b/llvm/lib/MC/MCFragment.cpp
index d8db55c9a5f5d..859ee73447aad 100644
--- a/llvm/lib/MC/MCFragment.cpp
+++ b/llvm/lib/MC/MCFragment.cpp
@@ -28,57 +28,90 @@ MCFragment::MCFragment(FragmentType Kind, bool HasInstructions)
: Kind(Kind), HasInstructions(HasInstructions), AlignToBundleEnd(false),
LinkerRelaxable(false), AllowAutoPadding(false) {}
-void MCFragment::destroy() {
- switch (Kind) {
- case FT_Align:
- cast<MCAlignFragment>(this)->~MCAlignFragment();
- return;
- case FT_Data:
- cast<MCDataFragment>(this)->~MCDataFragment();
- return;
- case FT_Fill:
- cast<MCFillFragment>(this)->~MCFillFragment();
- return;
- case FT_Nops:
- cast<MCNopsFragment>(this)->~MCNopsFragment();
- return;
- case FT_Relaxable:
- cast<MCRelaxableFragment>(this)->~MCRelaxableFragment();
- return;
- case FT_Org:
- cast<MCOrgFragment>(this)->~MCOrgFragment();
- return;
- case FT_Dwarf:
- cast<MCDwarfLineAddrFragment>(this)->~MCDwarfLineAddrFragment();
- return;
- case FT_DwarfFrame:
- cast<MCDwarfCallFrameFragment>(this)->~MCDwarfCallFrameFragment();
- return;
- case FT_LEB:
- cast<MCLEBFragment>(this)->~MCLEBFragment();
- return;
- case FT_BoundaryAlign:
- cast<MCBoundaryAlignFragment>(this)->~MCBoundaryAlignFragment();
- return;
- case FT_SymbolId:
- cast<MCSymbolIdFragment>(this)->~MCSymbolIdFragment();
- return;
- case FT_CVInlineLines:
- cast<MCCVInlineLineTableFragment>(this)->~MCCVInlineLineTableFragment();
- return;
- case FT_CVDefRange:
- cast<MCCVDefRangeFragment>(this)->~MCCVDefRangeFragment();
- return;
- case FT_PseudoProbe:
- cast<MCPseudoProbeAddrFragment>(this)->~MCPseudoProbeAddrFragment();
- return;
+const MCSymbol *MCFragment::getAtom() const {
+ return cast<MCSectionMachO>(Parent)->getAtom(LayoutOrder);
+}
+
+SmallVectorImpl<char> &MCEncodedFragment::getContentsForAppending() {
+ SmallVectorImpl<char> &S = getParent()->ContentStorage;
+ if (ContentSize == 0) {
+ ContentStart = S.size();
+ } else if (ContentStart + ContentSize != S.size()) {
+ // If not empty and not at the storage end, move to the storage end.
+ auto I = std::exchange(ContentStart, S.size());
+ S.reserve(S.size() + ContentSize);
+ S.append(S.begin() + I, S.begin() + I + ContentSize);
}
+ return S;
}
-const MCSymbol *MCFragment::getAtom() const {
- return cast<MCSectionMachO>(Parent)->getAtom(LayoutOrder);
+void MCEncodedFragment::doneAppending() {
+ ContentSize = getParent()->ContentStorage.size() - ContentStart;
}
+void MCEncodedFragment::appendContents(ArrayRef<char> Contents) {
+ getContentsForAppending().append(Contents.begin(), Contents.end());
+ doneAppending();
+}
+
+void MCEncodedFragment::appendContents(size_t Num, char Elt) {
+ getContentsForAppending().append(Num, Elt);
+ doneAppending();
+}
+
+void MCEncodedFragment::setContents(ArrayRef<char> Contents) {
+ auto &S = getParent()->ContentStorage;
+ if (Contents.size() > ContentSize) {
+ ContentStart = S.size();
+ S.resize_for_overwrite(S.size() + Contents.size());
+ }
+ ContentSize = Contents.size();
+ llvm::copy(Contents, S.begin() + ContentStart);
+}
+
+MutableArrayRef<char> MCEncodedFragment::getContents() {
+ return MutableArrayRef(getParent()->ContentStorage)
+ .slice(ContentStart, ContentSize);
+}
+
+ArrayRef<char> MCEncodedFragment::getContents() const {
+ return ArrayRef(getParent()->ContentStorage).slice(ContentStart, ContentSize);
+}
+
+void MCEncodedFragment::addFixup(MCFixup Fixup) { appendFixups({Fixup}); }
+
+void MCEncodedFragment::appendFixups(ArrayRef<MCFixup> Fixups) {
+ auto &S = getParent()->FixupStorage;
+ if (FixupSize == 0) {
+ FixupStart = S.size();
+ } else if (FixupStart + FixupSize != S.size()) {
+ // If not empty and not at the storage end, move to the storage end.
+ auto I = std::exchange(FixupStart, S.size());
+ S.reserve(S.size() + ContentSize);
+ S.append(S.begin() + I, S.begin() + I + FixupSize);
+ }
+ FixupSize += Fixups.size();
+ S.append(Fixups.begin(), Fixups.end());
+}
+
+void MCEncodedFragment::setFixups(ArrayRef<MCFixup> Fixups) {
+ auto &S = getParent()->FixupStorage;
+ if (Fixups.size() > FixupSize) {
+ FixupStart = S.size();
+ S.resize_for_overwrite(S.size() + Fixups.size());
+ }
+ FixupSize = Fixups.size();
+ llvm::copy(Fixups, S.begin() + FixupStart);
+}
+
+MutableArrayRef<MCFixup> MCEncodedFragment::getFixups() {
+ return MutableArrayRef(getParent()->FixupStorage)
+ .slice(FixupStart, FixupSize);
+}
+
+ArrayRef<MCFixup> MCEncodedFragment::getFixups() const {
+ return ArrayRef(getParent()->FixupStorage).slice(FixupStart, FixupSize);
+}
#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
LLVM_DUMP_METHOD void MCFragment::dump() const {
@@ -132,7 +165,7 @@ LLVM_DUMP_METHOD void MCFragment::dump() const {
const auto *F = cast<MCDataFragment>(this);
if (F->isLinkerRelaxable())
OS << " LinkerRelaxable";
- const SmallVectorImpl<char> &Contents = F->getContents();
+ auto Contents = F->getContents();
OS << " Size:" << Contents.size() << " [";
for (unsigned i = 0, e = Contents.size(); i != e; ++i) {
if (i) OS << ",";
diff --git a/llvm/lib/MC/MCObjectStreamer.cpp b/llvm/lib/MC/MCObjectStreamer.cpp
index 3b2bb594675f0..b46e06825ca11 100644
--- a/llvm/lib/MC/MCObjectStreamer.cpp
+++ b/llvm/lib/MC/MCObjectStreamer.cpp
@@ -63,7 +63,7 @@ void MCObjectStreamer::resolvePendingFixups() {
PendingFixup.Fixup.setOffset(PendingFixup.Sym->getOffset() +
PendingFixup.Fixup.getOffset());
- // If the location symbol to relocate is in MCEncodedFragmentWithFixups,
+ // If the location symbol to relocate is in MCEncodedFragment,
// put the Fixup into location symbol's fragment. Otherwise
// put into PendingFixup.DF
MCFragment *SymFragment = PendingFixup.Sym->getFragment();
@@ -71,15 +71,9 @@ void MCObjectStreamer::resolvePendingFixups() {
case MCFragment::FT_Relaxable:
case MCFragment::FT_Dwarf:
case MCFragment::FT_PseudoProbe:
- cast<MCEncodedFragmentWithFixups<8, 1>>(SymFragment)
- ->getFixups()
- .push_back(PendingFixup.Fixup);
- break;
case MCFragment::FT_Data:
case MCFragment::FT_CVDefRange:
- cast<MCEncodedFragmentWithFixups<32, 4>>(SymFragment)
- ->getFixups()
- .push_back(PendingFixup.Fixup);
+ cast<MCEncodedFragment>(SymFragment)->addFixup(PendingFixup.Fixup);
break;
default:
PendingFixup.DF->addFixup(PendingFixup.Fixup);
@@ -398,10 +392,9 @@ void MCObjectStreamer::emitInstToData(const MCInst &Inst,
getAssembler().getEmitter().encodeInstruction(Inst, Code, Fixups, STI);
auto CodeOffset = DF->getContents().size();
- for (MCFixup &Fixup : Fixups) {
+ for (MCFixup &Fixup : Fixups)
Fixup.setOffset(Fixup.getOffset() + CodeOffset);
- DF->addFixup(Fixup);
- }
+ DF->appendFixups(Fixups);
DF->setHasInstructions(STI);
DF->appendContents(Code);
}
@@ -414,8 +407,11 @@ void MCObjectStreamer::emitInstToFragment(const MCInst &Inst,
getContext().allocFragment<MCRelaxableFragment>(Inst, STI);
insert(IF);
- getAssembler().getEmitter().encodeInstruction(Inst, IF->getContents(),
- IF->getFixups(), STI);
+ SmallVector<MCFixup, 1> Fixups;
+ getAssembler().getEmitter().encodeInstruction(
+ Inst, IF->getContentsForAppending(), Fixups, STI);
+ IF->doneAppending();
+ IF->appendFixups(Fixups);
}
#ifndef NDEBUG
diff --git a/llvm/lib/MC/MCSection.cpp b/llvm/lib/MC/MCSection.cpp
index 9aad34e0d32ad..04e1cc5bcf41b 100644
--- a/llvm/lib/MC/MCSection.cpp
+++ b/llvm/lib/MC/MCSection.cpp
@@ -12,6 +12,7 @@
#include "llvm/MC/MCContext.h"
#include "llvm/MC/MCFragment.h"
#include "llvm/MC/MCSymbol.h"
+#include "llvm/Support/Casting.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"
@@ -37,10 +38,13 @@ MCSymbol *MCSection::getEndSymbol(MCContext &Ctx) {
bool MCSection::hasEnded() const { return End && End->isInSection(); }
MCSection::~MCSection() {
+ // If ~MCRelaxableFragment becomes trivial (no longer store a MCInst member),
+ // this dtor can be made empty.
for (auto &[_, Chain] : Subsections) {
for (MCFragment *X = Chain.Head, *Y; X; X = Y) {
Y = X->Next;
- X->destroy();
+ if (auto *F = dyn_cast<MCRelaxableFragment>(X))
+ F->~MCRelaxableFragment();
}
}
}
diff --git a/llvm/lib/MC/MachObjectWriter.cpp b/llvm/lib/MC/MachObjectWriter.cpp
index 9377a74b377ea..a43ca2eaf9f08 100644
--- a/llvm/lib/MC/MachObjectWriter.cpp
+++ b/llvm/lib/MC/MachObjectWriter.cpp
@@ -808,8 +808,8 @@ uint64_t MachObjectWriter::writeObject() {
MCSection *CGProfileSection = getContext().getMachOSection(
"__LLVM", "__cg_profile", 0, SectionKind::getMetadata());
auto &Frag = cast<MCDataFragment>(*CGProfileSection->begin());
- Frag.getContents().clear();
- raw_svector_ostream OS(Frag.getContents());
+ Frag.clearContents();
+ raw_svector_ostream OS(Frag.getContentsForAppending());
for (const MCObjectWriter::CGProfileEntry &CGPE : CGProfile) {
uint32_t FromIndex = CGPE.From->getSymbol().getIndex();
uint32_t ToIndex = CGPE.To->getSymbol().getIndex();
@@ -817,6 +817,7 @@ uint64_t MachObjectWriter::writeObject() {
support::endian::write(OS, ToIndex, W.Endian);
support::endian::write(OS, CGPE.Count, W.Endian);
}
+ Frag.doneAppending();
}
unsigned NumSections = Asm.end() - Asm.begin();
diff --git a/llvm/lib/MC/WinCOFFObjectWriter.cpp b/llvm/lib/MC/WinCOFFObjectWriter.cpp
index 2ed0aa3d20fc9..b534aa63ed1b3 100644
--- a/llvm/lib/MC/WinCOFFObjectWriter.cpp
+++ b/llvm/lib/MC/WinCOFFObjectWriter.cpp
@@ -1071,7 +1071,7 @@ uint64_t WinCOFFWriter::writeObject() {
auto *Sec = getContext().getCOFFSection(".llvm_addrsig",
COFF::IMAGE_SCN_LNK_REMOVE);
auto *Frag = cast<MCDataFragment>(Sec->curFragList()->Head);
- raw_svector_ostream OS(Frag->getContents());
+ raw_svector_ostream OS(Frag->getContentsForAppending());
for (const MCSymbol *S : OWriter.AddrsigSyms) {
if (!S->isRegistered())
continue;
@@ -1086,6 +1086,7 @@ uint64_t WinCOFFWriter::writeObject() {
"executePostLayoutBinding!");
encodeULEB128(SectionMap[TargetSection]->Symbol->getIndex(), OS);
}
+ Frag->doneAppending();
}
// Create the contents of the .llvm.call-graph-profile section.
@@ -1093,7 +1094,7 @@ uint64_t WinCOFFWriter::writeObject() {
auto *Sec = getContext().getCOFFSection(".llvm.call-graph-profile",
COFF::IMAGE_SCN_LNK_REMOVE);
auto *Frag = cast<MCDataFragment>(Sec->curFragList()->Head);
- raw_svector_ostream OS(Frag->getContents());
+ raw_svector_ostream OS(Frag->getContentsForAppending());
for (const auto &CGPE : OWriter.getCGProfile()) {
uint32_t FromIndex = CGPE.From->getSymbol().getIndex();
uint32_t ToIndex = CGPE.To->getSymbol().getIndex();
@@ -1101,6 +1102,7 @@ uint64_t WinCOFFWriter::writeObject() {
support::endian::write(OS, ToIndex, W.Endian);
support::endian::write(OS, CGPE.Count, W.Endian);
}
+ Frag->doneAppending();
}
assignFileOffsets();
diff --git a/llvm/lib/Target/LoongArch/MCTargetDesc/LoongArchAsmBackend.cpp b/llvm/lib/Target/LoongArch/MCTargetDesc/LoongArchAsmBackend.cpp
index f7bb33d2790a1..64192ed6632de 100644
--- a/llvm/lib/Target/LoongArch/MCTargetDesc/LoongArchAsmBackend.cpp
+++ b/llvm/lib/Target/LoongArch/MCTargetDesc/LoongArchAsmBackend.cpp
@@ -297,9 +297,8 @@ bool LoongArchAsmBackend::relaxDwarfLineAddr(MCDwarfLineAddrFragment &DF,
int64_t LineDelta = DF.getLineDelta();
const MCExpr &AddrDelta = DF.getAddrDelta();
- SmallVectorImpl<char> &Data = DF.getContents();
- SmallVectorImpl<MCFixup> &Fixups = DF.getFixups();
- size_t OldSize = Data.size();
+ SmallVector<MCFixup, 1> Fixups;
+ size_t OldSize = DF.getContents().size();
int64_t Value;
if (AddrDelta.evaluateAsAbsolute(Value, *Asm))
@@ -308,8 +307,7 @@ bool LoongArchAsmBackend::relaxDwarfLineAddr(MCDwarfLineAddrFragment &DF,
assert(IsAbsolute && "CFA with invalid expression");
(void)IsAbsolute;
- Data.clear();
- Fixups.clear();
+ SmallVector<char> Data;
raw_svector_ostream OS(Data);
// INT64_MAX is a signal that this is actually a DW_LNE_end_sequence.
@@ -354,6 +352,8 @@ bool LoongArchAsmBackend::relaxDwarfLineAddr(MCDwarfLineAddrFragment &DF,
OS << uint8_t(dwarf::DW_LNS_copy);
}
+ DF.setContents(Data);
+ DF.setFixups(Fixups);
WasRelaxed = OldSize != Data.size();
return true;
}
@@ -361,9 +361,8 @@ bool LoongArchAsmBackend::relaxDwarfLineAddr(MCDwarfLineAddrFragment &DF,
bool LoongArchAsmBackend::relaxDwarfCFA(MCDwarfCallFrameFragment &DF,
bool &WasRelaxed) const {
const MCExpr &AddrDelta = DF.getAddrDelta();
- SmallVectorImpl<char> &Data = DF.getContents();
- SmallVectorImpl<MCFixup> &Fixups = DF.getFixups();
- size_t OldSize = Data.size();
+ SmallVector<MCFixup, 2> Fixups;
+ size_t OldSize = DF.getContents().size();
int64_t Value;
if (AddrDelta.evaluateAsAbsolute(Value, *Asm))
@@ -372,14 +371,12 @@ bool LoongArchAsmBackend::relaxDwarfCFA(MCDwarfCallFrameFragment &DF,
assert(IsAbsolute && "CFA with invalid expression");
(void)IsAbsolute;
- Data.clear();
- Fixups.clear();
- raw_svector_ostream OS(Data);
-
assert(getContext().getAsmInfo()->getMinInstAlignment() == 1 &&
"expected 1-byte alignment");
if (Value == 0) {
- WasRelaxed = OldSize != Data.size();
+ DF.clearContents();
+ DF.clearFixups();
+ WasRelaxed = OldSize != DF.getContents().size();
return true;
}
@@ -391,6 +388,8 @@ bool LoongArchAsmBackend::relaxDwarfCFA(MCDwarfCallFrameFragment &DF,
Fixups.push_back(MCFixup::create(Offset, MBE.getRHS(), std::get<1>(FK)));
};
+ SmallVector<char, 8> Data;
+ raw_svector_ostream OS(Data);
if (isUIntN(6, Value)) {
OS << uint8_t(dwarf::DW_CFA_advance_loc);
AddFixups(0, getRelocPairForSize(6));
@@ -409,6 +408,8 @@ bool LoongArchAsmBackend::relaxDwarfCFA(MCDwarfCallFrameFragment &DF,
} else {
llvm_unreachable("unsupported CFA encoding");
}
+ DF.setContents(Data);
+ DF.setFixups(Fixups);
WasRelaxed = OldSize != Data.size();
return true;
diff --git a/llvm/lib/Target/RISCV/MCTargetDesc/RISCVAsmBackend.cpp b/llvm/lib/Target/RISCV/MCTargetDesc/RISCVAsmBackend.cpp
index 8ed0aa8b832cf..24525ef3ab865 100644
--- a/llvm/lib/Target/RISCV/MCTargetDesc/RISCVAsmBackend.cpp
+++ b/llvm/lib/Target/RISCV/MCTargetDesc/RISCVAsmBackend.cpp
@@ -275,17 +275,16 @@ bool RISCVAsmBackend::relaxDwarfLineAddr(MCDwarfLineAddrFragment &DF,
int64_t LineDelta = DF.getLineDelta();
const MCExpr &AddrDelta = DF.getAddrDelta();
- SmallVectorImpl<char> &Data = DF.getContents();
- SmallVectorImpl<MCFixup> &Fixups = DF.getFixups();
- size_t OldSize = Data.size();
+ SmallVector<MCFixup, 1> Fixups;
+ size_t OldSize = DF.getContents().size();
int64_t Value;
[[maybe_unused]] bool IsAbsolute =
AddrDelta.evaluateKnownAbsolute(Value, *Asm);
assert(IsAbsolute && "CFA with invalid expression");
- Data.clear();
Fixups.clear();
+ SmallVector<char> Data;
raw_svector_ostream OS(Data);
// INT64_MAX is a signal that this is actually a DW_LNE_end_sequence.
@@ -330,6 +329,8 @@ bool RISCVAsmBackend::relaxDwarfLineAddr(MCDwarfLineAddrFragment &DF,
OS << uint8_t(dwarf::DW_LNS_copy);
}
+ DF.setContents(Data);
+ DF.setFixups(Fixups);
WasRelaxed = OldSize != Data.size();
return true;
}
@@ -337,9 +338,8 @@ bool RISCVAsmBackend::relaxDwarfLineAddr(MCDwarfLineAddrFragment &DF,
bool RISCVAsmBackend::relaxDwarfCFA(MCDwarfCallFrameFragment &DF,
bool &WasRelaxed) const {
const MCExpr &AddrDelta = DF.getAddrDelta();
- SmallVectorImpl<char> &Data = DF.getContents();
- SmallVectorImpl<MCFixup> &Fixups = DF.getFixups();
- size_t OldSize = Data.size();
+ SmallVector<MCFixup, 2> Fixups;
+ size_t OldSize = DF.getContents().size();
int64_t Value;
if (AddrDelta.evaluateAsAbsolute(Value, *Asm))
@@ -348,14 +348,12 @@ bool RISCVAsmBackend::relaxDwarfCFA(MCDwarfCallFrameFragment &DF,
AddrDelta.evaluateKnownAbsolute(Value, *Asm);
assert(IsAbsolute && "CFA with invalid expression");
- Data.clear();
- Fixups.clear();
- raw_svector_ostream OS(Data);
-
assert(getContext().getAsmInfo()->getMinInstAlignment() == 1 &&
"expected 1-byte alignment");
if (Value == 0) {
- WasRelaxed = OldSize != Data.size();
+ DF.clearContents();
+ DF.clearFixups();
+ WasRelaxed = OldSize != DF.getContents().size();
return true;
}
@@ -366,6 +364,8 @@ bool RISCVAsmBackend::relaxDwarfCFA(MCDwarfCallFrameFragment &DF,
Fixups.push_back(MCFixup::create(Offset, MBE.getRHS(), std::get<1>(Fixup)));
};
+ SmallVector<char, 8> Data;
+ raw_svector_ostream OS(Data);
if (isUIntN(6, Value)) {
OS << uint8_t(dwarf::DW_CFA_advance_loc);
AddFixups(0, {ELF::R_RISCV_SET6, ELF::R_RISCV_SUB6});
@@ -384,6 +384,8 @@ bool RISCVAsmBackend::relaxDwarfCFA(MCDwarfCallFrameFragment &DF,
} else {
llvm_unreachable("unsupported CFA encoding");
}
+ DF.setContents(Data);
+ DF.setFixups(Fixups);
WasRelaxed = OldSize != Data.size();
return true;
diff --git a/llvm/lib/Target/X86/MCTargetDesc/X86AsmBackend.cpp b/llvm/lib/Target/X86/MCTargetDesc/X86AsmBackend.cpp
index 361d7b488d048..77e2011361162 100644
--- a/llvm/lib/Target/X86/MCTargetDesc/X86AsmBackend.cpp
+++ b/llvm/lib/Target/X86/MCTargetDesc/X86AsmBackend.cpp
@@ -847,7 +847,7 @@ bool X86AsmBackend::padInstructionViaRelaxation(MCRelaxableFragment &RF,
return false;
RF.setInst(Relaxed);
RF.setContents(Code);
- RF.getFixups() = Fixups;
+ RF.setFixups(Fixups);
RemainingSize -= Delta;
return true;
}
More information about the llvm-commits
mailing list