[lld] [lld][macho] Strip .__uniq. and .llvm. hashes in -order_file (PR #140670)
via llvm-commits
llvm-commits at lists.llvm.org
Wed May 21 11:13:05 PDT 2025
https://github.com/SharonXSharon updated https://github.com/llvm/llvm-project/pull/140670
>From a1d230aa577a16f55152f4516b332417310528a9 Mon Sep 17 00:00:00 2001
From: Sharon Xu <sharonxu at fb.com>
Date: Mon, 19 May 2025 17:39:06 -0700
Subject: [PATCH 1/2] [lld][macho] Strip .__uniq. and .llvm. hashes when doing
symbol matching in -order_file
---
lld/MachO/SectionPriorities.cpp | 10 ++-
lld/MachO/SectionPriorities.h | 5 ++
lld/test/MachO/order-file-strip-hashes.s | 101 +++++++++++++++++++++++
3 files changed, 114 insertions(+), 2 deletions(-)
create mode 100644 lld/test/MachO/order-file-strip-hashes.s
diff --git a/lld/MachO/SectionPriorities.cpp b/lld/MachO/SectionPriorities.cpp
index 7a4a5d8465f64..213623b338472 100644
--- a/lld/MachO/SectionPriorities.cpp
+++ b/lld/MachO/SectionPriorities.cpp
@@ -245,12 +245,18 @@ DenseMap<const InputSection *, int> CallGraphSort::run() {
return orderMap;
}
+StringRef macho::PriorityBuilder::getRootSymbol(StringRef Name) {
+ auto [P0, S0] = Name.rsplit(".llvm.");
+ auto [P1, S1] = P0.rsplit(".__uniq.");
+ return P1;
+}
+
std::optional<int>
macho::PriorityBuilder::getSymbolPriority(const Defined *sym) {
if (sym->isAbsolute())
return std::nullopt;
- auto it = priorities.find(sym->getName());
+ auto it = priorities.find(getRootSymbol(sym->getName()));
if (it == priorities.end())
return std::nullopt;
const SymbolPriorityEntry &entry = it->second;
@@ -330,7 +336,7 @@ void macho::PriorityBuilder::parseOrderFile(StringRef path) {
break;
}
}
- symbol = line.trim();
+ symbol = getRootSymbol(line.trim());
if (!symbol.empty()) {
SymbolPriorityEntry &entry = priorities[symbol];
diff --git a/lld/MachO/SectionPriorities.h b/lld/MachO/SectionPriorities.h
index 44fb101990c51..0bbf238c2c116 100644
--- a/lld/MachO/SectionPriorities.h
+++ b/lld/MachO/SectionPriorities.h
@@ -69,6 +69,11 @@ class PriorityBuilder {
std::optional<int> getSymbolPriority(const Defined *sym);
llvm::DenseMap<llvm::StringRef, SymbolPriorityEntry> priorities;
llvm::MapVector<SectionPair, uint64_t> callGraphProfile;
+ /// Symbols can be appended with "(.__uniq.xxxx)?.llvm.yyyy" where "xxxx" and
+ /// "yyyy" are numbers that could change between builds. We need to use the
+ /// root symbol name before this suffix so these symbols can be matched with
+ /// profiles which may have different suffixes.
+ llvm::StringRef getRootSymbol(llvm::StringRef Name);
};
extern PriorityBuilder priorityBuilder;
diff --git a/lld/test/MachO/order-file-strip-hashes.s b/lld/test/MachO/order-file-strip-hashes.s
new file mode 100644
index 0000000000000..d7e21371ad9ca
--- /dev/null
+++ b/lld/test/MachO/order-file-strip-hashes.s
@@ -0,0 +1,101 @@
+# RUN: rm -rf %t && split-file %s %t
+# RUN: llvm-mc -filetype=obj -triple=arm64-apple-darwin %t/a.s -o %t/a.o
+
+# RUN: %lld -arch arm64 -lSystem -e _main -o %t/a.out %t/a.o -order_file %t/ord-1
+# RUN: llvm-nm --numeric-sort --format=just-symbols %t/a.out | FileCheck %s
+
+
+# .text
+# CHECK: A
+# CHECK: B
+# CHECK: C
+# .section __DATA,__objc_const
+# CHECK: _OBJC_$_CATEGORY_INSTANCE_METHODS_Foo_$_Cat1
+# CHECK: _ALPHABETIC_SORT_FIRST
+# CHECK: _OBJC_$_CATEGORY_Foo_$_Cat1
+# CHECK: _OBJC_$_CATEGORY_SOME_$_FOLDED
+# CHECK: _OBJC_$_CATEGORY_CLASS_METHODS_Foo_$_Cat2
+# CHECK: _OBJC_$_CATEGORY_Foo_$_Cat2
+# .section __DATA,__objc_data
+# CHECK: _OBJC_CLASS_$_Baz
+# CHECK: _OBJC_CLASS_$_Bar
+# CHECK: _OBJC_CLASS_$_Foo
+# CHECK: _OBJC_CLASS_$_Baz2
+
+
+#--- a.s
+.text
+.globl _main, A, _B, C.__uniq.111111111111111111111111111111111111111.llvm.2222222222222222222
+
+_main:
+ ret
+A:
+ ret
+F:
+ add w0, w0, #3
+ bl C.__uniq.111111111111111111111111111111111111111.llvm.2222222222222222222
+ ret
+C.__uniq.111111111111111111111111111111111111111.llvm.2222222222222222222:
+ add w0, w0, #2
+ bl A
+ ret
+D:
+ add w0, w0, #2
+ bl B
+ ret
+B:
+ add w0, w0, #1
+ bl A
+ ret
+E:
+ add w0, w0, #2
+ bl C.__uniq.111111111111111111111111111111111111111.llvm.2222222222222222222
+ ret
+
+.section __DATA,__objc_const
+# test multiple symbols at the same address, which will be alphabetic sorted based symbol names
+_OBJC_$_CATEGORY_CLASS_METHODS_Foo_$_Cat2:
+ .quad 789
+
+_OBJC_$_CATEGORY_SOME_$_FOLDED:
+_OBJC_$_CATEGORY_Foo_$_Cat1:
+_ALPHABETIC_SORT_FIRST:
+ .quad 123
+
+_OBJC_$_CATEGORY_Foo_$_Cat2:
+ .quad 222
+
+_OBJC_$_CATEGORY_INSTANCE_METHODS_Foo_$_Cat1:
+ .quad 456
+
+.section __DATA,__objc_data
+_OBJC_CLASS_$_Foo:
+ .quad 123
+
+_OBJC_CLASS_$_Bar.llvm.1234:
+ .quad 456
+
+_OBJC_CLASS_$_Baz:
+ .quad 789
+
+_OBJC_CLASS_$_Baz2:
+ .quad 999
+
+.section __DATA,__objc_classrefs
+.quad _OBJC_CLASS_$_Foo
+.quad _OBJC_CLASS_$_Bar.llvm.1234
+.quad _OBJC_CLASS_$_Baz
+
+.subsections_via_symbols
+
+
+#--- ord-1
+# change order, parital covered
+A
+B
+C.__uniq.555555555555555555555555555555555555555.llvm.6666666666666666666
+_OBJC_CLASS_$_Baz
+_OBJC_CLASS_$_Bar.__uniq.12345
+_OBJC_CLASS_$_Foo.__uniq.123.llvm.123456789
+_OBJC_$_CATEGORY_INSTANCE_METHODS_Foo_$_Cat1
+_OBJC_$_CATEGORY_Foo_$_Cat1.llvm.1234567
>From e607206a397fdf866bbcbae4ff1ac2b072f987e9 Mon Sep 17 00:00:00 2001
From: Sharon Xu <sharonxu at fb.com>
Date: Wed, 21 May 2025 10:45:14 -0700
Subject: [PATCH 2/2] create a new lld/include/lld/Common/Utils.h
---
lld/Common/CMakeLists.txt | 1 +
lld/Common/Utils.cpp | 20 ++++++++++++
lld/ELF/BPSectionOrderer.cpp | 3 +-
lld/MachO/BPSectionOrderer.cpp | 2 +-
lld/MachO/SectionPriorities.cpp | 11 ++-----
lld/MachO/SectionPriorities.h | 5 ---
.../lld/Common/BPSectionOrdererBase.inc | 20 +++---------
lld/include/lld/Common/Utils.h | 28 +++++++++++++++++
lld/test/MachO/order-file-strip-hashes.s | 31 +++++++------------
9 files changed, 71 insertions(+), 50 deletions(-)
create mode 100644 lld/Common/Utils.cpp
create mode 100644 lld/include/lld/Common/Utils.h
diff --git a/lld/Common/CMakeLists.txt b/lld/Common/CMakeLists.txt
index 4f503d04f7844..a9e8d72fb5ecc 100644
--- a/lld/Common/CMakeLists.txt
+++ b/lld/Common/CMakeLists.txt
@@ -34,6 +34,7 @@ add_lld_library(lldCommon
Strings.cpp
TargetOptionsCommandFlags.cpp
Timer.cpp
+ Utils.cpp
VCSVersion.inc
Version.cpp
diff --git a/lld/Common/Utils.cpp b/lld/Common/Utils.cpp
new file mode 100644
index 0000000000000..55d25fde9879e
--- /dev/null
+++ b/lld/Common/Utils.cpp
@@ -0,0 +1,20 @@
+//===- Utils.cpp ------------------------------------------------*- C++-*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+// The file defines untils functions that can be shared across archs.
+//===----------------------------------------------------------------------===//
+
+#include "lld/Common/Utils.h"
+
+using namespace llvm;
+using namespace lld;
+
+StringRef lld::utils::getRootSymbol(StringRef name) {
+ name.consume_back(".Tgm");
+ auto [P0, S0] = name.rsplit(".llvm.");
+ auto [P1, S1] = P0.rsplit(".__uniq.");
+ return P1;
+}
diff --git a/lld/ELF/BPSectionOrderer.cpp b/lld/ELF/BPSectionOrderer.cpp
index 793176c7725a3..f464b1d4518a4 100644
--- a/lld/ELF/BPSectionOrderer.cpp
+++ b/lld/ELF/BPSectionOrderer.cpp
@@ -81,7 +81,8 @@ DenseMap<const InputSectionBase *, int> elf::runBalancedPartitioning(
if (!sec || sec->size == 0 || !sec->isLive() || sec->repl != sec ||
!orderer.secToSym.try_emplace(sec, d).second)
return;
- rootSymbolToSectionIdxs[CachedHashStringRef(getRootSymbol(sym.getName()))]
+ rootSymbolToSectionIdxs[CachedHashStringRef(
+ lld::utils::getRootSymbol(sym.getName()))]
.insert(sections.size());
sections.emplace_back(sec);
};
diff --git a/lld/MachO/BPSectionOrderer.cpp b/lld/MachO/BPSectionOrderer.cpp
index ca273f7f798dc..268fb89502a88 100644
--- a/lld/MachO/BPSectionOrderer.cpp
+++ b/lld/MachO/BPSectionOrderer.cpp
@@ -124,7 +124,7 @@ DenseMap<const InputSection *, int> lld::macho::runBalancedPartitioning(
size_t idx = sections.size();
sections.emplace_back(isec);
for (auto *sym : BPOrdererMachO::getSymbols(*isec)) {
- auto rootName = getRootSymbol(sym->getName());
+ auto rootName = lld::utils::getRootSymbol(sym->getName());
rootSymbolToSectionIdxs[CachedHashStringRef(rootName)].insert(idx);
if (auto linkageName =
BPOrdererMachO::getResolvedLinkageName(rootName))
diff --git a/lld/MachO/SectionPriorities.cpp b/lld/MachO/SectionPriorities.cpp
index 213623b338472..5faedd9b790a5 100644
--- a/lld/MachO/SectionPriorities.cpp
+++ b/lld/MachO/SectionPriorities.cpp
@@ -21,6 +21,7 @@
#include "lld/Common/Args.h"
#include "lld/Common/CommonLinkerContext.h"
#include "lld/Common/ErrorHandler.h"
+#include "lld/Common/Utils.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/MapVector.h"
#include "llvm/Support/Path.h"
@@ -245,18 +246,12 @@ DenseMap<const InputSection *, int> CallGraphSort::run() {
return orderMap;
}
-StringRef macho::PriorityBuilder::getRootSymbol(StringRef Name) {
- auto [P0, S0] = Name.rsplit(".llvm.");
- auto [P1, S1] = P0.rsplit(".__uniq.");
- return P1;
-}
-
std::optional<int>
macho::PriorityBuilder::getSymbolPriority(const Defined *sym) {
if (sym->isAbsolute())
return std::nullopt;
- auto it = priorities.find(getRootSymbol(sym->getName()));
+ auto it = priorities.find(utils::getRootSymbol(sym->getName()));
if (it == priorities.end())
return std::nullopt;
const SymbolPriorityEntry &entry = it->second;
@@ -336,7 +331,7 @@ void macho::PriorityBuilder::parseOrderFile(StringRef path) {
break;
}
}
- symbol = getRootSymbol(line.trim());
+ symbol = utils::getRootSymbol(line.trim());
if (!symbol.empty()) {
SymbolPriorityEntry &entry = priorities[symbol];
diff --git a/lld/MachO/SectionPriorities.h b/lld/MachO/SectionPriorities.h
index 0bbf238c2c116..44fb101990c51 100644
--- a/lld/MachO/SectionPriorities.h
+++ b/lld/MachO/SectionPriorities.h
@@ -69,11 +69,6 @@ class PriorityBuilder {
std::optional<int> getSymbolPriority(const Defined *sym);
llvm::DenseMap<llvm::StringRef, SymbolPriorityEntry> priorities;
llvm::MapVector<SectionPair, uint64_t> callGraphProfile;
- /// Symbols can be appended with "(.__uniq.xxxx)?.llvm.yyyy" where "xxxx" and
- /// "yyyy" are numbers that could change between builds. We need to use the
- /// root symbol name before this suffix so these symbols can be matched with
- /// profiles which may have different suffixes.
- llvm::StringRef getRootSymbol(llvm::StringRef Name);
};
extern PriorityBuilder priorityBuilder;
diff --git a/lld/include/lld/Common/BPSectionOrdererBase.inc b/lld/include/lld/Common/BPSectionOrdererBase.inc
index 51dfb6471644a..bb2e55af1eb35 100644
--- a/lld/include/lld/Common/BPSectionOrdererBase.inc
+++ b/lld/include/lld/Common/BPSectionOrdererBase.inc
@@ -20,6 +20,7 @@
//===----------------------------------------------------------------------===//
#include "lld/Common/ErrorHandler.h"
+#include "lld/Common/Utils.h"
#include "llvm/ADT/CachedHashString.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/MapVector.h"
@@ -147,19 +148,6 @@ static SmallVector<std::pair<unsigned, UtilityNodes>> getUnsForCompression(
return sectionUns;
}
-/// Symbols can be appended with "(.__uniq.xxxx)?(.llvm.yyyy)?(.Tgm)?" where
-/// "xxxx" and "yyyy" are numbers that could change between builds, and .Tgm is
-/// the global merge functions suffix
-/// (see GlobalMergeFunc::MergingInstanceSuffix). We need to use the root symbol
-/// name before this suffix so these symbols can be matched with profiles which
-/// may have different suffixes.
-inline StringRef getRootSymbol(StringRef name) {
- name.consume_back(".Tgm");
- auto [P0, S0] = name.rsplit(".llvm.");
- auto [P1, S1] = P0.rsplit(".__uniq.");
- return P1;
-}
-
template <class D>
auto BPOrderer<D>::computeOrder(
StringRef profilePath, bool forFunctionCompression, bool forDataCompression,
@@ -197,7 +185,7 @@ auto BPOrderer<D>::computeOrder(
for (size_t timestamp = 0; timestamp < trace.size(); timestamp++) {
auto [_, parsedFuncName] = getParsedIRPGOName(
reader->getSymtab().getFuncOrVarName(trace[timestamp]));
- parsedFuncName = getRootSymbol(parsedFuncName);
+ parsedFuncName = lld::utils::getRootSymbol(parsedFuncName);
auto sectionIdxsIt =
rootSymbolToSectionIdxs.find(CachedHashStringRef(parsedFuncName));
@@ -375,7 +363,7 @@ auto BPOrderer<D>::computeOrder(
// 4?
uint64_t lastPage = endAddress / pageSize;
StringRef rootSymbol = D::getSymName(*sym);
- rootSymbol = getRootSymbol(rootSymbol);
+ rootSymbol = lld::utils::getRootSymbol(rootSymbol);
symbolToPageNumbers.try_emplace(rootSymbol, firstPage, lastPage);
if (auto resolvedLinkageName = D::getResolvedLinkageName(rootSymbol))
symbolToPageNumbers.try_emplace(resolvedLinkageName.value(),
@@ -393,7 +381,7 @@ auto BPOrderer<D>::computeOrder(
auto traceId = trace.FunctionNameRefs[step];
auto [Filename, ParsedFuncName] =
getParsedIRPGOName(reader->getSymtab().getFuncOrVarName(traceId));
- ParsedFuncName = getRootSymbol(ParsedFuncName);
+ ParsedFuncName = lld::utils::getRootSymbol(ParsedFuncName);
auto it = symbolToPageNumbers.find(ParsedFuncName);
if (it != symbolToPageNumbers.end()) {
auto &[firstPage, lastPage] = it->getValue();
diff --git a/lld/include/lld/Common/Utils.h b/lld/include/lld/Common/Utils.h
new file mode 100644
index 0000000000000..867b82eb34683
--- /dev/null
+++ b/lld/include/lld/Common/Utils.h
@@ -0,0 +1,28 @@
+//===- Utils.h ------------------------------------------------*- C++-*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+// The file declares untils functions that can be shared across archs.
+//===----------------------------------------------------------------------===//
+
+#ifndef LLD_UTILS_H
+#define LLD_UTILS_H
+
+#include "llvm/ADT/StringRef.h"
+
+namespace lld {
+namespace utils {
+
+/// Symbols can be appended with "(.__uniq.xxxx)?(.llvm.yyyy)?(.Tgm)?" where
+/// "xxxx" and "yyyy" are numbers that could change between builds, and .Tgm is
+/// the global merge functions suffix
+/// (see GlobalMergeFunc::MergingInstanceSuffix). We need to use the root symbol
+/// name before this suffix so these symbols can be matched with profiles which
+/// may have different suffixes.
+llvm::StringRef getRootSymbol(llvm::StringRef Name);
+} // namespace utils
+} // namespace lld
+
+#endif
diff --git a/lld/test/MachO/order-file-strip-hashes.s b/lld/test/MachO/order-file-strip-hashes.s
index d7e21371ad9ca..ee0f876bb67ef 100644
--- a/lld/test/MachO/order-file-strip-hashes.s
+++ b/lld/test/MachO/order-file-strip-hashes.s
@@ -3,25 +3,6 @@
# RUN: %lld -arch arm64 -lSystem -e _main -o %t/a.out %t/a.o -order_file %t/ord-1
# RUN: llvm-nm --numeric-sort --format=just-symbols %t/a.out | FileCheck %s
-
-
-# .text
-# CHECK: A
-# CHECK: B
-# CHECK: C
-# .section __DATA,__objc_const
-# CHECK: _OBJC_$_CATEGORY_INSTANCE_METHODS_Foo_$_Cat1
-# CHECK: _ALPHABETIC_SORT_FIRST
-# CHECK: _OBJC_$_CATEGORY_Foo_$_Cat1
-# CHECK: _OBJC_$_CATEGORY_SOME_$_FOLDED
-# CHECK: _OBJC_$_CATEGORY_CLASS_METHODS_Foo_$_Cat2
-# CHECK: _OBJC_$_CATEGORY_Foo_$_Cat2
-# .section __DATA,__objc_data
-# CHECK: _OBJC_CLASS_$_Baz
-# CHECK: _OBJC_CLASS_$_Bar
-# CHECK: _OBJC_CLASS_$_Foo
-# CHECK: _OBJC_CLASS_$_Baz2
-
#--- a.s
.text
@@ -99,3 +80,15 @@ _OBJC_CLASS_$_Bar.__uniq.12345
_OBJC_CLASS_$_Foo.__uniq.123.llvm.123456789
_OBJC_$_CATEGORY_INSTANCE_METHODS_Foo_$_Cat1
_OBJC_$_CATEGORY_Foo_$_Cat1.llvm.1234567
+
+# .text
+# CHECK: A
+# CHECK: B
+# CHECK: C
+# .section __DATA,__objc_const
+# CHECK: _OBJC_$_CATEGORY_INSTANCE_METHODS_Foo_$_Cat1
+# CHECK: _OBJC_$_CATEGORY_Foo_$_Cat1
+# .section __DATA,__objc_data
+# CHECK: _OBJC_CLASS_$_Baz
+# CHECK: _OBJC_CLASS_$_Bar
+# CHECK: _OBJC_CLASS_$_Foo
More information about the llvm-commits
mailing list