[lld] [LLD][ELF] add bp-* options in ELF (PR #120514)
via llvm-commits
llvm-commits at lists.llvm.org
Sun Jan 5 22:58:38 PST 2025
https://github.com/Colibrow updated https://github.com/llvm/llvm-project/pull/120514
>From c0b5ee6864e4e7acd8272d968c78ebb59d3a0870 Mon Sep 17 00:00:00 2001
From: xupengying <xpy66swsry at gmail.com>
Date: Thu, 19 Dec 2024 11:15:02 +0800
Subject: [PATCH 1/5] Add new ELF linker options for profile-guided section
ordering optimizations
---
lld/ELF/BPSectionOrderer.cpp | 61 +++++
lld/ELF/BPSectionOrderer.h | 134 ++++++++++
lld/ELF/CMakeLists.txt | 1 +
lld/ELF/Config.h | 6 +
lld/ELF/Driver.cpp | 49 ++++
lld/ELF/Options.td | 18 ++
lld/ELF/Writer.cpp | 11 +
lld/test/ELF/bp-section-orderer-stress.s | 104 ++++++++
lld/test/ELF/bp-section-orderer.s | 317 +++++++++++++++++++++++
lld/test/ELF/incompatible.s | 13 +
10 files changed, 714 insertions(+)
create mode 100644 lld/ELF/BPSectionOrderer.cpp
create mode 100644 lld/ELF/BPSectionOrderer.h
create mode 100644 lld/test/ELF/bp-section-orderer-stress.s
create mode 100644 lld/test/ELF/bp-section-orderer.s
diff --git a/lld/ELF/BPSectionOrderer.cpp b/lld/ELF/BPSectionOrderer.cpp
new file mode 100644
index 00000000000000..1fc6036e5dd9df
--- /dev/null
+++ b/lld/ELF/BPSectionOrderer.cpp
@@ -0,0 +1,61 @@
+//===- BPSectionOrderer.cpp------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#include "BPSectionOrderer.h"
+#include "Config.h"
+#include "InputFiles.h"
+#include "InputSection.h"
+#include "lld/Common/BPSectionOrdererBase.h"
+#include "llvm/ADT/DenseMap.h"
+
+#include "SymbolTable.h"
+#include "Symbols.h"
+
+using namespace llvm;
+using namespace lld::elf;
+
+llvm::DenseMap<const lld::elf::InputSectionBase *, int>
+lld::elf::runBalancedPartitioning(Ctx &ctx, llvm::StringRef profilePath,
+ bool forFunctionCompression,
+ bool forDataCompression,
+ bool compressionSortStartupFunctions,
+ bool verbose) {
+ size_t highestAvailablePriority = std::numeric_limits<int>::max();
+ // Collect all InputSectionBase objects from symbols and wrap them as
+ // BPSectionELF instances for balanced partitioning which follow the way
+ // '--symbol-ordering-file' does.
+ SmallVector<std::unique_ptr<BPSectionBase>> sections;
+
+ for (Symbol *sym : ctx.symtab->getSymbols())
+ if (sym->getSize() > 0)
+ if (auto *d = dyn_cast<Defined>(sym))
+ if (auto *sec = dyn_cast_or_null<InputSectionBase>(d->section))
+ sections.emplace_back(std::make_unique<BPSectionELF>(
+ sec, std::make_unique<BPSymbolELF>(d)));
+
+ for (ELFFileBase *file : ctx.objectFiles)
+ for (Symbol *sym : file->getLocalSymbols())
+ if (sym->getSize() > 0)
+ if (auto *d = dyn_cast<Defined>(sym))
+ if (auto *sec = dyn_cast_or_null<InputSectionBase>(d->section))
+ sections.emplace_back(std::make_unique<BPSectionELF>(
+ sec, std::make_unique<BPSymbolELF>(d)));
+
+ auto reorderedSections = BPSectionBase::reorderSectionsByBalancedPartitioning(
+ highestAvailablePriority, profilePath, forFunctionCompression,
+ forDataCompression, compressionSortStartupFunctions, verbose, sections);
+
+ DenseMap<const InputSectionBase *, int> result;
+ for (const auto &[sec, priority] : reorderedSections) {
+ auto *elfSection = cast<BPSectionELF>(sec);
+ result.try_emplace(
+ static_cast<const InputSectionBase *>(elfSection->getSection()),
+ static_cast<int>(priority));
+ }
+ return result;
+}
diff --git a/lld/ELF/BPSectionOrderer.h b/lld/ELF/BPSectionOrderer.h
new file mode 100644
index 00000000000000..9a55c19f159a91
--- /dev/null
+++ b/lld/ELF/BPSectionOrderer.h
@@ -0,0 +1,134 @@
+//===- BPSectionOrderer.h -------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+///
+/// This file uses Balanced Partitioning to order sections to improve startup
+/// time and compressed size.
+///
+//===----------------------------------------------------------------------===//
+
+#ifndef LLD_ELF_BPSECTION_ORDERER_H
+#define LLD_ELF_BPSECTION_ORDERER_H
+
+#include "InputFiles.h"
+#include "InputSection.h"
+#include "Relocations.h"
+#include "Symbols.h"
+#include "lld/Common/BPSectionOrdererBase.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/StringRef.h"
+#include "llvm/BinaryFormat/ELF.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/xxhash.h"
+
+namespace lld::elf {
+
+class InputSection;
+
+class BPSymbolELF : public BPSymbol {
+ const Symbol *sym;
+
+public:
+ explicit BPSymbolELF(const Symbol *s) : sym(s) {}
+
+ llvm::StringRef getName() const override { return sym->getName(); }
+
+ const Defined *asDefined() const {
+ return llvm::dyn_cast_or_null<Defined>(sym);
+ }
+
+ std::optional<uint64_t> getValue() const override {
+ if (auto *d = asDefined())
+ return d->value;
+ return {};
+ }
+
+ std::optional<uint64_t> getSize() const override {
+ if (auto *d = asDefined())
+ return d->size;
+ return {};
+ }
+
+ InputSectionBase *getInputSection() const {
+ if (auto *d = asDefined())
+ return llvm::dyn_cast_or_null<InputSectionBase>(d->section);
+ return nullptr;
+ }
+
+ const Symbol *getSymbol() const { return sym; }
+};
+
+class BPSectionELF : public BPSectionBase {
+ const InputSectionBase *isec;
+ std::unique_ptr<BPSymbolELF> symbol;
+
+public:
+ explicit BPSectionELF(const InputSectionBase *sec,
+ std::unique_ptr<BPSymbolELF> sym)
+ : isec(sec), symbol(std::move(sym)) {}
+
+ const void *getSection() const override { return isec; }
+
+ uint64_t getSize() const override { return isec->getSize(); }
+
+ bool isCodeSection() const override {
+ return isec->flags & llvm::ELF::SHF_EXECINSTR;
+ }
+
+ bool hasValidData() const override {
+ return isec && !isec->content().empty();
+ }
+
+ SmallVector<std::unique_ptr<BPSymbol>> getSymbols() const override {
+ SmallVector<std::unique_ptr<BPSymbol>> symbols;
+ if (auto *d = symbol->asDefined())
+ symbols.emplace_back(std::make_unique<BPSymbolELF>(d));
+ return symbols;
+ }
+
+ std::optional<StringRef>
+ getResolvedLinkageName(llvm::StringRef name) const override {
+ return {};
+ }
+
+ void getSectionHashes(llvm::SmallVectorImpl<uint64_t> &hashes,
+ const llvm::DenseMap<const void *, uint64_t>
+ §ionToIdx) const override {
+ constexpr unsigned windowSize = 4;
+
+ // Calculate content hashes
+ size_t size = isec->content().size();
+ for (size_t i = 0; i < size; i++) {
+ auto window = isec->content().drop_front(i).take_front(windowSize);
+ hashes.push_back(xxHash64(window));
+ }
+
+ // TODO: Calculate relocation hashes.
+ // Since in ELF, relocations are complex, but the effect without them are
+ // good enough, we just use 0 as their hash.
+
+ llvm::sort(hashes);
+ hashes.erase(std::unique(hashes.begin(), hashes.end()), hashes.end());
+ }
+
+ static bool classof(const BPSectionBase *s) { return true; }
+};
+
+/// Run Balanced Partitioning to find the optimal function and data order to
+/// improve startup time and compressed size.
+///
+/// It is important that -ffunction-sections and -fdata-sections are used to
+/// ensure functions and data are in their own sections and thus can be
+/// reordered.
+llvm::DenseMap<const lld::elf::InputSectionBase *, int>
+runBalancedPartitioning(Ctx &ctx, llvm::StringRef profilePath,
+ bool forFunctionCompression, bool forDataCompression,
+ bool compressionSortStartupFunctions, bool verbose);
+} // namespace lld::elf
+
+#endif
diff --git a/lld/ELF/CMakeLists.txt b/lld/ELF/CMakeLists.txt
index 83d816ddb0601e..298443cd6ea42c 100644
--- a/lld/ELF/CMakeLists.txt
+++ b/lld/ELF/CMakeLists.txt
@@ -37,6 +37,7 @@ add_lld_library(lldELF
Arch/X86.cpp
Arch/X86_64.cpp
ARMErrataFix.cpp
+ BPSectionOrderer.cpp
CallGraphSort.cpp
DWARF.cpp
Driver.cpp
diff --git a/lld/ELF/Config.h b/lld/ELF/Config.h
index 5b6b332cd597df..0a82eeecd93fce 100644
--- a/lld/ELF/Config.h
+++ b/lld/ELF/Config.h
@@ -264,6 +264,12 @@ struct Config {
bool armBe8 = false;
BsymbolicKind bsymbolic = BsymbolicKind::None;
CGProfileSortKind callGraphProfileSort;
+ llvm::StringRef irpgoProfilePath;
+ bool bpStartupFunctionSort = false;
+ bool bpCompressionSortStartupFunctions = false;
+ bool bpFunctionOrderForCompression = false;
+ bool bpDataOrderForCompression = false;
+ bool bpVerboseSectionOrderer = false;
bool checkSections;
bool checkDynamicRelocs;
std::optional<llvm::DebugCompressionType> compressDebugSections;
diff --git a/lld/ELF/Driver.cpp b/lld/ELF/Driver.cpp
index 9240f29d98d614..084abb3f5dc7f5 100644
--- a/lld/ELF/Driver.cpp
+++ b/lld/ELF/Driver.cpp
@@ -1255,6 +1255,55 @@ static void readConfigs(Ctx &ctx, opt::InputArgList &args) {
ctx.arg.bsymbolic = BsymbolicKind::All;
}
ctx.arg.callGraphProfileSort = getCGProfileSortKind(ctx, args);
+ ctx.arg.irpgoProfilePath = args.getLastArgValue(OPT_irpgo_profile);
+ ctx.arg.bpCompressionSortStartupFunctions =
+ args.hasFlag(OPT_bp_compression_sort_startup_functions,
+ OPT_no_bp_compression_sort_startup_functions, false);
+ if (auto *arg = args.getLastArg(OPT_bp_startup_sort)) {
+ StringRef startupSortStr = arg->getValue();
+ if (startupSortStr == "function") {
+ ctx.arg.bpStartupFunctionSort = true;
+ } else if (startupSortStr != "none") {
+ ErrAlways(ctx) << "unknown value '" + startupSortStr + "' for " +
+ arg->getSpelling();
+ }
+ if (startupSortStr != "none")
+ if (args.hasArg(OPT_call_graph_ordering_file))
+ ErrAlways(ctx) << "--bp-startup-sort=function is incompatible with "
+ "--call-graph-ordering-file";
+ }
+ if (ctx.arg.irpgoProfilePath.empty()) {
+ if (ctx.arg.bpStartupFunctionSort)
+ ErrAlways(ctx) << "--bp-startup-sort=function must be used with "
+ "--irpgo-profile";
+ if (ctx.arg.bpCompressionSortStartupFunctions)
+ ErrAlways(ctx)
+ << "--bp-compression-sort-startup-functions must be used with "
+ "--irpgo-profile";
+ }
+
+ if (auto *arg = args.getLastArg(OPT_bp_compression_sort)) {
+ StringRef compressionSortStr = arg->getValue();
+ if (compressionSortStr == "function") {
+ ctx.arg.bpFunctionOrderForCompression = true;
+ } else if (compressionSortStr == "data") {
+ ctx.arg.bpDataOrderForCompression = true;
+ } else if (compressionSortStr == "both") {
+ ctx.arg.bpFunctionOrderForCompression = true;
+ ctx.arg.bpDataOrderForCompression = true;
+ } else if (compressionSortStr != "none") {
+ ErrAlways(ctx) << "unknown value '" + compressionSortStr + "' for " +
+ arg->getSpelling();
+ }
+ if (ctx.arg.bpDataOrderForCompression ||
+ ctx.arg.bpFunctionOrderForCompression) {
+ if (args.getLastArg(OPT_call_graph_ordering_file) != nullptr) {
+ ErrAlways(ctx) << "--bp-compression-sort is incompatible with "
+ "--call-graph-ordering-file";
+ }
+ }
+ }
+ ctx.arg.bpVerboseSectionOrderer = args.hasArg(OPT_verbose_bp_section_orderer);
ctx.arg.checkSections =
args.hasFlag(OPT_check_sections, OPT_no_check_sections, true);
ctx.arg.chroot = args.getLastArgValue(OPT_chroot);
diff --git a/lld/ELF/Options.td b/lld/ELF/Options.td
index c31875305952fb..1948a5a524170c 100644
--- a/lld/ELF/Options.td
+++ b/lld/ELF/Options.td
@@ -141,6 +141,24 @@ def call_graph_profile_sort: JJ<"call-graph-profile-sort=">,
def : FF<"no-call-graph-profile-sort">, Alias<call_graph_profile_sort>, AliasArgs<["none"]>,
Flags<[HelpHidden]>;
+defm irpgo_profile: Eq<"irpgo-profile",
+ "Read the IRPGO profile for use with -bp-startup-sort and other profile-guided optimizations">;
+
+def bp_startup_sort: JJ<"bp-startup-sort=">,
+ MetaVarName<"[none,function]">,
+ HelpText<"Order sections based on profile data to improve startup time">;
+
+defm bp_compression_sort_startup_functions: BB<"bp-compression-sort-startup-functions",
+ "Order startup functions by balanced partition to improve compressed size in addition to startup time",
+ "Do not order startup function for compression">;
+
+def bp_compression_sort: JJ<"bp-compression-sort=">,
+ MetaVarName<"[none,function,data,both]">,
+ HelpText<"Order sections by balanced partition to improve compressed size">;
+
+def verbose_bp_section_orderer: FF<"verbose-bp-section-orderer">,
+ HelpText<"Print information on how many sections were ordered by balanced partitioning and a measure of the expected number of page faults">;
+
// --chroot doesn't have a help text because it is an internal option.
def chroot: Separate<["--"], "chroot">;
diff --git a/lld/ELF/Writer.cpp b/lld/ELF/Writer.cpp
index d5581ca3e1c921..26c4916bfcccf6 100644
--- a/lld/ELF/Writer.cpp
+++ b/lld/ELF/Writer.cpp
@@ -9,6 +9,7 @@
#include "Writer.h"
#include "AArch64ErrataFix.h"
#include "ARMErrataFix.h"
+#include "BPSectionOrderer.h"
#include "CallGraphSort.h"
#include "Config.h"
#include "InputFiles.h"
@@ -1083,6 +1084,16 @@ static void maybeShuffle(Ctx &ctx,
// Builds section order for handling --symbol-ordering-file.
static DenseMap<const InputSectionBase *, int> buildSectionOrder(Ctx &ctx) {
DenseMap<const InputSectionBase *, int> sectionOrder;
+ if (ctx.arg.bpStartupFunctionSort || ctx.arg.bpFunctionOrderForCompression ||
+ ctx.arg.bpDataOrderForCompression) {
+ TimeTraceScope timeScope("Balanced Partitioning Section Orderer");
+ sectionOrder = runBalancedPartitioning(
+ ctx, ctx.arg.bpStartupFunctionSort ? ctx.arg.irpgoProfilePath : "",
+ ctx.arg.bpFunctionOrderForCompression,
+ ctx.arg.bpDataOrderForCompression,
+ ctx.arg.bpCompressionSortStartupFunctions,
+ ctx.arg.bpVerboseSectionOrderer);
+ }
// Use the rarely used option --call-graph-ordering-file to sort sections.
if (!ctx.arg.callGraphProfile.empty())
return computeCallGraphProfileOrder(ctx);
diff --git a/lld/test/ELF/bp-section-orderer-stress.s b/lld/test/ELF/bp-section-orderer-stress.s
new file mode 100644
index 00000000000000..a25afafdadca28
--- /dev/null
+++ b/lld/test/ELF/bp-section-orderer-stress.s
@@ -0,0 +1,104 @@
+# REQUIRES: aarch64
+
+# Generate a large test case and check that the output is deterministic.
+
+# RUN: %python %s %t.s %t.proftext
+
+# RUN: llvm-mc -filetype=obj -triple=aarch64 %t.s -o %t.o
+# RUN: llvm-profdata merge %t.proftext -o %t.profdata
+
+# RUN: ld.lld -e _main --icf=all -o - %t.o --irpgo-profile=%t.profdata --bp-startup-sort=function --bp-compression-sort-startup-functions --bp-compression-sort=both | llvm-nm --numeric-sort --format=just-symbols - > %t.order1.txt
+# RUN: ld.lld -e _main --icf=all -o - %t.o --irpgo-profile=%t.profdata --bp-startup-sort=function --bp-compression-sort-startup-functions --bp-compression-sort=both | llvm-nm --numeric-sort --format=just-symbols - > %t.order2.txt
+# RUN: diff %t.order1.txt %t.order2.txt
+
+import random
+import sys
+
+assembly_filepath = sys.argv[1]
+proftext_filepath = sys.argv[2]
+
+random.seed(1234)
+num_functions = 1000
+num_data = 100
+num_traces = 10
+
+function_names = [f"f{n}" for n in range(num_functions)]
+data_names = [f"d{n}" for n in range(num_data)]
+profiled_functions = function_names[: int(num_functions / 2)]
+
+function_contents = [
+ f"""
+{name}:
+ add w0, w0, #{i % 4096}
+ add w1, w1, #{i % 10}
+ add w2, w0, #{i % 20}
+ adrp x3, {name}
+ ret
+"""
+ for i, name in enumerate(function_names)
+]
+
+data_contents = [
+ f"""
+{name}:
+ .ascii "s{i % 2}-{i % 3}-{i % 5}"
+ .xword {name}
+"""
+ for i, name in enumerate(data_names)
+]
+
+trace_contents = [
+ f"""
+# Weight
+1
+{", ".join(random.sample(profiled_functions, len(profiled_functions)))}
+"""
+ for i in range(num_traces)
+]
+
+profile_contents = [
+ f"""
+{name}
+# Func Hash:
+{i}
+# Num Counters:
+1
+# Counter Values:
+1
+"""
+ for i, name in enumerate(profiled_functions)
+]
+
+with open(assembly_filepath, "w") as f:
+ f.write(
+ f"""
+.text
+.globl _main
+
+_main:
+ ret
+
+{"".join(function_contents)}
+
+.data
+{"".join(data_contents)}
+
+"""
+ )
+
+with open(proftext_filepath, "w") as f:
+ f.write(
+ f"""
+:ir
+:temporal_prof_traces
+
+# Num Traces
+{num_traces}
+# Trace Stream Size:
+{num_traces}
+
+{"".join(trace_contents)}
+
+{"".join(profile_contents)}
+"""
+ )
diff --git a/lld/test/ELF/bp-section-orderer.s b/lld/test/ELF/bp-section-orderer.s
new file mode 100644
index 00000000000000..470c01f6bce5f2
--- /dev/null
+++ b/lld/test/ELF/bp-section-orderer.s
@@ -0,0 +1,317 @@
+# REQUIRES: aarch64
+
+# RUN: rm -rf %t && split-file %s %t && cd %t
+# RUN: llvm-mc -filetype=obj -triple=aarch64 a.s -o a.o
+# RUN: llvm-profdata merge a.proftext -o a.profdata
+# RUN: ld.lld -e main -o a.out a.o --irpgo-profile=a.profdata --bp-startup-sort=function --verbose-bp-section-orderer --icf=all 2>&1 | FileCheck %s --check-prefix=STARTUP
+
+# STARTUP: Ordered 3 sections using balanced partitioning
+
+# RUN: ld.lld -e main -o - a.o --symbol-ordering-file a.orderfile --irpgo-profile=a.profdata --bp-startup-sort=function | llvm-nm --numeric-sort --format=just-symbols - | FileCheck %s --check-prefix=ORDERFILE
+# RUN: ld.lld -e main -o - a.o --symbol-ordering-file a.orderfile --bp-compression-sort=both | llvm-nm --numeric-sort --format=just-symbols - | FileCheck %s --check-prefix=ORDERFILE
+
+# Rodata
+# ORDERFILE: s2
+# ORDERFILE: s1
+# ORDERFILE-DAG: s3
+
+# Functions
+# ORDERFILE: A
+# ORDERFILE: F
+# ORDERFILE: E
+# ORDERFILE: D
+# ORDERFILE-DAG: main
+# ORDERFILE-DAG: B
+# ORDERFILE-DAG: C
+
+# Data
+# ORDERFILE: r3
+# ORDERFILE: r2
+# ORDERFILE-DAG: r1
+# ORDERFILE-DAG: r4
+
+# RUN: ld.lld -e main -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=function 2>&1 | FileCheck %s --check-prefix=COMPRESSION-FUNC
+# RUN: ld.lld -e main -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=data 2>&1 | FileCheck %s --check-prefix=COMPRESSION-DATA
+# RUN: ld.lld -e main -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=both 2>&1 | FileCheck %s --check-prefix=COMPRESSION-BOTH
+# RUN: ld.lld -e main -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=both --irpgo-profile=a.profdata --bp-startup-sort=function 2>&1 | FileCheck %s --check-prefix=COMPRESSION-BOTH
+
+# COMPRESSION-FUNC: Ordered 7 sections using balanced partitioning
+# COMPRESSION-DATA: Ordered 7 sections using balanced partitioning
+# COMPRESSION-BOTH: Ordered 14 sections using balanced partitioning
+
+#--- a.proftext
+:ir
+:temporal_prof_traces
+# Num Traces
+1
+# Trace Stream Size:
+1
+# Weight
+1
+A, B, C
+
+A
+# Func Hash:
+1111
+# Num Counters:
+1
+# Counter Values:
+1
+
+B
+# Func Hash:
+2222
+# Num Counters:
+1
+# Counter Values:
+1
+
+C
+# Func Hash:
+3333
+# Num Counters:
+1
+# Counter Values:
+1
+
+D
+# Func Hash:
+4444
+# Num Counters:
+1
+# Counter Values:
+1
+
+#--- a.orderfile
+A
+F
+E
+D
+s2
+s1
+r3
+r2
+
+#--- a.c
+const char s1[] = "hello world";
+const char s2[] = "i am a string";
+const char s3[] = "this is s3";
+const char* r1 = s1;
+const char** r2 = &r1;
+const char*** r3 = &r2;
+const char* r4 = s2;
+void A() {
+ return;
+}
+
+int B(int a) {
+ A();
+ return a + 1;
+}
+
+int C(int a) {
+ A();
+ return a + 2;
+}
+
+int D(int a) {
+ return B(a + 2);
+}
+
+int E(int a) {
+ return C(a + 2);
+}
+
+int F(int a) {
+ return C(a + 3);
+}
+
+int main() {
+ return 0;
+}
+#--- gen
+clang --target=aarch64-linux-gnu -O0 -ffunction-sections -fdata-sections -fno-exceptions -fno-rtti -fno-asynchronous-unwind-tables -S a.c -o -
+;--- a.s
+ .text
+ .file "a.c"
+ .section .text.A,"ax", at progbits
+ .globl A // -- Begin function A
+ .p2align 2
+ .type A, at function
+A: // @A
+// %bb.0:
+ ret
+.Lfunc_end0:
+ .size A, .Lfunc_end0-A
+ // -- End function
+ .section .text.B,"ax", at progbits
+ .globl B // -- Begin function B
+ .p2align 2
+ .type B, at function
+B: // @B
+// %bb.0:
+ sub sp, sp, #32
+ stp x29, x30, [sp, #16] // 16-byte Folded Spill
+ add x29, sp, #16
+ stur w0, [x29, #-4]
+ bl A
+ ldur w8, [x29, #-4]
+ add w0, w8, #1
+ ldp x29, x30, [sp, #16] // 16-byte Folded Reload
+ add sp, sp, #32
+ ret
+.Lfunc_end1:
+ .size B, .Lfunc_end1-B
+ // -- End function
+ .section .text.C,"ax", at progbits
+ .globl C // -- Begin function C
+ .p2align 2
+ .type C, at function
+C: // @C
+// %bb.0:
+ sub sp, sp, #32
+ stp x29, x30, [sp, #16] // 16-byte Folded Spill
+ add x29, sp, #16
+ stur w0, [x29, #-4]
+ bl A
+ ldur w8, [x29, #-4]
+ add w0, w8, #2
+ ldp x29, x30, [sp, #16] // 16-byte Folded Reload
+ add sp, sp, #32
+ ret
+.Lfunc_end2:
+ .size C, .Lfunc_end2-C
+ // -- End function
+ .section .text.D,"ax", at progbits
+ .globl D // -- Begin function D
+ .p2align 2
+ .type D, at function
+D: // @D
+// %bb.0:
+ sub sp, sp, #32
+ stp x29, x30, [sp, #16] // 16-byte Folded Spill
+ add x29, sp, #16
+ stur w0, [x29, #-4]
+ ldur w8, [x29, #-4]
+ add w0, w8, #2
+ bl B
+ ldp x29, x30, [sp, #16] // 16-byte Folded Reload
+ add sp, sp, #32
+ ret
+.Lfunc_end3:
+ .size D, .Lfunc_end3-D
+ // -- End function
+ .section .text.E,"ax", at progbits
+ .globl E // -- Begin function E
+ .p2align 2
+ .type E, at function
+E: // @E
+// %bb.0:
+ sub sp, sp, #32
+ stp x29, x30, [sp, #16] // 16-byte Folded Spill
+ add x29, sp, #16
+ stur w0, [x29, #-4]
+ ldur w8, [x29, #-4]
+ add w0, w8, #2
+ bl C
+ ldp x29, x30, [sp, #16] // 16-byte Folded Reload
+ add sp, sp, #32
+ ret
+.Lfunc_end4:
+ .size E, .Lfunc_end4-E
+ // -- End function
+ .section .text.F,"ax", at progbits
+ .globl F // -- Begin function F
+ .p2align 2
+ .type F, at function
+F: // @F
+// %bb.0:
+ sub sp, sp, #32
+ stp x29, x30, [sp, #16] // 16-byte Folded Spill
+ add x29, sp, #16
+ stur w0, [x29, #-4]
+ ldur w8, [x29, #-4]
+ add w0, w8, #3
+ bl C
+ ldp x29, x30, [sp, #16] // 16-byte Folded Reload
+ add sp, sp, #32
+ ret
+.Lfunc_end5:
+ .size F, .Lfunc_end5-F
+ // -- End function
+ .section .text.main,"ax", at progbits
+ .globl main // -- Begin function main
+ .p2align 2
+ .type main, at function
+main: // @main
+// %bb.0:
+ sub sp, sp, #16
+ mov w0, wzr
+ str wzr, [sp, #12]
+ add sp, sp, #16
+ ret
+.Lfunc_end6:
+ .size main, .Lfunc_end6-main
+ // -- End function
+ .type s1, at object // @s1
+ .section .rodata.s1,"a", at progbits
+ .globl s1
+s1:
+ .asciz "hello world"
+ .size s1, 12
+
+ .type s2, at object // @s2
+ .section .rodata.s2,"a", at progbits
+ .globl s2
+s2:
+ .asciz "i am a string"
+ .size s2, 14
+
+ .type s3, at object // @s3
+ .section .rodata.s3,"a", at progbits
+ .globl s3
+s3:
+ .asciz "this is s3"
+ .size s3, 11
+
+ .type r1, at object // @r1
+ .section .data.r1,"aw", at progbits
+ .globl r1
+ .p2align 3, 0x0
+r1:
+ .xword s1
+ .size r1, 8
+
+ .type r2, at object // @r2
+ .section .data.r2,"aw", at progbits
+ .globl r2
+ .p2align 3, 0x0
+r2:
+ .xword r1
+ .size r2, 8
+
+ .type r3, at object // @r3
+ .section .data.r3,"aw", at progbits
+ .globl r3
+ .p2align 3, 0x0
+r3:
+ .xword r2
+ .size r3, 8
+
+ .type r4, at object // @r4
+ .section .data.r4,"aw", at progbits
+ .globl r4
+ .p2align 3, 0x0
+r4:
+ .xword s2
+ .size r4, 8
+
+ .section ".note.GNU-stack","", at progbits
+ .addrsig
+ .addrsig_sym A
+ .addrsig_sym B
+ .addrsig_sym C
+ .addrsig_sym s1
+ .addrsig_sym s2
+ .addrsig_sym r1
+ .addrsig_sym r2
diff --git a/lld/test/ELF/incompatible.s b/lld/test/ELF/incompatible.s
index 0d25acd857610b..05edc69cda7c24 100644
--- a/lld/test/ELF/incompatible.s
+++ b/lld/test/ELF/incompatible.s
@@ -1,3 +1,16 @@
+# REQUIRES: aarch64
+# RUN: rm -rf %t
+# RUN: not ld.lld -o /dev/null %t --irpgo-profile=/dev/null --bp-startup-sort=function --call-graph-ordering-file=/dev/null 2>&1 | FileCheck %s --check-prefix=IRPGO-ERR
+# IRPGO-ERR: --bp-startup-sort=function is incompatible with --call-graph-ordering-file
+# RUN: not ld.lld -o /dev/null --bp-compression-sort=function --call-graph-ordering-file /dev/null 2>&1 | FileCheck %s --check-prefix=COMPRESSION-ERR
+# COMPRESSION-ERR: --bp-compression-sort is incompatible with --call-graph-ordering-file
+# RUN: not ld.lld -o /dev/null --bp-compression-sort=malformed 2>&1 | FileCheck %s --check-prefix=COMPRESSION-MALFORM
+# COMPRESSION-MALFORM: unknown value 'malformed' for --bp-compression-sort=
+# RUN: not ld.lld -o /dev/null --bp-startup-sort=function 2>&1 | FileCheck %s --check-prefix=STARTUP
+# STARTUP: --bp-startup-sort=function must be used with --irpgo-profile
+# RUN: not ld.lld -o /dev/null --bp-compression-sort-startup-functions 2>&1 | FileCheck %s --check-prefix=STARTUP-COMPRESSION
+# STARTUP-COMPRESSION: --bp-compression-sort-startup-functions must be used with --irpgo-profile
+
// REQUIRES: x86,aarch64
// RUN: llvm-mc -filetype=obj -triple=x86_64-unknown-linux %s -o %ta.o
// RUN: llvm-mc -filetype=obj -triple=i686-unknown-linux %s -o %tb.o
>From 10f532556fcd0a76054d83cf5a89d7e230919eb9 Mon Sep 17 00:00:00 2001
From: xupengying <xpy66swsry at gmail.com>
Date: Thu, 19 Dec 2024 13:05:42 +0800
Subject: [PATCH 2/5] fixup! Add new ELF linker options for profile-guided
section ordering optimizations
---
lld/ELF/BPSectionOrderer.cpp | 16 ++++++++++++++++
lld/ELF/BPSectionOrderer.h | 25 +++----------------------
lld/test/ELF/bp-section-orderer.s | 7 +++++++
lld/test/ELF/incompatible.s | 6 ------
4 files changed, 26 insertions(+), 28 deletions(-)
diff --git a/lld/ELF/BPSectionOrderer.cpp b/lld/ELF/BPSectionOrderer.cpp
index 1fc6036e5dd9df..69dfde7c9c6412 100644
--- a/lld/ELF/BPSectionOrderer.cpp
+++ b/lld/ELF/BPSectionOrderer.cpp
@@ -12,6 +12,7 @@
#include "InputSection.h"
#include "lld/Common/BPSectionOrdererBase.h"
#include "llvm/ADT/DenseMap.h"
+#include "llvm/Support/xxhash.h"
#include "SymbolTable.h"
#include "Symbols.h"
@@ -19,6 +20,21 @@
using namespace llvm;
using namespace lld::elf;
+void BPSectionELF::getSectionHashes(
+ llvm::SmallVectorImpl<uint64_t> &hashes,
+ const llvm::DenseMap<const void *, uint64_t> §ionToIdx) const {
+ constexpr unsigned windowSize = 4;
+
+ size_t size = isec->content().size();
+ for (size_t i = 0; i < size; i++) {
+ auto window = isec->content().drop_front(i).take_front(windowSize);
+ hashes.push_back(xxHash64(window));
+ }
+
+ llvm::sort(hashes);
+ hashes.erase(std::unique(hashes.begin(), hashes.end()), hashes.end());
+}
+
llvm::DenseMap<const lld::elf::InputSectionBase *, int>
lld::elf::runBalancedPartitioning(Ctx &ctx, llvm::StringRef profilePath,
bool forFunctionCompression,
diff --git a/lld/ELF/BPSectionOrderer.h b/lld/ELF/BPSectionOrderer.h
index 9a55c19f159a91..c039408591b22f 100644
--- a/lld/ELF/BPSectionOrderer.h
+++ b/lld/ELF/BPSectionOrderer.h
@@ -24,7 +24,6 @@
#include "llvm/ADT/StringRef.h"
#include "llvm/BinaryFormat/ELF.h"
#include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/xxhash.h"
namespace lld::elf {
@@ -38,9 +37,7 @@ class BPSymbolELF : public BPSymbol {
llvm::StringRef getName() const override { return sym->getName(); }
- const Defined *asDefined() const {
- return llvm::dyn_cast_or_null<Defined>(sym);
- }
+ const Defined *asDefined() const { return llvm::dyn_cast<Defined>(sym); }
std::optional<uint64_t> getValue() const override {
if (auto *d = asDefined())
@@ -98,23 +95,7 @@ class BPSectionELF : public BPSectionBase {
void getSectionHashes(llvm::SmallVectorImpl<uint64_t> &hashes,
const llvm::DenseMap<const void *, uint64_t>
- §ionToIdx) const override {
- constexpr unsigned windowSize = 4;
-
- // Calculate content hashes
- size_t size = isec->content().size();
- for (size_t i = 0; i < size; i++) {
- auto window = isec->content().drop_front(i).take_front(windowSize);
- hashes.push_back(xxHash64(window));
- }
-
- // TODO: Calculate relocation hashes.
- // Since in ELF, relocations are complex, but the effect without them are
- // good enough, we just use 0 as their hash.
-
- llvm::sort(hashes);
- hashes.erase(std::unique(hashes.begin(), hashes.end()), hashes.end());
- }
+ §ionToIdx) const override;
static bool classof(const BPSectionBase *s) { return true; }
};
@@ -125,7 +106,7 @@ class BPSectionELF : public BPSectionBase {
/// It is important that -ffunction-sections and -fdata-sections are used to
/// ensure functions and data are in their own sections and thus can be
/// reordered.
-llvm::DenseMap<const lld::elf::InputSectionBase *, int>
+llvm::DenseMap<const InputSectionBase *, int>
runBalancedPartitioning(Ctx &ctx, llvm::StringRef profilePath,
bool forFunctionCompression, bool forDataCompression,
bool compressionSortStartupFunctions, bool verbose);
diff --git a/lld/test/ELF/bp-section-orderer.s b/lld/test/ELF/bp-section-orderer.s
index 470c01f6bce5f2..7bd7f278564e7f 100644
--- a/lld/test/ELF/bp-section-orderer.s
+++ b/lld/test/ELF/bp-section-orderer.s
@@ -1,6 +1,13 @@
# REQUIRES: aarch64
# RUN: rm -rf %t && split-file %s %t && cd %t
+# RUN: not ld.lld -o /dev/null --bp-compression-sort=malformed 2>&1 | FileCheck %s --check-prefix=COMPRESSION-MALFORM
+# COMPRESSION-MALFORM: unknown value 'malformed' for --bp-compression-sort=
+# RUN: not ld.lld -o /dev/null --bp-startup-sort=function 2>&1 | FileCheck %s --check-prefix=IRPGO-STARTUP
+# IRPGO-STARTUP: --bp-startup-sort=function must be used with --irpgo-profile
+# RUN: not ld.lld -o /dev/null --bp-compression-sort-startup-functions 2>&1 | FileCheck %s --check-prefix=STARTUP-COMPRESSION
+# STARTUP-COMPRESSION: --bp-compression-sort-startup-functions must be used with --irpgo-profile
+
# RUN: llvm-mc -filetype=obj -triple=aarch64 a.s -o a.o
# RUN: llvm-profdata merge a.proftext -o a.profdata
# RUN: ld.lld -e main -o a.out a.o --irpgo-profile=a.profdata --bp-startup-sort=function --verbose-bp-section-orderer --icf=all 2>&1 | FileCheck %s --check-prefix=STARTUP
diff --git a/lld/test/ELF/incompatible.s b/lld/test/ELF/incompatible.s
index 05edc69cda7c24..7b6f557bdac2e0 100644
--- a/lld/test/ELF/incompatible.s
+++ b/lld/test/ELF/incompatible.s
@@ -4,12 +4,6 @@
# IRPGO-ERR: --bp-startup-sort=function is incompatible with --call-graph-ordering-file
# RUN: not ld.lld -o /dev/null --bp-compression-sort=function --call-graph-ordering-file /dev/null 2>&1 | FileCheck %s --check-prefix=COMPRESSION-ERR
# COMPRESSION-ERR: --bp-compression-sort is incompatible with --call-graph-ordering-file
-# RUN: not ld.lld -o /dev/null --bp-compression-sort=malformed 2>&1 | FileCheck %s --check-prefix=COMPRESSION-MALFORM
-# COMPRESSION-MALFORM: unknown value 'malformed' for --bp-compression-sort=
-# RUN: not ld.lld -o /dev/null --bp-startup-sort=function 2>&1 | FileCheck %s --check-prefix=STARTUP
-# STARTUP: --bp-startup-sort=function must be used with --irpgo-profile
-# RUN: not ld.lld -o /dev/null --bp-compression-sort-startup-functions 2>&1 | FileCheck %s --check-prefix=STARTUP-COMPRESSION
-# STARTUP-COMPRESSION: --bp-compression-sort-startup-functions must be used with --irpgo-profile
// REQUIRES: x86,aarch64
// RUN: llvm-mc -filetype=obj -triple=x86_64-unknown-linux %s -o %ta.o
>From 419de887a12ad61de8e50bf913292a63c7539427 Mon Sep 17 00:00:00 2001
From: xupengying <xpy66swsry at gmail.com>
Date: Thu, 19 Dec 2024 13:09:03 +0800
Subject: [PATCH 3/5] fixup! Add new ELF linker options for profile-guided
section ordering optimizations
---
lld/test/ELF/bp-section-orderer.s | 13 +++++++------
1 file changed, 7 insertions(+), 6 deletions(-)
diff --git a/lld/test/ELF/bp-section-orderer.s b/lld/test/ELF/bp-section-orderer.s
index 7bd7f278564e7f..433ebe9635f0dc 100644
--- a/lld/test/ELF/bp-section-orderer.s
+++ b/lld/test/ELF/bp-section-orderer.s
@@ -1,12 +1,6 @@
# REQUIRES: aarch64
# RUN: rm -rf %t && split-file %s %t && cd %t
-# RUN: not ld.lld -o /dev/null --bp-compression-sort=malformed 2>&1 | FileCheck %s --check-prefix=COMPRESSION-MALFORM
-# COMPRESSION-MALFORM: unknown value 'malformed' for --bp-compression-sort=
-# RUN: not ld.lld -o /dev/null --bp-startup-sort=function 2>&1 | FileCheck %s --check-prefix=IRPGO-STARTUP
-# IRPGO-STARTUP: --bp-startup-sort=function must be used with --irpgo-profile
-# RUN: not ld.lld -o /dev/null --bp-compression-sort-startup-functions 2>&1 | FileCheck %s --check-prefix=STARTUP-COMPRESSION
-# STARTUP-COMPRESSION: --bp-compression-sort-startup-functions must be used with --irpgo-profile
# RUN: llvm-mc -filetype=obj -triple=aarch64 a.s -o a.o
# RUN: llvm-profdata merge a.proftext -o a.profdata
@@ -46,6 +40,13 @@
# COMPRESSION-DATA: Ordered 7 sections using balanced partitioning
# COMPRESSION-BOTH: Ordered 14 sections using balanced partitioning
+# RUN: not ld.lld -o /dev/null --bp-compression-sort=malformed 2>&1 | FileCheck %s --check-prefix=COMPRESSION-MALFORM
+# COMPRESSION-MALFORM: unknown value 'malformed' for --bp-compression-sort=
+# RUN: not ld.lld -o /dev/null --bp-startup-sort=function 2>&1 | FileCheck %s --check-prefix=IRPGO-STARTUP
+# IRPGO-STARTUP: --bp-startup-sort=function must be used with --irpgo-profile
+# RUN: not ld.lld -o /dev/null --bp-compression-sort-startup-functions 2>&1 | FileCheck %s --check-prefix=STARTUP-COMPRESSION
+# STARTUP-COMPRESSION: --bp-compression-sort-startup-functions must be used with --irpgo-profile
+
#--- a.proftext
:ir
:temporal_prof_traces
>From 44c0aae8905710b69449fb1bad5198c05f6fa2c8 Mon Sep 17 00:00:00 2001
From: xupengying <xpy66swsry at gmail.com>
Date: Fri, 3 Jan 2025 16:45:00 +0800
Subject: [PATCH 4/5] fixup! Add new ELF linker options for profile-guided
section ordering optimizations
---
lld/test/ELF/bp-section-orderer-stress.s | 8 ++---
lld/test/ELF/bp-section-orderer.s | 43 +++++++++++-------------
2 files changed, 24 insertions(+), 27 deletions(-)
diff --git a/lld/test/ELF/bp-section-orderer-stress.s b/lld/test/ELF/bp-section-orderer-stress.s
index a25afafdadca28..04da718f1eda48 100644
--- a/lld/test/ELF/bp-section-orderer-stress.s
+++ b/lld/test/ELF/bp-section-orderer-stress.s
@@ -7,8 +7,8 @@
# RUN: llvm-mc -filetype=obj -triple=aarch64 %t.s -o %t.o
# RUN: llvm-profdata merge %t.proftext -o %t.profdata
-# RUN: ld.lld -e _main --icf=all -o - %t.o --irpgo-profile=%t.profdata --bp-startup-sort=function --bp-compression-sort-startup-functions --bp-compression-sort=both | llvm-nm --numeric-sort --format=just-symbols - > %t.order1.txt
-# RUN: ld.lld -e _main --icf=all -o - %t.o --irpgo-profile=%t.profdata --bp-startup-sort=function --bp-compression-sort-startup-functions --bp-compression-sort=both | llvm-nm --numeric-sort --format=just-symbols - > %t.order2.txt
+# RUN: ld.lld -e _start --icf=all -o - %t.o --irpgo-profile=%t.profdata --bp-startup-sort=function --bp-compression-sort-startup-functions --bp-compression-sort=both | llvm-nm --numeric-sort --format=just-symbols - > %t.order1.txt
+# RUN: ld.lld -e _start --icf=all -o - %t.o --irpgo-profile=%t.profdata --bp-startup-sort=function --bp-compression-sort-startup-functions --bp-compression-sort=both | llvm-nm --numeric-sort --format=just-symbols - > %t.order2.txt
# RUN: diff %t.order1.txt %t.order2.txt
import random
@@ -73,9 +73,9 @@ with open(assembly_filepath, "w") as f:
f.write(
f"""
.text
-.globl _main
+.globl _start
-_main:
+_start:
ret
{"".join(function_contents)}
diff --git a/lld/test/ELF/bp-section-orderer.s b/lld/test/ELF/bp-section-orderer.s
index 433ebe9635f0dc..aa460d4334a721 100644
--- a/lld/test/ELF/bp-section-orderer.s
+++ b/lld/test/ELF/bp-section-orderer.s
@@ -4,12 +4,12 @@
# RUN: llvm-mc -filetype=obj -triple=aarch64 a.s -o a.o
# RUN: llvm-profdata merge a.proftext -o a.profdata
-# RUN: ld.lld -e main -o a.out a.o --irpgo-profile=a.profdata --bp-startup-sort=function --verbose-bp-section-orderer --icf=all 2>&1 | FileCheck %s --check-prefix=STARTUP
+# RUN: ld.lld -e _start -o a.out a.o --irpgo-profile=a.profdata --bp-startup-sort=function --verbose-bp-section-orderer --icf=all 2>&1 | FileCheck %s --check-prefix=STARTUP
# STARTUP: Ordered 3 sections using balanced partitioning
-# RUN: ld.lld -e main -o - a.o --symbol-ordering-file a.orderfile --irpgo-profile=a.profdata --bp-startup-sort=function | llvm-nm --numeric-sort --format=just-symbols - | FileCheck %s --check-prefix=ORDERFILE
-# RUN: ld.lld -e main -o - a.o --symbol-ordering-file a.orderfile --bp-compression-sort=both | llvm-nm --numeric-sort --format=just-symbols - | FileCheck %s --check-prefix=ORDERFILE
+# RUN: ld.lld -e _start -o - a.o --symbol-ordering-file a.orderfile --irpgo-profile=a.profdata --bp-startup-sort=function | llvm-nm --numeric-sort --format=just-symbols - | FileCheck %s --check-prefix=ORDERFILE
+# RUN: ld.lld -e _start -o - a.o --symbol-ordering-file a.orderfile --bp-compression-sort=both | llvm-nm --numeric-sort --format=just-symbols - | FileCheck %s --check-prefix=ORDERFILE
# Rodata
# ORDERFILE: s2
@@ -21,7 +21,7 @@
# ORDERFILE: F
# ORDERFILE: E
# ORDERFILE: D
-# ORDERFILE-DAG: main
+# ORDERFILE-DAG: _start
# ORDERFILE-DAG: B
# ORDERFILE-DAG: C
@@ -31,21 +31,21 @@
# ORDERFILE-DAG: r1
# ORDERFILE-DAG: r4
-# RUN: ld.lld -e main -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=function 2>&1 | FileCheck %s --check-prefix=COMPRESSION-FUNC
-# RUN: ld.lld -e main -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=data 2>&1 | FileCheck %s --check-prefix=COMPRESSION-DATA
-# RUN: ld.lld -e main -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=both 2>&1 | FileCheck %s --check-prefix=COMPRESSION-BOTH
-# RUN: ld.lld -e main -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=both --irpgo-profile=a.profdata --bp-startup-sort=function 2>&1 | FileCheck %s --check-prefix=COMPRESSION-BOTH
+# RUN: ld.lld -e _start -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=function 2>&1 | FileCheck %s --check-prefix=COMPRESSION-FUNC
+# RUN: ld.lld -e _start -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=data 2>&1 | FileCheck %s --check-prefix=COMPRESSION-DATA
+# RUN: ld.lld -e _start -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=both 2>&1 | FileCheck %s --check-prefix=COMPRESSION-BOTH
+# RUN: ld.lld -e _start -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=both --irpgo-profile=a.profdata --bp-startup-sort=function 2>&1 | FileCheck %s --check-prefix=COMPRESSION-BOTH
# COMPRESSION-FUNC: Ordered 7 sections using balanced partitioning
# COMPRESSION-DATA: Ordered 7 sections using balanced partitioning
# COMPRESSION-BOTH: Ordered 14 sections using balanced partitioning
-# RUN: not ld.lld -o /dev/null --bp-compression-sort=malformed 2>&1 | FileCheck %s --check-prefix=COMPRESSION-MALFORM
-# COMPRESSION-MALFORM: unknown value 'malformed' for --bp-compression-sort=
-# RUN: not ld.lld -o /dev/null --bp-startup-sort=function 2>&1 | FileCheck %s --check-prefix=IRPGO-STARTUP
-# IRPGO-STARTUP: --bp-startup-sort=function must be used with --irpgo-profile
-# RUN: not ld.lld -o /dev/null --bp-compression-sort-startup-functions 2>&1 | FileCheck %s --check-prefix=STARTUP-COMPRESSION
-# STARTUP-COMPRESSION: --bp-compression-sort-startup-functions must be used with --irpgo-profile
+# RUN: not ld.lld --bp-compression-sort=malformed 2>&1 | FileCheck %s --check-prefix=COMPRESSION-MALFORM
+# COMPRESSION-MALFORM: error: unknown value 'malformed' for --bp-compression-sort=
+# RUN: not ld.lld --bp-startup-sort=function 2>&1 | FileCheck %s --check-prefix=IRPGO-STARTUP
+# IRPGO-STARTUP: error: --bp-startup-sort=function must be used with --irpgo-profile
+# RUN: not ld.lld --bp-compression-sort-startup-functions 2>&1 | FileCheck %s --check-prefix=STARTUP-COMPRESSION
+# STARTUP-COMPRESSION: error: --bp-compression-sort-startup-functions must be used with --irpgo-profile
#--- a.proftext
:ir
@@ -134,7 +134,7 @@ int F(int a) {
return C(a + 3);
}
-int main() {
+int _start() {
return 0;
}
#--- gen
@@ -247,19 +247,16 @@ F: // @F
.Lfunc_end5:
.size F, .Lfunc_end5-F
// -- End function
- .section .text.main,"ax", at progbits
- .globl main // -- Begin function main
+ .section .text._start,"ax", at progbits
+ .globl _start // -- Begin function _start
.p2align 2
- .type main, at function
-main: // @main
+ .type _start, at function
+_start: // @_start
// %bb.0:
- sub sp, sp, #16
mov w0, wzr
- str wzr, [sp, #12]
- add sp, sp, #16
ret
.Lfunc_end6:
- .size main, .Lfunc_end6-main
+ .size _start, .Lfunc_end6-_start
// -- End function
.type s1, at object // @s1
.section .rodata.s1,"a", at progbits
>From 8f3cbb17ed6cb086041d920ece283f72d1750bab Mon Sep 17 00:00:00 2001
From: xupengying <xpy66swsry at gmail.com>
Date: Mon, 6 Jan 2025 14:58:09 +0800
Subject: [PATCH 5/5] fixup! Add new ELF linker options for profile-guided
section ordering optimizations
---
lld/Common/BPSectionOrdererBase.cpp | 3 ---
lld/ELF/BPSectionOrderer.cpp | 4 +--
lld/ELF/BPSectionOrderer.h | 7 +-----
lld/MachO/BPSectionOrderer.h | 4 ---
lld/include/lld/Common/BPSectionOrdererBase.h | 1 -
lld/test/ELF/bp-section-orderer-stress.s | 6 ++---
lld/test/ELF/bp-section-orderer.s | 25 +++++++++++--------
lld/test/ELF/incompatible.s | 7 ------
8 files changed, 21 insertions(+), 36 deletions(-)
diff --git a/lld/Common/BPSectionOrdererBase.cpp b/lld/Common/BPSectionOrdererBase.cpp
index 7c5874f0d04131..4aedad987b94bb 100644
--- a/lld/Common/BPSectionOrdererBase.cpp
+++ b/lld/Common/BPSectionOrdererBase.cpp
@@ -109,9 +109,6 @@ BPSectionBase::reorderSectionsByBalancedPartitioning(
// Process input sections
for (const auto &isec : inputSections) {
- if (!isec->hasValidData())
- continue;
-
unsigned sectionIdx = sections.size();
sectionToIdx.try_emplace(isec->getSection(), sectionIdx);
sections.emplace_back(isec.get());
diff --git a/lld/ELF/BPSectionOrderer.cpp b/lld/ELF/BPSectionOrderer.cpp
index 69dfde7c9c6412..c01f6e28234472 100644
--- a/lld/ELF/BPSectionOrderer.cpp
+++ b/lld/ELF/BPSectionOrderer.cpp
@@ -26,7 +26,7 @@ void BPSectionELF::getSectionHashes(
constexpr unsigned windowSize = 4;
size_t size = isec->content().size();
- for (size_t i = 0; i < size; i++) {
+ for (size_t i = 0; i != size; ++i) {
auto window = isec->content().drop_front(i).take_front(windowSize);
hashes.push_back(xxHash64(window));
}
@@ -67,7 +67,7 @@ lld::elf::runBalancedPartitioning(Ctx &ctx, llvm::StringRef profilePath,
forDataCompression, compressionSortStartupFunctions, verbose, sections);
DenseMap<const InputSectionBase *, int> result;
- for (const auto &[sec, priority] : reorderedSections) {
+ for (const auto [sec, priority] : reorderedSections) {
auto *elfSection = cast<BPSectionELF>(sec);
result.try_emplace(
static_cast<const InputSectionBase *>(elfSection->getSection()),
diff --git a/lld/ELF/BPSectionOrderer.h b/lld/ELF/BPSectionOrderer.h
index c039408591b22f..868d8f330ab510 100644
--- a/lld/ELF/BPSectionOrderer.h
+++ b/lld/ELF/BPSectionOrderer.h
@@ -77,14 +77,9 @@ class BPSectionELF : public BPSectionBase {
return isec->flags & llvm::ELF::SHF_EXECINSTR;
}
- bool hasValidData() const override {
- return isec && !isec->content().empty();
- }
-
SmallVector<std::unique_ptr<BPSymbol>> getSymbols() const override {
SmallVector<std::unique_ptr<BPSymbol>> symbols;
- if (auto *d = symbol->asDefined())
- symbols.emplace_back(std::make_unique<BPSymbolELF>(d));
+ symbols.emplace_back(std::make_unique<BPSymbolELF>(symbol->getSymbol()));
return symbols;
}
diff --git a/lld/MachO/BPSectionOrderer.h b/lld/MachO/BPSectionOrderer.h
index 29b20c781c6b0d..8ba911fcc546bd 100644
--- a/lld/MachO/BPSectionOrderer.h
+++ b/lld/MachO/BPSectionOrderer.h
@@ -68,10 +68,6 @@ class BPSectionMacho : public BPSectionBase {
bool isCodeSection() const override { return macho::isCodeSection(isec); }
- bool hasValidData() const override {
- return isec && !isec->data.empty() && isec->data.data();
- }
-
SmallVector<std::unique_ptr<BPSymbol>> getSymbols() const override {
SmallVector<std::unique_ptr<BPSymbol>> symbols;
for (auto *sym : isec->symbols)
diff --git a/lld/include/lld/Common/BPSectionOrdererBase.h b/lld/include/lld/Common/BPSectionOrdererBase.h
index 6f483aff464aa8..e2cb41f69cc684 100644
--- a/lld/include/lld/Common/BPSectionOrdererBase.h
+++ b/lld/include/lld/Common/BPSectionOrdererBase.h
@@ -37,7 +37,6 @@ class BPSectionBase {
public:
virtual ~BPSectionBase() = default;
virtual uint64_t getSize() const = 0;
- virtual bool hasValidData() const = 0;
virtual bool isCodeSection() const = 0;
virtual llvm::SmallVector<std::unique_ptr<BPSymbol>> getSymbols() const = 0;
virtual const void *getSection() const = 0;
diff --git a/lld/test/ELF/bp-section-orderer-stress.s b/lld/test/ELF/bp-section-orderer-stress.s
index 04da718f1eda48..ff10a24e381c25 100644
--- a/lld/test/ELF/bp-section-orderer-stress.s
+++ b/lld/test/ELF/bp-section-orderer-stress.s
@@ -1,14 +1,14 @@
# REQUIRES: aarch64
-# Generate a large test case and check that the output is deterministic.
+## Generate a large test case and check that the output is deterministic.
# RUN: %python %s %t.s %t.proftext
# RUN: llvm-mc -filetype=obj -triple=aarch64 %t.s -o %t.o
# RUN: llvm-profdata merge %t.proftext -o %t.profdata
-# RUN: ld.lld -e _start --icf=all -o - %t.o --irpgo-profile=%t.profdata --bp-startup-sort=function --bp-compression-sort-startup-functions --bp-compression-sort=both | llvm-nm --numeric-sort --format=just-symbols - > %t.order1.txt
-# RUN: ld.lld -e _start --icf=all -o - %t.o --irpgo-profile=%t.profdata --bp-startup-sort=function --bp-compression-sort-startup-functions --bp-compression-sort=both | llvm-nm --numeric-sort --format=just-symbols - > %t.order2.txt
+# RUN: ld.lld --icf=all -o - %t.o --irpgo-profile=%t.profdata --bp-startup-sort=function --bp-compression-sort-startup-functions --bp-compression-sort=both | llvm-nm --numeric-sort --format=just-symbols - > %t.order1.txt
+# RUN: ld.lld --icf=all -o - %t.o --irpgo-profile=%t.profdata --bp-startup-sort=function --bp-compression-sort-startup-functions --bp-compression-sort=both | llvm-nm --numeric-sort --format=just-symbols - > %t.order2.txt
# RUN: diff %t.order1.txt %t.order2.txt
import random
diff --git a/lld/test/ELF/bp-section-orderer.s b/lld/test/ELF/bp-section-orderer.s
index aa460d4334a721..7a50619dd250ed 100644
--- a/lld/test/ELF/bp-section-orderer.s
+++ b/lld/test/ELF/bp-section-orderer.s
@@ -2,21 +2,26 @@
# RUN: rm -rf %t && split-file %s %t && cd %t
+# RUN: not ld.lld -o /dev/null %t --irpgo-profile=/dev/null --bp-startup-sort=function --call-graph-ordering-file=/dev/null 2>&1 | FileCheck %s --check-prefix=IRPGO-ERR
+# IRPGO-ERR: error: --bp-startup-sort=function is incompatible with --call-graph-ordering-file
+# RUN: not ld.lld -o /dev/null --bp-compression-sort=function --call-graph-ordering-file /dev/null 2>&1 | FileCheck %s --check-prefix=COMPRESSION-ERR
+# COMPRESSION-ERR: error: --bp-compression-sort is incompatible with --call-graph-ordering-file
+
# RUN: llvm-mc -filetype=obj -triple=aarch64 a.s -o a.o
# RUN: llvm-profdata merge a.proftext -o a.profdata
-# RUN: ld.lld -e _start -o a.out a.o --irpgo-profile=a.profdata --bp-startup-sort=function --verbose-bp-section-orderer --icf=all 2>&1 | FileCheck %s --check-prefix=STARTUP
+# RUN: ld.lld -o a.out a.o --irpgo-profile=a.profdata --bp-startup-sort=function --verbose-bp-section-orderer --icf=all 2>&1 | FileCheck %s --check-prefix=STARTUP
# STARTUP: Ordered 3 sections using balanced partitioning
-# RUN: ld.lld -e _start -o - a.o --symbol-ordering-file a.orderfile --irpgo-profile=a.profdata --bp-startup-sort=function | llvm-nm --numeric-sort --format=just-symbols - | FileCheck %s --check-prefix=ORDERFILE
-# RUN: ld.lld -e _start -o - a.o --symbol-ordering-file a.orderfile --bp-compression-sort=both | llvm-nm --numeric-sort --format=just-symbols - | FileCheck %s --check-prefix=ORDERFILE
+# RUN: ld.lld -o - a.o --symbol-ordering-file a.orderfile --irpgo-profile=a.profdata --bp-startup-sort=function | llvm-nm --numeric-sort --format=just-symbols - | FileCheck %s --check-prefix=ORDERFILE
+# RUN: ld.lld -o - a.o --symbol-ordering-file a.orderfile --bp-compression-sort=both | llvm-nm --numeric-sort --format=just-symbols - | FileCheck %s --check-prefix=ORDERFILE
-# Rodata
+## Rodata
# ORDERFILE: s2
# ORDERFILE: s1
# ORDERFILE-DAG: s3
-# Functions
+## Functions
# ORDERFILE: A
# ORDERFILE: F
# ORDERFILE: E
@@ -25,16 +30,16 @@
# ORDERFILE-DAG: B
# ORDERFILE-DAG: C
-# Data
+## Data
# ORDERFILE: r3
# ORDERFILE: r2
# ORDERFILE-DAG: r1
# ORDERFILE-DAG: r4
-# RUN: ld.lld -e _start -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=function 2>&1 | FileCheck %s --check-prefix=COMPRESSION-FUNC
-# RUN: ld.lld -e _start -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=data 2>&1 | FileCheck %s --check-prefix=COMPRESSION-DATA
-# RUN: ld.lld -e _start -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=both 2>&1 | FileCheck %s --check-prefix=COMPRESSION-BOTH
-# RUN: ld.lld -e _start -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=both --irpgo-profile=a.profdata --bp-startup-sort=function 2>&1 | FileCheck %s --check-prefix=COMPRESSION-BOTH
+# RUN: ld.lld -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=function 2>&1 | FileCheck %s --check-prefix=COMPRESSION-FUNC
+# RUN: ld.lld -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=data 2>&1 | FileCheck %s --check-prefix=COMPRESSION-DATA
+# RUN: ld.lld -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=both 2>&1 | FileCheck %s --check-prefix=COMPRESSION-BOTH
+# RUN: ld.lld -o a.out a.o --verbose-bp-section-orderer --bp-compression-sort=both --irpgo-profile=a.profdata --bp-startup-sort=function 2>&1 | FileCheck %s --check-prefix=COMPRESSION-BOTH
# COMPRESSION-FUNC: Ordered 7 sections using balanced partitioning
# COMPRESSION-DATA: Ordered 7 sections using balanced partitioning
diff --git a/lld/test/ELF/incompatible.s b/lld/test/ELF/incompatible.s
index 7b6f557bdac2e0..0d25acd857610b 100644
--- a/lld/test/ELF/incompatible.s
+++ b/lld/test/ELF/incompatible.s
@@ -1,10 +1,3 @@
-# REQUIRES: aarch64
-# RUN: rm -rf %t
-# RUN: not ld.lld -o /dev/null %t --irpgo-profile=/dev/null --bp-startup-sort=function --call-graph-ordering-file=/dev/null 2>&1 | FileCheck %s --check-prefix=IRPGO-ERR
-# IRPGO-ERR: --bp-startup-sort=function is incompatible with --call-graph-ordering-file
-# RUN: not ld.lld -o /dev/null --bp-compression-sort=function --call-graph-ordering-file /dev/null 2>&1 | FileCheck %s --check-prefix=COMPRESSION-ERR
-# COMPRESSION-ERR: --bp-compression-sort is incompatible with --call-graph-ordering-file
-
// REQUIRES: x86,aarch64
// RUN: llvm-mc -filetype=obj -triple=x86_64-unknown-linux %s -o %ta.o
// RUN: llvm-mc -filetype=obj -triple=i686-unknown-linux %s -o %tb.o
More information about the llvm-commits
mailing list