[ELF] Add BPSectionOrderer options #120514
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@llvm/pr-subscribers-lld-wasm @llvm/pr-subscribers-lld Author: Max (Colibrow) ChangesAdd new ELF linker options for profile-guided section ordering optimizations:
Thanks to the @ellishg, @thevinster, and their team's work. Patch is 24.95 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/120514.diff 10 Files Affected:
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",@progbits
+ .globl A // -- Begin function A
+ .p2align 2
+ .type A,@function
+A: // @A
+// %bb.0:
+ ret
+.Lfunc_end0:
+ .size A, .Lfunc_end0-A
+ // -- End function
+ .section .text.B,"ax",@progbits
+ .globl B // -- Begin function B
+ .p2align 2
+ .type B,@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",@progbits
+ .globl C // -- Begin function C
+ .p2align 2
+ .type C,@function
+C: ...
[truncated]
|
|
@MaskRay @ellishg I also tested the ld64.lld build on my Mac locally using the --bp-compression-sort=function option to compare the effect of relocation hashes. Below are the results:
In the lld-no-reloc version, I removed the following code responsible for calculating relocation hashes: llvm-project/lld/MachO/BPSectionOrderer.h Lines 104 to 115 in 1cc926b The results show that removing relocation hashes reduces the file size and gzip-compressed size slightly. I’m concerned that something might be wrong here—do you have any insights or thoughts on this? |
MaskRay
reviewed
Dec 19, 2024
MaskRay
reviewed
Dec 19, 2024
MaskRay
reviewed
Dec 19, 2024
MaskRay
reviewed
Dec 19, 2024
MaskRay
reviewed
Dec 19, 2024
Hmm I'm not sure why hashing relocations is not as effective on ELF. We know this is effective on our use cases on Mach-O and I think @dreampiggy measured some wins on Mach-O in this comment. It might depend on some features of the binary you are building. |
This relocation hashes seems has a little benefit in both firefox-iOS Demo and our internal project. But I've didn't test disabling this relocation hash to link lld or clang itself and compare the compression ratio. I agreed that this may diff in different binaries. If possible, make this hashing calculation under an optional flag? |
|
@ellishg Yes. I believe it may depend on the binary features. After discussing with @dreampiggy, I've concluded that the relocation aspects likely have minimal impact. Therefore, I prefer to maintain the current implementation to avoid introducing additional complexity in ELF relocations. How do you think? |
Seems ok for now. If someone finds that hashing relocations on ELF is useful in the future, they can look at the Mach-O implementation for reference and add it pretty easily |
|
@MaskRay Hi, Could you please review this PR? And Merry Christmas! |
MaskRay
reviewed
Jan 3, 2025
MaskRay
reviewed
Jan 3, 2025
MaskRay
reviewed
Jan 3, 2025
|
If passing -flto -fprofile-use to clang, it also passes -plugin-opt=cs-profile-path to the linker. |
|
@Andarwinux The IRPGO profile here is generated according to the RFC and is designed for coherence with the Mach-O implementation. As of now, I believe it is unrelated to the lto-cs-profile-path and not well-suited for reuse. What do you think? cc @ellishg |
|
I could be wrong but I think |
MaskRay
requested changes
Jan 5, 2025
MaskRay
reviewed
Jan 6, 2025
MaskRay
reviewed
Jan 6, 2025
|
@ellishg Hopefully I understand the features correctly...
|
Second this. With the typical CSPGO usage lld --lto-cs-profile-path is for context-sensitive profile generation ( It could also be utilized for CS profile use ( |
MaskRay
force-pushed
the
elf-add-bp-options
branch
2 times, most recently
from
3cd9781 to
e165e92
Compare
MaskRay
force-pushed
the
elf-add-bp-options
branch
from
e165e92 to
be9c54b
Compare
MaskRay
approved these changes
Feb 3, 2025
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/63/builds/3814 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/190/builds/13961 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/55/builds/6499 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/35/builds/6910 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/85/builds/5064 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/193/builds/5410 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/169/builds/8062 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/24/builds/4900 Here is the relevant piece of the build log for the reference |
Colibrow
added a commit
to Colibrow/llvm-project
that referenced
this pull request
Feb 3, 2025
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/25/builds/6143 Here is the relevant piece of the build log for the reference |
Colibrow
added a commit
to Colibrow/llvm-project
that referenced
this pull request
Feb 3, 2025
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/52/builds/5755 Here is the relevant piece of the build log for the reference |
Closed
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/14/builds/2527 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/146/builds/2219 Here is the relevant piece of the build log for the reference |
MaskRay
added a commit
that referenced
this pull request
Feb 3, 2025
Exposed by the test added in the reverted #120514 * Fix libstdc++/libc++ differences due to nth_element. #125450 (comment) * Fix LLVM_ENABLE_REVERSE_ITERATION=1 differences * Fix potential issue in `currentSize += D::getSize(*sections[*sectionIdxs.begin()])` where DenseSet was used, though not covered by a test
MaskRay
added a commit
to MaskRay/llvm-project
that referenced
this pull request
Feb 3, 2025
Add new ELF linker options for profile-guided section ordering
optimizations:
- `--irpgo-profile=<file>`: Read IRPGO profile data for use with startup
and compression optimizations
- `--bp-startup-sort={none,function}`: Order sections based on profile
data to improve star tup time
- `--bp-compression-sort={none,function,data,both}`: Order sections
using balanced partitioning to improve compressed size
- `--bp-compression-sort-startup-functions`: Additionally optimize
startup functions for compression
- `--verbose-bp-section-orderer`: Print statistics about balanced
partitioning section ordering
Thanks to the @ellishg, @thevinster, and their team's work.
---------
Reland after 2f6e3df fixed iteration
order issue and libstdc++/libc++ differences.
Co-authored-by: Fangrui Song <[email protected]>
MaskRay
added a commit
that referenced
this pull request
Feb 4, 2025
Reland #120514 after 2f6e3df fixed iteration order issue and libstdc++/libc++ differences. --- Both options instruct the linker to optimize section layout with the following goals: * `--bp-compression-sort=[data|function|both]`: Improve Lempel-Ziv compression by grouping similar sections together, resulting in a smaller compressed app size. * `--bp-startup-sort=function --irpgo-profile=<file>`: Utilize a temporal profile file to reduce page faults during program startup. The linker determines the section order by considering three groups: * Function sections ordered according to the temporal profile (`--irpgo-profile=`), prioritizing early-accessed and frequently accessed functions. * Function sections. Sections containing similar functions are placed together, maximizing compression opportunities. * Data sections. Similar data sections are placed together. Within each group, the sections are ordered using the Balanced Partitioning algorithm. The linker constructs a bipartite graph with two sets of vertices: sections and utility vertices. * For profile-guided function sections: + The number of utility vertices is determined by the symbol order within the profile file. + If `--bp-compression-sort-startup-functions` is specified, extra utility vertices are allocated to prioritize nearby function similarity. * For sections ordered for compression: Utility vertices are determined by analyzing k-mers of the section content and relocations. The call graph profile is disabled during this optimization. When `--symbol-ordering-file=` is specified, sections described in that file are placed earlier. Co-authored-by: Pengying Xu <[email protected]>
Icohedron
pushed a commit
to Icohedron/llvm-project
that referenced
this pull request
Feb 11, 2025
The ELF/bp-section-orderer.s test is failing on some buildbots due to what seems like non-determinism issues, see comments on the original PR and llvm#125450 Reverting to green the build. This reverts commit 0154dce and follow-up commits 046dd4b and c92f204.
Icohedron
pushed a commit
to Icohedron/llvm-project
that referenced
this pull request
Feb 11, 2025
Exposed by the test added in the reverted llvm#120514 * Fix libstdc++/libc++ differences due to nth_element. llvm#125450 (comment) * Fix LLVM_ENABLE_REVERSE_ITERATION=1 differences * Fix potential issue in `currentSize += D::getSize(*sections[*sectionIdxs.begin()])` where DenseSet was used, though not covered by a test
Icohedron
pushed a commit
to Icohedron/llvm-project
that referenced
this pull request
Feb 11, 2025
Reland llvm#120514 after 2f6e3df fixed iteration order issue and libstdc++/libc++ differences. --- Both options instruct the linker to optimize section layout with the following goals: * `--bp-compression-sort=[data|function|both]`: Improve Lempel-Ziv compression by grouping similar sections together, resulting in a smaller compressed app size. * `--bp-startup-sort=function --irpgo-profile=<file>`: Utilize a temporal profile file to reduce page faults during program startup. The linker determines the section order by considering three groups: * Function sections ordered according to the temporal profile (`--irpgo-profile=`), prioritizing early-accessed and frequently accessed functions. * Function sections. Sections containing similar functions are placed together, maximizing compression opportunities. * Data sections. Similar data sections are placed together. Within each group, the sections are ordered using the Balanced Partitioning algorithm. The linker constructs a bipartite graph with two sets of vertices: sections and utility vertices. * For profile-guided function sections: + The number of utility vertices is determined by the symbol order within the profile file. + If `--bp-compression-sort-startup-functions` is specified, extra utility vertices are allocated to prioritize nearby function similarity. * For sections ordered for compression: Utility vertices are determined by analyzing k-mers of the section content and relocations. The call graph profile is disabled during this optimization. When `--symbol-ordering-file=` is specified, sections described in that file are placed earlier. Co-authored-by: Pengying Xu <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Add new ELF linker options for profile-guided section ordering optimizations:
--irpgo-profile=<file>: Read IRPGO profile data for use with startup and compression optimizations--bp-startup-sort={none,function}: Order sections based on profile data to improve star tup time--bp-compression-sort={none,function,data,both}: Order sections using balanced partitioning to improve compressed size--bp-compression-sort-startup-functions: Additionally optimize startup functions for compression--verbose-bp-section-orderer: Print statistics about balanced partitioning section orderingThanks to the @ellishg, @thevinster, and their team's work.