diff --git a/lld/ELF/Config.h b/lld/ELF/Config.h --- a/lld/ELF/Config.h +++ b/lld/ELF/Config.h @@ -122,7 +122,8 @@ void inferMachineType(); void link(llvm::opt::InputArgList &args); template void compileBitcodeFiles(bool skipLinkedOutput); - + bool tryAddFatLTOFile(MemoryBufferRef mb, StringRef archiveName, + uint64_t offsetInArchive, bool lazy); // True if we are in --whole-archive and --no-whole-archive. bool inWholeArchive = false; @@ -202,6 +203,7 @@ callGraphProfile; bool cmseImplib = false; bool allowMultipleDefinition; + bool fatLTOObjects; bool androidPackDynRelocs = false; bool armHasBlx = false; bool armHasMovtMovw = false; diff --git a/lld/ELF/Driver.cpp b/lld/ELF/Driver.cpp --- a/lld/ELF/Driver.cpp +++ b/lld/ELF/Driver.cpp @@ -52,6 +52,7 @@ #include "llvm/Config/llvm-config.h" #include "llvm/LTO/LTO.h" #include "llvm/Object/Archive.h" +#include "llvm/Object/IRObjectFile.h" #include "llvm/Remarks/HotnessThresholdParser.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Compression.h" @@ -237,6 +238,19 @@ return identify_magic(mb.getBuffer()) == llvm::file_magic::bitcode; } +bool LinkerDriver::tryAddFatLTOFile(MemoryBufferRef mb, StringRef archiveName, + uint64_t offsetInArchive, bool lazy) { + if (!config->fatLTOObjects) + return false; + Expected fatLTOData = + IRObjectFile::findBitcodeInMemBuffer(mb); + if (errorToBool(fatLTOData.takeError())) + return false; + files.push_back( + make(*fatLTOData, archiveName, offsetInArchive, lazy)); + return true; +} + // Opens a file and create a file object. Path has to be resolved already. void LinkerDriver::addFile(StringRef path, bool withLOption) { using namespace sys::fs; @@ -261,7 +275,7 @@ for (const std::pair &p : members) { if (isBitcode(p.first)) files.push_back(make(p.first, path, p.second, false)); - else + else if (!tryAddFatLTOFile(p.first, path, p.second, false)) files.push_back(createObjFile(p.first, path)); } return; @@ -285,9 +299,10 @@ InputFile::isInGroup = true; for (const std::pair &p : members) { auto magic = identify_magic(p.first.getBuffer()); - if (magic == file_magic::elf_relocatable) - files.push_back(createObjFile(p.first, path, true)); - else if (magic == file_magic::bitcode) + if (magic == file_magic::elf_relocatable) { + if (!tryAddFatLTOFile(p.first, path, p.second, true)) + files.push_back(createObjFile(p.first, path, true)); + } else if (magic == file_magic::bitcode) files.push_back(make(p.first, path, p.second, true)); else warn(path + ": archive member '" + p.first.getBufferIdentifier() + @@ -319,7 +334,8 @@ files.push_back(make(mbref, "", 0, inLib)); break; case file_magic::elf_relocatable: - files.push_back(createObjFile(mbref, "", inLib)); + if (!tryAddFatLTOFile(mbref, "", 0, inLib)) + files.push_back(createObjFile(mbref, "", inLib)); break; default: error(path + ": unknown file type"); @@ -1138,6 +1154,8 @@ args.hasFlag(OPT_android_memtag_heap, OPT_no_android_memtag_heap, false); config->androidMemtagStack = args.hasFlag(OPT_android_memtag_stack, OPT_no_android_memtag_stack, false); + config->fatLTOObjects = + args.hasFlag(OPT_fat_lto_objects, OPT_no_fat_lto_objects, false); config->androidMemtagMode = getMemtagMode(args); config->auxiliaryList = args::getStrings(args, OPT_auxiliary); config->armBe8 = args.hasArg(OPT_be8); diff --git a/lld/ELF/Options.td b/lld/ELF/Options.td --- a/lld/ELF/Options.td +++ b/lld/ELF/Options.td @@ -644,6 +644,10 @@ def thinlto_single_module_eq: JJ<"thinlto-single-module=">, HelpText<"Specify a single module to compile in ThinLTO mode, for debugging only">; +defm fat_lto_objects: BB<"fat-lto-objects", + "Use the embedded bitcode in the .llvm.lto section of the object file", + "Use the assembly in the object file (default)">; + def: J<"plugin-opt=O">, Alias, HelpText<"Alias for --lto-O">; def: F<"plugin-opt=debug-pass-manager">, Alias, HelpText<"Alias for --lto-debug-pass-manager">; diff --git a/lld/docs/ReleaseNotes.rst b/lld/docs/ReleaseNotes.rst --- a/lld/docs/ReleaseNotes.rst +++ b/lld/docs/ReleaseNotes.rst @@ -31,6 +31,10 @@ * ``PT_RISCV_ATTRIBUTES`` is added to include the SHT_RISCV_ATTRIBUTES section. (`D152065 `_) +- Support LTO with LLVM FatLTO objects by adding the ``--fat-lto-objects`` option. + Without ``--fat-lto-objects``, LLD will link FatLTO objects without LTO. + (`D146778 `_) + Breaking changes ---------------- diff --git a/lld/docs/ld.lld.1 b/lld/docs/ld.lld.1 --- a/lld/docs/ld.lld.1 +++ b/lld/docs/ld.lld.1 @@ -621,6 +621,10 @@ (default) means all of concurrent threads supported. .Cm 1 disables multi-threading. +.It Fl --no-fat-lto-objects +Select the .text object code section of fat lto object files(use traditional linking). +.It Fl --fat-lto-objects +Select the .llvm.lto bitcode section of fat lto object files(use LTO). .It Fl -time-trace Record time trace. .It Fl -time-trace-file Ns = Ns Ar file diff --git a/lld/test/ELF/fatlto/fatlto.invalid.s b/lld/test/ELF/fatlto/fatlto.invalid.s new file mode 100644 --- /dev/null +++ b/lld/test/ELF/fatlto/fatlto.invalid.s @@ -0,0 +1,10 @@ +# RUN: llvm-mc -filetype=obj -triple=x86_64-unknown-linux %s -o %t +# RUN: not ld.lld %t -o /dev/null --fat-lto-objects 2>&1 | FileCheck %s + +# CHECK: Invalid bitcode signature + +.section .llvm.lto,"e",@progbits +.Lllvm.embedded.object: + .asciz "BC\300\3365\000" + .size .Lllvm.embedded.object, 12 + diff --git a/lld/test/ELF/fatlto/fatlto.test b/lld/test/ELF/fatlto/fatlto.test new file mode 100644 --- /dev/null +++ b/lld/test/ELF/fatlto/fatlto.test @@ -0,0 +1,110 @@ +;; Basic FatLTO tests. +; REQUIRES: x86 + +; RUN: rm -rf %t && split-file %s %t + +;; Ensure that input files contain .llvm.lto section. +; RUN: llc %t/a-LTO.ll --filetype=obj -o %t/a-fatLTO.o +; RUN: llvm-as %t/a-LTO.ll -o %t/a-fatLTO.bc +; RUN: llvm-objcopy --add-section=.llvm.lto=%t/a-fatLTO.bc --set-section-flags=.llvm.lto=exclude %t/a-fatLTO.o + +; RUN: llc %t/main-LTO.ll --filetype=obj -o %t/main-fatLTO.o +; RUN: llvm-as %t/main-LTO.ll -o %t/main-fatLTO.bc +; RUN: llvm-objcopy --add-section=.llvm.lto=%t/main-fatLTO.bc --set-section-flags=.llvm.lto=exclude %t/main-fatLTO.o + +;; Final executable should not have .llvm.lto section no matter what the target is. +; RUN: ld.lld -o %t/foo-fatLTO %t/a-fatLTO.o %t/main-fatLTO.o --fat-lto-objects +; RUN: llvm-readobj -S %t/foo-fatLTO | FileCheck --check-prefix=CHECK-LTO-TARGET %s + +;; Check that fat objects work w/ --start-lib. +; RUN: ld.lld -o %t/foo-fatLTO.start_lib --start-lib %t/a-fatLTO.o %t/main-fatLTO.o --fat-lto-objects +; RUN: llvm-readobj -S %t/foo-fatLTO.start_lib | FileCheck --check-prefix=CHECK-LTO-TARGET %s + +;; Check if .llvm.lto section gets aggregated in LTO target. +; CHECK-LTO-TARGET-NOT: Name: .llvm.lto + +;; Final executable should not have .llvm.lto section no matter what the target is. +; RUN: ld.lld -o %t/foo-fatNoLTO %t/a-fatLTO.o %/t/main-fatLTO.o +; RUN: llvm-readobj -S %t/foo-fatNoLTO | FileCheck --check-prefix=CHECK-NON-LTO-TARGET %s + +;; Check if .llvm.lto section gets aggregated in non-LTO target. +; CHECK-NON-LTO-TARGET-NOT: Name: .llvm.lto + +;; Check if the LTO target executable produced from FatLTO object file is +;; identical to the one produced from LTO modules. +; RUN: llvm-as %t/a-LTO.ll -o %t/a-LTO.bc +; RUN: llvm-as %t/main-LTO.ll -o %t/main-LTO.bc +; RUN: ld.lld -o %t/foo-LTO %t/a-LTO.bc %t/main-LTO.bc +; RUN: cmp %t/foo-fatLTO %t/foo-LTO + +;; Check if the no-LTO target executable produced from FatLTO object file is +;; identical to the one produced from regular object files. +; RUN: llc %t/a-LTO.ll --filetype=obj -o %t/a.o +; RUN: llc %t/main-LTO.ll --filetype=obj -o %t/main.o +; RUN: ld.lld -o %t/foo-noLTO %t/a.o %t/main.o +; RUN: cmp %t/foo-fatNoLTO %t/foo-noLTO + +;; Check archive support. +; RUN: llvm-ar rcs %t/a.a %t/a-fatLTO.o +; RUN: ld.lld -o %t/foo-fatLTO.archive %t/a.a %t/main-LTO.bc --fat-lto-objects +; RUN: cmp %t/foo-fatLTO.archive %t/foo-LTO + +;--- a-LTO.ll +target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128" +target triple = "x86_64-unknown-linux-gnu" + +; Function Attrs: noinline nounwind uwtable +define dso_local i32 @_start() #0 { +entry: + ret i32 0 +} + +attributes #0 = { noinline nounwind uwtable } + +!llvm.module.flags = !{!0, !1, !2, !3, !4, !5, !6} + +!0 = !{i32 1, !"wchar_size", i32 4} +!1 = !{i32 7, !"PIC Level", i32 2} +!2 = !{i32 7, !"PIE Level", i32 2} +!3 = !{i32 7, !"uwtable", i32 2} +!4 = !{i32 7, !"frame-pointer", i32 2} +!5 = !{i32 1, !"ThinLTO", i32 0} +!6 = !{i32 1, !"EnableSplitLTOUnit", i32 1} + +^0 = module: (path: "a-LTO.bc", hash: (0, 0, 0, 0, 0)) +^1 = gv: (name: "_start", summaries: (function: (module: ^0, flags: (linkage: external, visibility: default, notEligibleToImport: 1, live: 0, dsoLocal: 1, canAutoHide: 0), insts: 1, funcFlags: (readNone: 0, readOnly: 0, noRecurse: 0, returnDoesNotAlias: 0, noInline: 1, alwaysInline: 0, noUnwind: 1, mayThrow: 0, hasUnknownCall: 0, mustBeUnreachable: 0)))) ; guid = 12157170054180749580 +^2 = flags: 8 +^3 = blockcount: 1 + +;--- main-LTO.ll +target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128" +target triple = "x86_64-unknown-linux-gnu" + +; Function Attrs: noinline nounwind uwtable +define dso_local i32 @main() #0 { +entry: + %retval = alloca i32, align 4 + store i32 0, ptr %retval, align 4 + %call = call i32 (...) @_start() + ret i32 %call +} + +declare i32 @_start(...) + +attributes #0 = { noinline nounwind uwtable } + +!llvm.module.flags = !{!0, !1, !2, !3, !4, !5, !6} + +!0 = !{i32 1, !"wchar_size", i32 4} +!1 = !{i32 7, !"PIC Level", i32 2} +!2 = !{i32 7, !"PIE Level", i32 2} +!3 = !{i32 7, !"uwtable", i32 2} +!4 = !{i32 7, !"frame-pointer", i32 2} +!5 = !{i32 1, !"ThinLTO", i32 0} +!6 = !{i32 1, !"EnableSplitLTOUnit", i32 1} + +^0 = module: (path: "main-LTO.bc", hash: (0, 0, 0, 0, 0)) +^1 = gv: (name: "_start") ; guid = 12157170054180749580 +^2 = gv: (name: "main", summaries: (function: (module: ^0, flags: (linkage: external, visibility: default, notEligibleToImport: 1, live: 0, dsoLocal: 1, canAutoHide: 0), insts: 4, funcFlags: (readNone: 0, readOnly: 0, noRecurse: 0, returnDoesNotAlias: 0, noInline: 1, alwaysInline: 0, noUnwind: 1, mayThrow: 0, hasUnknownCall: 0, mustBeUnreachable: 0), calls: ((callee: ^1))))) ; guid = 15822663052811949562 +^3 = flags: 8 +^4 = blockcount: 1