[lld-macho] Add support for LTO optimization level

[lgrey]

Summary

Everything (including test) modified from ELF/COFF. Using the same syntax (--lto-O3, etc) as ELF.

[thakis]

Nice!

lld/MachO/Config.h

129

Is this the same default ld64 seems to use?

[lgrey]

lld/MachO/Config.h

129

AFAICT, it doesn't optimize.

$ cat opt_check.c
void foo() {
    return;
}

int main(int argc, char** argv) {
    foo();
    return 0;
}
$ clang -flto=thin opt_check.c && nm -pa a.out
0000000100003f80 t _foo
0000000100000000 T __mh_execute_header
0000000100003f90 T _main
                 U dyld_stub_binder

Should we change it to 0 for compat?

[int3]

lld/MachO/Config.h

129

ld64 has an undocumented -flto-codegen-only flag. I haven't looked into what exactly it does, but it seems relevant...

[thakis]

lld/MachO/Config.h

129

Should we change it to 0 for compat?

Yes, I think so.

-flto-codegen-only

I think that's probably something else. That ends up calling thinlto_codegen_set_codegen_only() in llvm/tools/lto/lto.cpp.

…but lto_file.cpp in ld64 doesn't seem to pass an optimization flag, and OptLevel in lto.cpp defaults to 2. Strange that ld64 doesn't get an opt level of 2 then…

[tschuett]

Does this test -O2 or --lto-O2? Do you need two files for checking lto, i.e. it inlines foo across two TUs?

[lgrey]

In D105223#2851229, @tschuett wrote:

Does this test -O2 or --lto-O2? Do you need two files for checking lto, i.e. it inlines foo across two TUs?

Effectively both, since what it really wants to test is the passthrough, but I don't think there's a way to sense for that directly. This just adapts lld/test/COFF/lto-opt-level.ll, but if there's some way to mock lto::LTO and assert things about its config, that seems like it would be good. Is there?

I think two files aren't necessary since the single file inlining is enough to demonstrate the flag was passed.

lld/MachO/Config.h

129

Maybe the legacy LTO implementation bitrotted? The docs bake optimization in:

% clang -flto=thin -O2 file1.c file2.c -c
% clang -flto=thin -O2 file1.o file2.o -o a.out

I tried changing to 0, but this breaks existing behavior, since the default appears to be 2 when omitting opt level from the config. Are we OK with that?

[steven_wu]

lld/MachO/Config.h

129

You cannot control optimization level for ld64. It is always on and the pipeline is configured in libLTO.
-flto-codegen-only will turn off all the optimization on LLVM IR and only run the code generation pipeline with all optimization on. This is only used for debugging or apple bitcode.

-O flag will control how much local optimization is done for each individual .o file (bitcode) before passing to linker for LTO.

[thakis]

Thanks for chiming in, Steven! Sounds like this is all good as-is then, so landed :)

lld/MachO/Config.h

struct Configuration {
   uint32_t dylibCurrentVersion = 0;
   uint32_t timeTraceGranularity = 500;
   std::string progName;
   llvm::StringRef installName;
   llvm::StringRef mapFile;
   llvm::StringRef outputFile;
   llvm::StringRef ltoObjPath;
   llvm::StringRef thinLTOJobs;
+  uint32_t ltoo = 2;
   bool deadStripDylibs = false;
   bool demangle = false;
   bool deadStrip = false;
   PlatformInfo platformInfo;
   NamespaceKind namespaceKind = NamespaceKind::twolevel;
   UndefinedSymbolTreatment undefinedSymbolTreatment =
       UndefinedSymbolTreatment::error;
   ICFLevel icfLevel = ICFLevel::none;

lld/MachO/Driver.cpp

if (config->outputType != MH_BUNDLE)
       error("-bundle_loader can only be used with MachO bundle output");
     addFile(arg->getValue(), /*forceLoadArchive=*/false, /*isExplicit=*/false,
             /*isBundleLoader=*/true);
   }
   config->ltoObjPath = args.getLastArgValue(OPT_object_path_lto);
   config->ltoNewPassManager =
       args.hasFlag(OPT_no_lto_legacy_pass_manager, OPT_lto_legacy_pass_manager,
                    LLVM_ENABLE_NEW_PASS_MANAGER);
+  config->ltoo = args::getInteger(args, OPT_lto_O, 2);
+  if (config->ltoo > 3)
+    error("--lto-O: invalid optimization level: " + Twine(config->ltoo));
   config->runtimePaths = args::getStrings(args, OPT_rpath);
   config->allLoad = args.hasArg(OPT_all_load);
   config->forceLoadObjC = args.hasArg(OPT_ObjC);
   config->forceLoadSwift = args.hasArg(OPT_force_load_swift_libs);
   config->deadStripDylibs = args.hasArg(OPT_dead_strip_dylibs);
   config->demangle = args.hasArg(OPT_demangle);
   config->implicitDylibs = !args.hasArg(OPT_no_implicit_dylibs);
   config->emitFunctionStarts =

lld/MachO/LTO.cpp

 //===- LTO.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 "LTO.h"
 #include "Config.h"
 #include "Driver.h"
 #include "InputFiles.h"
 #include "Symbols.h"
 #include "Target.h"
 
+#include "lld/Common/Args.h"
 #include "lld/Common/ErrorHandler.h"
 #include "lld/Common/Strings.h"
 #include "lld/Common/TargetOptionsCommandFlags.h"
 #include "llvm/LTO/LTO.h"
 #include "llvm/Support/FileSystem.h"
 #include "llvm/Support/Path.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Transforms/ObjCARC.h"
static lto::Config createConfig() {
   c.CPU = getCPUStr();
   c.MAttrs = getMAttrs();
   c.UseNewPM = config->ltoNewPassManager;
   c.PreCodeGenPassesHook = [](legacy::PassManager &pm) {
     pm.add(createObjCARCContractPass());
   };
   c.TimeTraceEnabled = config->timeTraceEnabled;
   c.TimeTraceGranularity = config->timeTraceGranularity;
+  c.OptLevel = config->ltoo;
+  c.CGOptLevel = args::getCGOptLevel(config->ltoo);
   if (config->saveTemps)
     checkError(c.addSaveTemps(config->outputFile.str() + ".",
                               /*UseInputModulePath=*/true));
   return c;
 }
 
 BitcodeCompiler::BitcodeCompiler() {
   lto::ThinBackend backend = lto::createInProcessThinBackend(

lld/MachO/Options.td

def deduplicate_literals: Flag<["--"], "deduplicate-literals">,
     Group<grp_lld>;
 def print_dylib_search: Flag<["--"], "print-dylib-search">,
     HelpText<"Print which paths lld searched when trying to find dylibs">,
     Group<grp_lld>;
 def icf_eq: Joined<["--"], "icf=">,
     HelpText<"Set level for identical code folding (default: none)">,
     MetaVarName<"[none,safe,all]">,
     Group<grp_lld>;
+def lto_O: Joined<["--"], "lto-O">,
+    HelpText<"Set optimization level for LTO (default: 2)">,
+    MetaVarName<"<opt-level>">,
+    Group<grp_lld>;
 
 // This is a complete Options.td compiled from Apple's ld(1) manpage
 // dated 2018-03-07 and cross checked with ld64 source code in repo
 // https://github.com/apple-opensource/ld64 at git tag "512.4" dated
 // 2018-03-18.
 
 // Flags<[HelpHidden]> marks options that are not yet ported to lld,
 // and serve as a scoreboard for annotating our progress toward