[lld] r269605 - [LTO] Add the ability to specify a subset of passes to run.
Rui Ueyama via llvm-commits
llvm-commits at lists.llvm.org
Mon May 16 13:59:47 PDT 2016
It still can be improved as the current error check is done at a very late
stage, which is after all symbol are resolved. Theoretically we could
determine the target machine when we see the first object file (or see -m
option).
However I'm not sure if we want to do that. As long as linking time is not
that long, it's not that bad user experience.
On Mon, May 16, 2016 at 1:54 PM, Chandler Carruth <chandlerc at gmail.com>
wrote:
> We could add a validatePassPipeline API to the PassBuilder maybe... My
> only hesitation is that I'd like for targets to be able to register passes
> here, and that might make it hard to validate without a TargetMachine.
> Fundamentally, I suspect that we will want to do pipeline validation based
> on the actual target machine that will be used, so it may not make sense to
> hoist stuff past a point where you have that.
>
> On Mon, May 16, 2016 at 1:43 PM Rui Ueyama <ruiu at google.com> wrote:
>
>> I'm not sure if you can instantiate these classes in the driver. In the
>> driver, we don't even know what architecture the target machine is.
>>
>> Can you call parsePassPipeline with a dummy TargetMachine and an empty
>> PassManager to verify a string?
>>
>> On Mon, May 16, 2016 at 1:10 PM, Davide Italiano <davide at freebsd.org>
>> wrote:
>>
>>> On Mon, May 16, 2016 at 10:39 AM, Rui Ueyama <ruiu at google.com> wrote:
>>> > On Sun, May 15, 2016 at 12:29 PM, Davide Italiano via llvm-commits
>>> > <llvm-commits at lists.llvm.org> wrote:
>>> >>
>>> >> Author: davide
>>> >> Date: Sun May 15 14:29:38 2016
>>> >> New Revision: 269605
>>> >>
>>> >> URL: http://llvm.org/viewvc/llvm-project?rev=269605&view=rev
>>> >> Log:
>>> >> [LTO] Add the ability to specify a subset of passes to run.
>>> >>
>>> >> Differential Revision: http://reviews.llvm.org/D20267
>>> >>
>>> >> Added:
>>> >> lld/trunk/test/ELF/lto/ltopasses-custom.ll
>>> >> Modified:
>>> >> lld/trunk/ELF/CMakeLists.txt
>>> >> lld/trunk/ELF/Config.h
>>> >> lld/trunk/ELF/Driver.cpp
>>> >> lld/trunk/ELF/LTO.cpp
>>> >> lld/trunk/ELF/Options.td
>>> >>
>>> >> Modified: lld/trunk/ELF/CMakeLists.txt
>>> >> URL:
>>> >>
>>> http://llvm.org/viewvc/llvm-project/lld/trunk/ELF/CMakeLists.txt?rev=269605&r1=269604&r2=269605&view=diff
>>> >>
>>> >>
>>> ==============================================================================
>>> >> --- lld/trunk/ELF/CMakeLists.txt (original)
>>> >> +++ lld/trunk/ELF/CMakeLists.txt Sun May 15 14:29:38 2016
>>> >> @@ -31,6 +31,7 @@ add_lld_library(lldELF
>>> >> Linker
>>> >> Object
>>> >> Option
>>> >> + Passes
>>> >> MC
>>> >> Support
>>> >> Target
>>> >>
>>> >> Modified: lld/trunk/ELF/Config.h
>>> >> URL:
>>> >>
>>> http://llvm.org/viewvc/llvm-project/lld/trunk/ELF/Config.h?rev=269605&r1=269604&r2=269605&view=diff
>>> >>
>>> >>
>>> ==============================================================================
>>> >> --- lld/trunk/ELF/Config.h (original)
>>> >> +++ lld/trunk/ELF/Config.h Sun May 15 14:29:38 2016
>>> >> @@ -44,6 +44,7 @@ struct Configuration {
>>> >> llvm::StringRef Emulation;
>>> >> llvm::StringRef Fini;
>>> >> llvm::StringRef Init;
>>> >> + llvm::StringRef LtoNewPmPasses;
>>> >> llvm::StringRef OutputFile;
>>> >> llvm::StringRef SoName;
>>> >> llvm::StringRef Sysroot;
>>> >>
>>> >> Modified: lld/trunk/ELF/Driver.cpp
>>> >> URL:
>>> >>
>>> http://llvm.org/viewvc/llvm-project/lld/trunk/ELF/Driver.cpp?rev=269605&r1=269604&r2=269605&view=diff
>>> >>
>>> >>
>>> ==============================================================================
>>> >> --- lld/trunk/ELF/Driver.cpp (original)
>>> >> +++ lld/trunk/ELF/Driver.cpp Sun May 15 14:29:38 2016
>>> >> @@ -337,6 +337,7 @@ void LinkerDriver::readConfigs(opt::Inpu
>>> >> Config->Entry = getString(Args, OPT_entry);
>>> >> Config->Fini = getString(Args, OPT_fini, "_fini");
>>> >> Config->Init = getString(Args, OPT_init, "_init");
>>> >> + Config->LtoNewPmPasses = getString(Args, OPT_lto_newpm_passes);
>>> >> Config->OutputFile = getString(Args, OPT_o);
>>> >> Config->SoName = getString(Args, OPT_soname);
>>> >> Config->Sysroot = getString(Args, OPT_sysroot);
>>> >> @@ -495,6 +496,8 @@ template <class ELFT> void LinkerDriver:
>>> >> Symtab.scanVersionScript();
>>> >>
>>> >> Symtab.addCombinedLtoObject();
>>> >> + if (HasError)
>>> >> + return;
>>> >>
>>> >> for (auto *Arg : Args.filtered(OPT_wrap))
>>> >> Symtab.wrap(Arg->getValue());
>>> >>
>>> >> Modified: lld/trunk/ELF/LTO.cpp
>>> >> URL:
>>> >>
>>> http://llvm.org/viewvc/llvm-project/lld/trunk/ELF/LTO.cpp?rev=269605&r1=269604&r2=269605&view=diff
>>> >>
>>> >>
>>> ==============================================================================
>>> >> --- lld/trunk/ELF/LTO.cpp (original)
>>> >> +++ lld/trunk/ELF/LTO.cpp Sun May 15 14:29:38 2016
>>> >> @@ -13,6 +13,9 @@
>>> >> #include "Error.h"
>>> >> #include "InputFiles.h"
>>> >> #include "Symbols.h"
>>> >> +#include "llvm/Analysis/AliasAnalysis.h"
>>> >> +#include "llvm/Analysis/CGSCCPassManager.h"
>>> >> +#include "llvm/Analysis/LoopPassManager.h"
>>> >> #include "llvm/Analysis/TargetLibraryInfo.h"
>>> >> #include "llvm/Analysis/TargetTransformInfo.h"
>>> >> #include "llvm/Bitcode/ReaderWriter.h"
>>> >> @@ -20,13 +23,16 @@
>>> >> #include "llvm/CodeGen/ParallelCG.h"
>>> >> #include "llvm/IR/AutoUpgrade.h"
>>> >> #include "llvm/IR/LegacyPassManager.h"
>>> >> +#include "llvm/IR/PassManager.h"
>>> >> #include "llvm/Linker/IRMover.h"
>>> >> +#include "llvm/Passes/PassBuilder.h"
>>> >> #include "llvm/Support/StringSaver.h"
>>> >> #include "llvm/Support/TargetRegistry.h"
>>> >> #include "llvm/Target/TargetMachine.h"
>>> >> #include "llvm/Transforms/IPO.h"
>>> >> #include "llvm/Transforms/IPO/PassManagerBuilder.h"
>>> >> #include "llvm/Transforms/Utils/ModuleUtils.h"
>>> >> +#include <llvm/IR/Verifier.h>
>>> >>
>>> >> using namespace llvm;
>>> >> using namespace llvm::object;
>>> >> @@ -56,10 +62,44 @@ static void saveBCFile(Module &M, String
>>> >> WriteBitcodeToFile(&M, OS, /* ShouldPreserveUseListOrder */ true);
>>> >> }
>>> >>
>>> >> -// Run LTO passes.
>>> >> -// Note that the gold plugin has a similar piece of code, so
>>> >> -// it is probably better to move this code to a common place.
>>> >> -static void runLTOPasses(Module &M, TargetMachine &TM) {
>>> >> +static void runNewCustomLtoPasses(Module &M, TargetMachine &TM) {
>>> >> + PassBuilder PB(&TM);
>>> >> +
>>> >> + AAManager AA;
>>> >> + LoopAnalysisManager LAM;
>>> >> + FunctionAnalysisManager FAM;
>>> >> + CGSCCAnalysisManager CGAM;
>>> >> + ModuleAnalysisManager MAM;
>>> >> +
>>> >> + // Register the AA manager first so that our version is the one
>>> used.
>>> >> + FAM.registerPass([&] { return std::move(AA); });
>>> >> +
>>> >> + // Register all the basic analyses with the managers.
>>> >> + PB.registerModuleAnalyses(MAM);
>>> >> + PB.registerCGSCCAnalyses(CGAM);
>>> >> + PB.registerFunctionAnalyses(FAM);
>>> >> + PB.registerLoopAnalyses(LAM);
>>> >> + PB.crossRegisterProxies(LAM, FAM, CGAM, MAM);
>>> >> +
>>> >> + ModulePassManager MPM;
>>> >> + if (!Config->DisableVerify)
>>> >> + MPM.addPass(VerifierPass());
>>> >> +
>>> >> + // Now, add all the passes we've been requested to.
>>> >> + if (!PB.parsePassPipeline(MPM, Config->LtoNewPmPasses)) {
>>> >
>>> >
>>> > Is there any way to validate a string earlier than this? It would not
>>> only
>>> > make error handling simpler but also make it more user friendly as
>>> user will
>>> > get an error message immediately rather than after all symbols are
>>> resolved.
>>> >
>>>
>>> Hmm, I agree it's very annoying detecting mistakes so late in the
>>> linking process. I originally thought about it.
>>> The real issue is that you need to instantiate an object `PassBuilder`
>>> on which you call the parse* functions, and the parse functions add
>>> passes to the pipeline on the fly. We could probably move the
>>> validation to the stage of command line parsing, build the pipeline
>>> there, add PB as part of Config, and use that instead of the string in
>>> runNewCustomLtoPasses.
>>> Would you like it better?
>>>
>>> --
>>> Davide
>>>
>>> "There are no solved problems; there are only problems that are more
>>> or less solved" -- Henri Poincare
>>>
>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160516/caf93be4/attachment.html>
More information about the llvm-commits
mailing list