In PGO, a C++ external linkage function `foo` has a private counter
`__profc_foo` and a private `__profd_foo` in a `comdat nodeduplicate`.
A `__attribute__((weak))` function `foo` has a weak hidden counter `__profc_foo`
and a private `__profd_foo` in a `comdat nodeduplicate`.
In `ld.lld a.o b.o`, say a.o defines an external linkage `foo` and b.o
defines a weak `foo`. Currently we treat `comdat nodeduplicate` as `comdat any`,
ld.lld will incorrectly consider `b.o:__profc_foo` non-prevailing. In the worst
case when `b.o:__profd_foo` is retained and `b.o:__profc_foo` isn't, there will
be dangling reference causing an `undefined hidden symbol` error.
Add SelectionKind to `Comdat` in IRSymtab and let linkers ignore nodeduplicate comdat.
Differential Revision: https://reviews.llvm.org/D106228
This patch adds initial support for using the new pass manager when
doing ThinLTO via libLTO.
Reviewed By: steven_wu
Differential Revision: https://reviews.llvm.org/D102627
This patch implements first part of Flow Sensitive SampleFDO (FSAFDO).
It has the following changes:
(1) disable current discriminator encoding scheme,
(2) new hierarchical discriminator for FSAFDO.
For this patch, option "-enable-fs-discriminator=true" turns on the new
functionality. Option "-enable-fs-discriminator=false" (the default)
keeps the current SampleFDO behavior. When the fs-discriminator is
enabled, we insert a flag variable, namely, llvm_fs_discriminator, to
the object. This symbol will checked by create_llvm_prof tool, and used
to generate a profile with FS-AFDO discriminators enabled. If this
happens, for an extbinary format profile, create_llvm_prof tool
will add a flag to profile summary section.
Differential Revision: https://reviews.llvm.org/D102246
in this patch we add a new libLTO API to specify debug options independent of an lto_code_gen_t.
This allows clients to pass codegen flags (through libLTO) which otherwise today are ignored.
Reviewed By: steven_wu
Differential Revision: https://reviews.llvm.org/D92611
another one for distributed mode.
Currently during module importing, ThinLTO opens all the source modules,
collect functions to be imported and append them to the destination module,
then leave all the modules open through out the lto backend pipeline. This
patch refactors it in the way that one source module will be closed before
another source module is opened. All the source modules will be closed after
importing phase is done. It will save some amount of memory when there are
many source modules to be imported.
Note that this patch only changes the distributed thinlto mode. For in
process thinlto mode, one source module is shared acorss different thinlto
backend threads so it is not changed in this patch.
Differential Revision: https://reviews.llvm.org/D99554
This version of the patch includes a fix for the cfi failures.
(undoes the revert commit 7db390cc7738a9ba0ed7d4ca59ab6ea2e69c47e9)
It also undoes reverts of follow-up patches that also needed reverting
originally:
* [LTO] Add option enable NewPM with LTOCodeGenerator.
(undoes revert commit 0a17664b47c153aa26a0d31b4835f26375440ec6)
* [LTOCodeGenerator] Use lto::Config for options (NFC)."
(undoes revert commit b0a8e41cfff717ff067bf63412d6edb0280608cd)
This patch removes some options that have been duplicated in
LTOCodeGenerator and instead use lto::Config directly to manage the
options.
This is a cleanup after 6a59f0560648.
Reviewed By: tejohnson
Differential Revision: https://reviews.llvm.org/D95738
This patch adds an option to enable the new pass manager in
LTOCodeGenerator. It also updates a few tests with legacy PM specific
tests, which started failing after 6a59f0560648 when
LLVM_ENABLE_NEW_PASS_MANAGER=true.
This patch updates LTOCodeGenerator to use the utilities provided by
LTOBackend to run middle-end optimizations and backend code generation.
This is a first step towards unifying the code used by libLTO's C API
and the newer, C++ interface (see PR41541).
The immediate motivation is to allow using the new pass manager when
doing LTO using libLTO's C API, which is used on Darwin, among others.
With the changes, there are no codegen/stats differences when building
MultiSource/SPEC2000/SPEC2006 on Darwin X86 with LTO, compared
to without the patch.
Reviewed By: steven_wu
Differential Revision: https://reviews.llvm.org/D94487
splitCodeGen does not need to take ownership of the module, as it
currently clones the original module for each split operation.
There is an ~4 year old fixme to change that, but until this is
addressed, the function can just take a reference to the module.
This makes the transition of LTOCodeGenerator to use LTOBackend a bit
easier, because under some circumstances, LTOCodeGenerator needs to
write the original module back after codegen.
Reviewed By: tejohnson
Differential Revision: https://reviews.llvm.org/D95222
Identify dynamically exported symbols (--export-dynamic[-symbol=],
--dynamic-list=, or definitions needed to preempt shared objects) and
prevent their LTO visibility from being upgraded.
This helps avoid use of whole program devirtualization when there may
be overrides in dynamic libraries.
Differential Revision: https://reviews.llvm.org/D91583
Imported functions and variable get the visibility from the module supplying the
definition. However, non-imported definitions do not get the visibility from
(ELF) the most constraining visibility among all modules (Mach-O) the visibility
of the prevailing definition.
This patch
* adds visibility bits to GlobalValueSummary::GVFlags
* computes the result visibility and propagates it to all definitions
Protected/hidden can imply dso_local which can enable some optimizations (this
is stronger than GVFlags::DSOLocal because the implied dso_local can be
leveraged for ELF -shared while default visibility dso_local has to be cleared
for ELF -shared).
Note: we don't have summaries for declarations, so for ELF if a declaration has
the most constraining visibility, the result visibility may not be that one.
Differential Revision: https://reviews.llvm.org/D92900
To simplify the transition to using LTOBackend, move DisableVerify to
the LTOCodeGenerator class, like most/all other options.
Reviewed By: tejohnson
Differential Revision: https://reviews.llvm.org/D95223
The target features are obtained as a list of features/attributes.
Instead of storing them in a single string, store the vector. This
matches lto::Config's behavior and simplifies the transition to
lto::backend().
Reviewed By: tejohnson
Differential Revision: https://reviews.llvm.org/D95224
Run the ObjCARCContractPass during LTO. The legacy LTO backend (under
LTO/ThinLTOCodeGenerator.cpp) already does this; this diff just adds that
behavior to the new LTO backend. Without that pass, the objc.clang.arc.use
intrinsic will get passed to the instruction selector, which doesn't know how to
handle it.
In order to test both the new and old pass managers, I've also added support for
the `--[no-]lto-legacy-pass-manager` flags.
P.S. Not sure if the ordering of the pass within the pipeline matters...
Reviewed By: fhahn
Differential Revision: https://reviews.llvm.org/D94547
This patch removes some ancient options as a clean-up before moving
code-gen to use LTOBackend in D94487.
I think it would preferable to remove those ancient options, because
1. There are no corresponding options in LTOBackend based tools,
2. There are no unit tests for them,
3. They are not passed through by Clang,
4. At least for GNVLoadPRE, users could just use GVN's `enable-load-pre`.
Alternatively we could add support for those options to lto::Config &
co, but I think it would be better to remove them, unless they are
actually used in practice.
Reviewed By: steven_wu, tejohnson
Differential Revision: https://reviews.llvm.org/D94783
This is the #1 of 2 changes that make remarks hotness threshold option
available in more tools. The changes also allow the threshold to sync with
hotness threshold from profile summary with special value 'auto'.
This change modifies the interface of lto::setupLLVMOptimizationRemarks() to
accept remarks hotness threshold. Update all the tools that use it with remarks
hotness threshold options:
* lld: '--opt-remarks-hotness-threshold='
* llvm-lto2: '--pass-remarks-hotness-threshold='
* llvm-lto: '--lto-pass-remarks-hotness-threshold='
* gold plugin: '-plugin-opt=opt-remarks-hotness-threshold='
Differential Revision: https://reviews.llvm.org/D85809
This reverts 9b5b3050237db3642ed7ab1bdb3ffa2202511b99 and fixes the unwanted re-ordering when generating ThinLTO indexes.
The goal of this patch is to better balance thread utilization during ThinLTO in-process linking (in llvm-lto2 or in LLD). Before this patch, large modules would often be scheduled late during execution, taking a long time to complete, thus starving the thread pool.
We now sort modules in descending order, based on each module's bitcode size, so that larger modules are processed first. By doing so, smaller modules have a better chance to keep the thread pool active, and thus avoid starvation when the bitcode compilation is almost complete.
In our case (on dual Intel Xeon Gold 6140, Windows 10 version 2004, two-stage build), this saves 15 sec when linking `clang.exe` with LLD & -flto=thin, /opt:lldltojobs=all, no ThinLTO cache, -DLLVM_INTEGRATED_CRT_ALLOC=d:\git\rpmalloc.
Before patch: 100 sec
After patch: 85 sec
Inspired by the work done by David Callahan in D60495.
Differential Revision: https://reviews.llvm.org/D87966
This reverts commit 6537004913f3009d896bc30856698e7d22199ba7. This is causing test failures internally, and while a few of the cases turned out to be bad user code (relying on a specific order of static initialization across translation units), some cases are less clear. Temporarily reverting for now, and Teresa is going to follow up with more details.
This completes the circle, complementing -lto-embed-bitcode
(specifically, post-merge-pre-opt). Using -thinlto-assume-merged skips
function importing. The index file is still needed for the other data it
contains.
Differential Revision: https://reviews.llvm.org/D87949
Re-use an optimizition from the old LTO API (used by ld64).
This sorts modules in ascending order, based on bitcode size, so that larger modules are processed first. This allows for smaller modules to be process last, and better fill free threads 'slots', and thusly allow for better multi-thread load balancing.
In our case (on dual Intel Xeon Gold 6140, Windows 10 version 2004, two-stage build), this saves 15 sec when linking `clang.exe` with LLD & `-flto=thin`, `/opt:lldltojobs=all`, no ThinLTO cache, -DLLVM_INTEGRATED_CRT_ALLOC=d:\git\rpmalloc.
Before patch: 102 sec
After patch: 85 sec
Inspired by the work done by David Callahan in D60495.
Differential Revision: https://reviews.llvm.org/D87966
This will embed bitcode after (Thin)LTO merge, but before optimizations.
In the case the thinlto backend is called from clang, the .llvmcmd
section is also produced. Doing so in the case where the caller is the
linker doesn't yet have a motivation, and would require plumbing through
command line args.
Differential Revision: https://reviews.llvm.org/D87636
Fixes an issue with missing nul-terminators and saves us some string
copying, compared to a version which would insert nul-terminators.
Differential Revision: https://reviews.llvm.org/D82033
This change introduces an LLD switch --thinlto-single-module to allow compiling only a part of the input modules. This is specifically enables:
1. Fast investigating/debugging modules of interest without spending time on compiling unrelated modules.
2. Compiler debug dump with -mllvm -debug-only= for specific modules.
It will be useful for large applications which has 1K+ input modules for thinLTO.
The switch can be combined with `--lto-obj-path=` or `--lto-emit-asm` to obtain intermediate object files or assembly files. So far the module name matching is implemented as a fuzzy name lookup where the modules with name containing the switch value are compiled.
E.g,
Command:
ld.lld main.o thin.a --thinlto-single-module=thin.a --lto-obj-path=single.o
log:
[ThinLTO] Selecting thin.a(thin1.o at 168) to compile
[ThinLTO] Selecting thin.a(thin2.o at 228) to compile
Command:
ld.lld main.o thin.a --thinlto-single-module=thin1.o --lto-obj-path=single.o
log:
[ThinLTO] Selecting thin.a(thin1.o at 168) to compile
Differential Revision: https://reviews.llvm.org/D80406
Summary:
That unless the user requested an output object (--lto-obj-path), the an
unused empty combined module is not emitted.
This changed is helpful for some target (ex. RISCV-V) which encoded the
ABI info in IR module flags (target-abi). Empty unused module has no ABI
info so the linker would get the linking error during merging
incompatible ABIs.
Reviewers: tejohnson, espindola, MaskRay
Subscribers: emaste, inglorion, arichardson, hiraditya, simoncook, MaskRay, steven_wu, dexonsmith, PkmX, dang, lenary, s.egerton, luismarques, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D78988
This should make both static and dynamic NewPM plugins work with LTO.
And as a bonus, it makes static linking of OldPM plugins more reliable
for plugins with both an OldPM and NewPM interface.
I only implemented the command-line flag to specify NewPM plugins in
llvm-lto2, to show it works. Support can be added for other tools later.
Differential Revision: https://reviews.llvm.org/D76866
Before this patch, it wasn't possible to extend the ThinLTO threads to all SMT/CMT threads in the system. Only one thread per core was allowed, instructed by usage of llvm::heavyweight_hardware_concurrency() in the ThinLTO code. Any number passed to the LLD flag /opt:lldltojobs=..., or any other ThinLTO-specific flag, was previously interpreted in the context of llvm::heavyweight_hardware_concurrency(), which means SMT disabled.
One can now say in LLD:
/opt:lldltojobs=0 -- Use one std::thread / hardware core in the system (no SMT). Default value if flag not specified.
/opt:lldltojobs=N -- Limit usage to N threads, regardless of usage of heavyweight_hardware_concurrency().
/opt:lldltojobs=all -- Use all hardware threads in the system. Equivalent to /opt:lldltojobs=$(nproc) on Linux and /opt:lldltojobs=%NUMBER_OF_PROCESSORS% on Windows. When an affinity mask is set for the process, threads will be created only for the cores selected by the mask.
When N > number-of-hardware-threads-in-the-system, the threads in the thread pool will be dispatched equally on all CPU sockets (tested only on Windows).
When N <= number-of-hardware-threads-on-a-CPU-socket, the threads will remain on the CPU socket where the process started (only on Windows).
Differential Revision: https://reviews.llvm.org/D75153
Tools working with object files on Darwin (e.g. lipo) may need to know
properties like the CPU type and subtype of a bitcode file. The logic of
converting a triple to a Mach-O CPU_(SUB_)TYPE should be provided by
LLVM instead of relying on tools to re-implement it.
Differential Revision: https://reviews.llvm.org/D75067
The goal of this patch is to maximize CPU utilization on multi-socket or high core count systems, so that parallel computations such as LLD/ThinLTO can use all hardware threads in the system. Before this patch, on Windows, a maximum of 64 hardware threads could be used at most, in some cases dispatched only on one CPU socket.
== Background ==
Windows doesn't have a flat cpu_set_t like Linux. Instead, it projects hardware CPUs (or NUMA nodes) to applications through a concept of "processor groups". A "processor" is the smallest unit of execution on a CPU, that is, an hyper-thread if SMT is active; a core otherwise. There's a limit of 32-bit processors on older 32-bit versions of Windows, which later was raised to 64-processors with 64-bit versions of Windows. This limit comes from the affinity mask, which historically is represented by the sizeof(void*). Consequently, the concept of "processor groups" was introduced for dealing with systems with more than 64 hyper-threads.
By default, the Windows OS assigns only one "processor group" to each starting application, in a round-robin manner. If the application wants to use more processors, it needs to programmatically enable it, by assigning threads to other "processor groups". This also means that affinity cannot cross "processor group" boundaries; one can only specify a "preferred" group on start-up, but the application is free to allocate more groups if it wants to.
This creates a peculiar situation, where newer CPUs like the AMD EPYC 7702P (64-cores, 128-hyperthreads) are projected by the OS as two (2) "processor groups". This means that by default, an application can only use half of the cores. This situation could only get worse in the years to come, as dies with more cores will appear on the market.
== The problem ==
The heavyweight_hardware_concurrency() API was introduced so that only *one hardware thread per core* was used. Once that API returns, that original intention is lost, only the number of threads is retained. Consider a situation, on Windows, where the system has 2 CPU sockets, 18 cores each, each core having 2 hyper-threads, for a total of 72 hyper-threads. Both heavyweight_hardware_concurrency() and hardware_concurrency() currently return 36, because on Windows they are simply wrappers over std:🧵:hardware_concurrency() -- which can only return processors from the current "processor group".
== The changes in this patch ==
To solve this situation, we capture (and retain) the initial intention until the point of usage, through a new ThreadPoolStrategy class. The number of threads to use is deferred as late as possible, until the moment where the std::threads are created (ThreadPool in the case of ThinLTO).
When using hardware_concurrency(), setting ThreadCount to 0 now means to use all the possible hardware CPU (SMT) threads. Providing a ThreadCount above to the maximum number of threads will have no effect, the maximum will be used instead.
The heavyweight_hardware_concurrency() is similar to hardware_concurrency(), except that only one thread per hardware *core* will be used.
When LLVM_ENABLE_THREADS is OFF, the threading APIs will always return 1, to ensure any caller loops will be exercised at least once.
Differential Revision: https://reviews.llvm.org/D71775
This adds some of LLD specific scopes and picks up optimisation scopes
via LTO/ThinLTO. Makes use of TimeProfiler multi-thread support added in
77e6bb3c.
Differential Revision: https://reviews.llvm.org/D71060
This extends the RemarkStreamer to allow for other emitters (e.g.
frontends, SIL, etc.) to emit remarks through a common interface.
See changes in llvm/docs/Remarks.rst for motivation and design choices.
Differential Revision: https://reviews.llvm.org/D73676
This is how it should've been and brings it more in line with
std::string_view. There should be no functional change here.
This is mostly mechanical from a custom clang-tidy check, with a lot of
manual fixups. It uncovers a lot of minor inefficiencies.
This doesn't actually modify StringRef yet, I'll do that in a follow-up.
This restores 59733525d37cf9ad88b5021b33ecdbaf2e18911c (D71913), along
with bot fix 19c76989bb505c3117730c47df85fd3800ea2767.
The bot failure should be fixed by D73418, committed as
af954e441a5170a75687699d91d85e0692929d43.
I also added a fix for non-x86 bot failures by requiring x86 in new test
lld/test/ELF/lto/devirt_vcall_vis_public.ll.
Summary:
Third part in series to support Safe Whole Program Devirtualization
Enablement, see RFC here:
http://lists.llvm.org/pipermail/llvm-dev/2019-December/137543.html
This patch adds type test metadata under -fwhole-program-vtables,
even for classes without hidden visibility. It then changes WPD to skip
devirtualization for a virtual function call when any of the compatible
vtables has public vcall visibility.
Additionally, internal LLVM options as well as lld and gold-plugin
options are added which enable upgrading all public vcall visibility
to linkage unit (hidden) visibility during LTO. This enables the more
aggressive WPD to kick in based on LTO time knowledge of the visibility
guarantees.
Support was added to all flavors of LTO WPD (regular, hybrid and
index-only), and to both the new and old LTO APIs.
Unfortunately it was not simple to split the first and second parts of
this part of the change (the unconditional emission of type tests and
the upgrading of the vcall visiblity) as I needed a way to upgrade the
public visibility on legacy WPD llvm assembly tests that don't include
linkage unit vcall visibility specifiers, to avoid a lot of test churn.
I also added a mechanism to LowerTypeTests that allows dropping type
test assume sequences we now aggressively insert when we invoke
distributed ThinLTO backends with null indexes, which is used in testing
mode, and which doesn't invoke the normal ThinLTO backend pipeline.
Depends on D71907 and D71911.
Reviewers: pcc, evgeny777, steven_wu, espindola
Subscribers: emaste, Prazek, inglorion, arichardson, hiraditya, MaskRay, dexonsmith, dang, davidxl, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D71913
The lto::Config object saved on the global LTO object should not be
updated by any of the LTO backends. Otherwise we could run into
interference between threads utilizing it. Motivated by some proposed
changes that would have caused it to get modified in the ThinLTO
backends.
down to pass builder in ltobackend.
Currently CodeGenOpts like UnrollLoops/VectorizeLoop/VectorizeSLP in clang
are not passed down to pass builder in ltobackend when new pass manager is
used. This is inconsistent with the behavior when new pass manager is used
and thinlto is not used. Such inconsistency causes slp vectorization pass
not being enabled in ltobackend for O3 + thinlto right now. This patch
fixes that.
Differential Revision: https://reviews.llvm.org/D72386
