ICF was not able to merge equivalent sections because of relocations to
sections ineligible for ICF that use alternative symbols, e.g. symbol
aliases or section relative relocations.
Merging in this scenario has been enabled by giving the sections that
are ineligible for ICF a unique ID, i.e. an equivalence class of their
own. This approach also provides another benefit as it improves the
hashing that is used to perform the initial equivalance grouping for
ICF. This is because the ICF ineligible sections can now contribute a
unique value towards the hashes instead of the same value of zero. This
has been seen to reduce link time with ICF by ~68% for objects compiled
with -fprofile-instr-generate.
In order to facilitate this use of a unique ID, the existing
inconsistent approach to the setting of the InputSection eqClass in ICF
has been changed so that there is a clear distinction between the
eqClass values of ICF eligible sections and those of the ineligible
sections that have a unique ID. This inconsistency could have caused
incorrect equivalence class equality in the past, although it appears
that no issues were encountered in actual use.
Differential Revision: https://reviews.llvm.org/D88830
Fix PR36272 and PR46835
A .eh_frame FDE references a text section and (optionally) a LSDA (in
.gcc_except_table). Even if two text sections have identical content and
relocations (e.g. a() and b()), we cannot fold them if their LSDA are different.
```
void foo();
void a() {
try { foo(); } catch (int) { }
}
void b() {
try { foo(); } catch (float) { }
}
```
Scan .eh_frame pieces with LSDA and disallow referenced text sections to be
folded. If two .gcc_except_table have identical semantics (usually identical
content with PC-relative encoding), we will lose folding opportunity.
For ClickHouse (an exception-heavy application), this can reduce --icf=all efficiency
from 9% to 5%. There may be some percentage we can reclaim without affecting
correctness, if we analyze .eh_frame and .gcc_except_table sections.
gold 2.24 implemented a more complex fix (resolution to
https://sourceware.org/bugzilla/show_bug.cgi?id=21066) which combines the
checksum of .eh_frame CIE/FDE pieces.
Reviewed By: grimar
Differential Revision: https://reviews.llvm.org/D84610
Essentially takes the lld/Common/Threads.h wrappers and moves them to
the llvm/Support/Paralle.h algorithm header.
The changes are:
- Remove policy parameter, since all clients use `par`.
- Rename the methods to `parallelSort` etc to match LLVM style, since
they are no longer C++17 pstl compatible.
- Move algorithms from llvm::parallel:: to llvm::, since they have
"parallel" in the name and are no longer overloads of the regular
algorithms.
- Add range overloads
- Use the sequential algorithm directly when 1 thread is requested
(skips task grouping)
- Fix the index type of parallelForEachN to size_t. Nobody in LLVM was
using any other parameter, and it made overload resolution hard for
for_each_n(par, 0, foo.size(), ...) because 0 is int, not size_t.
Remove Threads.h and update LLD for that.
This is a prerequisite for parallel public symbol processing in the PDB
library, which is in LLVM.
Reviewed By: MaskRay, aganea
Differential Revision: https://reviews.llvm.org/D79390
--no-threads is a name copied from gold.
gold has --no-thread, --thread-count and several other --thread-count-*.
There are needs to customize the number of threads (running several lld
processes concurrently or customizing the number of LTO threads).
Having a single --threads=N is a straightforward replacement of gold's
--no-threads + --thread-count.
--no-threads is used rarely. So just delete --no-threads instead of
keeping it for compatibility for a while.
If --threads= is specified (ELF,wasm; COFF /threads: is similar),
--thinlto-jobs= defaults to --threads=,
otherwise all available hardware threads are used.
There is currently no way to override a --threads={1,2,...}. It is still
a debate whether we should use --threads=all.
Reviewed By: rnk, aganea
Differential Revision: https://reviews.llvm.org/D76885
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 makes it clear `ELF/**/*.cpp` files define things in the `lld::elf`
namespace and simplifies `elf::foo` to `foo`.
Reviewed By: atanasyan, grimar, ruiu
Differential Revision: https://reviews.llvm.org/D68323
llvm-svn: 373885
ICF is performed after EhInputSections and MergeInputSections were
eliminated from inputSections. Every element of inputSections is an
InputSection.
llvm-svn: 371744
Recommit r370635 (reverted by r371202), with one change: move addOrphanSections() before ICF.
Before, orphan sections in two different partitions may be folded and
moved to the main partition.
Now, InputSection->OutputSection assignment for orphans happens before
ICF. ICF does not fold input sections with different output sections.
With the PR43241 reproduce,
`llvm-objcopy --extract-partition libvr.so libchrome__combined.so libvr.so` => no error
Updated description:
Fixes PR39418. Complements D47241 (the non-linker-script case).
processSectionCommands() assigns input sections to output sections.
ICF is called before it, so .text.foo and .text.bar may be folded even if
their output sections are made different by SECTIONS commands.
```
markLive<ELFT>()
doIcf<ELFT>() // During ICF, we don't know the output sections
writeResult()
combineEhSections<ELFT>()
script->processSectionCommands() // InputSection -> OutputSection assignment
```
This patch splits processSectionCommands() into processSectionCommands()
and processSymbolAssignments(), and moves
processSectionCommands()/addOrphanSections() before ICF:
```
markLive<ELFT>()
combineEhSections<ELFT>()
script->processSectionCommands()
script->addOrphanSections();
doIcf<ELFT>() // should remove folded input sections
writeResult()
script->processSymbolAssignments()
```
An alternative approach is to unfold a section `sec` in
processSectionCommands() when we find `sec` and `sec->repl` belong to
different output sections. I feel this patch is superior because this
can fold more sections and the decouple of
SectionCommand/SymbolAssignment gives flexibility:
* An ExprValue can't be evaluated before its section is assigned to an
output section -> we can delete getOutputSectionVA and simplify
another place where we had to check if the output section is null.
Moreover, a case in linkerscript/early-assign-symbol.s can be handled
now.
* processSectionCommands/processSymbolAssignments can be freely moved
around.
llvm-svn: 371216
Fixes PR39418. Complements D47241 (the non-linker-script case).
processSectionCommands() assigns input sections to output sections.
ICF is called before it, so .text.foo and .text.bar may be folded even if
their output sections are made different by SECTIONS commands.
```
markLive<ELFT>()
doIcf<ELFT>() // During ICF, we don't know the output sections
writeResult()
combineEhSections<ELFT>()
script->processSectionCommands() // InputSection -> OutputSection assignment
```
This patch splits processSectionCommands() into processSectionCommands() and
processSymbolAssignments(), and moves processSectionCommands() before ICF:
```
markLive<ELFT>()
combineEhSections<ELFT>()
script->processSectionCommands()
doIcf<ELFT>() // should remove folded input sections
writeResult()
script->processSymbolAssignments()
```
An alternative approach is to unfold a section `sec` in
processSectionCommands() when we find `sec` and `sec->repl` belong to
different output sections. I feel this patch is superior because this
can fold more sections and the decouple of
SectionCommand/SymbolAssignment gives flexibility:
* An ExprValue can't be evaluated before its section is assigned to an
output section -> we can delete getOutputSectionVA and simplify
another place where we had to check if the output section is null.
Moreover, a case in linkerscript/early-assign-symbol.s can be handled
now.
* processSectionCommands/processSymbolAssignments can be freely moved
around.
Reviewed By: ruiu
Differential Revision: https://reviews.llvm.org/D66717
llvm-svn: 370635
This patch does the same thing as r365595 to other subdirectories,
which completes the naming style change for the entire lld directory.
With this, the naming style conversion is complete for lld.
Differential Revision: https://reviews.llvm.org/D64473
llvm-svn: 365730
This patch is mechanically generated by clang-llvm-rename tool that I wrote
using Clang Refactoring Engine just for creating this patch. You can see the
source code of the tool at https://reviews.llvm.org/D64123. There's no manual
post-processing; you can generate the same patch by re-running the tool against
lld's code base.
Here is the main discussion thread to change the LLVM coding style:
https://lists.llvm.org/pipermail/llvm-dev/2019-February/130083.html
In the discussion thread, I proposed we use lld as a testbed for variable
naming scheme change, and this patch does that.
I chose to rename variables so that they are in camelCase, just because that
is a minimal change to make variables to start with a lowercase letter.
Note to downstream patch maintainers: if you are maintaining a downstream lld
repo, just rebasing ahead of this commit would cause massive merge conflicts
because this patch essentially changes every line in the lld subdirectory. But
there's a remedy.
clang-llvm-rename tool is a batch tool, so you can rename variables in your
downstream repo with the tool. Given that, here is how to rebase your repo to
a commit after the mass renaming:
1. rebase to the commit just before the mass variable renaming,
2. apply the tool to your downstream repo to mass-rename variables locally, and
3. rebase again to the head.
Most changes made by the tool should be identical for a downstream repo and
for the head, so at the step 3, almost all changes should be merged and
disappear. I'd expect that there would be some lines that you need to merge by
hand, but that shouldn't be too many.
Differential Revision: https://reviews.llvm.org/D64121
llvm-svn: 365595
This change causes us to read partition specifications from partition
specification sections and split output sections into partitions according
to their reachability from partition entry points.
This is only the first step towards a full implementation of partitions. Later
changes will add additional synthetic sections to each partition so that
they can be loaded independently.
Differential Revision: https://reviews.llvm.org/D60353
llvm-svn: 361925
Make some small adjustment while touching the code: make parameters
const, use less_first(), etc.
Differential Revision: https://reviews.llvm.org/D60989
llvm-svn: 358943
LLD's performance on PGO instrumented Windows binaries was still not
great even with the fix in D56955; out of the 2m41s linker runtime,
around 2 minutes were still being spent in ICF. I looked into this more
closely and discovered that the vast majority of the runtime was being
spent segregating .pdata sections with the following relocation chain:
.pdata -> identical .text -> unique PGO counter (not eligible for ICF)
This patch causes us to perform 2 rounds of relocation hash
propagation, which allows the hash for the .pdata sections to
incorporate the identifier from the PGO counter. With that, the amount
of time spent in ICF was reduced to about 2 seconds. I also found that
the same change led to a significant ICF performance improvement in a
regular release build of Chromium's chrome_child.dll, where ICF time
was reduced from around 1s to around 700ms.
With the same change applied to the ELF linker, median of 100 runs
for lld-speed-test/chrome reduced from 4.53s to 4.45s on my machine.
I also experimented with increasing the number of propagation rounds
further, but I did not observe any further significant performance
improvements linking Chromium or Firefox.
Differential Revision: https://reviews.llvm.org/D56986
llvm-svn: 351899
It turns out that sections in PGO instrumented object files on Windows
contain a large number of relocations pointing to themselves. With
r347429 this can cause many sections to receive the same hash (usually
zero) as a result of a section's hash being xor'ed with itself.
This patch causes the COFF and ELF linkers to avoid this problem
by adding the hash of the relocated section instead of xor'ing it.
On my machine this causes the regressing test case
provided by Mozilla to terminate in 2m41s.
Differential Revision: https://reviews.llvm.org/D56955
llvm-svn: 351898
to reflect the new license.
We understand that people may be surprised that we're moving the header
entirely to discuss the new license. We checked this carefully with the
Foundation's lawyer and we believe this is the correct approach.
Essentially, all code in the project is now made available by the LLVM
project under our new license, so you will see that the license headers
include that license only. Some of our contributors have contributed
code under our old license, and accordingly, we have retained a copy of
our old license notice in the top-level files in each project and
repository.
llvm-svn: 351636
On my machine this reduced median link time of lld-speed-test/chrome
from 2.68s to 2.41s. It also reduces link time of Chrome for Android
with a prototype compiler change that causes the compiler to create
large numbers of identical (modulo relocations) sections from >15
minutes to a few seconds.
Differential Revision: https://reviews.llvm.org/D54773
llvm-svn: 347594
Previously, we uncompress all compressed sections before doing anything.
That works, and that is conceptually simple, but that could results in
a waste of CPU time and memory if uncompressed sections are then
discarded or just copied to the output buffer.
In particular, if .debug_gnu_pub{names,types} are compressed and if no
-gdb-index option is given, we wasted CPU and memory because we
uncompress them into newly allocated bufers and then memcpy the buffers
to the output buffer. That temporary buffer was redundant.
This patch changes how to uncompress sections. Now, compressed sections
are uncompressed lazily. To do that, `Data` member of `InputSectionBase`
is now hidden from outside, and `data()` accessor automatically expands
an compressed buffer if necessary.
If no one calls `data()`, then `writeTo()` directly uncompresses
compressed data into the output buffer. That eliminates the redundant
memory allocation and redundant memcpy.
This patch significantly reduces memory consumption (20 GiB max RSS to
15 Gib) for an executable whose .debug_gnu_pub{names,types} are in total
5 GiB in an uncompressed form.
Differential Revision: https://reviews.llvm.org/D52917
llvm-svn: 343979
These symbols are declared early with the same value, so they otherwise
appear identical to ICF.
Differential Revision: https://reviews.llvm.org/D51376
llvm-svn: 340998
This patch does the same thing as r338153 for COFF.
Note that this patch affects only the order of log messages.
The output file is already deterministic.
Differential Revision: https://reviews.llvm.org/D50023
llvm-svn: 338406
The xxHash64 function has been made unsigned-char-independent, so
we can reland this change now.
Original commit message:
> The icf-safe.s test currently fails on 32-bit platforms because it uses
> the --print-icf-sections flag and depends on the output appearing in
> a specific order. However, this flag causes the output to depend on
> the order of the sections in the Sections array, which depends on the
> hash values returned from hash_combine, which happen to be different
> for that test between 32-bit and 64-bit platforms.
>
> This change makes the output deterministic by using xxHash64 instead of
> hash_combine.
Differential Revision: https://reviews.llvm.org/D49877
llvm-svn: 338153
The icf-safe.s test currently fails on 32-bit platforms because it uses
the --print-icf-sections flag and depends on the output appearing in
a specific order. However, this flag causes the output to depend on
the order of the sections in the Sections array, which depends on the
hash values returned from hash_combine, which happen to be different
for that test between 32-bit and 64-bit platforms.
This change makes the output deterministic by using xxHash64 instead of
hash_combine.
Differential Revision: https://reviews.llvm.org/D49877
llvm-svn: 338088
We are already ICF'ing these sections as a unit with their dependent
sections, so they don't need to be considered for ICF individually.
This change also "fixes" slowness caused by our quadratic-in-group-size
relocation segregation algorithm on 32-bit ARM platforms with unwind
data and ICF on rodata. In this scenario almost every function's
.ARM.exidx is identical except for the targets of the relocations
that refer to the function and its .ARM.extab, which causes almost
all of the program's .ARM.exidx sections to be initially added to the
same class, which causes us to compare every such section with every
other such section.
Differential Revision: https://reviews.llvm.org/D49716
llvm-svn: 337967
The gold behaviour with regard to --keep-unique is arguably a bug.
I also noticed a bug in my patch, which is that we mislink the
following program with --icf=safe by merging f3 and f4:
void f1() {}
void f2() {}
__attribute__((weak)) void* f3() { return f1; }
__attribute__((weak)) void* f4() { return f2; }
int main() {
printf("%p %p\n", f3(), f4());
}
llvm-svn: 337729
Under --icf=all we now only apply KeepUnique to non-executable
address-significant sections. This has the effect of making --icf=all
mean unsafe ICF for executable sections and safe ICF for non-executable
sections.
With this change the meaning of the KeepUnique bit changes to
"does the current ICF mode (together with the --keep-unique and
--ignore-data-address-equality flags) require this section to be
kept unique".
Differential Revision: https://reviews.llvm.org/D49626
llvm-svn: 337640
The only restriction is that we cannot merge more than one KeepUnique
section together. This matches gold's behaviour and reduces code size
when using --icf=safe.
Differential Revision: https://reviews.llvm.org/D49622
llvm-svn: 337638
Code is dead. We use only InputSections when building the list of
sections elegible for the ICF:
https://github.com/llvm-mirror/lld/blob/master/ELF/ICF.cpp#L439
And 'isEligible' filters out SyntheticSections as well for us.
That way the only Kind we have in the Sections vector is SectionBase::Regular,
so we do not need to check sections kind at all, it is always the same.
llvm-svn: 335460
Change removes the excessive comparsion of
the relocation arrays sizes.
This code was dead, because at the higer level,
equalsConstant function contains the following check:
`A->NumRelocations != B->NumRelocations`
where NumRelocations contains the size of the relocations array.
So removed check did the same job twice.
This was found with use of code coverage analysis.
llvm-svn: 335346
A user program may enumerate sections named with a C identifier using
__start_* and __stop_* symbols. We cannot ICF any such sections because
that could change program semantics.
Differential Revision: https://reviews.llvm.org/D47242
llvm-svn: 333054
Note that this doesn't do the right thing in the case where there is
a linker script. We probably need to move output section assignment
before ICF to get the correct behaviour here.
Differential Revision: https://reviews.llvm.org/D47241
llvm-svn: 333052
The --keep-unique <symbol> option is taken from gold. The intention is that
<symbol> will be prevented from being folded by ICF. Although not
specifically mentioned in the documentation <symbol> only matches
global symbols, with a warning if the symbol is not found.
The implementation finds the Section defining <symbol> and removes it from
the set of sections considered for ICF.
Differential Revision: https://reviews.llvm.org/D46755
llvm-svn: 332332
Summary:
r328610 fixed a data race in the COFF linker. This change makes a
similar fix to the ELF linker.
Reviewers: ruiu, pcc, rnk
Reviewed By: ruiu
Subscribers: emaste, llvm-commits, arichardson
Differential Revision: https://reviews.llvm.org/D45192
llvm-svn: 329088
The Data member of synthetic section's is not valid and empty. The Data
member is required to be valid by ICF as it is used by ICF to determine
the equality of section contents. Therefore, exclude synthetic sections
from ICF.
Fixes bug PR36910.
Differential Revision: https://reviews.llvm.org/D44923
llvm-svn: 328624
Latest patch version now.
Original commit message:
[ELF] - Do not crash with --emit-relocs and --icf=all together.
Previously we would crash because did not mark .rel[a] sections
as dead and they tried to access parent which was not live
after ICF and therefore was null.
Differential revision: https://reviews.llvm.org/D43241
llvm-svn: 325879
Andrew Ng