In order to keep signal:noise high for the `__eh_frame` diff, I have teased-out the NFC changes and put them here.
Differential Revision: https://reviews.llvm.org/D114017
The `r_address` field of `relocation_info` is only 4 bytes, so our
offset field (which is the `r_address` field adjusted for subsection
splitting) also only needs to be 4 bytes. This reduces the structure
size from 32 bytes to 24 bytes.
Combined with https://reviews.llvm.org/D113813, this is a minor perf
improvement for linking an internal app, tested on two machines:
```
smol-relocs baseline difference (95% CI)
sys_time 7.367 ± 0.138 7.543 ± 0.157 [ +0.9% .. +3.8%]
user_time 21.843 ± 0.351 21.861 ± 0.450 [ -1.3% .. +1.4%]
wall_time 20.301 ± 0.307 20.556 ± 0.324 [ +0.1% .. +2.4%]
samples 16 16
smol-relocs baseline difference (95% CI)
sys_time 2.923 ± 0.050 2.992 ± 0.018 [ +1.4% .. +3.4%]
user_time 10.345 ± 0.039 10.448 ± 0.023 [ +0.8% .. +1.2%]
wall_time 12.068 ± 0.071 12.229 ± 0.021 [ +1.0% .. +1.7%]
samples 15 12
```
More importantly though, this change by itself reduces our maximum
resident set size by 220 MB (2.75%, from 7.85 GB to 7.64 GB) on the
first machine. On the second machine, it reduces it by 125 MB (1.94%,
from 6.31 GB to 6.19 GB).
Reviewed By: #lld-macho, int3
Differential Revision: https://reviews.llvm.org/D113818
This is an NFC diff that prepares for pruning & relocating `__eh_frame`.
Along the way, I made the following changes to ...
* clarify usage of `section` vs. `subsection`
* remove `map` & `vec` from type names
* disambiguate class `Section` from template parameter `SectionHeader`.
Differential Revision: https://reviews.llvm.org/D113241
autohide symbols behaves similarly to private_extern symbols.
However, LD64 allows exporting autohide symbols. LLD currently does not.
This patch allows LLD to export them.
Differential Revision: https://reviews.llvm.org/D113167
Dedup'ing unwind info is tricky because each CUE contains a different
function address, if ICF operated naively and compared the entire
contents of each CUE, entries with identical unwind info but belonging
to different functions would never be considered identical. To work
around this problem, we slice away the function address before
performing ICF. We rely on `relocateCompactUnwind()` to correctly handle
these truncated input sections.
Here are the numbers before and after D109944, D109945, and this diff
were applied, as tested on my 3.2 GHz 16-Core Intel Xeon W:
Without any optimizations:
base diff difference (95% CI)
sys_time 0.849 ± 0.015 0.896 ± 0.012 [ +4.8% .. +6.2%]
user_time 3.357 ± 0.030 3.512 ± 0.023 [ +4.3% .. +5.0%]
wall_time 3.944 ± 0.039 4.032 ± 0.031 [ +1.8% .. +2.6%]
samples 40 38
With `-dead_strip`:
base diff difference (95% CI)
sys_time 0.847 ± 0.010 0.896 ± 0.012 [ +5.2% .. +6.5%]
user_time 3.377 ± 0.014 3.532 ± 0.015 [ +4.4% .. +4.8%]
wall_time 3.962 ± 0.024 4.060 ± 0.030 [ +2.1% .. +2.8%]
samples 47 30
With `-dead_strip` and `--icf=all`:
base diff difference (95% CI)
sys_time 0.935 ± 0.013 0.957 ± 0.018 [ +1.5% .. +3.2%]
user_time 3.472 ± 0.022 6.531 ± 0.046 [ +87.6% .. +88.7%]
wall_time 4.080 ± 0.040 5.329 ± 0.060 [ +30.0% .. +31.2%]
samples 37 30
Unsurprisingly, ICF is now a lot slower, likely due to the much larger
number of input sections it needs to process. But the rest of the
linker only suffers a mild slowdown.
Note that the compact-unwind-bad-reloc.s test was expanded because we
now handle the relocation for CUE's function address in a separate code
path from the rest of the CUE relocations. The extended test covers both
code paths.
Reviewed By: #lld-macho, oontvoo
Differential Revision: https://reviews.llvm.org/D109946
Clang seems to emit all functionAddress relocs as section relocs, but
`ld -r` can turn those relocs into symbol ones. It turns out that we
weren't handling that case correctly when the symbol was a weak def
whose definition did not prevail.
Reviewed By: #lld-macho, oontvoo
Differential Revision: https://reviews.llvm.org/D113702
Clang seems to emit all functionAddress relocs as section relocs, but
`ld -r` can turn those relocs into symbol ones. It turns out that we
weren't handling that case correctly when the symbol was a weak def
whose definition did not prevail.
Reviewed By: #lld-macho, oontvoo
Differential Revision: https://reviews.llvm.org/D113702
In one of our links lld was reading 760k files, but the unique number of
files was only 1500. This takes that link from 30 seconds to 8.
This seems like a heavy hammer, especially since some things don't need
to be cached, like the filelist arguments and the passed static
archives (the latter is already cached as a one off), but it seems ld64
does something similar here to short circuit these duplicate reads:
82e429e186/src/ld/InputFiles.cpp (L644-L665)
Of the types of files being read for our iOS app, the biggest problem
was constantly re-reading small tbd files:
```
% wc -l /tmp/read.txt
761414 /tmp/read.txt
% cat /tmp/read.txt | sort -u | wc -l
1503
% cat /tmp/read.txt | grep "\.a$" | wc -l
43721
% cat /tmp/read.txt | grep "\.tbd$" | wc -l
717656
```
We could likely hoist this logic up to not cache at this level, but it
would be a more invasive change to make sure all callers that needed it
cached the results.
I could see this being an issue with OOMs, and I'm not a linker expert so
maybe there's another way we should solve this problem? Feedback welcome!
Reviewed By: int3, #lld-macho
Differential Revision: https://reviews.llvm.org/D113153
By default with ld64, architecture mismatches are just warnings, then
this flag can be passed to make these fail. This matches that behavior.
Reviewed By: int3, #lld-macho
Differential Revision: https://reviews.llvm.org/D113082
Broken by a9353dbe51.
Now that the functions point to the compact unwind entries, instead of
the other way around, we need to perform the "invalid reference" check
in a different place.
This change was originally part of the stacked diff D109946, but should
have been included as part of D109945.
Compact unwind entries (CUEs) contain pointers to their respective
function symbols. However, during the link process, it's far more useful
to have pointers from the function symbol to the CUE than vice versa.
This diff adds that pointer in the form of `Defined::compactUnwind`.
In particular, when doing dead-stripping, we want to mark CUEs live when
their function symbol is live; and when doing ICF, we want to dedup
sections iff the symbols in that section have identical CUEs. In both
cases, we want to be able to locate the symbols within a given section,
as well as locate the CUEs belonging to those symbols. So this diff also
adds `InputSection::symbols`.
The ultimate goal of this refactor is to have ICF support dedup'ing
functions with unwind info, but that will be handled in subsequent
diffs. This diff focuses on simplifying `-dead_strip` --
`findFunctionsWithUnwindInfo` is no longer necessary, and
`Defined::isLive()` is now a lot simpler. Moreover, UnwindInfoSection no
longer has to check for dead CUEs -- we simply avoid adding them in the
first place.
Additionally, we now support stripping of dead LSDAs, which follows
quite naturally since `markLive()` can now reach them via the CUEs.
Reviewed By: #lld-macho, gkm
Differential Revision: https://reviews.llvm.org/D109944
In Driver.cpp, addFramework used std::string instance to represent the path of a framework, which will be freed after the function returns. However, this string is stored in loadedArchive, which will be used later to compare with path of newly added frameworks. This caused https://bugs.llvm.org/show_bug.cgi?id=52133. A test is included in this commit to reproduce this bug.
Now resolveDylibPath returns a StringRef instance, and it uses StringSaver to save its data, then returns it to functions on the top. This ensures the resolved framework path is still valid after LC_LINKER_OPTION is parsed.
Reviewed By: int3, #lld-macho, oontvoo
Differential Revision: https://reviews.llvm.org/D111706
... instead of constructing a new one each time. This allows us
to take advantage of {D105305}.
I didn't see a substantial difference when linking chromium_framework,
but this paves the way for reusing similar logic for splitting compact
unwind entries into sections. There are a lot more of those, so the
performance impact is significant.
Differential Revision: https://reviews.llvm.org/D109895
The previous logic was duplicated between symbol-initiated
archive loads versus flag-initiated loads (i.e. `-force_load` and
`-ObjC`). This resulted in code duplication as well as redundant work --
we would create Archive instances twice whenever we had one of those
flags; once in `getArchiveMembers` and again when we constructed the
ArchiveFile.
This was motivated by an upcoming diff where we load archive members
containing ObjC-related symbols before loading those containing
ObjC-related sections, as well as before performing symbol resolution.
Without this refactor, it would be difficult to do that while avoiding
loading the same archive member twice.
Differential Revision: https://reviews.llvm.org/D108780
Address post follow up comment in D108016. Avoid creating isec for
LLVM segments since we are skipping over it.
Reviewed By: #lld-macho, int3
Differential Revision: https://reviews.llvm.org/D108167
There was an instance of a third-party archive containing multiple
_llvm symbols from different files that clashed with each other
producing duplicate symbols. Symbols under the LLVM segment
don't seem to be producing any meaningful value, so just ignore them.
Reviewed By: #lld-macho, int3
Differential Revision: https://reviews.llvm.org/D108016
ld64 seems to handle common symbols in bitcode rather
bizarrely. They follow entirely different precedence rules from their
non-bitcode counterparts. I initially tried to emulate ld64 in D106597,
but I'm not sure the extra complexity is worth it, especially given that
common symbols are not, well, very common.
This diff accords common bitcode symbols the same precedence as regular
common symbols, just as we treat all other pairs of bitcode and
non-bitcode symbol types. The tests document ld64's behavior in detail,
just in case we want to revisit this.
Reviewed By: #lld-macho, thakis
Differential Revision: https://reviews.llvm.org/D107027
Ported from COFF/ELF; test is adapted from
test/COFF/thinlto-archivecollision.ll
LTO expects every bitcode file to have a unique name. If given multiple bitcode
files with the same name, it errors with "Expected at most one ThinLTO module
per bitcode file".
This change incorporates the archive name, to disambiguate members with the
same name in different archives and the offset in archive to disambiguate
members with the same name in the same archive.
Differential Revision: https://reviews.llvm.org/D106179
The mappings we were using had a small number of keys, so a vector is
probably better. This allows us to remove the last usage of std::map in
our codebase.
I also used `removeSimulator` to simplify the code a bit further.
Reviewed By: #lld-macho, thakis
Differential Revision: https://reviews.llvm.org/D105786
This is one of two flags clang passes to the linker when giving calling
clang with multiple -arch flags.
I think it'd make sense to also use finalOutput instead of outputFile
in CodeSignatureSection() and when replacing @executable_path, but
ld64 doesn't do that, so I'll at least put those in separate commits.
Differential Revision: https://reviews.llvm.org/D105449
clang and gcc both seem to emit relocations in reverse order of
address. That means we can match relocations to their containing
subsections in `O(relocs + subsections)` rather than the `O(relocs *
log(subsections))` that our previous binary search implementation
required.
Unfortunately, `ld -r` can still emit unsorted relocations, so we have a
fallback code path for that (less common) case.
Numbers for linking chromium_framework on my 3.2 GHz 16-Core Intel Xeon W:
N Min Max Median Avg Stddev
x 20 4.04 4.11 4.075 4.0775 0.018027756
+ 20 3.95 4.02 3.98 3.985 0.020900768
Difference at 95.0% confidence
-0.0925 +/- 0.0124919
-2.26855% +/- 0.306361%
(Student's t, pooled s = 0.0195172)
Reviewed By: #lld-macho, thakis
Differential Revision: https://reviews.llvm.org/D105410
We have been creating many ConcatInputSections with identical values due
to .subsections_via_symbols. This diff factors out the identical values
into a Shared struct, to reduce memory consumption and make copying
cheaper.
I also changed `callSiteCount` from a uint32_t to a 31-bit field to save an
extra word.
All in all, this takes InputSection from 120 to 72 bytes (and
ConcatInputSection from 160 to 112 bytes), i.e. 30% size reduction in
ConcatInputSection.
Numbers for linking chromium_framework on my 3.2 GHz 16-Core Intel Xeon W:
N Min Max Median Avg Stddev
x 20 4.14 4.24 4.18 4.183 0.027548999
+ 20 4.04 4.11 4.075 4.0775 0.018027756
Difference at 95.0% confidence
-0.1055 +/- 0.0149005
-2.52211% +/- 0.356215%
(Student's t, pooled s = 0.0232803)
Reviewed By: #lld-macho, thakis
Differential Revision: https://reviews.llvm.org/D105305
`__cfstring` is a special literal section, so instead of breaking it up
at symbol boundaries, we break it up at fixed-width boundaries (since
each literal is the same size). Symbols can only occur at one of those
boundaries, so this is strictly more powerful than
`.subsections_via_symbols`.
With that in place, we then run the section through ICF.
This change is about perf-neutral when linking chromium_framework.
Reviewed By: #lld-macho, gkm
Differential Revision: https://reviews.llvm.org/D105045
SymtabSection::emitStabs() writes the symbol table in the order
of externalSymbols, which has the order of symtab->getSymbols(),
which is just the order symbols are added to the symbol table.
In practice, symbols in the symbol files of input .o files are
sorted, but since that's not guaranteed we sort them in
ObjFile::parseSymbols(). To make sure several symbols with the same
address keep the order they're in the input file, we have to use
stable_sort().
In practice, std::sort() on already-sorted inputs won't change the order
of just adjacent elements, and while in theory std::sort() could use a
random pivot, in practice the code should be deterministic as it was
previously too.
But now lld/test/MachO/stabs.s passes with LLVM_ENABLE_EXPENSIVE_CHECKS=ON
(the last test that was failing with that set).
Fixes a regression from D99972.
While here, remove an empty section in stabs.s and move
.subsections_via_symbols to the end where it usually is (this part no
behavior change).
Differential Revision: https://reviews.llvm.org/D105071
Fixes PR50637.
Downstream bug: https://crbug.com/1218958
Currently, we split __cstring along symbol boundaries with .subsections_via_symbols
when not deduplicating, and along null bytes when deduplicating. This change splits
along null bytes unconditionally, and preserves original alignment in the non-
deduplicated case.
Removing subsections-section-relocs.s because with this change, __cstring
is never reordered based on the order file.
Differential Revision: https://reviews.llvm.org/D104919
Previously, we asserted that such a case was invalid, but in fact
`ld -r` can emit such symbols if the input contained a (true) private
extern, or if it contained a symbol started with "L".
Non-extern symbols marked as private extern are essentially equivalent
to regular TU-scoped symbols, so no new functionality is needed.
Reviewed By: #lld-macho, thakis
Differential Revision: https://reviews.llvm.org/D104502
ICF = Identical C(ode|OMDAT) Folding
This is the LLD ELF/COFF algorithm, adapted for MachO. So far, only `-icf all` is supported. In order to support `-icf safe`, we will need to port address-significance tables (`.addrsig` directives) to MachO, which will come in later diffs.
`check-{llvm,clang,lld}` have 0 regressions for `lld -icf all` vs. baseline ld64.
We only run ICF on `__TEXT,__text` for reasons explained in the block comment in `ConcatOutputSection.cpp`.
Here is the perf impact for linking `chromium_framekwork` on a Mac Pro (16-core Xeon W) for the non-ICF case vs. pre-ICF:
```
N Min Max Median Avg Stddev
x 20 4.27 4.44 4.34 4.349 0.043029977
+ 20 4.37 4.46 4.405 4.4115 0.025188761
Difference at 95.0% confidence
0.0625 +/- 0.0225658
1.43711% +/- 0.518873%
(Student's t, pooled s = 0.0352566)
```
Reviewed By: #lld-macho, int3
Differential Revision: https://reviews.llvm.org/D103292
It's a warning in ld64. While having LLD be stricter would be nice, it
makes it harder for it to be a drop-in replacement into existing builds.
Reviewed By: #lld-macho, thakis
Differential Revision: https://reviews.llvm.org/D104333
This is motivated by an upcoming diff in which the
WordLiteralInputSection ctor sets itself up based on the value of its
section flags. As such, it needs to be passed the `flags` value as part
of its ctor parameters, instead of having them assigned after the fact
in `parseSection()`. While refactoring code to make that possible, I
figured it would make sense for the other InputSections to also take
their initial values as ctor parameters.
Reviewed By: #lld-macho, thakis
Differential Revision: https://reviews.llvm.org/D103978
Conceptually, the implementation is pretty straightforward: we put each
literal value into a hashtable, and then write out the keys of that
hashtable at the end.
In contrast with ELF, the Mach-O format does not support variable-length
literals that aren't strings. Its literals are either 4, 8, or 16 bytes
in length. LLD-ELF dedups its literals via sorting + uniq'ing, but since
we don't need to worry about overly-long values, we should be able to do
a faster job by just hashing.
That said, the implementation right now is far from optimal, because we
add to those hashtables serially. To parallelize this, we'll need a
basic concurrent hashtable (only needs to support concurrent writes w/o
interleave reads), which shouldn't be to hard to implement, but I'd like
to punt on it for now.
Numbers for linking chromium_framework on my 3.2 GHz 16-Core Intel Xeon W:
N Min Max Median Avg Stddev
x 20 4.27 4.39 4.315 4.3225 0.033225703
+ 20 4.36 4.82 4.44 4.4845 0.13152846
Difference at 95.0% confidence
0.162 +/- 0.0613971
3.74783% +/- 1.42041%
(Student's t, pooled s = 0.0959262)
This corresponds to binary size savings of 2MB out of 335MB, or 0.6%.
It's not a great tradeoff as-is, but as mentioned our implementation can
be signficantly optimized, and literal dedup will unlock more
opportunities for ICF to identify identical structures that reference
the same literals.
Reviewed By: #lld-macho, gkm
Differential Revision: https://reviews.llvm.org/D103113
This is important for Frameworks, which are usually symlinks.
ld64 gets this right for @rpath that's replaced with @loader_path, but not for
bare @loader_path -- ld64's code calls realpath() in that case too, but ignores
the result.
ld64 somehow manages to find libbar1.dylib in the test without the
explicit `-rpath` in Foo1. I don't understand why or how. But this
change is a step forward and fixes an immediate problem I'm having,
so let's start with this :)
Differential Revision: https://reviews.llvm.org/D103990
Our implementation draws heavily from LLD-ELF's, which in turn delegates
its string deduplication to llvm-mc's StringTableBuilder. The messiness of
this diff is largely due to the fact that we've previously assumed that
all InputSections get concatenated together to form the output. This is
no longer true with CStringInputSections, which split their contents into
StringPieces. StringPieces are much more lightweight than InputSections,
which is important as we create a lot of them. They may also overlap in
the output, which makes it possible for strings to be tail-merged. In
fact, the initial version of this diff implemented tail merging, but
I've dropped it for reasons I'll explain later.
**Alignment Issues**
Mergeable cstring literals are found under the `__TEXT,__cstring`
section. In contrast to ELF, which puts strings that need different
alignments into different sections, clang's Mach-O backend puts them all
in one section. Strings that need to be aligned have the `.p2align`
directive emitted before them, which simply translates into zero padding
in the object file.
I *think* ld64 extracts the desired per-string alignment from this data
by preserving each string's offset from the last section-aligned
address. I'm not entirely certain since it doesn't seem consistent about
doing this; but perhaps this can be chalked up to cases where ld64 has
to deduplicate strings with different offset/alignment combos -- it
seems to pick one of their alignments to preserve. This doesn't seem
correct in general; we can in fact can induce ld64 to produce a crashing
binary just by linking in an additional object file that only contains
cstrings and no code. See PR50563 for details.
Moreover, this scheme seems rather inefficient: since unaligned and
aligned strings are all put in the same section, which has a single
alignment value, it doesn't seem possible to tell whether a given string
doesn't have any alignment requirements. Preserving offset+alignments
for strings that don't need it is wasteful.
In practice, the crashes seen so far seem to stem from x86_64 SIMD
operations on cstrings. X86_64 requires SIMD accesses to be
16-byte-aligned. So for now, I'm thinking of just aligning all strings
to 16 bytes on x86_64. This is indeed wasteful, but implementation-wise
it's simpler than preserving per-string alignment+offsets. It also
avoids the aforementioned crash after deduplication of
differently-aligned strings. Finally, the overhead is not huge: using
16-byte alignment (vs no alignment) is only a 0.5% size overhead when
linking chromium_framework.
With these alignment requirements, it doesn't make sense to attempt tail
merging -- most strings will not be eligible since their overlaps aren't
likely to start at a 16-byte boundary. Tail-merging (with alignment) for
chromium_framework only improves size by 0.3%.
It's worth noting that LLD-ELF only does tail merging at `-O2`. By
default (at `-O1`), it just deduplicates w/o tail merging. @thakis has
also mentioned that they saw it regress compressed size in some cases
and therefore turned it off. `ld64` does not seem to do tail merging at
all.
**Performance Numbers**
CString deduplication reduces chromium_framework from 250MB to 242MB, or
about a 3.2% reduction.
Numbers for linking chromium_framework on my 3.2 GHz 16-Core Intel Xeon W:
N Min Max Median Avg Stddev
x 20 3.91 4.03 3.935 3.95 0.034641016
+ 20 3.99 4.14 4.015 4.0365 0.0492336
Difference at 95.0% confidence
0.0865 +/- 0.027245
2.18987% +/- 0.689746%
(Student's t, pooled s = 0.0425673)
As expected, cstring merging incurs some non-trivial overhead.
When passing `--no-literal-merge`, it seems that performance is the
same, i.e. the refactoring in this diff didn't cost us.
N Min Max Median Avg Stddev
x 20 3.91 4.03 3.935 3.95 0.034641016
+ 20 3.89 4.02 3.935 3.9435 0.043197831
No difference proven at 95.0% confidence
Reviewed By: #lld-macho, gkm
Differential Revision: https://reviews.llvm.org/D102964
When a library "host"'s reexports change their installName with
`$ld$os10.11$install_name$host`, we used to write a load command for "host" but
write the version numbers of the reexport instead of "host". This fixes that.
I first thought that the rule is to take the version numbers from the library
that originally had that install name (implemented in D103819), but that's not
what ld64 seems to be doing: It takes the version number from the first dylib
with that install name it loads, and it loads the reexporting library before
the reexports. We already did most of that, we just added reexports before the
reexporter. After this change, we add the reexporter before the reexports.
Addresses https://bugs.llvm.org/show_bug.cgi?id=49800#c11 part 1.
(ld64 seems to add reexports after processing _all_ files on the command line,
while we add them right after the reexporter. For the common case of reexport +
$ld$ symbol changing back to the exporter name, this doesn't make a difference,
but you can construct a case where it does. I expect this to not make a
difference in practice though.)
Differential Revision: https://reviews.llvm.org/D103821
Also adjust a few comments, and move the DylibFile comment talking about
umbrella next to the parameter again.
Differential Revision: https://reviews.llvm.org/D103783
Greg McGary