Summary:
This commit contains the following changes:
- Rewrite vfile close/read/write packet handlers with portable routines from lldb.
This removes #if(s) and allows the handlers to work on Windows.
- Fix a bug in File::Write. This is intended to write data at an offset to a file
but actually writes at the current position of the file.
- Add a default boolean argument 'should_close_fd' to FileSystem::Open to
let the user decide whether to close the fd or not.
Reviewers: zturner, llvm-commits, labath
Reviewed By: zturner
Subscribers: Hui, labath, abidh, lldb-commits
Differential Revision: https://reviews.llvm.org/D56231
llvm-svn: 353446
R_X86_64_PC{8,16} relocations are sign-extended, so when we check
for relocation overflow, we had to use checkInt instead of checkUInt.
I confirmed that GNU linkers create the same output for the test case.
llvm-svn: 353437
Summary:
Before this change, check-profile would run, but all tests would be
marked unsupported on Windows. This is the new status of 'check-profile'
after this change:
Testing Time: 6.66s
Expected Passes : 29
Expected Failures : 5
Unsupported Tests : 39
I moved many tests that exercise posix-y features like dlopen and DSOs
into the Posix subdirectory, and ran the tests on Linux to validate my
changes.
These are the remaining tests that I handled on a case by case basis:
- instrprof-path.c
Passes, Fixed some path portability issues
- instrprof-gcov-exceptions.test
Passes, the FileCheck actually succeeds on Windows, so I RUNX'd it
- instrprof-icall-promo.test
XFAILed, probably due to C++ ABI differences in vtables
- instrprof-merge-match.test
- instrprof-merge.c
- instrprof-merging.cpp
XFAILed, These seem like real bugs that need fixing
- instrprof-version-mismatch.c
XFAILed, Overriding the weak version symbol doesn't work
- instrprof-without-libc.c
UNSUPPORTED, test needs an executable symbol table, Windows has none
Reviewers: davidxl, wmi, void
Subscribers: fedor.sergeev, #sanitizers, llvm-commits
Tags: #sanitizers
Differential Revision: https://reviews.llvm.org/D57853
llvm-svn: 353435
Summary:
When compiling with profile data, ensure the split cold function gets
cold function_entry_count metadata (just use 0 since it should be cold).
Otherwise with function sections it will not be placed in the unlikely
text section with other cold code.
Reviewers: vsk
Subscribers: sebpop, hiraditya, davidxl, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D57900
llvm-svn: 353434
I noticed that we are missing this canonicalization in IR:
rL352515
...and then realized that we don't get this right in SDAG either,
so this has to be fixed first regardless of what we choose to do in IR.
The existing fold was limited to scalars and using the wrong predicate
to guard the transform. We have a boolean contents TLI query that can
be used to decide which direction to fold.
This may eventually lead back to the problems/question in:
https://bugs.llvm.org/show_bug.cgi?id=40486
...but it makes no difference to that yet.
Differential Revision: https://reviews.llvm.org/D57401
llvm-svn: 353433
Introduce a new function which handles instructions with multiple type
indices, but have the same number of vector elements.
Also legalize v2s16 shifts when applicable.
llvm-svn: 353432
Valid OpenCL C code should still compile in C++ mode.
This change enables extensions and OpenCL types.
Differential Revision: https://reviews.llvm.org/D57824
llvm-svn: 353431
This is intentionally a small step because it's hard to know exactly
where we might introduce a conflicting transform with the code that
tries to form wider shuffles. But I think this is safe - if we have
a wide shuffle with 2 operands, then we should do better with an
extract + narrow shuffle.
Differential Revision: https://reviews.llvm.org/D57867
llvm-svn: 353427
Summary:
Currently `TestImportBase` is derived from `ParameterizedTestsFixture`
which explicitly states that the gtest parameter can be only an
`ArgVector`. This is a limitation when we want to create tests which may
have different parameters.
E.g. we would like to create tests where we can combine different test
parameters. So, for example we'd like gtest to be able to provide
parameters of `<std::tuple<ArgVector, const char *>` instead of a simple
`ArgVector`.
Reviewers: a_sidorin, shafik, a.sidorin
Subscribers: rnkovacs, dkrupp, Szelethus, gamesh411, cfe-commits
Differential Revision: https://reviews.llvm.org/D57322
llvm-svn: 353425
Summary:
When adding one thin archive to another, we currently chop off the relative path to the flattened members. For instance, when adding `foo/child.a` (which contains `x.txt`) to `parent.a`, when flattening it we should add it as `foo/x.txt` (which exists) instead of `x.txt` (which does not exist).
As a note, this also undoes the `IsNew` parameter of handling relative paths in r288280. The unit test there still passes.
This was reported as part of testing the kernel build with llvm-ar: https://patchwork.kernel.org/patch/10767545/ (see the second point).
Reviewers: mstorsjo, pcc, ruiu, davide, david2050
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D57842
llvm-svn: 353424
Summary:
clangd is using as many threads as logical cores for BackgroundIndex
by default. We observed that it increases latency of foreground tasks.
This patch aims to change that default to number of physical cores to get rid of
that extra latency.
Reviewers: ilya-biryukov
Reviewed By: ilya-biryukov
Subscribers: ioeric, MaskRay, jkorous, arphaman, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D57819
llvm-svn: 353422
The "signal" argument was removed from the MonitorCallback function, but
not from the log statements within it. This wasn't noticed because the
name "signal" suddenly started referring to the libc function with that
name.
This fixes that.
llvm-svn: 353419
Summary:
The standalone Scudo version is being built with `-Werror` which can be
tripped by extraneous command line arguments. We have little control over
those as they can be passed down to us by `CMAKE_C(XX)_FLAGS`, the reported
scenario involving `-stdlib=libc++` (see https://reviews.llvm.org/D57412#1384504).
To work around this, disable `-Wunused-command-line-argument`.
Reviewers: eugenis, vitalybuka, Eugene.Zelenko
Reviewed By: eugenis
Subscribers: mgorny, delcypher, #sanitizers, llvm-commits
Tags: #llvm, #sanitizers
Differential Revision: https://reviews.llvm.org/D57757
llvm-svn: 353418
Summary:
Memory usage for a sample TU:
Without Dex: 17.9M
With Dex: 24.4M
The memory increase is considerable but seems tolerable.
Subscribers: ilya-biryukov, MaskRay, jkorous, arphaman, kadircet, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D57878
llvm-svn: 353413
When type streams with forward references were merged using GHashes, cycles
were introduced in the debug info. This was caused by
GlobalTypeTableBuilder::insertRecordAs() not inserting the record on the second
pass, thus yielding an empty ArrayRef at that record slot. Later on, upon PDB
emission, TpiStreamBuilder::commit() would skip that empty record, thus
offseting all indices that came after in the stream.
This solution comes in two steps:
1. Fix the hash calculation, by doing a multiple-step resolution, iff there are
forward references in the input stream.
2. Fix merge by resolving with multiple passes, therefore moving records with
forward references at the end of the stream.
This patch also adds support for llvm-readoj --codeview-ghash.
Finally, fix dumpCodeViewMergedTypes() which previously could reference deleted
memory.
Fixes PR40221
Differential Revision: https://reviews.llvm.org/D57790
llvm-svn: 353412
The patch in r350643 incorrectly sets the COFF emission based on bits
instead of bytes. This patch converts the 32 via CharUnits to bits to
compare the correct values.
Change-Id: Icf38a16470ad5ae3531374969c033557ddb0d323
llvm-svn: 353411
The tests are failing on windows because the paths in the symbol file
are parsed using the host path style. I'm working on a patch to have
SymbolFileBreakpad auto-detect the correct path style (similar to dwarf
r351328).
I originally wanted to make this a part of the initial line-table patch,
but then I simply forgot.
llvm-svn: 353410
func_entry_exit.cc is for __tsan_func_entry/exit (spends ~75% there),
we don't yet have any.
mop.cc is for memory access functions, as compared to mini_bench_local/shared.cc
this benchmark passes through deduplication logic (ContainsSameAccess).
llvm-svn: 353407
The loop searching for use_lldb_suite_root had a bug where if the marker
file happened to be missing, it would enter an infinite loop. While this
shouldn't happen in normal circumstances, it can happen accidentally,
and debugging it is not very pleasant.
The loop had an exit condition, but it was incorrent (os.path.dirname
returning None). This will never happen as dirname will just return the
same folder over and over again once it reaches the root folder. This
fixes the exit condition to account for that.
llvm-svn: 353406
GetIndexForObjectImpl generated a bunch of "conversion casts away
constness warnings". Change the function to use "const void *" (and
static_cast, while I'm at it), to avoid this.
Driver.cpp: unused variable "replay" (this was actually caused by a
subsequent partial revert of this patch). I just finish the revert by
removing the variable completely.
llvm-svn: 353405
Summary:
This patch teaches SymbolFileBreakpad to parse the line information in
breakpad files and present it to lldb.
The trickiest question here was what kind of "compile units" to present
to lldb, as there really isn't enough information in breakpad files to
correctly reconstruct those.
A couple of options were considered
- have the entire file be one compile unit
- have one compile unit for each FILE record
- have one compile unit for each FUNC record
The main drawback of the first approach is that all of the files would
be considered "headers" by lldb, and so they wouldn't be searched if
target.inline-breakpoint-strategy=never. The single compile unit would
also be huge, and there isn't a good way to name it.
The second approach will create mostly correct compile units for cpp
files, but it will still be wrong for headers. However, the biggest
drawback here seemed to be the fact that this can cause a compile unit
to change mid-function (for example when a function from another file is
inlined or another file is #included into a function). While I don't
know of any specific thing that would break in this case, it does sound
like a thing that we should avoid.
In the end, we chose the third option, as it didn't seem to have any
major disadvantages, though it was not ideal either. One disadvantage
here is that this generates a large number of compile units, and there
is still a question on how to name it. We chose to simply name it after
the first line record in that function. This should be correct 99.99% of
the time, though it can produce somewhat strange results if the very
first line record comes from an #included file.
Reviewers: clayborg, zturner, lemo, markmentovai
Subscribers: lldb-commits
Differential Revision: https://reviews.llvm.org/D56595
llvm-svn: 353404
Modify GenerateConstantOffsetsImpl to create offsets that can be used
by indexed addressing modes. If formulae can be generated which
result in the constant offset being the same size as the recurrence,
we can generate a pre-indexed access. This allows the pointer to be
updated via the single pre-indexed access so that (hopefully) no
add/subs are required to update it for the next iteration. For small
cores, this can significantly improve performance DSP-like loops.
Differential Revision: https://reviews.llvm.org/D55373
llvm-svn: 353403
- Manually unwind code in MemoryAccessImpl1() because clang do not optimize it
- Check for .rodata section only in read operations
- Place LIKELY/UNLIKELY on fast paths
This speeds up synthetic memory access benchmarks by 10-20%.
[dvyukov: fixed up consts in check_analyze.sh]
Author: yuri (Yuri Per)
Reviewed in: https://reviews.llvm.org/D57882
Context: https://reviews.llvm.org/D54889
llvm-svn: 353401
Sam Clegg