We support all __sync_val_compare_and_swap_* builtins (only 64-bit on 64-bit
targets) on all cores, and should define the corresponding
__GCC_HAVE_SYNC_COMPARE_AND_SWAP_* macros, just as GCC does. As it turns out,
this is really important because they're needed to prevent a bad ODR violation
with libstdc++'s std::shared_ptr (this is well explained in PR12730).
We were doing this only for P8, but this is necessary on all PPC systems.
llvm-svn: 249009
This reverts commit r248982 as it was breaking the ARM buildbots and the fix didn't work.
This reverts commit r248984, the fix that didn't work.
llvm-svn: 249005
With this patch we erase cached results for regions that are invalid
as early as possible. If we do not (as before), it is possible that
two expanded regions will have the same address and the tracked
results for both are mixed. Currently this would "only" cause
pessimism in later passes but that will change when we allow invariant
loads in the SCoP. Additionally, it triggers non-deterministic
results as we might dismiss a later region because of results cached
for an earlier one.
There is no test case as the problem occurs only non-deterministically.
llvm-svn: 249000
This allows clang-format to align identifiers in consecutive
declarations. This is useful for increasing the readability of the code
in the same way the alignment of assignations is.
The code is a slightly modified version of the consecutive assignment
alignment code. Currently only the identifiers are aligned, and there is
no support of alignment of the pointer star or reference symbol.
The patch also solve the issue of alignments not being possible due to
the ColumnLimit for both the existing AlignConsecutiveAligments and the
new AlignConsecutiveDeclarations.
Patch by Beren Minor, thank you.
Review: http://reviews.llvm.org/D12362
llvm-svn: 248999
Before this check, I would get the following error:
Updating ./misc/CMakeLists.txt...
Creating ./misc/NoReinterpret_castCheck.h...
Creating ./misc/NoReinterpret_castCheck.cpp...
Updating ./misc/MiscTidyModule.cpp...
Creating ../test/clang-tidy/misc-no-reinterpret_cast.cpp...
Creating ../docs/clang-tidy/checks/misc-no-reinterpret_cast.rst...
Traceback (most recent call last):
File "./add_new_check.py", line 271, in <module>
main()
File "./add_new_check.py", line 267, in main
update_checks_list(module_path)
File "./add_new_check.py", line 220, in update_checks_list
os.listdir('docs/clang-tidy/checks')))
OSError: [Errno 2] No such file or directory: 'docs/clang-tidy/checks'
Patch by Matthias Gehre!
Differential revision: http://reviews.llvm.org/D13272
llvm-svn: 248997
This diff requires http://reviews.llvm.org/D13079 to be applied first. I wasn't sure about how to make patch series in Phabricator, and I wanted to keep the two separate for clarity.
It looks like that most cases can be supported with this patch. I'm not totally sure about the actual coverage though. I think that the matchers are very generic, but I'm still not totally fluent with the AST.
Patch by Beren Minor!
Differential revision: http://reviews.llvm.org/D13081
llvm-svn: 248996
Summary:
This fixes https://llvm.org/bugs/show_bug.cgi?id=22196 .
Also add a non-trivially copyable type to fix some tests that were meant to be about using const-refs, but were changed in r248438.
Reviewers: klimek
Subscribers: alexfh, cfe-commits
Differential Revision: http://reviews.llvm.org/D13292
llvm-svn: 248994
Summary:
After a developer builds LLDB from source on Windows (assuming they've built it with Python support enabled), they may be somewhat flustered when it fails to launch with a cryptic error.
{F890625}
This happens because Windows can't find python27.dll (or python27_d.dll in case LLDB was built in debug mode). Many developers may have previously installed a release build of Python 2.7 and will not notice anything is amiss when they run a release build of LLDB because Windows will load the python27.dll from one of the system directories or `PATH` (rather than the one that the LLDB build instructions tell them to build). The issue tends to be more pronounced with debug builds of LLDB, since fewer developers probably have python27_d.dll sitting in one of the Windows system directories.
To ensure Windows loads the correct custom built Python DLL when launching LLDB I've added a post-build event that copies the relevant DLL (based on the LLDB build configuration) from `PYTHON_HOME` to the directory in which the LLDB executable is generated.
Patch by Vadim Macagon. Thanks!
Reviewers: brucem, zturner
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D13237
llvm-svn: 248992
Summary:
Previously `CMAKE_BUILD_TYPE` was used to determine whether to link in `python27.lib` or `python27_d.lib`, unfortunately this only works reliably when using a CMake generator that generates a single build configuration (e.g. Ninja). The Visual Studio CMake generator generates four build configurations at once (`Debug`, `Release`, `RelWithDebInfo`, `MinSizeRel`), so if `CMAKE_BUILD_TYPE` is set to `Debug` all four build configurations end up linking in `python27_d.lib`, this is clearly undesirable.
To ensure that the correct Python lib is used for each build configuration the value of `PYTHON_LIBRARY` is now determined using generator expressions that evaluate to either the debug or release Python lib. The values of `PYTHON_EXECUTABLE` and `PYTHON_DLL` are now likewise determined using generator expressions.
Note that these changes only apply to the Windows build.
Patch by Vadim Macagon. Thanks!
Reviewers: zturner, brucem
Subscribers: zturner, lldb-commits
Differential Revision: http://reviews.llvm.org/D13234
llvm-svn: 248991
introduced by r235737 but I didn't look into it too closely).
A dSYM can have a per-UUID plist in it which tells lldb where
to find an executable binary for the dSYM (DBGSymbolRichExecutable)
- other information can be included in this plist, like how to
remap the source file paths from their build pathnames to their
long-term storage pathnames.
This per-UUID plist is a unusual; it is used probably exclusively
inside apple with our build system. It is not created by default
in normal dSYMs.
The problem was like this:
1. lldb wants to find an executable, given only a UUID
(this happens when lldb is doing cross-host debugging
and doesn't have a copy of the target system's binaries)
2. It eventually calls LocateMacOSXFilesUsingDebugSymbols
which does a spotlight search for the dSYM on the local
system, and failing that, tries the DBGShellCommands
command to find the dSYM.
3. It gets a dSYM. It reads the per-UUID plist in the dSYM.
The dSYM has a DBGSymbolRichExecutable kv pair pointing to
the binary on a network filesystem.
4. Using the binary on the network filesystem, lldb now goes
to find the dSYM.
5. It starts by looking for a dSYM next to the binary it found.
6. lldb is now reading the dSYM over a network filesystem,
ignoring the one it found on its local filesystem earlier.
Everything still *works* but it's much slower.
This would be a tricky one to write up in a testsuite case;
you really need the binary to not exist on the local system.
And LocateMacOSXFilesUsingDebugSymbols will only compile on
Mac OS X - even if I found a way to write up a test case, it
would not run anywhere but on a mac.
One change Greg wanted while I was touching this code was to
have LocateMacOSXFilesUsingDebugSymbols (which could be asked
to find a binary OR find a dSYM) to instead return a ModuleSpec
with the sum total of everything it could find. This
change of passing around a ModuleSpec instead of a FileSpec
was percolated up into ModuleList::GetSharedModule.
The changes to LocateMacOSXFilesUsingDebugSymbols look larger
than they really are - there's a lot of simple whitespace changes
in there.
I ran the testsuites on mac, no new regressions introduced
<rdar://problem/21993813>
llvm-svn: 248985
Summary:
The instructions SeenExprs records may be deleted during rewriting.
FindClosestMatchingDominator should ignore these deleted instructions.
Fixes PR24301.
Reviewers: grosser
Subscribers: grosser, llvm-commits
Differential Revision: http://reviews.llvm.org/D13315
llvm-svn: 248983
Summary:
Without this patch, the memory manager would call `mprotect` on every memory
region it ever allocated whenever it wanted to finalize memory (i.e. not just
the ones it just allocated). This caused terrible performance problems for
long running memory managers. In one particular compile heavy julia benchmark,
we were spending 50% of time in `mprotect` if running under MCJIT.
Fix this by splitting allocated memory blocks into those on which memory
permissions have been set and those on which they haven't and only running
`mprotect` on the latter.
Reviewers: lhames
Subscribers: reames, llvm-commits
Differential Revision: http://reviews.llvm.org/D13156
llvm-svn: 248981
recently when we started using direct conversion to model sign
extension. The __v16qi type we use for SSE v16i8 vectors is defined in
terms of 'char' which may or may not be signed! This causes us to
generate pmovsx and pmovzx depending on the setting of -funsigned-char.
This patch just forms an explicitly signed type and uses that to
formulate the sign extension. While this gets the correct behavior
(which we now verify with the enhanced test) this is just the tip of the
ice berg. Now that I know what to look for, I have found errors of this
sort *throughout* our vector code. Fortunately, this is the only
specific place where I know of users actively having their code
miscompiled by Clang due to this, so I'm keeping the fix for those users
minimal and targeted.
I'll be sending a proper email for discussion of how to fix these
systematically, what the implications are, and just how widely broken
this is... From what I can tell, we have never shipped a correct set of
builtin headers for x86 when users rely on -funsigned-char. Oops.
llvm-svn: 248980
Make sure the output filepath supplied to createUniqueFile() in HTMLDiagnostics::ReportDiag() is absolute.
Summary: Make sure the output filepath supplied to createUniqueFile() in HTMLDiagnostics::ReportDiag() is absolute.
Reviewers: rsmith, akyrtzi
Subscribers: cfe-commits
Differential Revision: http://reviews.llvm.org/D12774
llvm-svn: 248977
This reverts commit r248963.
Seems there's some standard libraries (and libcxxabi implementations)
that aren't -Wdeprecated clean... hrm.
llvm-svn: 248972
- Trim spaces.
- Use nullptr in place of 0 for pointer variables.
- Use '!p' in place of 'p == 0' for null pointer checks.
- Add blank lines to separate function definitions.
- Add 'extern "C"' or 'namespace foo' comments after the appropriate
closing brackets
This is a continuation of work from 409b7b82. The focus here is on the
various sanitizers (not sanitizer_common, as before).
Patch by Eugene Zelenko!
Differential Revision: http://reviews.llvm.org/D13225
llvm-svn: 248966
The custom code produces incorrect results if later reassociated.
Since r221657, on x86, vNi32 uitofp is lowered using an optimized
sequence:
movdqa LCPI0_0(%rip), %xmm1 ## xmm1 = [65535, ...]
pand %xmm0, %xmm1
por LCPI0_1(%rip), %xmm1 ## [0x4b000000, ...]
psrld $16, %xmm0
por LCPI0_2(%rip), %xmm0 ## [0x53000000, ...]
addps LCPI0_3(%rip), %xmm0 ## [float -5.497642e+11, ...]
addps %xmm1, %xmm0
Since r240361, the machine combiner opportunistically reassociates
2-instruction sequences (with -ffast-math). In the new code sequence,
the ADDPS' are eligible. In isolation, for simple examples (without
reassociable users), this makes no performance difference (the goal
being to enable reassociation of longer chains).
In the trivial example (just one uitofp), the reassociation doesn't
happen, because (I think) it would require the emission of a separate
movaps for a constantpool load (instead of folding it into addps).
However, when we have multiple uitofp sequences, and the constantpool
loads are CSE'd earlier, the machine combiner can do the reassociation.
When the ADDPS' are reassociated, the resulting sequence isn't correct
anymore, as we'd be adding large (2**39) constants with comparatively
smaller values (~2**23). Given that two of the three inputs are powers
of 2 larger than 2**16, and that ulp(2**39) == 2**(39-24) == 2**15,
the reassociated chain will produce 0 for any input in [0, 2**14[.
In my testing, it also produces wrong results for 99.5% of [0, 2**32[.
Avoid this by disabling the new lowering when -ffast-math. It does
mean that we'll get slower code than without it, but at least we
won't get egregiously incorrect code.
One might argue that, considering -ffast-math is all but meaningless,
uitofp producing wrong results isn't a compiler bug. But it really is.
Fixes PR24512.
...though this is really more of a workaround.
Ideally, we'd have some sort of Machine FMF, but that's a problem
that's not worth tackling until we do more with machine IR.
llvm-svn: 248965
- Trim spaces.
- Use nullptr in place of 0 for pointer variables.
- Use '!p' in place of 'p == 0' for null pointer checks.
Patch by Eugene Zelenko!
Differential Revision: http://reviews.llvm.org/D13310
llvm-svn: 248964
This particularly helps enforce the C++ Rule of 5 (for new move ops this
is already an error, but for a type only using C++98 features (copy
ctor/assign, dtor) it is only deprecated, not invalid)
Applying the flag for any GCC compatible compiler - GCC doesn't warn on
the Rule of 5 cases that C++11 deprecates, but it doesn't have other
false positives so far as I could see (compiling with GCC 4.8 didn't
produce any -Wdeprecated warnings I could spot).
Reviewers: aaron.ballman
Differential Revision: http://reviews.llvm.org/D13314
llvm-svn: 248963
This is meant to support languages that have a scripting mode with top-level code that acts as global
For now, this flag only controls whether 'frame variable' will attempt to treat globals as locals when within such a function
llvm-svn: 248960
The Win64 unwinder disassembles forwards from each PC to try to
determine if this PC is in an epilogue. If so, it skips calling the EH
personality function for that frame. Typically, this means you cannot
catch an exception in the same frame that you threw it, because 'throw'
calls a noreturn runtime function.
Previously we avoided this problem with the TrapUnreachable
TargetOption, but that's a much bigger hammer than we need. All we need
is a 1 byte non-epilogue instruction right after the call. Instead,
what we got was an unconditional branch to a shared block containing the
ud2, potentially 7 bytes instead of 1. So, this reverts r206684, which
added TrapUnreachable, and replaces it with something better.
The new code pattern matches for invoke/call followed by unreachable and
inserts an int3 into the DAG. To be 100% watertight, we would need to
insert SEH_Epilogue instructions into all basic blocks ending in a call
with no terminators or successors, but in practice this is unlikely to
come up.
llvm-svn: 248959
Hal Finkel