Justin Lebar [Wed, 15 Jun 2016 23:20:15 +0000 (23:20 +0000)]
[IR] [DAE] Copy comdats during DAE, and don't copy comdats in GlobalObject::copyAttributesFrom.
Summary: This reverts the changes to Globals.cpp and IRMover.cpp in
"[IR] Copy comdats in GlobalObject::copyAttributesFrom" (D20631,
rL270743).
The DeadArgElim test is left unchanged, and we change DAE to explicitly
copy comdats.
The reverted change breaks copyAttributesFrom when the destination lives
in a different module from the source. The decision in D21255 was to
revert this patch and handle comdat copying separately from
copyAttributesFrom.
Justin Lebar [Wed, 15 Jun 2016 23:20:12 +0000 (23:20 +0000)]
[Bugpoint] Erase comdat annotations after removing a global's initializer.
Summary:
This is necessary to keep the verifier happy after bugpoint removes an
initializer from a global variable with a comdat annotation, because
globals without initializers may not have comdats.
Kevin Enderby [Wed, 15 Jun 2016 21:14:01 +0000 (21:14 +0000)]
Fix llvm-objdump when disassembling a stripped Mach-O binary with the -macho option.
It was printing out nothing in this case.
llvm-objdump tries to disassemble sections a symbol at a time. In the case of a
fully stripped Mach-O executable the only symbol remaining in the (__TEXT,__text)
section is the special linker defined symbol __mh_execute_header . This
symbol is special in that while it is N_SECT symbol in the (__TEXT,__text)
its address is before the start of the (__TEXT,__text). It’s address is the
start of the __TEXT segment which is where the mach header is statically
linked. So the code in DisassembleMachO() needs to deal with this case specially.
[CFLAA] Ignore non-pointers, move Attrs to graph nodes.
This patch makes CFLAA ignore non-pointer values, since we can now
sanely do that with the escaping/unknown attributes. Additionally,
StratifiedAttrs make more sense to sit on nodes than edges (since
they're properties of values, and ultimately end up on the nodes of
StratifiedSets). So, this patch puts said attributes on nodes.
Tim Northover [Wed, 15 Jun 2016 20:33:36 +0000 (20:33 +0000)]
AArch64: stop trying to use 32-bit MOVZs when expanding patchpoints.
Of course the assembly was right but because the opcode was MOVZWi it was
encoded as "movz w16, #65535, lsl #32" which is an unallocated encoding and
would go horribly wrong on a CPU.
No idea how this bug survived this long. It seems nobody is using that aspect
of patchpoints.
Sanjay Patel [Wed, 15 Jun 2016 20:26:58 +0000 (20:26 +0000)]
[x86] add folds for x86 vector compare nodes (PR27924)
Ideally, we can get rid of most x86 LLVM intrinsics by transforming them to IR (and some of that happened
with http://reviews.llvm.org/rL272807), but it doesn't cost much to have some simple folds in the backend
too while we're working on that and as a backstop.
This fixes:
https://llvm.org/bugs/show_bug.cgi?id=27924
Matthias Braun [Wed, 15 Jun 2016 20:19:16 +0000 (20:19 +0000)]
Statistic: Add machine parseable json output
- We lacked a short unique identifier for a statistics, so I renamed the
current "Name" field that just contained the DEBUG_TYPE name of the
current file to DebugType and added a new "Name" field that contains
the C++ identifier of the statistic variable.
- Add the -stats-json option which outputs statistics in json format.
Sanjay Patel [Wed, 15 Jun 2016 17:17:27 +0000 (17:17 +0000)]
[x86, SSE] remove the GCCBuiltins from the integer min/max intrinsics
This allows us to emit native IR in Clang (next commit).
Also, update the intrinsic tests to show that codegen already knows how to handle
the IR that Clang will soon produce.
Sean Silva [Wed, 15 Jun 2016 10:51:40 +0000 (10:51 +0000)]
Work around MSVC "friend" semantics.
The error on clang-x86-win2008-selfhost is:
C:\buildbot\slave-config\clang-x86-win2008-selfhost\llvm\lib\Transforms\Vectorize\SLPVectorizer.cpp(955) : error C2248: 'llvm::slpvectorizer::BoUpSLP::ScheduleData' : cannot access private struct declared in class 'llvm::slpvectorizer::BoUpSLP'
C:\buildbot\slave-config\clang-x86-win2008-selfhost\llvm\lib\Transforms\Vectorize\SLPVectorizer.cpp(608) : see declaration of 'llvm::slpvectorizer::BoUpSLP::ScheduleData'
C:\buildbot\slave-config\clang-x86-win2008-selfhost\llvm\lib\Transforms\Vectorize\SLPVectorizer.cpp(337) : see declaration of 'llvm::slpvectorizer::BoUpSLP'
I reproduced this locally with both MSVC 2013 and MSVC 2015.
Patrik Hagglund [Wed, 15 Jun 2016 10:32:00 +0000 (10:32 +0000)]
Use FPasses in opt exactly when it is initialized.
Previously, there was a discrepancy between the population of function
passes in FPasses, and their invocation. Function passes specified on
the command line, after an optimizaton level was simply discared. This
fix PR27509.
Sean Silva [Wed, 15 Jun 2016 09:00:33 +0000 (09:00 +0000)]
Speculative buildbot fix.
This wasn't failing for me with clang as the compiler. I think GCC may
disagree with clang about whether a friend declaration introduces a
declaration in the enclosing namespace (or something).
Example error:
/home/uweigand/sandbox/buildbot/clang-s390x-linux/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp:950:77: error: ‘llvm::raw_ostream& llvm::slpvectorizer::operator<<(llvm::raw_ostream&, const llvm::slpvectorizer::BoUpSLP::ScheduleData&)’ should have been declared inside ‘llvm::slpvectorizer’
const BoUpSLP::ScheduleData &SD) {
^
Sean Silva [Wed, 15 Jun 2016 08:43:40 +0000 (08:43 +0000)]
[PM] Port SLPVectorizer to the new PM
This uses the "runImpl" approach to share code with the old PM.
Porting to the new PM meant abandoning the anonymous namespace enclosing
most of SLPVectorizer.cpp which is a bit of a bummer (but not a big deal
compared to having to pull the pass class into a header which the new PM
requires since it calls the constructor directly).
Summary:
This fixes two related bugs. First, the generic optimization passes
unfortunately generate negative constant offsets but the hardware treats
SOffset as an unsigned value.
Second, there is a hardware bug on SI and CI, where address clamping in MUBUF
instructions does not work correctly when SOffset is larger than the buffer
size. This patch works around this bug by never using SOffset.
An alternative workaround would be to do the clamping manually when SOffset
is too large, but generating the required code sequence during instruction
selection would be rather involved, and in any case the resulting code would
probably be worse.
Sanjoy Das [Wed, 15 Jun 2016 05:35:14 +0000 (05:35 +0000)]
Don't force SP-relative addressing for statepoints
Summary:
... when the offset is not statically known.
Prioritize addresses relative to the stack pointer in the stackmap, but
fallback gracefully to other modes of addressing if the offset to the
stack pointer is not a known constant.
Dan Liew [Wed, 15 Jun 2016 01:40:02 +0000 (01:40 +0000)]
[LibFuzzer] Fix ``FuzzerMutate.ShuffleBytes2`` unit test on OSX.
The ``FuzzerMutate.ShuffleBytes2`` unit test was failing on
OSX due to the implementation of ``std::random_shuffle()``
being different between libcxx and libstdc++.
@kcc has decided (see http://reviews.llvm.org/D21218) it is acceptable
for there to be different mutation behavior on different platforms so
this commit just adjusts the test to perform the minimum number of
iterations (that is a power of 2) to see all the mutations the unit test
is looking for.
Recommit [LV] Enable vectorization of loops where the IV has an external use
r272715 broke libcxx because it did not correctly handle cases where the
last iteration of one IV is the second-to-last iteration of another.
Original commit message:
Vectorizing loops with "escaping" IVs has been disabled since r190790, due to
PR17179. This re-enables it, with support for external use of both
"post-increment" (last iteration) and "pre-increment" (second-to-last iteration)
IVs.
David Majnemer [Wed, 15 Jun 2016 00:19:56 +0000 (00:19 +0000)]
[LoopUnroll] Don't crash trying to unroll loop with EH pad exit
We do not support splitting cleanuppad or catchswitches. This is
problematic for passes which assume that a loop is in loop simplify
form (the loop would have a dedicated exit block instead of sharing it).
While it isn't great that we don't support this for cleanups, we still
cannot make loop-simplify form an assertable precondition because
indirectbr will also disable these sorts of CFG cleanups.
David Majnemer [Wed, 15 Jun 2016 00:19:09 +0000 (00:19 +0000)]
Remove the ScalarReplAggregates pass
Nearly all the changes to this pass have been done while maintaining and
updating other parts of LLVM. LLVM has had another pass, SROA, which
has superseded ScalarReplAggregates for quite some time.
Dehao Chen [Tue, 14 Jun 2016 22:27:17 +0000 (22:27 +0000)]
Set machine block placement hot prob threshold for both static and runtime profile.
Summary: With runtime profile, we have more confidence in branch probability, thus during basic block layout, we set a lower hot prob threshold so that blocks can be layouted optimally.
Vitaly Buka [Tue, 14 Jun 2016 22:09:00 +0000 (22:09 +0000)]
Revert "Enable libFuzzer's afl_driver to append stderr to a file."
Crashes with AddressSanitizer: SEGV on unknown address
http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fuzzer/builds/9924/steps/annotate/logs/stdio
Amaury Sechet [Tue, 14 Jun 2016 22:04:16 +0000 (22:04 +0000)]
Rename AttributeSetImpl::NumAttrs and AttributeSetImpl::getNumAttributes to reflect that they work on slots rather than attributes. NFC
Summary: The current naming not only doesn't convey the meaning of what this does, but worse, it convey the wrong meaning. This was a major source of confusion understanding the code, so I'm applying the boy scout rule here and making it better after I leave.
[LV] Enable vectorization of loops where the IV has an external use
Vectorizing loops with "escaping" IVs has been disabled since r190790, due to
PR17179. This re-enables it, with support for external use of both
"post-increment" (last iteration) and "pre-increment" (second-to-last iteration)
IVs.
If a local_unnamed_addr attribute is attached to a global, the address
is known to be insignificant within the module. It is distinct from the
existing unnamed_addr attribute in that it only describes a local property
of the module rather than a global property of the symbol.
This attribute is intended to be used by the code generator and LTO to allow
the linker to decide whether the global needs to be in the symbol table. It is
possible to exclude a global from the symbol table if three things are true:
- This attribute is present on every instance of the global (which means that
the normal rule that the global must have a unique address can be broken without
being observable by the program by performing comparisons against the global's
address)
- The global has linkonce_odr linkage (which means that each linkage unit must have
its own copy of the global if it requires one, and the copy in each linkage unit
must be the same)
- It is a constant or a function (which means that the program cannot observe that
the unique-address rule has been broken by writing to the global)
Although this attribute could in principle be computed from the module
contents, LTO clients (i.e. linkers) will normally need to be able to compute
this property as part of symbol resolution, and it would be inefficient to
materialize every module just to compute it.
See:
http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20160509/356401.html
http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20160516/356738.html
for earlier discussion.
Vitaly Buka [Tue, 14 Jun 2016 20:42:05 +0000 (20:42 +0000)]
Enable libFuzzer's afl_driver to append stderr to a file.
Summary:
[libFuzzer] Enable afl_driver to append stderr to a user specified file.
Append stderr of afl_driver to the file specified by the environmental variable
AFL_DRIVER_STDERR_DUPLICATE_FILENAME if it is set. This lets users see outputs
on crashes without rerunning crashing test cases (which won't work for crashes
that are difficult to reproduce). Before this patch, stderr would only be sent to afl-fuzz
and users would have no way of seeing it.
Tom Stellard [Tue, 14 Jun 2016 20:29:59 +0000 (20:29 +0000)]
AMDGPU/SI: Refactor fixup handling for constant addrspace variables
Summary:
We now use a standard fixup type applying the pc-relative address of
constant address space variables, and we have the GlobalAddress lowering
code add the required 4 byte offset to the global address rather than
doing it as part of the fixup.
This refactoring will make it easier to use the same code for global
address space variables and also simplifies the code.
Re-commit this after fixing a bug where we were trying to use a
reference to a Triple object that had already been destroyed.
Amaury Sechet [Tue, 14 Jun 2016 20:27:35 +0000 (20:27 +0000)]
Make sure attribute kind and attributes are named respectively Kind and Attr consistently. Historically they used to be the same the terminology is very confused in the codebase. NFC.
Sanjoy Das [Tue, 14 Jun 2016 20:23:16 +0000 (20:23 +0000)]
[ValueTracking] Calls to @llvm.assume always return
This change teaches llvm::isGuaranteedToTransferExecutionToSuccessor
that calls to @llvm.assume always terminate. Most other relevant
intrinsics should be covered by the "CS.onlyReadsMemory() ||
CS.onlyAccessesArgMemory()" bit but we were missing @llvm.assumes
because we state that it clobbers memory.
Added an LICM test case, but this change is not specific to LICM.
Dan Liew [Tue, 14 Jun 2016 19:15:13 +0000 (19:15 +0000)]
[LibFuzzer] Disable the ``fuzzer-trace-pc.test`` test on non-linux platforms.
On OSX this test sometimes fails due to the
``LLVMFuzzer-FullCoverageSetTest-TracePC`` program going over the
default 2GiB memory limit. This shouldn't be happening and needs
investigating. For now just disable the test so we can set up an
OSX buildbot.
Wei Mi [Tue, 14 Jun 2016 18:53:20 +0000 (18:53 +0000)]
[X86] Reduce the width of multiplification when its operands are extended from i8 or i16
For <N x i32> type mul, pmuludq will be used for targets without SSE41, which
often introduces many extra pack and unpack instructions in vectorized loop
body because pmuludq generates <N/2 x i64> type value. However when the operands
of <N x i32> mul are extended from smaller size values like i8 and i16, the type
of mul may be shrunk to use pmullw + pmulhw/pmulhuw instead of pmuludq, which
generates better code. For targets with SSE41, pmulld is supported so no
shrinking is needed.
projects / llvm / log