Simon Pilgrim [Sat, 2 Jul 2016 12:13:29 +0000 (12:13 +0000)]
[X86][AVX512] Add fast-isel shuffle tests
Its not worth trying to write out tests for all the avx512f builtins yet, just adding tests for lowering of generic IR as we transition to it (shuffles mainly right now).
Summary:
Add renderscript32 and renderscript64 ArchTypes. This is to configure
the ABI requirement on 32-bit RenderScript that 'long' types have 64-bit
size and alignment. 64-bit RenderScript is the same as AArch64, but is
added here for completeness.
Matt Arsenault [Fri, 1 Jul 2016 22:55:55 +0000 (22:55 +0000)]
AMDGPU: Expand unaligned accesses early
Due to visit order problems, in the case of an unaligned copy
the legalized DAG fails to eliminate extra instructions introduced
by the expansion of both unaligned parts.
[msan] Fix __msan_maybe_ for non-standard type sizes.
Fix incorrect calculation of the type size for __msan_maybe_warning_N
call that resulted in an invalid (narrowing) zext instruction and
"Assertion `castIsValid(op, S, Ty) && "Invalid cast!"' failed."
Only happens in very large functions (with more than 3500 MSan
checks) operating on integer types that are not power-of-two.
[codeview] Don't record UDTs for anonymous structs
MSVC makes up names for these anonymous structs, but we don't (yet).
Eventually Clang should use getTypedefNameForAnonDecl() to put some name
in the debug info, and we can update the test case when that happens.
Address two correctness issues in LoadStoreVectorizer
Summary:
GetBoundryInstruction returns the last instruction as the instruction which follows or end(). Otherwise the last instruction in the boundry set is not being tested by isVectorizable().
Partially solve reordering of instructions. More extensive solution to follow.
[codeview] Assert that our CV type records are valid
We were asserting that our type records were valid when emitting
assembly, but not when emitting an object file.
I've been seeing lots of LNK1285 errors (corrupt PDB) during incremental
debug self-host builds with the MSVC linker, and hopefully this will
catch some of them earlier.
Matt Arsenault [Fri, 1 Jul 2016 18:03:46 +0000 (18:03 +0000)]
AMDGPU/SI: Enable testing several variants for si scheduler
Enable testing different scheduling variants if sgpr usage
is very high. It was previously disabled because of a bug
in handleMove, but it has been fixed since.
John Brawn [Fri, 1 Jul 2016 17:05:58 +0000 (17:05 +0000)]
Make extract_symbols.py be compatible with Python 3
This involved running 2to3 on it and adjusting all uses of subprocess to use
universal_newlines=True so the output is text instead of binary. It remains
compatible with Python 2.7.
CodeGen: Use MachineInstr& in ScheduleDAGIntrs, NFC
Use MachineInstr& to avoid implicit conversions from
MachineBasicBlock::iterator to MachineInstr*. In one case, this could
use a range-based for loop, but the other loops iterated in reverse
order.
One of the reverse-loops checked the MachineInstr* for nullptr, a
condition that is provably unreachable. (And even if my proof has a
flaw, UBSan would catch the bug.)
Use MachineInstr& instead of MachineInstr* in RegAllocFast to avoid
implicit conversions from MachineInstrBundleIterator. RAFast::spillAll
and RAFast::spillVirtReg still take iterators, since their argument may
be an end iterator from MachineBasicBlock::getFirstTerminator.
John Brawn [Fri, 1 Jul 2016 14:22:52 +0000 (14:22 +0000)]
[CMake] Add LLVM_BUILD_32_BITS to LLVMConfig.cmake
Previously out-of-tree passes could detect if LLVM was built with
LLVM_BUILD_32_BITS by looking for -m32 in LLVM_DEFINITIONS, but as of r271871
it no longer appears there. Resolve this by instead emitting LLVM_BUILD_32_BITS
in LLVMConfig so it can be checked for directly.
Nikolay Haustov [Fri, 1 Jul 2016 10:00:58 +0000 (10:00 +0000)]
Resubmit r268719 - AMDGPU/SI: Add amdgpu_kernel calling convention. Part 2.
This was reverted in r268740 because of problems with corresponding Clang change.
Clang change was updated and resubmitted in r274220.
Check calling convention in AMDGPUMachineFunction::isKernel
This will be used for AMDGPU_HSA_KERNEL symbol type in output ELF.
Also, in the future unused non-kernels may be optimized.
[CodeGen] Cleanup getVectorShuffle a bit to take advantage of its new ArrayRef argument and its begin/end iterators. Also use 'int' type for number of elements and loop iterators to remove several typecasts. No functional change intended.
[CodeGen,Target] Remove the version of DAG.getVectorShuffle that takes a pointer to a mask array. Convert all callers to use the ArrayRef version. No functional change intended.
For the most part this simplifies all callers. There were two places in X86 that needed an explicit makeArrayRef to shorten a statically sized array.
Add support for allowing us to create uniquely identified "COMDAT" or "ELF
Group" sections while lowering. In particular, for ELF sections this is
useful for creating function-specific groups that get merged into the
same named section.
Also use const Twine& instead of StringRef for the getELF functions
while we're here.
Summary:
This represents the adjustment applied to the implicit 'this' parameter
in the prologue of a virtual method in the MS C++ ABI. The adjustment is
always zero unless multiple inheritance is involved.
This increases the size of DISubprogram by 8 bytes, unfortunately. The
adjustment really is a signed 32-bit integer. If this size increase is
too much, we could probably win it back by splitting out a subclass with
info specific to virtual methods (virtuality, vindex, thisadjustment,
containingType).
This reverts commit r274305, since it breaks self-hosting:
http://lab.llvm.org:8080/green/job/clang-stage1-configure-RA_build/22349/
http://lab.llvm.org:8011/builders/clang-x86_64-linux-selfhost-modules/builds/17232
Note that the blamelist on lab.llvm.org:8011 is incorrect. The previous
build was r274299, but somehow r274305 wasn't included in the blamelist:
http://lab.llvm.org:8011/builders/clang-x86_64-linux-selfhost-modules
CodeGen: Use MachineInstr& in LiveVariables API, NFC
Change all the methods in LiveVariables that expect non-null
MachineInstr* to take MachineInstr& and update the call sites. This
clarifies the API, and designs away a class of iterator to pointer
implicit conversions.
CodeGen: Use MachineInstr& in PrologEpilogInserter, NFC
Use MachineInstr& over MachineInstr* to avoid implicit iterator to
pointer conversions. MachineInstr*-as-nullptr was being used as a flag
for whether the for loop terminated normally; I added an explicit `bool`
instead.
Sebastian Pop [Fri, 1 Jul 2016 00:24:31 +0000 (00:24 +0000)]
code hoisting pass based on GVN
This pass hoists duplicated computations in the program. The primary goal of
gvn-hoist is to reduce the size of functions before inline heuristics to reduce
the total cost of function inlining.
Pass written by Sebastian Pop, Aditya Kumar, Xiaoyu Hu, and Brian Rzycki.
Important algorithmic contributions by Daniel Berlin under the form of reviews.
Target: Remove unused arguments from overrideSchedPolicy, NFC
TargetSubtargetInfo::overrideSchedPolicy takes two MachineInstr*
arguments (begin and end) that invite implicit conversions from
MachineInstrBundleIterator. One option would be to change their type to
an iterator, but since they don't seem to have been used since the API
was added in 2010, I'm deleting the dead code.
Switch to a range-based for in IfConverter::PredicateBlock and take
MachineInstr& in MaySpeculate to avoid an implicit conversion from
MachineBasicBlock::iterator to MachineInstr*.
This is a mechanical change to make TargetLowering API take MachineInstr&
(instead of MachineInstr*), since the argument is expected to be a valid
MachineInstr. In one case, changed a parameter from MachineInstr* to
MachineBasicBlock::iterator, since it was used as an insertion point.
As a side effect, this removes a bunch of MachineInstr* to
MachineBasicBlock::iterator implicit conversions, a necessary step
toward fixing PR26753.
Mike Aizatsky [Thu, 30 Jun 2016 20:43:06 +0000 (20:43 +0000)]
[libFuzzer] Let user specify extra stats file.
Summary: If AFL_DRIVER_EXTRA_STATS_FILENAME is set and valid, write to it peak_rss_mb and slowest_unit_time_sec. These are both stats that libFuzzer can print but afl cannot.
Vassil Vassilev [Thu, 30 Jun 2016 20:35:56 +0000 (20:35 +0000)]
[CMake] Module builds depend on target intrinsics_gen to be built first.
When compiling with modules, header A and B can be in the same module M.
B depends on intrinsics_gen and A doesn't. Compiling a source file #include-ing
header A, we implicitly request module M to be built. It puts header A and B in
the same TU and tries to build them. Since B depends on intrinsics_gen (which
might not be built yet) we run into build failures.
Chandler Carruth [Thu, 30 Jun 2016 20:27:54 +0000 (20:27 +0000)]
Introduce a *draft* of a code of conduct for the LLVM community and the
associated reporting guide.
I want to emphasize that at this point these are just drafts!
This is the result of very extended discussion on the mailing lists on
several different threads:
http://lists.llvm.org/pipermail/llvm-dev/2015-October/091218.html
http://lists.llvm.org/pipermail/llvm-dev/2016-May/099120.html
http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20151019/307070.html
The reporting guide in particular I anticipate will be shaped somewhat
by the advisory committee when they are selected. But hopefully this
serves as a good starting point and good guidance while the advisory
committee is being sorted out.
I'd like to thank all the folks who contributed to this. Many, *many*
people worked to help with drafting, wording, suggestions, and edits.
Also, this is based on widely used existing codes of coduct as mentioned
in the text, and the original authors of those deserve many thanks as
well.
Wei Mi [Thu, 30 Jun 2016 18:42:56 +0000 (18:42 +0000)]
Refine the set of UniformAfterVectorization instructions.
Except the seed uniform instructions (conditional branch and consecutive ptr
instructions), dependencies to be added into uniform set should only be used
by existing uniform instructions or intructions outside of current loop.
projects / llvm / log