In failure cases it's not guaranteed that the PHI we're inspecting is actually in the successor block! In this case we need to bail out early, and never query getIncomingValueForBlock() as that will cause an assert.
llvm-svn: 280794
After the reformat, the unittests do not compile due to missing due to redefinition errors
between PosixApi.h and ucrt/direct.h. This is a bit of a shot in the dark, as I have not tested
it on windows, but I am restoring the original include order, so it should hopefully fix it.
llvm-svn: 280793
I should have realised this the first time around, but if we're avoiding sinking stores where the operands come from allocas so they don't create selects, we also have to do the same for loads because SROA will be just as defective looking at loads of selected addresses as stores.
Fixes PR30188 (again).
llvm-svn: 280792
In the top-level CMakeLists.txt, we set CMAKE_BUILD_WITH_INSTALL_RPATH to ON,
and then for the unit tests we set it to <test>/../../lib. This works for tests
that live in unittest/<whatever>, but not for those that live in subdirectories
e.g. unittest/Transforms/IPO or unittest/ExecutionEngine/Orc. When building
with BUILD_SHARED_LIBRARIES, such tests don't manage to find their libraries.
Since the tests are run from the build directory, it makes sense to set their
RPATH for the build tree, rather than the install tree. This is the default in
CMake since 2.6, so all we have to do is set CMAKE_BUILD_WITH_INSTALL_RPATH to
OFF for the unit tests.
llvm-svn: 280791
PR30292 showed a case where our PHI checking wasn't correct. We were checking that all values were used by the same PHI before deciding to sink, but we weren't checking that the incoming values for that PHI were what we expected. As a result, we had to bail out after block splitting which caused us to never reach a steady state in SimplifyCFG.
Fixes PR30292.
llvm-svn: 280790
When folding an addi into a memory access that can take an immediate offset, we
were implicitly assuming that the existing offset was zero. This was incorrect.
If we're dealing with an addi with a plain constant, we can add it to the
existing offset (assuming that doesn't overflow the immediate, etc.), but if we
have anything else (i.e. something that will become a relocation expression),
we'll go back to requiring the existing immediate offset to be zero (because we
don't know what the requirements on that relocation expression might be - e.g.
maybe it is paired with some addis in some relevant way).
On the other hand, when dealing with a plain addi with a regular constant
immediate, the alignment restrictions (from the TOC base pointer, etc.) are
irrelevant.
I've added the test case from PR30280, which demonstrated the bug, but also
demonstrates a missed optimization opportunity (i.e. we don't need the memory
accesses at all).
Fixes PR30280.
llvm-svn: 280789
OpenCL requires __ENDIAN_LITTLE__ be set for little endian targets.
The default for targets was also apparently big endian, so AMDGPU
was incorrectly reported as big endian. Set this from the triple
so targets don't have another place to set the endianness.
llvm-svn: 280787
The previous commit (r280368 - https://reviews.llvm.org/D23313) does not cover AVX-512F, KNL set.
FNEG(x) operation is lowered to (bitcast (vpxor (bitcast x), (bitcast constfp(0x80000000))).
It happens because FP XOR is not supported for 512-bit data types on KNL and we use integer XOR instead.
I added pattern match for integer XOR.
Differential Revision: https://reviews.llvm.org/D24221
llvm-svn: 280785
copy-initialization. We previously got this wrong in a couple of ways:
- we only looked for copy / move constructors and constructor templates for
this copy, and thus would fail to copy in cases where doing so should use
some other constructor (but see core issue 670),
- we mishandled the special case for disabling user-defined conversions that
blocks infinite recursion through repeated application of a copy constructor
(applying it in slightly too many cases) -- though as far as I can tell,
this does not ever actually affect the result of overload resolution, and
- we misapplied the special-case rules for constructors taking a parameter
whose type is a (reference to) the same class type by incorrectly assuming
that only happens for copy/move constructors (it also happens for
constructors instantiated from templates and those inherited from base
classes).
These changes should only affect strange corner cases (for instance, where the
copy constructor exists but has a non-const-qualified parameter type), so for
the most part it only causes us to produce more 'candidate' notes, but see the
test changes for other cases whose behavior is affected.
llvm-svn: 280776
This patch fixes PR30260 by using a (void*) cast on the placement argument
to placement new to casts away the const. See also http://llvm.org/PR30260.
As a drive by change this patch also changes the header guard for
<experimental/optional> to _LIBCPP_EXPERIMENTAL_OPTIONAL from _LIBCPP_OPTIONAL.
llvm-svn: 280775
When libc++experimental was originally created it was empty and therefore there
was no reason to install it. Now that the library contains
<experimental/memory_resource> and <experimental/filesystem> there is a good
reason to install it.
Specifically this patch enables the installation whenever LIBCXX_INSTALL_LIBRARY
is true and LIBCPP_ENABLE_EXPERIMENTAL_LIBRARY is true.
llvm-svn: 280773
This was originally submitted in r280549, and reverted in r280577
due to breaking one MSVC buildbot. The issue is that MSVC 2013
doesn't synthesize move constructors. So even though i was
writing std::move(A) it was copying it, leading to a bogus ArrayRef.
The solution here is to simply remove the std::vector<> from the
type, since it is unused and unnecessary. This way the ArrayRef
continues to point into the original memory backing the CVType.
llvm-svn: 280769
This commit improves compatibility with the perl version of scan-build.
The perl version of scan-build produces output report directories with
increasing lexicographic ordering. This ordering is relied on by the CmpRuns.py
tool in utils/analyzer when comparing results for build commands with multiple
steps. That tool tries to line up the output directory for each step between
different runs of the analyzer based on the increasing directory name.
The python version of scan-build uses file.mkdtemp() with a time stamp
prefix to create report directories. The timestamp has a 1-second precision.
This means that when analysis of a single build step takes less than a second
the ordering property that CmpRuns.py expects will sometimes not hold,
depending on the timing and the random suffix generated by mkdtemp(). Ultimately
this causes CmpRuns to incorrectly correlate results from build steps and report
spurious differences between runs.
This commit increases the precision of the timestamp used in scan-build-py to
the microsecond level. This approach still has the same underlying issue -- but
in practice analysis of any build step is unlikely to take less than a
millisecond.
Differential Revision: https://reviews.llvm.org/D24163
llvm-svn: 280768
I might have called this "r246507, the sequel". It fixes the same issue, as the
issue has cropped up in a few more places. The underlying problem is that
isSetCCEquivalent can pick up select_cc nodes with a result type that is not
legal for a setcc node to have, and if we use that type to create new setcc
nodes, nothing fixes that (and so we've violated the contract that the
infrastructure has with the backend regarding setcc node types).
Fixes PR30276.
For convenience, here's the commit message from r246507, which explains the
problem is greater detail:
[DAGCombine] Fixup SETCC legality checking
SETCC is one of those special node types for which operation actions (legality,
etc.) is keyed off of an operand type, not the node's value type. This makes
sense because the value type of a legal SETCC node is determined by its
operands' value type (via the TLI function getSetCCResultType). When the
SDAGBuilder creates SETCC nodes, it either creates them with an MVT::i1 value
type, or directly with the value type provided by TLI.getSetCCResultType.
The first problem being fixed here is that DAGCombine had several places
querying TLI.isOperationLegal on SETCC, but providing the return of
getSetCCResultType, instead of the operand type directly. This does not mean
what the author thought, and "luckily", most in-tree targets have SETCC with
Custom lowering, instead of marking them Legal, so these checks return false
anyway.
The second problem being fixed here is that two of the DAGCombines could create
SETCC nodes with arbitrary (integer) value types; specifically, those that
would simplify:
(setcc a, b, op1) and|or (setcc a, b, op2) -> setcc a, b, op3
(which is possible for some combinations of (op1, op2))
If the operands of the and|or node are actual setcc nodes, then this is not an
issue (because the and|or must share the same type), but, the relevant code in
DAGCombiner::visitANDLike and DAGCombiner::visitORLike actually calls
DAGCombiner::isSetCCEquivalent on each operand, and that function will
recognise setcc-like select_cc nodes with other return types. And, thus, when
creating new SETCC nodes, we need to be careful to respect the value-type
constraint. This is even true before type legalization, because it is quite
possible for the SELECT_CC node to have a legal type that does not happen to
match the corresponding TLI.getSetCCResultType type.
To be explicit, there is nothing that later fixes the value types of SETCC
nodes (if the type is legal, but does not happen to match
TLI.getSetCCResultType). Creating SETCCs with an MVT::i1 value type seems to
work only because, either MVT::i1 is not legal, or it is what
TLI.getSetCCResultType returns if it is legal. Fixing that is a larger change,
however. For the time being, restrict the relevant transformations to produce
only SETCC nodes with a value type matching TLI.getSetCCResultType (or MVT::i1
prior to type legalization).
Fixes PR24636.
llvm-svn: 280767
Use the same color for counts and percentages. There doesn't seem to be
a reason for them to be different, and the summary looks more consistent
this way.
llvm-svn: 280765
This wasn't actually a problem with the reformat, but rather a
problem with Visual Studio 2015 Update 3, which uses some c++14
features in its standard libraries. So we had to change -std=c++11
to -std=c++14.
llvm-svn: 280759
Most of these issues arose as a result of header re-ordering, but
it turned up a real bug, which is that MSVC doesn't support
__attribute__((packed)) or __attribute__((aligned)). This was
working before because there's a Windows header that #defines
__attribute__(x) to nothing. We should fix this by removing
that #define entirely, and dealing with the fallout separately
which may turn up even more bugs.
I fixed this by replacing them with the corresponding LLVM
macros which understand how to do these operations on all the
different compilers.
llvm-svn: 280757
This patch removes the `<cstdlib>` include from exception where it is no longer
needed. Unlike my previous attempt this patch also adds <cstdlib> where needed
in other headers like <new> and <typeinfo>.
This won't fix the Firefox build issues discussed on IRC but it is more correct
for libc++.
llvm-svn: 280754
Apparently I missed a number of additional include which need to be added.
Reverting so I can recommit as a single patch with all of the required includes.
llvm-svn: 280752
*** to conform to clang-format’s LLVM style. This kind of mass change has
*** two obvious implications:
Firstly, merging this particular commit into a downstream fork may be a huge
effort. Alternatively, it may be worth merging all changes up to this commit,
performing the same reformatting operation locally, and then discarding the
merge for this particular commit. The commands used to accomplish this
reformatting were as follows (with current working directory as the root of
the repository):
find . \( -iname "*.c" -or -iname "*.cpp" -or -iname "*.h" -or -iname "*.mm" \) -exec clang-format -i {} +
find . -iname "*.py" -exec autopep8 --in-place --aggressive --aggressive {} + ;
The version of clang-format used was 3.9.0, and autopep8 was 1.2.4.
Secondly, “blame” style tools will generally point to this commit instead of
a meaningful prior commit. There are alternatives available that will attempt
to look through this change and find the appropriate prior commit. YMMV.
llvm-svn: 280751
This cleanup removes the need for the native support library to have its own target. That target was only needed because makefile builds were tripping over each other if two tablegen targets were building at the same time. This causes problems because the parallel make invocations through CMake can't communicate with each other. This is fixed by invoking make directly instead of through CMake which is how we handle this in External Project invocations.
The other part of the cleanup is to mark the custom commands as USES_TERMINAL. This is a bit of a hack, but we need to ensure that Ninja generators don't invoke multiple tablegen targets in the same build dir in parallel, because that too would be bad.
Marking as USES_TERMINAL does have some downside for Ninja because it results in decreased parallelism, but correct builds are worth the minor loss and LLVM_OPTIMZIED_TABLEGEN is such a huge win, it is worth it.
llvm-svn: 280748
- Implemented amdgpu-flat-work-group-size attribute
- Implemented amdgpu-num-active-waves-per-eu attribute
- Implemented amdgpu-num-sgpr attribute
- Implemented amdgpu-num-vgpr attribute
- Dynamic LDS constraints are in a separate patch
Patch by Tom Stellard and Konstantin Zhuravlyov
Differential Revision: https://reviews.llvm.org/D21562
llvm-svn: 280747
Summary:
I put this code here, because I want to re-use it in a few other places.
This supersedes some of the immediate folding code we have in SIFoldOperands.
I think the peephole optimizers is probably a better place for folding
immediates into copies, since it does some register coalescing in the same time.
This will also make it easier to transition SIFoldOperands into a smarter pass,
where it looks at all uses of instruction at once to determine the optimal way to
fold operands. Right now, the pass just considers one operand at a time.
Reviewers: arsenm
Subscribers: wdng, nhaehnle, arsenm, llvm-commits, kzhuravl
Differential Revision: https://reviews.llvm.org/D23402
llvm-svn: 280744
James Molloy