Compress by factor 4x, takes about 10ms per 8 MiB block.
Depends on D114498.
Reviewed By: morehouse
Differential Revision: https://reviews.llvm.org/D114503
The first 8b of each raw profile section need to be aligned to 8b since
the first item in each section is a u64 count of the number of items in
the section.
Summary of changes:
* Assert alignment when reading counts.
* Update test to check alignment, relax some size checks to allow padding.
* Update raw binary inputs for llvm-profdata tests.
Differential Revision: https://reviews.llvm.org/D114826
Allow toggling of -fnew-infallible so last instance takes precedence
Testing:
ninja check-all
Reviewed By: bruno
Differential Revision: https://reviews.llvm.org/D113523
Add Compression::Test type which just pretends packing,
but does nothing useful. It's only called from test for now.
Depends on D114493.
Reviewed By: kstoimenov
Differential Revision: https://reviews.llvm.org/D114494
Due to the debug info merging, there may have some contexts with zero probe id, we should truncate the context to avoid misleading pre-inliner.
Reviewed By: hoy, wenlei
Differential Revision: https://reviews.llvm.org/D114284
Tools such as `llvm-objdump` or `llvm-readobj` support indirect symbol
tables. Here, support it for `obj2yaml` and `yaml2obj`.
Reviewed By: jhenderson, drodriguez
Differential Revision: https://reviews.llvm.org/D114410
In order to support generating profile with FS discriminator, three kind of changes are done in llvm-profgen:
1) Dissassemble .rodata section to check if FS discriminator var ('"__llvm_fs_discriminator__"') exists and set the corresponding flag in the binary.
2) Change the discriminator decoding in `getBaseDiscriminator` and `getDuplicationFactor`.
3) set true for `FunctionSamples::ProfileIsFS` to enable FS functionality in ProfileData.
Reviewed By: xur, hoy, wenlei
Differential Revision: https://reviews.llvm.org/D113296
When OpenMP is compiled as a part runtimes for multiple targets, openmp
is compiled under build/runtimes/runtimes-arch-unknown-linux-gnu-bins
directory. Old implementation treats this directory name as errors.
This patch adds a guard like "[Uu]known[^-]".
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D114346
A quick fix last week to the shared library build caused
the predicate IsCoarray(const Symbol &) to be moved from
Semantics to Evaluate. This patch completes that move in
a way that properly combines the existing IsCoarray() tests
for expressions and other object with the test for a symbol.
Differential Revision: https://reviews.llvm.org/D114806
This reverts commit 29a50c5864.
After LLVM lowering, the original patch incorrectly moved alignment
information across an unconstrained GEP operation. This is only correct
for some index offsets in the GEP. It seems that the best approach is,
in fact, to rely on LLVM to propagate information from the llvm.assume()
to users.
Thanks to Thomas Raoux for catching this.
We've found that when profiling, counts are only generated for the real definition of constructor aliases (C2 in mangled name). However, when compiling the C1 version is present at the callsite and leads to a lack of counts due to this aliasing. This causes us to miss out on inlining an otherwise hot constructor.
-mconstructor-aliases is AFAICT an optimization, so having a disabling flag if wanted seems valuable.
Testing:
ninja check-all
Reviewed By: wenlei
Differential Revision: https://reviews.llvm.org/D114130
A big-endian version of vpermxor, named vpermxor_be, is added to LLVM
and Clang. vpermxor_be can be called directly on both the little-endian
and the big-endian platforms.
Reviewed By: nemanjai
Differential Revision: https://reviews.llvm.org/D114540
We would like to use TLS to store the ThreadState object (or at least a
reference ot it), but on Darwin accessing TLS via __thread or manually
by using pthread_key_* is problematic, because there are several places
where interceptors are called when TLS is not accessible (early process
startup, thread cleanup, ...).
Previously, we used a "poor man's TLS" implementation, where we use the
shadow memory of the pointer returned by pthread_self() to store a
pointer to the ThreadState object.
The problem with that was that certain operations can populate shadow
bytes unbeknownst to TSan, and we later interpret these non-zero bytes
as the pointer to our ThreadState object and crash on when dereferencing
the pointer.
This patch changes the storage location of our reference to the
ThreadState object to "real" TLS. We make this work by artificially
keeping this reference alive in the pthread_key destructor by resetting
the key value with pthread_setspecific().
This change also fixes the issue were the ThreadState object is
re-allocated after DestroyThreadState() because intercepted functions
can still get called on the terminating thread after the
THREAD_TERMINATE event.
Radar-Id: rdar://problem/72010355
Reviewed By: dvyukov
Differential Revision: https://reviews.llvm.org/D110236
Create a new flang/runtime/support.cpp module to hold miscellaneous
runtime APIs to support lowering, and define an API IsContiguous() to
wrap the member function predicate Descriptor::IsContiguous().
And do a little clean-up of other API headers that don't need to expose
Runtime/descriptor.h.
Differential Revision: https://reviews.llvm.org/D114752
Remove assertion that disallows getting a zero-extended value from a
zero-width APInt. This check is too restrictive and makes it difficult
to use APInt to model zero-width things, e.g., zero-width wires in the
CIRCT project.
Signed-off-by: Schuyler Eldridge <schuyler.eldridge@sifive.com>
Reviewed By: lattner, darthscsi, nikic
Differential Revision: https://reviews.llvm.org/D114768
The below commit is causing assertion when insert element type is not integer
type such as half. This is because the transformation is creating zext before
doing bitwise OR, and the zext is supported only for integer types
80ab06c599
Reviewed By: spatel
Differential Revision: https://reviews.llvm.org/D114734
Symbol table parsing has evolved over the years and many plug-ins contained duplicate code in the ObjectFile::GetSymtab() that used to be pure virtual. With this change, the "Symbtab *ObjectFile::GetSymtab()" is no longer virtual and will end up calling a new "void ObjectFile::ParseSymtab(Symtab &symtab)" pure virtual function to actually do the parsing. This helps centralize the code for parsing the symbol table and allows the ObjectFile base class to do all of the common work, like taking the necessary locks and creating the symbol table object itself. Plug-ins now just need to parse when they are asked to parse as the ParseSymtab function will only get called once.
This is a retry of the original patch https://reviews.llvm.org/D113965 which was reverted. There was a deadlock in the Manual DWARF indexing code during symbol preloading where the module was asked on the main thread to preload its symbols, and this would in turn cause the DWARF manual indexing to use a thread pool to index all of the compile units, and if there were relocations on the debug information sections, these threads could ask the ObjectFile to load section contents, which could cause a call to ObjectFileELF::RelocateSection() which would ask for the symbol table from the module and it would deadlock. We can't lock the module in ObjectFile::GetSymtab(), so the solution I am using is to use a llvm::once_flag to create the symbol table object once and then lock the Symtab object. Since all APIs on the symbol table use this lock, this will prevent anyone from using the symbol table before it is parsed and finalized and will avoid the deadlock I mentioned. ObjectFileELF::GetSymtab() was never locking the module lock before and would put off creating the symbol table until somewhere inside ObjectFileELF::GetSymtab(). Now we create it one time inside of the ObjectFile::GetSymtab() and immediately lock it which should be safe enough. This avoids the deadlocks and still provides safety.
Differential Revision: https://reviews.llvm.org/D114288
INQUIRE(POSITION=)'s results need to reflect the POSITION=
specifier used for the OPEN statement until the unit has been
repositioned. Preserve the POSITION= from OPEN and used it
for INQUIRE(POSITION=) until is becomes obsolete.
INQUIRE(PAD=) is implemented here in the case of an unconnected unit
with Fortran 2018 semantics; i.e., "UNDEFINED", rather than Fortran 90's
"YES"/"NO" (see 4.3.6 para 2). Apparent failures with F'90-only tests
will persist with INQUIRE(PAD=); these discrepancies don't seem to warrant
an option or environment variable.
To make the implementation of INQUIRE more closely match the language
in the standard, rename IsOpen() to IsConnected(), and use it explicitly
for the various INQUIRE specifiers.
Differential Revision: https://reviews.llvm.org/D114755
Proper test for sparse tensor outputs is a single condition throughout
the whole tensor index expression (not a general conjunction, since this
may include other conditions that cause cancellation).
Reviewed By: bixia
Differential Revision: https://reviews.llvm.org/D114810
To accommodate triangular implied DO loops in DATA statements, in which
the bounds of nested implied DO loops might depend on the values of the
indices of outer implied DO loops in the same DATA statement set, it
is necessary to run them through constant folding each time they are
encountered.
Differential Revision: https://reviews.llvm.org/D114754
Test failed on x86 platforms due to a calling convention mismatch
when member function was called like a free function. In this patch,
member function is marked static to address this.
727bd89b60 broke the UBSan decorator. The decorator compiles a custom
source code snippet that exposes UB and verifies the presence of a UBSan
symbol in the generated binary. The aforementioned commit broke both by
compiling a snippet without UB and discarding the result.
Max(), MIN(), and their specific variants are defined with an unlimited
number of dummy arguments named A1=, A2=, &c. whose names are almost never
used in practice but should be allowed for and properly checked for the
usual errors when they do appear. The intrinsic table's entries otherwise
have fixed numbers of dummy argument definitions, so add some special
case handling in a few spots for MAX/MIN/&c. checking and procedure
characteristics construction.
Differential Revision: https://reviews.llvm.org/D114750
This revision reintroduces tensor.insert_slice verification which seems
to have vanished over time: a verifier was initially introduced in cf9503c1b7
but for some reason the invalid.mlir was not properly updated; as time passed the verifier was not called anymore and later the code was deleted.
As a consequence, a non-negligible portion of tests has run astray using invalid
tensor.insert_slice semantics and needed to be fixed.
Also, extract isRankReducedType from TensorOps for better reuse
Originally, this facility was used by both tensor and memref forms but
it got copied around as dialects were split.
Differential Revision: https://reviews.llvm.org/D114715
The canonical type of the result of the `memref.subview` needs to make
sure that the previously dropped unit-dimensions are the ones dropped
for the canonicalized type as well. This means the generic
`inferRankReducedResultType` cannot be used. Instead the current
dropped dimensions need to be querried and the same need to be dropped.
Reviewed By: nicolasvasilache, ThomasRaoux
Differential Revision: https://reviews.llvm.org/D114751
Previously, this caused GlobalISel to emit invalid IR (a gpr32 to gpr64
copy) and fail during verification.
While this shift is not defined (returns poison), it should not crash
codegen, as it may appear inside dead code (for example, a select
instruction), and it is legal IR input, as long as the value is unused.
Discovered while trying to build Julia with LLVM v13:
https://github.com/JuliaLang/julia/pull/42602.
Reviewed By: aemerson
Differential Revision: https://reviews.llvm.org/D114389
The memprof profile reader tests rely on binary data which is generated
from and meant to be interpreted on little endian architectures. Add a
REQUIRES: x86_64-linux clause to both tests to ensure they don't fail on big
endian targets such as ppc.
The memprof unit tests use an older version of gmock (included in the
repo) which does not build cleanly with -pedantic:
https://github.com/google/googletest/issues/2650
For now just silence the warning by disabling pedantic and add the
appropriate flags for gcc and clang.
For a 1x1 weight and stride of 1, the input/weight can be reshaped and passed into a fully connected op then reshaped back
Reviewed By: rsuderman
Differential Revision: https://reviews.llvm.org/D114757
It seems clearer to me that this would check for *any of* instead of
*all of* these option categories, as it looks to me like that was the
intent. But apparently this logic has always has been inverted, and
possibly never fully used?
Differential Revision: https://reviews.llvm.org/D114572
Up until now, all references to `errno` were marked with `NOLINT`, since
it was technically calling an external function. This fixes the lint
rules so that `errno`, as well as `malloc`, `calloc`, `realloc`, and
`free` are all allowed to be called as external functions. All of the
relevant `NOLINT` comments have been removed, and the documentation has
been updated.
Reviewed By: sivachandra, lntue, aaron.ballman
Differential Revision: https://reviews.llvm.org/D113946
This reverts the PPC64PCRelLongBranchThunk part from D86706.
PPC64PCRelLongBranchThunk is the same as PPC64R12SetupStub.
Use `__gep_setup_` instead of `__long_branch_pcrel_` for the stub symbol name
as it more closely indicates the operation.
(Note: GNU ld uses `*.long_branch.*` and `*.plt_branch.*`).
Reviewed By: NeHuang, nemanjai
Differential Revision: https://reviews.llvm.org/D114656
Vitaly Buka