This change updates the clang front end to add symbols to llvm.used
when they have explicit export_name attribute.
Differential Revision: https://reviews.llvm.org/D71493
Summary:
In https://reviews.llvm.org/D62550 @rsmith pointed out that there are
many situations in which a coroutine body statement may be
transformed/rebuilt as part of a template instantiation, and my naive
check whether the coroutine was a generic lambda was insufficient.
This is indeed true, as I've learned by reading more of the
TreeTransform code. Most transformations are written in a way that
doesn't assume the resulting types are not dependent types. So the
assertion in 'TransformCoroutineBodyStmt', that the promise type must no
longer be dependent, is out of place.
This patch removes the assertion, spruces up some code comments, and
adds a test that would have failed with my naive check from
https://reviews.llvm.org/D62550.
Reviewers: GorNishanov, rsmith, lewissbaker
Reviewed By: rsmith
Subscribers: junparser, EricWF, rsmith, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D70579
If you set LLDB_TABLEGEN_EXE in a CMake cache file or in the CMake
invocation line, your setting isn't respected. Setting up the tablegen
for the host will overwrite the value that we set LLDB_TABLEGEN_EXE to,
which defeats the whole point of setting it in the first place.
This change introduces the __default_init_tag to memory, and a corresponding
element constructor to allow for default initialization of either of the pair
values. This is useful for classes such as std::string where most (all)
constructors explicitly initialize the values in the constructor.
Patch by Martijn Vels (mvels@google.com)
Reviewed as https://reviews.llvm.org/D70617
This reverts commit b1e542f302.
The original 'hack' didn't chop out fp-16 to double conversions, so
systems that use FP16ConversionIntrinsics end up in IR-CodeGen with an
i16 type isntead of a float type (like PPC64-BE). The bots noticed
this.
Reverting until I figure out how to fix this
Summary:
For users with 2FA Github password challenge can be rather cumbersome
and I personally use key based authentication through SSH. This fixes
the SSH feature that was present and exposes it as a command line
option.
Reviewers: tstellar
Subscribers: llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D71534
This skips calling `pthread_self` when `main_thread_identity` hasn't
been initialized yet. `main_thread_identity` is only ever assigned in
`__tsan::InitializePlatform`. This change should be relatively safe; we
are not changing behavior other than skipping the call to `pthread_self`
when `main_thread_identity == 0`.
rdar://57822138
Reviewed By: kubamracek
Differential Revision: https://reviews.llvm.org/D71559
dependent template names.
Apparently we didn't test this in the test suite because we have
a lot of redundant ways of representing this situation that kick
in in the more common situations. For example, DependentTST stores
a qualifier + identifier pair rather than a TemplateName.
The FP-classification builtins (__builtin_isfinite, etc) use variadic
packs in the definition file to mean an overload set. Because of that,
floats were converted to doubles, which is incorrect. There WAS a patch
to remove the cast after the fact.
THis patch switches these builtins to just be custom type checking,
calls the implicit conversions for the integer members, and makes sure
the correct L->R casts are put into place, then does type checking like
normal.
A future direction (that wouldn't be NFC) would consider making
conversions for the floating point parameter legal.
Summary:
This patch restricts loop fusion to only consider rotated loops as valid candidates.
This simplifies the analysis and transformation and aligns with other loop optimizations.
Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel
Reviewed By: Meinersbur
Subscribers: ormris, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D71025
Summary:
The instructions were originally implemented via builtins and
intrinsics so users would have to explicitly opt-in to using
them. This was useful while were validating whether these instructions
should have been merged into the spec proposal. Now that they have
been, we can use normal codegen patterns, so the intrinsics and
builtins are no longer useful.
Reviewers: aheejin
Subscribers: dschuff, sbc100, jgravelle-google, hiraditya, sunfish, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D71500
The %update_cc_test_checks substitution only gets added if python3
is on path, so the test fails if it isn't. Don't run the test
when it would fail.
Also include the '%' in the arg to add_update_script_substition(),
to help greppability.
This change only affects EM_386. relOff can be computed from `index`
easily, so it is unnecessarily passed as a parameter.
Both in.plt and in.iplt entries are written by writePLT. For in.iplt,
the instruction `push reloc_offset` will change because `index` is now
different. Fortunately, this does not matter because `push; jmp` is only
used by PLT. IPLT does not need the code sequence.
Reviewed By: grimar, ruiu
Differential Revision: https://reviews.llvm.org/D71518
According to OpenMP 5.0, if clause can be used in for simd directive. If
condition in the if clause if false, the non-vectorized version of the
loop must be executed.
Now that the machine verifier will check for cases of register/immediate
MachineOperands and their correspondence to the MC instruction descriptor,
this patch adds the operand types to the descriptors where they were
previously missing. All MCOI::OPERAND_UNKNOWN operand types have been handled
to get a known type, except for G_... (global isel) instructions.
Review: Ulrich Weigand
https://reviews.llvm.org/D71494
Summary:
llvm-cxxfilt wasn't correctly demangle COFF import thunk in those two
cases before:
* demangle in split mode (multiple words from commandline)
* the import thunk prefix was added no matter the later part of the
string can be demangled or not
Now llvm-cxxfilt should handle both case correctly.
Reviewers: compnerd, erik.pilkington, jhenderson
Reviewed By: jhenderson
Subscribers: jkorous, dexonsmith, ributzka, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D71425
Summary:
Follow-on to D66428 and D71193, to build the TLI per-function so
that -fno-builtin* handling can be migrated to use function attributes.
See discussion on D61634 for background. This is an enabler for fixing
handling of these options for LTO, for example.
With D71193, the -fno-builtin* flags are converted to function
attributes, so we can now set this information per-function on the TLI.
In this patch, the TLI constructor is changed to take a Function, which
can be used to override the available builtins. The TLI is augmented
with an array that can be used to specify which builtins are not
available for the corresponding function. The available function checks
are changed to consult this override before checking the underlying
module level baseline TLII. New code is added to set this override
array based on the attributes.
I also removed the code that sets availability in the TLII in clang from
the options, which is no longer needed.
I removed a per-Triple caching of TLII objects in the analysis object,
as it is based on the Module's Triple which is the same for all
functions in any case. Is there a case where we would be compiling
multiple Modules with different Triples in one compilation?
Finally, I have changed the legacy analysis wrapper to create and use
the new PM analysis class (TargetLibraryAnalysis) in getTLI. This is
consistent with the behavior of getTTI for the legacy
TargetTransformInfo analysis. This change means that getTLI now creates
a new TLI on each call (although that should be very cheap as we cache
the module level TLII, and computing the per-function
attribute based availability should also be reasonably efficient).
I measured the compile time for a large C++ file with tens of thousands
of functions and as expected there was no increase.
Reviewers: chandlerc, hfinkel, gchatelet
Subscribers: mehdi_amini, dexonsmith, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D67923
Summary:
Add unwinding support for 64-bit RISC-V.
This is from the FreeBSD implementation with the following minor
changes:
- Renamed and renumbered DWARF registers to match the RISC-V ABI [1]
- Use the ABI mneumonics in getRegisterName() instead of the exact
register names
- Include checks for __riscv_xlen == 64 to facilitate adding the 32-bit
ABI in the future.
[1] https://github.com/riscv/riscv-elf-psabi-doc/blob/master/riscv-elf.md
Patch by Mitchell Horne (mhorne)
Reviewers: lenary, luismarques, compnerd, phosek
Reviewed By: lenary, luismarques
Subscribers: arichardson, sameer.abuasal, abidh, asb, aprantl, krytarowski, simoncook, kito-cheng, christof, shiva0217, rogfer01, rkruppe, PkmX, psnobl, benna, lenary, s.egerton, luismarques, emaste, cfe-commits
Differential Revision: https://reviews.llvm.org/D68362
If a function requires optnone to trigger a crash, it must also have noline,
otherwise it will fail a verifier check.
Differential revision: https://reviews.llvm.org/D69522
Summary:
This reverts commit dd8a7fcdd7.
Alexey reports undefined symbols for the new inline functions defined in target_impl.h
This does not reproduce for me for nvptx, or amdgcn, under release or debug builds.
I believe the patch is fine, based on:
- the semantics of an inline function in C++ (the cuda INLINE functions end
up as linkonce_odr in IR), which are only legal to drop if they have no uses
- the code generated from a debug build of clang 9 does not show these undef symbols
- the tests pass
- the code is trivial
To progress from here I either need:
- A tie break - someone to play the role of CI in determining whether the patch works
- Alexey to provide sufficient information about his build for me to reproduce the failure
- Alexey to debug why the symbols are disappearing for him and report back
Reviewers: ABataev, jdoerfert, grokos
Subscribers: jvesely, openmp-commits
Tags: #openmp
Differential Revision: https://reviews.llvm.org/D71502
According to OpenMP 5.0, if clause can be used in for simd directive. If
condition in the if clause if false, the non-vectorized version of the
loop must be executed.
As brought up in D71467, a group of floating point builtins
automatically promoted floats to doubles because they used the variadic
builtin tag to support an overload set. The result is that the
parameters were treated as a variadic pack, which always promots
float->double.
This resulted in the wrong answer being given in cases with certain
values of NaN.
Handle -march=native in systemz::getSystemZTargetCPU, similar to
how this is done on other platforms. Also change the return type
to std::string instead of const char *.
Fangrui Song