The order of evaluation of a sum (e.g., `a() + b()`) is unspecified in
C++. clang evaluates left-to-right. gcc evaluate right-to-left. This led
to slighly different (but equivalent) affine_map in a test and the
FileCheck did not match anymore.
Implementation of Pass and Dialect Plugins that mirrors LLVM Pass Plugin
implementation from the new pass manager.
Currently the implementation only supports using the pass-pipeline option
for adding passes. This restriction is imposed by the `PassPipelineCLParser`
variable in mlir/lib/Tools/mlir-opt/MlirOptMain.cpp:114 that loads the
parse options statically before parsing the cmd line args.
```
mlir-opt stanalone-plugin.mlir --load-dialect-plugin=lib/libStandalonePlugin.so --pass-pipeline="builtin.module(standalone-switch-bar-foo)"
```
Reviewed By: rriddle, mehdi_amini
Differential Revision: https://reviews.llvm.org/D147053
Implementation of Pass and Dialect Plugins that mirrors LLVM Pass Plugin implementation from the new pass manager.
Currently the implementation only supports using the pass-pipeline option for adding passes. This restriction is imposed by the `PassPipelineCLParser` variable in mlir/lib/Tools/mlir-opt/MlirOptMain.cpp:114 that loads the parse options statically before parsing the cmd line args.
```
mlir-opt stanalone-plugin.mlir --load-dialect-plugin=lib/libStandalonePlugin.so --pass-pipeline="builtin.module(standalone-switch-bar-foo)"
```
Reviewed By: rriddle
Differential Revision: https://reviews.llvm.org/D147053
set_inactive is actually a kind of operation that is passing certain
value from active threads to inactive threads. In later WWM operation,
the activated threads which were disabled before would read such
values passed to them by set_inactive operation. So I think the
set_inactive is a convergent operation.
Differential Revision: https://reviews.llvm.org/D147683
The set_inactive was originally inside a if branch, but was incorrectly
hoisted out of the loop. So the lanes that were disabled later after the
if() branch inside the loop were not set correctly. It should stay in
the if branch.
Differential Revision: https://reviews.llvm.org/D147682
initializer list.
This previously led to some weird behaviors where we would unwrap the
initializer list expression and then try to use the DesignatedInitExprs
as constructor arguments.
Under the C++20 language rules, it's not valid to initialize a
reference-to-aggregate from a designated initializer list, but we have
historically accepted that, as do other compilers, and we continue to
accept that with this change. I've asked WG21 whether this should be
considered a wording defect.
The result of sub + setcc is 0 or 1 for all bits.
The sra instruction get the same result.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D147538
With this patch, whenever we emit a `DW_AT_type` for some declaration
and the type is a template class with a `clang::PreferredNameAttr`, we
will emit the typedef that the attribute refers to instead. I.e.,
```
0x123 DW_TAG_variable
DW_AT_name "var"
DW_AT_type (0x123 "basic_string<char>")
0x124 DW_TAG_structure_type
DW_AT_name "basic_string<char>"
```
...becomes
```
0x123 DW_TAG_variable
DW_AT_name "var"
DW_AT_type (0x124 "std::string")
0x124 DW_TAG_structure_type
DW_AT_name "basic_string<char>"
0x125 DW_TAG_typedef
DW_AT_name "std::string"
DW_AT_type (0x124 "basic_string<char>")
```
We do this by returning the preferred name typedef `DIType` when
we create a structure definition. In some cases, e.g., with `-gmodules`,
we don't complete the structure definition immediately but do so later
via `completeClassData`, which overwrites the `TypeCache`. In such cases
we don't actually want to rewrite the cache with the preferred name. We
handle this by returning both the definition and the preferred typedef
from `CreateTypeDefinition` and let the callee decide what to do with
it.
Essentially we set up the types as:
```
TypeCache[Record] => DICompositeType
ReplaceMap[Record] => DIDerivedType(baseType: DICompositeType)
```
For now we keep this behind LLDB tuning.
**Testing**
- Added clang unit-test
- `check-llvm`, `check-clang` pass
- Confirmed that this change correctly repoints
`basic_string` references in some of my test programs.
- Will add follow-up LLDB API tests
Differential Revision: https://reviews.llvm.org/D145803
Introduce `strxfrm` and unit tests. The current implementation is
introduced without locale support.
The simplified function performs a `memcpy` if the `n` value is large
enough to store the source len + '\0', otherwise `dest` is unmodified.
Ticket: https://fxbug.dev/124217
Differential Revision: https://reviews.llvm.org/D147478
Make adjustExtracts/needToDelay lambdas members of ShuffleInstructionBuilder to allow to overload them later for cost model.
Differential Revision: https://reviews.llvm.org/D147730
The original loop has O(MxN) since `is_contained` iterates over
all incoming values. This change makes it so only the phis
which use the value as an incoming value are iterated over so
it is now O(M).
Differential Revision: https://reviews.llvm.org/D146999
This re-lands commit db987b9589be1eb604fcb74c85b410469e31485f,
previously reverted in commit 1e0709167f5edd330889f51bb203c458bdb5e359,
with a fix for a pre-existing bug in how we write constructor names into
lookup tables that was exposed by that change.
As yet, we've been unable to produce a reduced testcase for this bug,
but we understand the problem: if we write out a lookup table for a
class which has two constructor names, and we only have local lookup
results for one of those constructor names, then we will write out a
lookup table where the "all the constructors" lookup result contains
results for only one of the constructor names.
Add an LLDB_DEPRECATED macro similar to LLVM_DEPRECATED. We cannot
directly reuse the LLVM one, because it's defined in Compiler.h which is
not something we can include in the SB API.
For the SB API, the macro is undef'd if:
1. We're building the SWIG bindings. We might still want to expose
deprecated methods to our scripting users.
2. We're targeting anything older than C++14 as the [[deprecated]]
attribute was added for C++14 and the SB API can be used by
projects compiling against an older language version.
Differential revision: https://reviews.llvm.org/D147736
The name `module.modulemap` is convention.
> Clang will also search for a file named `module.map`. This behavior is deprecated and
> we plan to eventually remove it.
Conditionally setting MaskForGaps is only needed for loads. This avoid
re-computing MaskForGaps for stores.
Suggested as independent cleanup in D147467.
Without explicitly unregistering you will get
```
'cuMemHostRegister(ptr, sizeBytes, 0)' failed with 'CUDA_ERROR_HOST_MEMORY_ALREADY_REGISTERED'
```
in CUDA (for example) after repeated runs (e.g., during benchmarking the same kernel).
Reviewed By: ftynse
Differential Revision: https://reviews.llvm.org/D147277
The math arithmetic expansions do not support vectorized types.
Updated the lowerings so that they support vectorized types. This
includes a different implementation for `math.ctlz` to be a binary
search and not have variable termination time.
Reviewed By: jpienaar, NatashaKnk
Differential Revision: https://reviews.llvm.org/D147289
bf16 has a trivial truncation/extension behavior with F32 that
can be described in elementary arith operations. Include some
expansions to efficiently convert including rounding towards
infinity for f32 to bf16 truncation.
Reviewed By: jpienaar
Differential Revision: https://reviews.llvm.org/D147585
Note that technically both M68000/010 can use M68881, despite the fact
that usually only M68020 and newer ISAs are equipped with M68881/2.
M68040 and newer processors have builtin M68882.
Differential Revision: https://reviews.llvm.org/D147479
Currently, SBTarget::SetModuleLoadAddress does not accept large slides
needed to load images in high memory. This function should always have
taken an unsigned as the slide, as it immediately passes it to
Target::SetSectionLoadAddress which takes an unsigned.
This patch adds an overload and exposes that to SWIG instead of the
signed variant. I've marked the signed variant as deprecated and added
check that the slide is positive.
rdar://101355155
Differential revision: https://reviews.llvm.org/D147482
If the value is used in the expression, need to adjust the mask before
applying the mask. Plus, need to fix the analysis of the phi nodes for
reused scalars.
This also reverts 282cae0b9a602267ad7ef622f770066491332a11 as the
particular crash is now handled by the new code.
Before this change Clang would always leave declarations inside the
type-locs as `null` if the declarator had an invalid type. This patch
populates declarations even for invalid types if the structure of the
type and the type-locs match.
There are certain cases that may still cause crashes. These happen when
Clang recovers the type in a way that is not reflected in the
declarator's structure, e.g. adding a pointer when it was not present in
the code for ObjC interfaces or ignoring pointers written in the code
in C++ with auto return type (`auto* foo() -> int`). Those cases look
fixable with a better recovery strategy and I plan to follow up with
more patches to address those.
The first attempt caused 31 tests from `check-clang` to crash due to
different structure of the types and type-locs after certain errors. The
good news is that the failure is localized and mismatch in structures is
discovered by assertions inside `DeclaratorLocFiller`. Some notable
cases caught by existing tests:
- Invalid chunks when type is fully ignored and replace with int or now.
Crashed in `C/C2x/n2838.c`.
- Invalid return types in lambdas. Crashed in `CXX/drs/dr6xx.cpp`.
- Invalid member pointers. Crashed in `CXX/dcl.dcl/dcl.spec/dcl.type/dcl.spec.auto/p3-generic-lambda-1y.cpp`
- ObjC recovery that adds pointers. Crashed in `SemaObjC/blocks.m`
This change also updates the output of `Index/complete-blocks.m`.
Not entirely sure what causes the change, but the new function signature
is closer to the source code, so this seems like an improvement.
Reviewed By: aaron.ballman, erichkeane
Differential Revision: https://reviews.llvm.org/D146971
As of Xcode 14.3 it passes -reproducible by default to ld64. It seems
this flag was added in ld64 with Xcode 14.0, but it is not documented.
Through my testing the only thing I have seen it do is the same as
ZERO_AR_DATE, but it's possible it does more, or will do more in the
future. Since we already default to this option, this is more about
handling the command line flag to maintain Xcode compatibility than
anything else.
Differential Revision: https://reviews.llvm.org/D147663
Loads of fixed length vectors with irregular element counts are
sometimes emitted as a scalar load + scalar_to_vector.
Previously the scalar_to_vector wasn't legal and so was scalarized
further. This patch handles it by lowering it to a vmv.s.x.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D147608
I was investigating why getMemoryOpCost was so expensive for these fixed length vector types, so I've added these tests to see what was going on.
It looks like there's some room for codegen improvement before we adjust the cost model.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D147607
Due to an improvement to name preservation, a lot of InstCombine
tests now show spurious diffs when regenerated.
Rather than regenerating individual files when they get touched,
mass-regenerate all UTC-based InstCombine tests. I have then reset
a number of files showing suspicious diffs where the UTC output
has clearly been manually adjusted. I apologize if I missed
anything in the mass of changes.
Implementation of Pass and Dialect Plugins that mirrors LLVM Pass Plugin implementation from the new pass manager.
Currently the implementation only supports using the pass-pipeline option for adding passes. This restriction is imposed by the `PassPipelineCLParser` variable in mlir/lib/Tools/mlir-opt/MlirOptMain.cpp:114 that loads the parse options statically before parsing the cmd line args.
```
mlir-opt stanalone-plugin.mlir --load-dialect-plugin=lib/libStandalonePlugin.so --pass-pipeline="builtin.module(standalone-switch-bar-foo)"
```
Reviewed By: rriddle
Differential Revision: https://reviews.llvm.org/D147053
ORC's ELF platform support prefers the newer libunwind registration functions
(__unw_add_dynamic_eh_frame_section, __unw_remove_dynamic_eh_frame_section) when
they're available, and falls back to the older registration functions
(__register_frame, __deregister_frame) when they're not.
Until now the choice of registration functions has been made on the controller
side in ELFNixPlatform: The platform JITDylib was searched for the registration
functions and aliases set depending on which ones were found. This patch drops
that selection logic from ELFNixPlatform and instead uses weak imports of the
registration functions in elfnix_platform.cpp to identify which ones are
available and choose which ones to use.
This has a few small benefits:
(1) The registration functions don't need to be defined in the same JITDylib as
the ORC runtime -- it's sufficient for them to be defined in a JITDylib that the
ORC runtime's JITDylib links against.
(2) THe elfnix_platfrom code is more readable, as we don't have to dig into
ELFNixPlatform.cpp on the controller side to discover the definition of the
registration aliases.
(3) We may save a separate round-trip to look up the registration APIs (the
lookup will be folded into the ordinary external symbol lookup when linking the
runtime).
Currently the use of bare pointer calling convention is controlled
globally through use of an option in the `LLVMTypeConverter`. To allow
more fine-grained control use an attribute on a function to drive the
calling convention to use.
Reviewed By: ftynse
Differential Revision: https://reviews.llvm.org/D147494
Made the condition for the erasing of the gathered extractelements
stricter, remove it only if it has single vectorized use, otherwise
leave it for instcombiner/instsimplify analysis.
generateUnrolledLoop was assuming that the yielded value is always generated in
the Block corresponding to the loop being unrolled. Thus, it was updating the
last yielded values with it's cloned value. However, if the yielded value is not
generated in the same Block then the cloned value and it's corresponding mapping
won't exist, resulting in a crash. Fix this.
Reviewed By: bondhugula
Differential Revision: https://reviews.llvm.org/D146931
