This reverts commit 9211e67da36782db44a46ccb9ac06734ccf2570f.
Summary:
This seemed to crash one one of the CUDA math tests. Revert until it can
be fixed.
Summary:
Currently, the linker wrapper sorts input files into different link
jobs according to their architectures. Here we assume each architecture
is a unique and incompatible link job unless they are specifically
marked compatible. This patch simply adds an `all` target to represent
an architecture that should be linked against every single other
architecture.
This will be useful for modelling generic IR such as the ROCm device
libraries or the NVPTX libdevice.
Summary:
The `__nvvm_reflect` function is used to guard invalid code that varies
between architectures. One problem with this feature is that if it is
used without optimizations, it will leave invalid code in the module
that will then make it to the backend. The `__nvvm_reflect` pass is
already mandatory, so it should do some trivial branch removal to ensure
that constants are handled correctly. This dead branch elimination only
works in the trivial case of a compare on a branch and does not touch
any conditionals that were not realted to the `__nvvm_reflect` call in
order to preserve `O0` semantics as much as possible. This should allow
the following to work on NVPTX targets
```c
int foo() {
if (__nvvm_reflect("__CUDA_ARCH") >= 700)
asm("valid;\n");
}
```
The current implementation of aliases tries to remove all the aliases in
the module to prevent the generic version of `AsmPrinter` from emitting
them incorrectly. Unfortunately, if the aliases are used this will fail.
Instead let's override the function to print aliases directly.
In addition, the declarations of the alias functions must occur before
the uses. To fix this we emit alias declarations as part of
`emitDeclarations` and only emit the `.alias` directives at the end
(where we can assume the aliasee has also already been declared).
The custom printer for `fir.global` was eluding all the attributes
present on the op when printing the attribute dictionary. So any
attribute that is not part of the pretty printing was therefore
discarded.
This patch fix the printer and also make use of the getters for the
attribute names when they are hardcoded.
Otherwise we will crash since target intrinsics don't have their types
legalized. Let the mgather get legalized first, then do the combine on
the legal type.
Fixes#81088
Co-authored-by: Craig Topper <craig.topper@sifive.com>
This code was correct as written prior to C++17, which allowed bases to
appear in the initializer list.
This was observable by creating non-constant aggregate initialization at
file scope in a compound literal, but since that behavior will change
soon if we implement support for dynamic initialization, I also added a
unit test for `isConstantInitializer`.
This fixes at least one part of issue #80510 .
---------
Co-authored-by: Aaron Ballman <aaron@aaronballman.com>
We run CheckLargeFunctions pass in non-relocation mode to prevent the
emission of functions that later could not be written to the output due
to their large size. The main reason behind the pass is to prevent the
emission of metadata for such functions since this metadata becomes
incorrect if the function is left unmodified.
Currently, the pass is enabled in non-relocation mode only when debug
info output is also enabled. As we emit increasingly more kinds of
metadata, e.g. for the Linux kernel, it becomes more challenging to
track metadata that needs to be fixed. Hence, I'm enabling the pass to
always run in non-relocation mode.
DWARFLinkerImpl::DWARFLinkerImpl initializes
DebugStrStrings/DebugLineStrStrings/CommonSections using GlobalData
but GlobalData is initialized after the three members.
Move GlobalData before.
Fix#81110
Currently symbol info is generated from a linking section or from export
names. This PR generates symbols in a WasmObjectFile from the name
section as well, which allows tools like objdump and nm to show useful
information for more linked binaries. There are some limitations:
most notably that we don't assume any particular ABI, so we don't get
detailed information about data symbols if the segments are merged
(which is the default).
Covers most of the desired functionality from #76107
Deprecate `TestVectors`, since no one uses it.
This affects the output order of ExecVectors.
The current impl emits sorted by binary value of ExecVector. This impl
emits along the traversal of `buildTestVector()`.
LLDB has a setting (symbols.enable-background-lookup) that calls
dsymForUUID on a background thread for images as they appear in the
current backtrace. Originally, the laziness of only looking up symbols
for images in the backtrace only existed to bring the number of
dsymForUUID calls down to a manageable number.
Users have requesting the same functionality but blocking. This gives
them the same user experience as enabling dsymForUUID globally, but
without the massive upfront cost of having to download all the images,
the majority of which they'll likely not need.
This patch renames the setting to have a more generic name
(symbols.auto-download) and changes its values from a boolean to an
enum. Users can now specify "off", "background" and "foreground". The
default remains "off" although I'll probably change that in the near
future.
This adjusts the isSubVectorExtractCheap callback to consider any
extract which fits entirely within the first VLEN bits of the src vector
(and uses a 5 bit immediate for the slide) as cheap. These can be done
via a single m1 vslide1down.vi instruction.
This allows our generic DAG combine logic to kick in and recognize a few
more cases where shuffle source is longer than the dest, but that using
a wider shuffle is still profitable. (Or as shown in the test diff, we
can split the wider source and do two narrower shuffles.)
Currently, a method exists to get the count of the operation objects
which are still alive. This helps for sanity checking, but isn't
terribly useful for debugging. This new method returns the actual
operation objects which are still alive.
This allows Python code like the following:
```
gc.collect()
live_ops = ir.Context.current._get_live_operation_objects()
for op in live_ops:
print(f"Warning: {op} is still live. Referrers:")
for referrer in gc.get_referrers(op)[0]:
print(f" {referrer}")
```
Add a SPIR-V target-specific intrinsic for creating handles, which is
used for lowering HLSL resources types like RWBuffer.
`llvm/lib/TargetParser/Triple.cpp`: SPIR-V intrinsics use "spv" as the
target prefix, not "spirv". As far as I can tell, this is the first one
that is used via the `CGBuiltin` codepath, which relies on
`getArchTypePrefix`, so I've corrected it here.
`clang/lib/Basic/Targets/SPIR.h`: When records are laid out in the
lowering from AST to IR, they were incorrectly offset because these
Pointer attributes were defaulting to 32.
Related to #81036
Fixes https://github.com/llvm/llvm-project/issues/80910.
Per the documentation in ISDOpcodes.h, for BUILD_VECTOR "The types of
the operands must match the vector element type, except that integer
types are allowed to be larger than the element type, in which case the
operands are implicitly truncated."
This transform was assuming that the scalar operand type matched the
result type. This resulted in essentially performing a truncate before a
binop, instead of after. As demonstrated by the test case changes, this
is often not legal.
LLDB has a setting (symbols.enable-background-lookup) that calls
dsymForUUID on a background thread for images as they appear in the
current backtrace. Originally, the laziness of only looking up symbols
for images in the backtrace only existed to bring the number of
dsymForUUID calls down to a manageable number.
Users have requesting the same functionality but blocking. This gives
them the same user experience as enabling dsymForUUID globally, but
without the massive upfront cost of having to download all the images,
the majority of which they'll likely not need.
This patch renames the setting to have a more generic name
(symbols.auto-download) and changes its values from a boolean to an
enum. Users can now specify "off", "background" and "foreground". The
default remains "off" although I'll probably change that in the near
future.
The following test case in
`clang-tools-extra/test/clang-tidy/infrastructure/diagnostic.cpp` is
failing:
```
#ifdef PR64602 // Should not crash
template <class T = void>
struct S
{
auto foo(auto);
};
template <>
auto S<>::foo(auto)
{
return 1;
}
// CHECK8: error: template parameter list matching the non-templated nested type 'S<>' should be empty ('template<>') [clang-diagnostic-error]
#endif
```
#80864 fixes a bug where we would (incorrectly) append invented template
parameters to empty template parameter lists, which causes this test to
fail.
The MemoryBuffer is created using `RequiresNullTerminator`, so we can
safely skip the `CurPtr != CurBuf.end()` check. The redundant check
causes a cppcheck report. In addition, elsewhere, including `*CurPtr ==
'#'` below, makes the null terminator assumption as well.
Close#81120
This commit adds a new broadcast bit to the debugger. When in use, it
will be listened to for progress events that will be delivered and kept
track of by category as opposed to the current behaviour of coming in
one by one.
Both LLVM_LINK_LLVM_DYLIB and LLVM_PARALLEL_LINK_JOBS help with some
common gotchas. It seems worth documenting them here explicitly.
Based on a review comment, also "refactor" the documentation to avoid duplication.
The cv specializations for `numeric_limits` inherited privately for some
reason. We can simplify the implementation by inheriting publicly and
removing the members that just replicate the values from the base class.
This reverts commit bdde5f9bea75e897bcc31a95b9c3376988c211cc.
Two situations that are tripping a few buildbots:
https://lab.llvm.org/buildbot/#/builders/205/builds/25126
Here, polly is currently presenting a DebugLoc attached to a debugging
intrinsic as a "true" source location in a user report, something that's
unreliable.
https://lab.llvm.org/buildbot/#/builders/184/builds/10242
These HWAsan failures are probably (97% confidence) because in
StackInfoBuilder::visit we're not observing DPValues attached to lifetime
intrinsics because they're delt with higher up the function.
But it's late-o'clock here, so revert for now.
This patch provides more information to the
`PPCallbacks::InclusionDirective()` hook. We now always pass the
suggested module, regardless of whether it was actually imported or not.
The extra `bool ModuleImported` parameter then denotes whether the
header `#include` will be automatically translated into import the the
module.
The main change is in `clang/lib/Lex/PPDirectives.cpp`, where we take
care to not modify `SuggestedModule` after it's been populated by
`LookupHeaderIncludeOrImport()`. We now exclusively use the `SM`
(`ModuleToImport`) variable instead, which has been equivalent to
`SuggestedModule` until now. This allows us to use the original
non-modified `SuggestedModule` for the callback itself.
(This patch turns out to be necessary for
https://github.com/apple/llvm-project/pull/8011).
According to [dcl.fct] p23:
> An abbreviated function template can have a _template-head_. The
invented _template-parameters_ are appended to the
_template-parameter-list_ after the explicitly declared
_template-parameters_.
`template<>` is not a _template-head_ -- a _template-head_ must have at
least one _template-parameter_. This patch corrects our current behavior
of appending the invented template parameters to the innermost template
parameter list, regardless of whether it is empty. Example:
```
template<typename T>
struct A
{
void f(auto);
};
template<>
void A<int>::f(auto); // ok
template<>
template<> // warning: extraneous template parameter list in template specialization
void A<int>::f(auto);
```
This is a first simple patch to introduce a new FIR attribute to carry
the CUDA variable attribute information to hlfir.declare and fir.declare
operations. It currently lowers this information for local variables.
The texture attribute is omitted since it is rejected by semantic and
will not make its way to MLIR.
This new attribute is added as optional attribute to the hlfir.declare
and fir.declare operations.
Fixes#81136 - we might be loading from a constant pool entry wider than the destination register bitwidth, affecting the vextload scale calculation.
ConvertToBroadcastAVX512 doesn't yet set an explicit bitwidth (it will default to the constant pool bitwidth) due to difficulties in looking up the original register width through the fold tables, but as we only use rebuildSplatCst this shouldn't cause any miscompilations, although it might prevent folding to broadcast if only the lower bits match a splatable pattern.
Refactors logic in `ParseInternal` that was previously calling
`GetFormatFromCString` twice, once with `partial_match_ok` set to false,
and the second time set to true.
With this change, lldb formats (ie `%@`, `%S`, etc) are checked first.
If a format is not one of those, then `GetFormatFromCString` is called
once, and now always checks for partial matches.
This formatter
https://github.com/llvm/llvm-project/pull/78609
was originally passing the signed seconds (which can refer to times in
the past) with an unsigned printf formatter, and had tests that expected
to see negative values from the printf which always failed on macOS. I'm
not clear how they ever passed on any platform.
Fix the printf to print seconds as a signed value, and re-enable the
tests.
