Current ASTContext.getAttributedType() takes attribute kind,
ModifiedType and EquivType as the hash to decide whether an AST node
has been generated or note. But this is not enough for btf_type_tag
as the attribute might have the same ModifiedType and EquivType, but
still have different string associated with attribute.
For example, for a data structure like below,
struct map_value {
int __attribute__((btf_type_tag("tag1"))) __attribute__((btf_type_tag("tag3"))) *a;
int __attribute__((btf_type_tag("tag2"))) __attribute__((btf_type_tag("tag4"))) *b;
};
The current ASTContext.getAttributedType() will produce
an AST similar to below:
struct map_value {
int __attribute__((btf_type_tag("tag1"))) __attribute__((btf_type_tag("tag3"))) *a;
int __attribute__((btf_type_tag("tag1"))) __attribute__((btf_type_tag("tag3"))) *b;
};
and this is incorrect.
It is very difficult to use the current AttributedType as it is hard to
get the tag information. To fix the problem, this patch introduced
BTFTagAttributedType which is similar to AttributedType
in many ways but with an additional BTFTypeTagAttr. The tag itself can
be retrieved with BTFTypeTagAttr.
With the new BTFTagAttributed type, the debuginfo code can be greatly
simplified compared to previous TypeLoc based approach.
Differential Revision: https://reviews.llvm.org/D120296
Sending output to /dev/stdout on AIX gets an llc permission denied error, so this patch removes this from the tests.
Reviewed By: simoll, hubert.reinterpretcast
Differential Revision: https://reviews.llvm.org/D121799
Report warnings and errors through events instead of printing directly
the to the debugger's error stream. By using events, IDEs such as Xcode
can report these issues in the UI instead of having them show up in the
debugger console.
The new diagnostic events are handled by the default event loop. If a
diagnostic is reported while nobody is listening for the new event
types, it is printed directly to the debugger's error stream.
Differential revision: https://reviews.llvm.org/D121511
Fix prefix emission order to emit REX immediately before the opcode (SDM vol2,
2.1, Figure 2-1). According to SDM vol2 2.2.1, "Other placements are ignored".
This fix has a side effect of outputting segment override prefix in a different
order than previously (benign).
Follow-up to https://reviews.llvm.org/D120592
Reviewed By: skan, craig.topper
Differential Revision: https://reviews.llvm.org/D120871
Print and emit redundant Address-Size override prefix if it's set on the
instruction.
Reviewed By: skan
Differential Revision: https://reviews.llvm.org/D120592
Add the rest of intrinsics to clang except intrinsics using vector mask
registers.
Reviewed By: simoll
Differential Revision: https://reviews.llvm.org/D121586
This patch adds support for lowering of the `set_exponent` intrinsic
from Fortran to the FIR dialect of MLIR.
This is part of the upstreaming effort from the `fir-dev` branch in [1].
[1] https://github.com/flang-compiler/f18-llvm-project
Differential Revision: https://reviews.llvm.org/D121805
Co-authored-by: Jean Perier <jperier@nvidia.com>
Co-authored-by: Eun Jung Park <ejpark@lanl.gov>
NFC. Switch from calculations based on dwords to bits, to be more
flexible.
Reviewed By: rampitec
Differential Revision: https://reviews.llvm.org/D121730
Allow fir.rebox input and output element type to differ for characters
if:
- Any of the character type is dynamic. Fortran allows making pointer
assignments between deferred and constant lengths entities, making this
case useful (if the input length is dynamic and the output length constant,
it is a user requirement that the length matches at runtime. There is no
option to check this at runtime, but it could be added as an option to
fir.rebox codegen later if desired).
- Or, there is a slice in the fir.rebox (the fir.rebox can implement a
substring view, hence the constant output and input lengths).
This is only a verifier constraint change, the fir.rebox codegen is not
impacted and already support those cases.
Add related FIR parsing, error, and codegen tests.
Differential Revision: https://reviews.llvm.org/D121710
When a load extends past the extent of the alloca, SROA will
restrict the slice size to extend to the end of the alloca only.
However, presplitting was asserting that the load size and the
slice size match exactly, which does not hold in this case.
Relax the assertion to only require that the load size is greater
or equal than the slice size.
This removes a restriction wrt. scf.for loops during One-Shot Bufferization. Such IR was previously rejected. It is still rejected by default because the bufferized IR could be slow. But such IR can now be bufferized with `allow-return-allocs`.
Differential Revision: https://reviews.llvm.org/D121529
New buffer allocations can now be returned/yielded from blocks with `allow-return-allocs`. One-Shot Bufferize deallocates all buffers at the end of the block. If this is not possible (because the buffer escapes the block), this is now done by the existing BufferDeallocation pass.
Differential Revision: https://reviews.llvm.org/D121527
CT_Dependent
When compile following code without -std=c++17, clang will abort by
llvm_unreachable:
class A {
public:
static const char X;
};
const char A::X = 0;
template<typename U> void func() noexcept(U::X);
template<class... B, char x>
void foo(void(B...) noexcept(x)) {}
void bar()
{
foo(func<A>);
}
So, my solution is to let EST_Uninstantiated in
FunctionProtoType::canThrow return CT_Dependent
Differential Revision: https://reviews.llvm.org/D121498
* Implement RegionBranchOpInterface: The op has a region, but it is conceptually not entered. The region just describes the semantics of the (monolithic) op.
* Linalg structured ops do not allocate memory.
Differential Revision: https://reviews.llvm.org/D121798
This patch adds lowering for the count intrinsic.
This patch is part of the upstreaming effort from fir-dev branch.
Reviewed By: jeanPerier
Differential Revision: https://reviews.llvm.org/D121782
Co-authored-by: Jean Perier <jperier@nvidia.com>
Co-authored-by: mleair <leairmark@gmail.com>
No need to schedule entry nodes where all instructions are not memory
read/write instructions and their operands are either constants, or
arguments, or phis, or instructions from others blocks, or their users
are phis or from the other blocks.
The resulting vector instructions can be placed at
the beginning of the basic block without scheduling (if operands does
not need to be scheduled) or at the end of the block (if users are
outside of the block).
It may save some compile time and scheduling resources.
Differential Revision: https://reviews.llvm.org/D121121
The original design of custom operands support assumed that most GPUs
have the same or very similar operand names end encodings. This is
no longer the case. As a result the support code becomes over-complicated
and difficult to maintain.
This change implements a different design with the following benefits:
- support of aliases;
- support of operands with overlapped encodings;
- identification of defined but unsupported operands.
Differential Revision: https://reviews.llvm.org/D121696
In 7fb40e1569, I changed the availability for bad_optional_access and
friends from macOS 10.14 to 10.13 after conducting an investigation on
old dylibs. It turns out that macOS 10.13 did have bad_optional_access,
however the dylib on iOS didn't match the dylib on macOS, so those
exception classes were only introduced in iOS 12.
Thanks to Aditya Kumar for noticing this.
Differential Revision: https://reviews.llvm.org/D121735
The BinaryEmitter uses opts::AlignText value to align the hot text
section. Also check that the opts::AlignText is at least
equal opts::AlignFunctions for the same reason, as described in D121392.
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
Differential Revision: https://reviews.llvm.org/D121728
It's the role of the C++ ABI library to install its own headers, not libc++.
This fixes an existing issue causing spurious CI failures where both libc++
and libc++abi would try to install <cxxabi.h> & friends in the same location,
leading to failures during the installation step.
Differential Revision: https://reviews.llvm.org/D121706
The code for traversing precompiled dependencies is somewhat complicated and contains a dangling iterator bug.
This patch simplifies the code and fixes the bug.
Reviewed By: dexonsmith
Differential Revision: https://reviews.llvm.org/D121533
When pruning header search paths (to reduce the number of modules we need to build explicitly), we can't prune the search paths used in (transitive) dependencies of a module. Otherwise, we could end up with either of the following dependency graphs:
```
X:<hash1> -> Y:<hash2>
X:<hash1> -> Y:<hash3>
```
depending on the search paths of the translation unit we discovered `X` and `Y` from.
This patch fixes that.
Depends on D121295.
Reviewed By: dexonsmith
Differential Revision: https://reviews.llvm.org/D121303
The iterator is not needed after the loop body anymore, meaning we can use more terse range-based for loop.
Depends on D121295.
Reviewed By: dexonsmith
Differential Revision: https://reviews.llvm.org/D121685
To reduce the number of modules we build in explicit builds (which use strict context hash), we prune unused header search paths. This essentially merges parts of the dependency graph.
Determining whether a search path was used to discover a module (through implicit module maps) proved to be somewhat complicated. Initial support landed in D102923, while D113676 attempts to fix some bugs.
However, now that we don't use implicit module maps in explicit builds (since D120465), we don't need to consider such search paths as used anymore. Modules are no longer discovered through the header search mechanism, so we can drop such search paths (provided they are not needed for other reasons).
This patch removes whatever support for detecting such usage we had, since it's buggy and not required anymore.
Depends on D120465.
Reviewed By: dexonsmith
Differential Revision: https://reviews.llvm.org/D121295
This option is added in both `flang-new` (the compiler driver) and
`flang-new -fc1` (the frontend driver). The semantics are consistent
with `clang` and `clang -cc1`.
As Flang does not run any LLVM passes when invoked with `-emit-llvm`
(i.e. `flang-new -S -emit-llvm <file>`), the tests use
`-S`/`-c`/`-emit-obj` instead. These options require an LLVM backend to
be run by the driver to generate the output (this makese `-mllvm`
relevant here).
Differential Revision: https://reviews.llvm.org/D121374
This patch adds lowering for the `adjustl` and `adjustr`
intrinsics.
This patch is part of the upstreaming effort from fir-dev branch.
Reviewed By: jeanPerier
Differential Revision: https://reviews.llvm.org/D121780
Co-authored-by: V Donaldson <vdonaldson@nvidia.com>
Co-authored-by: Jean Perier <jperier@nvidia.com>
This patch adds lowering for the `transfer` intrinsic.
The calls are lowered to runtime function calls.
This patch is part of the upstreaming effort from fir-dev branch.
Reviewed By: jeanPerier
Differential Revision: https://reviews.llvm.org/D121777
Co-authored-by: Jean Perier <jperier@nvidia.com>
Co-authored-by: mleair <leairmark@gmail.com>
This patch adds support for lowering of the `ibset` intrinsic from Fortran
to the FIR dialect of MLIR.
This is part of the upstreaming effort from the `fir-dev` branch in [1].
[1] https://github.com/flang-compiler/f18-llvm-project
Co-authored-by: Jean Perier <jperier@nvidia.com>
Co-authored-by: Valentin Clement <clementval@gmail.com>
Co-authored-by: V Donaldson <vdonaldson@nvidia.com>
Differential Revision: https://reviews.llvm.org/D121717
fneg instruction isel and tests. We do this also in preparation of fused
negatate-multiple-add fp operations.
Reviewed By: kaz7
Differential Revision: https://reviews.llvm.org/D121620
This patch adds initial argmemonly inference, by checking the underlying
objects of locations returned by MemoryLocation.
I think this should cover most cases, except function calls to other
argmemonly functions.
I'm not sure if there's a reason why we don't infer those yet.
Additional argmemonly can improve codegen in some cases. It also makes
it easier to come up with a C reproducer for 7662d1687b (already fixed,
but I'm trying to see if C/C++ fuzzing could help to uncover similar
issues.)
Compile-time impact:
NewPM-O3: +0.01%
NewPM-ReleaseThinLTO: +0.03%
NewPM-ReleaseLTO+g: +0.05%
https://llvm-compile-time-tracker.com/compare.php?from=067c035012fc061ad6378458774ac2df117283c6&to=fe209d4aab5b593bd62d18c0876732ddcca1614d&stat=instructions
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D121415
Nikita Popov