This ensures that attributes such as the index bitwidth propagate
correctly to the AMDGPUToROCDL patterns.
Differential Revision: https://reviews.llvm.org/D125320
Previously we took the old name and always appended a numberic suffix.
Since we're doing a 1:1 replacement, it's clearer to keep the original
name exactly.
Reviewed By: fhahn
Differential Revision: https://reviews.llvm.org/D125281
This makes the output IR more readable since we're doing a one to
one replacement.
Reviewed By: fhahn
Differential Revision: https://reviews.llvm.org/D125280
Fortran 2018 requires that a compiler allow objects whose rank + corank
is 15, and that's our maximum; detect and diagnose violations.
Differential Revision: https://reviews.llvm.org/D125153
The controlling expression of a _Generic selection expression undergoes
lvalue conversion, array conversion, and function conversion before
picking the association. This means that array types, function types,
and qualified types are all unreachable code if they're used as an
association. I've been caught by this twice in the past few months and
I figure that if a WG14 member can't seem to remember this rule, users
are also likely to struggle with it. So this adds an on-by-default
unreachable code diagnostic for generic selection expression
associations.
Note, we don't have to worry about function types as those are already
a constraint violation which generates an error.
Differential Revision: https://reviews.llvm.org/D125259
Evaluate real-valued references to the intrinsic functions MODULO
and MOD at compilation time without recourse to an external math
library.
Differential Revision: https://reviews.llvm.org/D125151
This patch updates calls to AnalysisState::getBuffer() so that we return
early with a failure if the call does not succeed.
Reviewed By: springerm
Differential Revision: https://reviews.llvm.org/D125251
In case of placement new, if we do not know the alignment of the
operand, we can't assume it has the preferred alignment. It might be
e.g. a pointer to a struct member which follows ABI alignment rules.
This makes UBSAN no longer report "constructor call on misaligned
address" when constructing a double into a struct field of type double
on i686. The psABI specifies an alignment of 4 bytes, but the preferred
alignment used by Clang is 8 bytes.
We now use ABI alignment for allocating new as well, as the preferred
alignment should be used for over-aligning e.g. local variables, which
isn't relevant for ABI code dealing with operator new. AFAICT there
wouldn't be problems either way though.
Fixes#54845.
Differential Revision: https://reviews.llvm.org/D124736
By analogy with the NVGPU dialect, introduce an AMDGPU dialect for
AMD-specific intrinsic wrappers.
The dialect initially includes wrappers around the raw buffer intrinsics.
On AMD GPUs, a memref can be converted to a "buffer descriptor" that
allows more precise control of memory access, such as by allowing for
out of bounds loads/stores to be replaced by 0/ignored without adding
additional conditional logic, which is important for performance.
The repository currently contains a limited conversion from
transfer_read/transfer_write to Mubuf intrinsics, which are an older,
deprecated intrinsic for the same functionality.
The new amdgpu.raw_buffer_* ops allow these operations to be used
explicitly and for including metadata such as whether the target
chipset is an RDNA chip or not (which impacts the interpretation of
some bits in the buffer descriptor), while still maintaining an
MLIR-like interface.
(This change also exposes the floating-point atomic add intrinsic.)
Reviewed By: ThomasRaoux
Differential Revision: https://reviews.llvm.org/D122765
This is generally a better default for tools other than the compiler, which
shouldn't assume a PCH file on disk is something they can consume.
Preserve the old behavior in places associated with libclang/c-index-test
(including ASTUnit) as there are tests relying on it and most important
consumers are out-of-tree. It's unclear whether the tests are specifically
trying to test this functionality, and what the downstream implications of
removing it are. Hopefully someone more familiar can clean this up in future.
Differential Revision: https://reviews.llvm.org/D125149
Fold references to the intrinsic function DIM with constant real
arguments. And clean up folding of comparisons with NaNs to address
a problem noticed in testing -- NaNs should successfully compare
unequal to all values, including themselves, instead of failing all
comparisons.
Differential Revision: https://reviews.llvm.org/D125146
A typical problem with missing a token is that the missing
token is at the end of a line. The problem with this is that
the error message gets reported on the start of the following
line (which is where the next / invalid token is) which can
be confusing.
Handle this by noticing this case and backing up to the end of
the previous line.
Differential Revision: https://reviews.llvm.org/D125295
This patch adds 32-bit AIX RUN lines to several test cases, along with the
addition of one new test case, to prepare for future codegen changes involving
the PPCISD::SCALAR_TO_VECTOR_PERMUTED node on 32-bit mode.
Only fold for uniform values on pre-GFX9 chips. GFX9+ allow us
to keep the calculation entirely on the SALU.
For subtargets where integer multiplication isn't full-rate, avoid
folding if the multiply has too many uses.
Finally, we expand 64x32 and 64x64 multiplies here as well, if they
feed into an addition. This results in better code generation than
the generic expansion for such multiplies because we end up using
the accumulator of the MAD instructions.
Differential Revision: https://reviews.llvm.org/D123835
When we terminate due to an exception being uncaught, libc++abi prints
a message saying "terminating with uncaught exception [...]". This patch
changes that to say "terminating due to uncaught exception [...]" instead,
which is a bit clearer. Indeed, I've seen some people being confused and
thinking that libc++abi was the component throwing the exception.
Differential Revision: https://reviews.llvm.org/D125245
Currently for SVE2 ACLE builtins, single tests are used to verify both
clang code generation (when the feature is available) and semantic
error/warning messages (when the feature is unavailable). This
patch moves the semantic testing for the target feature flag into
dedicated Sema tests.
Differential Revision: https://reviews.llvm.org/D124850
Currently for SVE ACLE builtins, single tests are used to verify both
clang code generation (when the feature is available) and semantic
error/warning messages (when the feature is unavailable). This
patch moves the semantic testing into dedicated Sema tests.
Differential Revision: https://reviews.llvm.org/D124924
This issue reproduces in the context of LoopDeletion, because the
bitcast does not get simplified away there. For a plain -inst-simplify
run the bitcast would get folded away first.
Fixes https://github.com/llvm/llvm-project/issues/54615.
The EnableReuseStorageInFrame option is designed for testing only.
But it is better to use *_PASS_WITH_PARAMS macro to keep consistent with
other passes.
30a12f3f6322399185fdceffe176152a58bb84ae switched the type check
to use the GEP result type rather than the GEP operand type.
However, the GEP result types may match even if the operand types
don't, in case GEPs with scalar/vector base and vector index
are compared.
Fixes https://github.com/llvm/llvm-project/issues/55363.
This halves the size of LargeTest, dropping time to compile this
file locally from 14s to 5.5s. Hopefully this will also fix the
persistent timeouts in pre-merge checks.
Differential Revision: https://reviews.llvm.org/D124237
Building libMLIR.so currently fails with:
> /usr/bin/ld: /tmp/ccNzulEA.ltrans39.ltrans.o: in function `(anonymous namespace)::SerializeToHsacoPass::optimizeLlvm(llvm::Module&, llvm::TargetMachine&)':
> /builddir/build/BUILD/llvm-project-15.0.0.src/mlir/lib/Dialect/GPU/Transforms/SerializeToHsaco.cpp:328: undefined reference to `mlir::makeOptimizingTransformer(unsigned int, unsigned int, llvm::TargetMachine*)'
This is because MLIRGPUTransforms depends on MLIRExecutionEngine in
61bb2e4ea8/mlir/lib/Dialect/GPU/Transforms/SerializeToHsaco.cpp (L328),
but MLIRExecutionEngine is marked as excluded from libMLIR.so.
However, this code doesn't require the full execution engine: It
only performs middle-end optimization, and does not need any of
the JIT/codegen infrastructure. As such, split off a separate
library MLIRExecutionEngineUtils, which only contains that part
and is not excluded from libMLIR.so.
Fixes https://github.com/llvm/llvm-project/issues/54242.
Differential Revision: https://reviews.llvm.org/D125214
Summary:
By evaluating both children states, now we are capable of discovering
infeasible parent states. In this patch, `assume` is implemented in the terms
of `assumeDuali`. This might be suboptimal (e.g. where there are adjacent
assume(true) and assume(false) calls, next patches addresses that). This patch
fixes a real CRASH.
Fixes https://github.com/llvm/llvm-project/issues/54272
Differential Revision:
https://reviews.llvm.org/D124758
In some cases a parent State is already infeasible, but we recognize
this only if an additonal constraint is added. This patch is the first
of a series to address this issue. In this patch `assumeDual` is changed
to clone the parent State but with an `Infeasible` flag set, and this
infeasible-parent is returned both for the true and false case. Then
when we add a new transition in the exploded graph and the destination
is marked as infeasible, the node will be a sink node.
Related bug:
https://github.com/llvm/llvm-project/issues/50883
Actually, this patch does not solve that bug in the solver, rather with
this patch we can handle the general parent-infeasible cases.
Next step would be to change the State API and require all checkers to
use the `assume*Dual` API and deprecate the simple `assume` calls.
Hopefully, the next patch will introduce `assumeInBoundDual` and will
solve the CRASH we have here:
https://github.com/llvm/llvm-project/issues/54272
Differential Revision: https://reviews.llvm.org/D124674
While it originally did, this option no longer affects the cc1
interface. For the cc1 interface, -no-opaque-pointers has to be
passed, there is no cmake option.
Looks like this test exposes some non-determinism and fails with
auth-generated checks. Recommited with sole check it didn't crash.
Will investigate why it happens.
This is a following cleanup for the previous work D123918. I missed
serveral places which still use legacy pass managers. This patch tries
to remove them.
