[flang][cuda] Support shape shift in data transfer op. (#115929)

When an array is declared with a non default lower bound, the declare op
`getShape` will return a `ShapeShiftOp`. This result is us

[flang][cuda] Support shape shift in data transfer op. (#115929)

When an array is declared with a non default lower bound, the declare op
`getShape` will return a `ShapeShiftOp`. This result is used in data
transfer operation to compute the number of bytes to transfer. Update
the op to support `ShapeShiftOp`.

show more ...

[flang][cuda] Relax the verifier for cuf.register_kernel op (#112585)

Relax the verifier since the `gpu.func` might be converted to
`llvm.func` before `cuf.register_kernel` is converted.

Reland '[flang][cuda] Add cuf.register_kernel operation' (#112389)

The operation will be used in the CUF constructor to register the kernel
functions. This allow to delay this until codegen when th

Reland '[flang][cuda] Add cuf.register_kernel operation' (#112389)

The operation will be used in the CUF constructor to register the kernel
functions. This allow to delay this until codegen when the gpu.binary
will be available.

Reland of #112268 with correct shared library build support.

show more ...

Revert "[flang][cuda] Add cuf.register_kernel operation" (#112306)

Reverts llvm/llvm-project#112268

[flang][cuda] Add cuf.register_kernel operation (#112268)

The operation will be used in the CUF constructor to register the kernel
functions. This allow to delay this until codegen when the gpu.bin

[flang][cuda] Add cuf.register_kernel operation (#112268)

The operation will be used in the CUF constructor to register the kernel
functions. This allow to delay this until codegen when the gpu.binary
will be available.

show more ...

[flang][cuda] Simplify data transfer when possible (#106120)

When possible, avoid using descriptors and use the reference and the
shape for data_transfer.

[flang][cuda] Add shape to cuf.data_transfer operation (#104631)

When doing data transfer with dynamic sized array, we are currently
generating a data transfer between two descriptors. If the shape

[flang][cuda] Add shape to cuf.data_transfer operation (#104631)

When doing data transfer with dynamic sized array, we are currently
generating a data transfer between two descriptors. If the shape values
can be provided, we can keep the data transfer between two references.
This patch adds the shape operands to the operation.

This will be exploited in lowering in a follow up patch.

show more ...

[flang][cuda] Use cuf.alloc/cuf.free for local descriptor (#98518)

Local descriptor for cuda allocatable need to be handled on host and
device. One solution is to duplicate the descriptor (one on

[flang][cuda] Use cuf.alloc/cuf.free for local descriptor (#98518)

Local descriptor for cuda allocatable need to be handled on host and
device. One solution is to duplicate the descriptor (one on the host and
one on the device) and keep them in sync or have the descriptor in
managed/unified memory so we don't to take care of any sync.
The second solution is probably the one we will implement. In order to
have more flexibility on how descriptor representing cuda allocatable
are allocated, this patch updates the lowering to use the cuf operations
alloc and free to managed them.

show more ...

mlir/LogicalResult: move into llvm (#97309)

This patch is part of a project to move the Presburger library into
LLVM.

[flang][cuda] Relax cuf.data_transfer verifier (#95974)

Allow data transfer between array reference and array described by a
descriptor.

[flang] Add reductions for CUF Kernels: Lowering (#95184)

* Add reductionOperands and reductionAttrs to cuf's KernelOp.
* Parsing is already working and the tree has the info: here I make the
Bri

[flang] Add reductions for CUF Kernels: Lowering (#95184)

* Add reductionOperands and reductionAttrs to cuf's KernelOp.
* Parsing is already working and the tree has the info: here I make the
Bridge emit the updated KernelOp with reduction information added.
* Check |reductionAttrs| = |reductionOperands| in verifier
* Add a test
@clementval @vzakhari

---------

Co-authored-by: Iman Hosseini <imanh@nvidia.com>
Co-authored-by: Valentin Clement (バレンタイン クレメン) <clementval@gmail.com>

show more ...

[flang][cuda] Accept constant as src for cuf.data_tranfer (#92951)

Assignment of a constant (host) to a device variable is a special case
that can be further lowered to `cudaMemset` or similar func

[flang][cuda] Accept constant as src for cuf.data_tranfer (#92951)

Assignment of a constant (host) to a device variable is a special case
that can be further lowered to `cudaMemset` or similar functions. This
patch update the lowering to avoid the creation of a temporary when we
assign a constant to a device variable.

show more ...

[flang][cuda][NFC] Silence warning triggered in buildbot

[flang][cuda] Enable data transfer for descriptors (#92804)

Remove the TODO when data transfer is done with descriptor variables.

[flang][cuda] Move CUDA Fortran operations to a CUF dialect (#92317)

The number of operations dedicated to CUF grew and where all still in
FIR. In order to have a better organization, the CUF opera

[flang][cuda] Move CUDA Fortran operations to a CUF dialect (#92317)

The number of operations dedicated to CUF grew and where all still in
FIR. In order to have a better organization, the CUF operations,
attributes and code is moved into their specific dialect and files. CUF
dialect is tightly coupled with HLFIR/FIR and their types.

The CUF attributes are bundled into their own library since some
HLFIR/FIR operations depend on them and the CUF dialect depends on the
FIR types. Without having the attributes into a separate library there
would be a dependency cycle.

show more ...