'mpi' Dialect

This dialect models the Message Passing Interface (MPI), version 4.0. It is meant to serve as an interfacing dialect that is targeted by higher-level dialects. The MPI dialect itself can be lowered to multiple MPI implementations and hide differences in ABI. The dialect models the functions of the MPI specification as close to 1:1 as possible while preserving SSA value semantics where it makes sense, and uses memref types instead of bare pointers.

This dialect is under active development, and while stability is an eventual goal, it is not guaranteed at this juncture. Given the early state, it is recommended to inquire further prior to using this dialect.

For an in-depth documentation of the MPI library interface, please refer to official documentation such as the OpenMPI online documentation.

Operations ¶

source

mpi.allreduce (mpi::AllReduceOp) ¶

Equivalent to MPI_Allreduce(sendbuf, recvbuf, op, comm)

Syntax:

operation ::= `mpi.allreduce` `(` $sendbuf `,` $recvbuf `,` $op `,` $comm `)` attr-dict `:` type($sendbuf) `,` type($recvbuf) (`->` type($retval)^)?

MPI_Allreduce performs a reduction operation on the values in the sendbuf array and stores the result in the recvbuf array. The operation is performed across all processes in the communicator.

The op attribute specifies the reduction operation to be performed. Currently only the MPI_Op predefined in the standard (e.g. MPI_SUM) are supported.

This operation can optionally return an !mpi.retval value that can be used to check for errors.

Attributes: ¶

Operands: ¶

Results: ¶

mpi.barrier (mpi::Barrier) ¶

Equivalent to MPI_Barrier(comm)

Syntax:

operation ::= `mpi.barrier` `(` $comm `)` attr-dict
              (`->` type($retval)^)?

MPI_Barrier blocks execution until all processes in the communicator have reached this routine.

This operation can optionally return an !mpi.retval value that can be used to check for errors.

Operands: ¶

Results: ¶

mpi.comm_rank (mpi::CommRankOp) ¶

Get the current rank, equivalent to MPI_Comm_rank(comm, &rank)

Syntax:

operation ::= `mpi.comm_rank` `(` $comm `)` attr-dict `:` type(results)

This operation can optionally return an !mpi.retval value that can be used to check for errors.

Operands: ¶

Results: ¶

mpi.comm_size (mpi::CommSizeOp) ¶

Get the size of the group associated to the communicator, equivalent to MPI_Comm_size(comm, &size)

Syntax:

operation ::= `mpi.comm_size` `(` $comm `)` attr-dict `:` type(results)

This operation can optionally return an !mpi.retval value that can be used to check for errors.

Operands: ¶

Results: ¶

mpi.comm_split (mpi::CommSplitOp) ¶

Partition the group associated with the given communicator into disjoint subgroups

Syntax:

operation ::= `mpi.comm_split` `(` $comm `,` $color `,` $key `)` attr-dict `:` type(results)

This operation splits the communicator into multiple sub-communicators. The color value determines the group of processes that will be part of the new communicator. The key value determines the rank of the calling process in the new communicator.

This operation can optionally return an !mpi.retval value that can be used to check for errors.

Operands: ¶

Results: ¶

mpi.comm_world (mpi::CommWorldOp) ¶

Get the World communicator, equivalent to MPI_COMM_WORLD

Syntax:

operation ::= `mpi.comm_world` attr-dict `:` type(results)

This operation returns the predefined MPI_COMM_WORLD communicator.

Results: ¶

mpi.error_class (mpi::ErrorClassOp) ¶

Get the error class from an error code, equivalent to the MPI_Error_class function

Syntax:

operation ::= `mpi.error_class` $val attr-dict `:` type($val)

MPI_Error_class maps return values from MPI calls to a set of well-known MPI error classes.

Operands: ¶

Results: ¶

mpi.finalize (mpi::FinalizeOp) ¶

Finalize the MPI library, equivalent to MPI_Finalize()

Syntax:

operation ::= `mpi.finalize` attr-dict (`:` type($retval)^)?

This function cleans up the MPI state. Afterwards, no MPI methods may be invoked (excpet for MPI_Get_version, MPI_Initialized, and MPI_Finalized). Notably, MPI_Init cannot be called again in the same program.

This operation can optionally return an !mpi.retval value that can be used to check for errors.

Results: ¶

mpi.init (mpi::InitOp) ¶

Initialize the MPI library, equivalent to MPI_Init(NULL, NULL)

Syntax:

operation ::= `mpi.init` attr-dict (`:` type($retval)^)?

This operation must preceed most MPI calls (except for very few exceptions, please consult with the MPI specification on these).

Passing &argc, &argv is not supported currently.

This operation can optionally return an !mpi.retval value that can be used to check for errors.

Results: ¶

mpi.irecv (mpi::IRecvOp) ¶

Equivalent to MPI_Irecv(ptr, size, dtype, source, tag, comm, &req)

Syntax:

operation ::= `mpi.irecv` `(` $ref `,` $tag `,` $source `,` $comm`)` attr-dict `:` type($ref) `,` type($tag) `,` type($source)`->` type(results)

MPI_Irecv begins a non-blocking receive of size elements of type dtype from rank source. The tag value and communicator enables the library to determine the matching of multiple sends and receives between the same ranks.

This operation can optionally return an !mpi.retval value that can be used to check for errors.

Operands: ¶

Results: ¶

mpi.isend (mpi::ISendOp) ¶

Equivalent to MPI_Isend(ptr, size, dtype, dest, tag, comm)

Syntax:

operation ::= `mpi.isend` `(` $ref `,` $tag `,` $dest `,` $comm`)` attr-dict `:` type($ref) `,` type($tag) `,` type($dest) `->` type(results)

MPI_Isend begins a non-blocking send of size elements of type dtype to rank dest. The tag value and communicator enables the library to determine the matching of multiple sends and receives between the same ranks.

This operation can optionally return an !mpi.retval value that can be used to check for errors.

Operands: ¶

Results: ¶

mpi.recv (mpi::RecvOp) ¶

Equivalent to MPI_Recv(ptr, size, dtype, source, tag, comm, MPI_STATUS_IGNORE)

Syntax:

operation ::= `mpi.recv` `(` $ref `,` $tag `,` $source `,` $comm `)` attr-dict `:` type($ref) `,` type($tag) `,` type($source) (`->` type($retval)^)?

MPI_Recv performs a blocking receive of size elements of type dtype from rank source. The tag value and communicator enables the library to determine the matching of multiple sends and receives between the same ranks.

The MPI_Status is set to MPI_STATUS_IGNORE, as the status object is not yet ported to MLIR.

This operation can optionally return an !mpi.retval value that can be used to check for errors.

Operands: ¶

Results: ¶

mpi.retval_check (mpi::RetvalCheckOp) ¶

Check an MPI return value against an error class

Syntax:

operation ::= `mpi.retval_check` $val `=` $errclass attr-dict `:` type($res)

This operation compares MPI status codes to known error class constants such as MPI_SUCCESS, or MPI_ERR_COMM.

Attributes: ¶

Operands: ¶

Results: ¶

mpi.send (mpi::SendOp) ¶

Equivalent to MPI_Send(ptr, size, dtype, dest, tag, comm)

Syntax:

operation ::= `mpi.send` `(` $ref `,` $tag `,` $dest `,` $comm `)` attr-dict `:` type($ref) `,` type($tag) `,` type($dest)(`->` type($retval)^)?

MPI_Send performs a blocking send of size elements of type dtype to rank dest. The tag value and communicator enables the library to determine the matching of multiple sends and receives between the same ranks.

This operation can optionally return an !mpi.retval value that can be used to check for errors.

Operands: ¶

Results: ¶

mpi.wait (mpi::Wait) ¶

Equivalent to MPI_Wait(req, MPI_STATUS_IGNORE)

Syntax:

operation ::= `mpi.wait` `(` $req `)` attr-dict `:` type($req) (`->` type($retval) ^)?

MPI_Wait blocks execution until the request has completed.

The MPI_Status is set to MPI_STATUS_IGNORE, as the status object is not yet ported to MLIR.

This operation can optionally return an !mpi.retval value that can be used to check for errors.

Operands: ¶

Results: ¶

Attributes ¶

MPI_ErrorClassEnumAttr ¶

MPI error class name

Syntax:

#mpi.errclass<
  ::mlir::mpi::MPI_ErrorClassEnum   # value
>

Parameters: ¶

Types ¶

CommType ¶

MPI communicator handler

Syntax: !mpi.comm

This type represents a handler for the MPI communicator.

RequestType ¶

MPI asynchronous request handler

Syntax: !mpi.request

This type represents a handler to an asynchronous request.

RetvalType ¶

MPI function call return value (!mpi.retval)

Syntax: !mpi.retval

This type represents a return value from an MPI function call. This value can be MPI_SUCCESS, MPI_ERR_IN_STATUS, or any error code.

This return value can be compared agains the known MPI error classes represented by #mpi.errclass using the mpi.retval_check operation.

StatusType ¶

MPI reception operation status type

Syntax: !mpi.status

This type represents the status of a reception operation.

Enums ¶

MPI_ErrorClassEnum ¶

MPI error class name

Cases: ¶

MPI_OpClassEnum ¶

MPI operation class

Cases: ¶