gs_device_nccl.F90 File Reference

Go to the source code of this file.

Data Types
type	gs_device_nccl::gs_device_nccl_buf_t
	Buffers for non-blocking communication and packing/unpacking. More...
type	gs_device_nccl::gs_device_nccl_t
	Gather-scatter communication using NCCL The arrays are indexed per PE like send_pe and @ recv_pe. More...
interface	gs_device_nccl::hip_gs_pack
interface	gs_device_nccl::hip_gs_unpack
interface	gs_device_nccl::hip_gs_pack_vec
interface	gs_device_nccl::hip_gs_unpack_vec
interface	gs_device_nccl::device_nccl_sendrecv

Modules
module	gs_device_nccl
	Defines NCCL based gather-scatter communication.

Functions/Subroutines
subroutine	gs_device_nccl::gs_device_nccl_buf_init (this, pe_order, dof_stack, mark_dupes)
subroutine	gs_device_nccl::gs_device_nccl_buf_free (this)
subroutine	gs_device_nccl::gs_device_nccl_init (this, send_pe, recv_pe)
	Initialise NCCL based communication method.
subroutine	gs_device_nccl::gs_device_nccl_free (this)
	Deallocate MPI based communication method.
subroutine	gs_device_nccl::gs_device_nccl_nbsend (this, u, n, tag, deps, strm)
	Post non-blocking send operations.
subroutine	gs_device_nccl::gs_device_nccl_nbrecv (this, tag)
	Post non-blocking receive operations.
subroutine	gs_device_nccl::gs_device_nccl_nbwait (this, u, n, op, strm)
	Wait for non-blocking operations.
subroutine	gs_device_nccl::gs_device_nccl_nbsend_vec (this, u, n, nc, tag, deps, strm)
	Fused nc-component send. u is the compact shared device buffer (component-outer, per-component stride n = nshared). Packs each peer's nc-component slab on its own stream.
subroutine	gs_device_nccl::gs_device_nccl_nbrecv_vec (this, tag, nc)
	No-op: send/recv and unpack happen in nbwait_vec.
subroutine	gs_device_nccl::gs_device_nccl_nbwait_vec (this, u, n, nc, op, strm)
	Fused nc-component send/recv + unpack.