Neko 1.99.5
A portable framework for high-order spectral element flow simulations
Loading...
Searching...
No Matches
gs_device_nccl Module Reference

Defines NCCL based gather-scatter communication.

Data Types

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

Functions/Subroutines

subroutine gs_device_nccl_buf_init (this, pe_order, dof_stack, mark_dupes)
 
subroutine gs_device_nccl_buf_free (this)
 
subroutine gs_device_nccl_init (this, send_pe, recv_pe)
 Initialise NCCL based communication method.
 
subroutine gs_device_nccl_free (this)
 Deallocate MPI based communication method.
 
subroutine gs_device_nccl_nbsend (this, u, n, tag, deps, strm)
 Post non-blocking send operations.
 
subroutine gs_device_nccl_nbrecv (this, tag)
 Post non-blocking receive operations.
 
subroutine gs_device_nccl_nbwait (this, u, n, op, strm)
 Wait for non-blocking operations.
 
subroutine 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_nbrecv_vec (this, tag, nc)
 No-op: send/recv and unpack happen in nbwait_vec.
 
subroutine gs_device_nccl_nbwait_vec (this, u, n, nc, op, strm)
 Fused nc-component send/recv + unpack.
 

Function/Subroutine Documentation

◆ gs_device_nccl_buf_free()

subroutine gs_device_nccl::gs_device_nccl_buf_free ( class(gs_device_nccl_buf_t), intent(inout this)
private

Definition at line 249 of file gs_device_nccl.F90.

Here is the call graph for this function:

◆ gs_device_nccl_buf_init()

subroutine gs_device_nccl::gs_device_nccl_buf_init ( class(gs_device_nccl_buf_t), intent(inout this,
integer, dimension(:), intent(inout), allocatable  pe_order,
type(stack_i4_t), dimension(:), intent(inout), allocatable  dof_stack,
logical, intent(in mark_dupes 
)
private

Definition at line 176 of file gs_device_nccl.F90.

Here is the call graph for this function:

◆ gs_device_nccl_free()

subroutine gs_device_nccl::gs_device_nccl_free ( class(gs_device_nccl_t), intent(inout this)
private

Definition at line 294 of file gs_device_nccl.F90.

Here is the call graph for this function:

◆ gs_device_nccl_init()

subroutine gs_device_nccl::gs_device_nccl_init ( class(gs_device_nccl_t), intent(inout this,
type(stack_i4_t), intent(inout send_pe,
type(stack_i4_t), intent(inout recv_pe 
)
private

Definition at line 261 of file gs_device_nccl.F90.

Here is the call graph for this function:

◆ gs_device_nccl_nbrecv()

subroutine gs_device_nccl::gs_device_nccl_nbrecv ( class(gs_device_nccl_t), intent(inout this,
integer, intent(in tag 
)
private

Definition at line 355 of file gs_device_nccl.F90.

◆ gs_device_nccl_nbrecv_vec()

subroutine gs_device_nccl::gs_device_nccl_nbrecv_vec ( class(gs_device_nccl_t), intent(inout this,
integer, intent(in tag,
integer, intent(in nc 
)
private

Definition at line 452 of file gs_device_nccl.F90.

◆ gs_device_nccl_nbsend()

subroutine gs_device_nccl::gs_device_nccl_nbsend ( class(gs_device_nccl_t), intent(inout this,
real(kind=rp), dimension(n), intent(inout u,
integer, intent(in n,
integer, intent(in tag,
type(c_ptr), intent(inout deps,
type(c_ptr), intent(inout strm 
)
private

Definition at line 316 of file gs_device_nccl.F90.

Here is the call graph for this function:

◆ gs_device_nccl_nbsend_vec()

subroutine gs_device_nccl::gs_device_nccl_nbsend_vec ( class(gs_device_nccl_t), intent(inout this,
real(kind=rp), dimension(nc*n), intent(inout u,
integer, intent(in n,
integer, intent(in nc,
integer, intent(in tag,
type(c_ptr), intent(inout deps,
type(c_ptr), intent(inout strm 
)
private

Definition at line 422 of file gs_device_nccl.F90.

Here is the call graph for this function:

◆ gs_device_nccl_nbwait()

subroutine gs_device_nccl::gs_device_nccl_nbwait ( class(gs_device_nccl_t), intent(inout this,
real(kind=rp), dimension(n), intent(inout u,
integer, intent(in n,
integer  op,
type(c_ptr), intent(inout strm 
)
private

Definition at line 365 of file gs_device_nccl.F90.

Here is the call graph for this function:

◆ gs_device_nccl_nbwait_vec()

subroutine gs_device_nccl::gs_device_nccl_nbwait_vec ( class(gs_device_nccl_t), intent(inout this,
real(kind=rp), dimension(nc*n), intent(inout u,
integer, intent(in n,
integer, intent(in nc,
integer  op,
type(c_ptr), intent(inout strm 
)
private

Definition at line 458 of file gs_device_nccl.F90.

Here is the call graph for this function: