Skip to content

osc/rdma: add support for network AMOs #2045

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Sep 8, 2016
Merged

osc/rdma: add support for network AMOs #2045

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 11 additions & 4 deletions ompi/mca/osc/rdma/osc_rdma.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,7 +8,7 @@
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2007-2015 Los Alamos National Security, LLC. All rights
* Copyright (c) 2007-2016 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2010 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2012-2013 Sandia National Laboratories. All rights reserved.
Expand Down Expand Up @@ -86,6 +86,12 @@ struct ompi_osc_rdma_component_t {
/** Default value of the no_locks info key for new windows */
bool no_locks;

/** Accumulate operations will only operate on a single intrinsic datatype */
bool acc_single_intrinsic;

/** Use network AMOs when available */
bool acc_use_amo;

/** Priority of the osc/rdma component */
unsigned int priority;

Expand Down Expand Up @@ -121,12 +127,13 @@ struct ompi_osc_rdma_module_t {
/** value of same_size info key for this window */
bool same_size;

/** window should have accumulate ordering... */
bool accumulate_ordering;

/** passive-target synchronization will not be used in this window */
bool no_locks;

bool acc_single_intrinsic;

bool acc_use_amo;

/** flavor of this window */
int flavor;

Expand Down
547 changes: 416 additions & 131 deletions ompi/mca/osc/rdma/osc_rdma_accumulate.c
View file
Open in desktop

Large diffs are not rendered by default.

64 changes: 25 additions & 39 deletions ompi/mca/osc/rdma/osc_rdma_accumulate.h
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2014-2015 Los Alamos National Security, LLC. All rights
* Copyright (c) 2014-2016 Los Alamos National Security, LLC. All rights
* reserved.
* $COPYRIGHT$
*
Expand All @@ -14,44 +14,30 @@

#include "osc_rdma.h"

int ompi_osc_rdma_compare_and_swap (const void *origin_addr, const void *compare_addr,
void *result_addr, struct ompi_datatype_t *dt,
int target, OPAL_PTRDIFF_TYPE target_disp,
struct ompi_win_t *win);

int ompi_osc_rdma_accumulate (const void *origin_addr, int origin_count,
struct ompi_datatype_t *origin_dt,
int target, OPAL_PTRDIFF_TYPE target_disp,
int target_count, struct ompi_datatype_t *target_dt,
struct ompi_op_t *op, struct ompi_win_t *win);
int ompi_osc_rdma_fetch_and_op (const void *origin_addr, void *result_addr,
struct ompi_datatype_t *dt, int target,
OPAL_PTRDIFF_TYPE target_disp,
struct ompi_op_t *op, struct ompi_win_t *win);

int ompi_osc_rdma_get_accumulate (const void *origin_addr, int origin_count,
struct ompi_datatype_t *origin_datatype,
void *result_addr, int result_count,
struct ompi_datatype_t *result_datatype,
int target_rank, MPI_Aint target_disp,
int target_count, struct ompi_datatype_t *target_datatype,
struct ompi_op_t *op, struct ompi_win_t *win);

int ompi_osc_rdma_raccumulate (const void *origin_addr, int origin_count,
struct ompi_datatype_t *origin_dt,
int target, OPAL_PTRDIFF_TYPE target_disp,
int target_count, struct ompi_datatype_t *target_dt,
struct ompi_op_t *op, struct ompi_win_t *win,
struct ompi_request_t **request);

int ompi_osc_rdma_rget_accumulate (const void *origin_addr, int origin_count,
struct ompi_datatype_t *origin_datatype,
void *result_addr, int result_count,
struct ompi_datatype_t *result_datatype,
int target_rank, MPI_Aint target_disp,
int target_count, struct ompi_datatype_t *target_datatype,
struct ompi_op_t *op, struct ompi_win_t *win,
struct ompi_request_t **request);
int ompi_osc_rdma_compare_and_swap (const void *origin_addr, const void *compare_addr, void *result_addr,
ompi_datatype_t *dt, int target_rank, OPAL_PTRDIFF_TYPE target_disp,
ompi_win_t *win);

int ompi_osc_rdma_accumulate (const void *origin_addr, int origin_count, ompi_datatype_t *origin_datatype, int target_rank,
OPAL_PTRDIFF_TYPE target_disp, int target_count, ompi_datatype_t *target_datatype, ompi_op_t *op,
ompi_win_t *win);

int ompi_osc_rdma_fetch_and_op (const void *origin_addr, void *result_addr, ompi_datatype_t *dt, int target_rank,
OPAL_PTRDIFF_TYPE target_disp, ompi_op_t *op, ompi_win_t *win);

int ompi_osc_rdma_get_accumulate (const void *origin_addr, int origin_count, ompi_datatype_t *origin_datatype,
void *result_addr, int result_count, ompi_datatype_t *result_datatype,
int target_rank, MPI_Aint target_disp, int target_count, ompi_datatype_t *target_datatype,
ompi_op_t *op, ompi_win_t *win);

int ompi_osc_rdma_raccumulate (const void *origin_addr, int origin_count, ompi_datatype_t *origin_datatype, int target_rank,
OPAL_PTRDIFF_TYPE target_disp, int target_count, ompi_datatype_t *target_datatype, ompi_op_t *op,
ompi_win_t *win, ompi_request_t **request);

int ompi_osc_rdma_rget_accumulate (const void *origin_addr, int origin_count, ompi_datatype_t *origin_datatype,
void *result_addr, int result_count, ompi_datatype_t *result_datatype,
int target_rank, MPI_Aint target_disp, int target_count, ompi_datatype_t *target_datatype,
ompi_op_t *op, ompi_win_t *win, ompi_request_t **request);


#endif /* OSC_RDMA_ACCUMULATE_H */
28 changes: 14 additions & 14 deletions ompi/mca/osc/rdma/osc_rdma_comm.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -772,9 +772,9 @@ static int ompi_osc_rdma_get_contig (ompi_osc_rdma_sync_t *sync, ompi_osc_rdma_p
}

static inline int ompi_osc_rdma_put_w_req (ompi_osc_rdma_sync_t *sync, const void *origin_addr, int origin_count,
struct ompi_datatype_t *origin_datatype, ompi_osc_rdma_peer_t *peer,
ompi_datatype_t *origin_datatype, ompi_osc_rdma_peer_t *peer,
OPAL_PTRDIFF_TYPE target_disp, int target_count,
struct ompi_datatype_t *target_datatype, ompi_osc_rdma_request_t *request)
ompi_datatype_t *target_datatype, ompi_osc_rdma_request_t *request)
{
ompi_osc_rdma_module_t *module = sync->module;
mca_btl_base_registration_handle_t *target_handle;
Expand Down Expand Up @@ -807,9 +807,9 @@ static inline int ompi_osc_rdma_put_w_req (ompi_osc_rdma_sync_t *sync, const voi
ompi_osc_rdma_put_contig, false);
}

static inline int ompi_osc_rdma_get_w_req (ompi_osc_rdma_sync_t *sync, void *origin_addr, int origin_count, struct ompi_datatype_t *origin_datatype,
static inline int ompi_osc_rdma_get_w_req (ompi_osc_rdma_sync_t *sync, void *origin_addr, int origin_count, ompi_datatype_t *origin_datatype,
ompi_osc_rdma_peer_t *peer, OPAL_PTRDIFF_TYPE source_disp, int source_count,
struct ompi_datatype_t *source_datatype, ompi_osc_rdma_request_t *request)
ompi_datatype_t *source_datatype, ompi_osc_rdma_request_t *request)
{
ompi_osc_rdma_module_t *module = sync->module;
mca_btl_base_registration_handle_t *source_handle;
Expand Down Expand Up @@ -841,9 +841,9 @@ static inline int ompi_osc_rdma_get_w_req (ompi_osc_rdma_sync_t *sync, void *ori
source_handle, source_count, source_datatype, request,
module->selected_btl->btl_get_limit, ompi_osc_rdma_get_contig, true);
}
int ompi_osc_rdma_put (const void *origin_addr, int origin_count, struct ompi_datatype_t *origin_datatype,
int ompi_osc_rdma_put (const void *origin_addr, int origin_count, ompi_datatype_t *origin_datatype,
int target_rank, OPAL_PTRDIFF_TYPE target_disp, int target_count,
struct ompi_datatype_t *target_datatype, ompi_win_t *win)
ompi_datatype_t *target_datatype, ompi_win_t *win)
{
ompi_osc_rdma_module_t *module = GET_MODULE(win);
ompi_osc_rdma_peer_t *peer;
Expand All @@ -862,10 +862,10 @@ int ompi_osc_rdma_put (const void *origin_addr, int origin_count, struct ompi_da
target_count, target_datatype, NULL);
}

int ompi_osc_rdma_rput (const void *origin_addr, int origin_count, struct ompi_datatype_t *origin_datatype,
int ompi_osc_rdma_rput (const void *origin_addr, int origin_count, ompi_datatype_t *origin_datatype,
int target_rank, OPAL_PTRDIFF_TYPE target_disp, int target_count,
struct ompi_datatype_t *target_datatype, struct ompi_win_t *win,
struct ompi_request_t **request)
ompi_datatype_t *target_datatype, ompi_win_t *win,
ompi_request_t **request)
{
ompi_osc_rdma_module_t *module = GET_MODULE(win);
ompi_osc_rdma_peer_t *peer;
Expand Down Expand Up @@ -897,9 +897,9 @@ int ompi_osc_rdma_rput (const void *origin_addr, int origin_count, struct ompi_d
return OMPI_SUCCESS;
}

int ompi_osc_rdma_get (void *origin_addr, int origin_count, struct ompi_datatype_t *origin_datatype,
int ompi_osc_rdma_get (void *origin_addr, int origin_count, ompi_datatype_t *origin_datatype,
int source_rank, OPAL_PTRDIFF_TYPE source_disp, int source_count,
struct ompi_datatype_t *source_datatype, struct ompi_win_t *win)
ompi_datatype_t *source_datatype, ompi_win_t *win)
{
ompi_osc_rdma_module_t *module = GET_MODULE(win);
ompi_osc_rdma_peer_t *peer;
Expand All @@ -918,10 +918,10 @@ int ompi_osc_rdma_get (void *origin_addr, int origin_count, struct ompi_datatype
source_disp, source_count, source_datatype, NULL);
}

int ompi_osc_rdma_rget (void *origin_addr, int origin_count, struct ompi_datatype_t *origin_datatype,
int ompi_osc_rdma_rget (void *origin_addr, int origin_count, ompi_datatype_t *origin_datatype,
int source_rank, OPAL_PTRDIFF_TYPE source_disp, int source_count,
struct ompi_datatype_t *source_datatype, struct ompi_win_t *win,
struct ompi_request_t **request)
ompi_datatype_t *source_datatype, ompi_win_t *win,
ompi_request_t **request)
{
ompi_osc_rdma_module_t *module = GET_MODULE(win);
ompi_osc_rdma_peer_t *peer;
Expand Down
20 changes: 10 additions & 10 deletions ompi/mca/osc/rdma/osc_rdma_comm.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -96,23 +96,23 @@ static inline int osc_rdma_get_remote_segment (ompi_osc_rdma_module_t *module, o

/* prototypes for implementations of MPI RMA window functions. these will be called from the
* mpi interface (ompi/mpi/c) */
int ompi_osc_rdma_put (const void *origin_addr, int origin_count, struct ompi_datatype_t *origin_dt,
int ompi_osc_rdma_put (const void *origin_addr, int origin_count, ompi_datatype_t *origin_dt,
int target, OPAL_PTRDIFF_TYPE target_disp, int target_count,
struct ompi_datatype_t *target_dt, struct ompi_win_t *win);
ompi_datatype_t *target_dt, ompi_win_t *win);

int ompi_osc_rdma_get (void *origin_addr, int origin_count, struct ompi_datatype_t *origin_dt,
int ompi_osc_rdma_get (void *origin_addr, int origin_count, ompi_datatype_t *origin_dt,
int target, OPAL_PTRDIFF_TYPE target_disp, int target_count,
struct ompi_datatype_t *target_dt, struct ompi_win_t *win);
ompi_datatype_t *target_dt, ompi_win_t *win);

int ompi_osc_rdma_rput (const void *origin_addr, int origin_count, struct ompi_datatype_t *origin_dt,
int ompi_osc_rdma_rput (const void *origin_addr, int origin_count, ompi_datatype_t *origin_dt,
int target, OPAL_PTRDIFF_TYPE target_disp, int target_count,
struct ompi_datatype_t *target_dt, struct ompi_win_t *win,
struct ompi_request_t **request);
ompi_datatype_t *target_dt, ompi_win_t *win,
ompi_request_t **request);

int ompi_osc_rdma_rget (void *origin_addr, int origin_count, struct ompi_datatype_t *origin_dt,
int ompi_osc_rdma_rget (void *origin_addr, int origin_count, ompi_datatype_t *origin_dt,
int target, OPAL_PTRDIFF_TYPE target_disp, int target_count,
struct ompi_datatype_t *target_dt, struct ompi_win_t *win,
struct ompi_request_t **request);
ompi_datatype_t *target_dt, ompi_win_t *win,
ompi_request_t **request);

/**
* @brief read data from a remote memory region (blocking)
Expand Down
Loading