Skip to content

Commit 7c8e769

Browse files
hjelmnhjelmn
authored
Merge pull request open-mpi#2045 from hjelmn/osc_rdma_atomics
osc/rdma: add support for network AMOs
2 parents e84b352 + 1ce5847 commit 7c8e769

9 files changed

+530
-244
lines changed

‎ompi/mca/osc/rdma/osc_rdma.h

+11-4
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,7 @@
88
* University of Stuttgart. All rights reserved.
99
* Copyright (c) 2004-2005 The Regents of the University of California.
1010
* All rights reserved.
11-
* Copyright (c) 2007-2015 Los Alamos National Security, LLC. All rights
11+
* Copyright (c) 2007-2016 Los Alamos National Security, LLC. All rights
1212
* reserved.
1313
* Copyright (c) 2010 Cisco Systems, Inc. All rights reserved.
1414
* Copyright (c) 2012-2013 Sandia National Laboratories. All rights reserved.
@@ -86,6 +86,12 @@ struct ompi_osc_rdma_component_t {
8686
/** Default value of the no_locks info key for new windows */
8787
bool no_locks;
8888

89+
/** Accumulate operations will only operate on a single intrinsic datatype */
90+
bool acc_single_intrinsic;
91+
92+
/** Use network AMOs when available */
93+
bool acc_use_amo;
94+
8995
/** Priority of the osc/rdma component */
9096
unsigned int priority;
9197

@@ -121,12 +127,13 @@ struct ompi_osc_rdma_module_t {
121127
/** value of same_size info key for this window */
122128
bool same_size;
123129

124-
/** window should have accumulate ordering... */
125-
bool accumulate_ordering;
126-
127130
/** passive-target synchronization will not be used in this window */
128131
bool no_locks;
129132

133+
bool acc_single_intrinsic;
134+
135+
bool acc_use_amo;
136+
130137
/** flavor of this window */
131138
int flavor;
132139

‎ompi/mca/osc/rdma/osc_rdma_accumulate.c

+416-131
Large diffs are not rendered by default.

‎ompi/mca/osc/rdma/osc_rdma_accumulate.h

+25-39
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,6 @@
11
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
22
/*
3-
* Copyright (c) 2014-2015 Los Alamos National Security, LLC. All rights
3+
* Copyright (c) 2014-2016 Los Alamos National Security, LLC. All rights
44
* reserved.
55
* $COPYRIGHT$
66
*
@@ -14,44 +14,30 @@
1414

1515
#include "osc_rdma.h"
1616

17-
int ompi_osc_rdma_compare_and_swap (const void *origin_addr, const void *compare_addr,
18-
void *result_addr, struct ompi_datatype_t *dt,
19-
int target, OPAL_PTRDIFF_TYPE target_disp,
20-
struct ompi_win_t *win);
21-
22-
int ompi_osc_rdma_accumulate (const void *origin_addr, int origin_count,
23-
struct ompi_datatype_t *origin_dt,
24-
int target, OPAL_PTRDIFF_TYPE target_disp,
25-
int target_count, struct ompi_datatype_t *target_dt,
26-
struct ompi_op_t *op, struct ompi_win_t *win);
27-
int ompi_osc_rdma_fetch_and_op (const void *origin_addr, void *result_addr,
28-
struct ompi_datatype_t *dt, int target,
29-
OPAL_PTRDIFF_TYPE target_disp,
30-
struct ompi_op_t *op, struct ompi_win_t *win);
31-
32-
int ompi_osc_rdma_get_accumulate (const void *origin_addr, int origin_count,
33-
struct ompi_datatype_t *origin_datatype,
34-
void *result_addr, int result_count,
35-
struct ompi_datatype_t *result_datatype,
36-
int target_rank, MPI_Aint target_disp,
37-
int target_count, struct ompi_datatype_t *target_datatype,
38-
struct ompi_op_t *op, struct ompi_win_t *win);
39-
40-
int ompi_osc_rdma_raccumulate (const void *origin_addr, int origin_count,
41-
struct ompi_datatype_t *origin_dt,
42-
int target, OPAL_PTRDIFF_TYPE target_disp,
43-
int target_count, struct ompi_datatype_t *target_dt,
44-
struct ompi_op_t *op, struct ompi_win_t *win,
45-
struct ompi_request_t **request);
46-
47-
int ompi_osc_rdma_rget_accumulate (const void *origin_addr, int origin_count,
48-
struct ompi_datatype_t *origin_datatype,
49-
void *result_addr, int result_count,
50-
struct ompi_datatype_t *result_datatype,
51-
int target_rank, MPI_Aint target_disp,
52-
int target_count, struct ompi_datatype_t *target_datatype,
53-
struct ompi_op_t *op, struct ompi_win_t *win,
54-
struct ompi_request_t **request);
17+
int ompi_osc_rdma_compare_and_swap (const void *origin_addr, const void *compare_addr, void *result_addr,
18+
ompi_datatype_t *dt, int target_rank, OPAL_PTRDIFF_TYPE target_disp,
19+
ompi_win_t *win);
20+
21+
int ompi_osc_rdma_accumulate (const void *origin_addr, int origin_count, ompi_datatype_t *origin_datatype, int target_rank,
22+
OPAL_PTRDIFF_TYPE target_disp, int target_count, ompi_datatype_t *target_datatype, ompi_op_t *op,
23+
ompi_win_t *win);
24+
25+
int ompi_osc_rdma_fetch_and_op (const void *origin_addr, void *result_addr, ompi_datatype_t *dt, int target_rank,
26+
OPAL_PTRDIFF_TYPE target_disp, ompi_op_t *op, ompi_win_t *win);
27+
28+
int ompi_osc_rdma_get_accumulate (const void *origin_addr, int origin_count, ompi_datatype_t *origin_datatype,
29+
void *result_addr, int result_count, ompi_datatype_t *result_datatype,
30+
int target_rank, MPI_Aint target_disp, int target_count, ompi_datatype_t *target_datatype,
31+
ompi_op_t *op, ompi_win_t *win);
32+
33+
int ompi_osc_rdma_raccumulate (const void *origin_addr, int origin_count, ompi_datatype_t *origin_datatype, int target_rank,
34+
OPAL_PTRDIFF_TYPE target_disp, int target_count, ompi_datatype_t *target_datatype, ompi_op_t *op,
35+
ompi_win_t *win, ompi_request_t **request);
36+
37+
int ompi_osc_rdma_rget_accumulate (const void *origin_addr, int origin_count, ompi_datatype_t *origin_datatype,
38+
void *result_addr, int result_count, ompi_datatype_t *result_datatype,
39+
int target_rank, MPI_Aint target_disp, int target_count, ompi_datatype_t *target_datatype,
40+
ompi_op_t *op, ompi_win_t *win, ompi_request_t **request);
5541

5642

5743
#endif /* OSC_RDMA_ACCUMULATE_H */

‎ompi/mca/osc/rdma/osc_rdma_comm.c

+14-14
Original file line numberDiff line numberDiff line change
@@ -772,9 +772,9 @@ static int ompi_osc_rdma_get_contig (ompi_osc_rdma_sync_t *sync, ompi_osc_rdma_p
772772
}
773773

774774
static inline int ompi_osc_rdma_put_w_req (ompi_osc_rdma_sync_t *sync, const void *origin_addr, int origin_count,
775-
struct ompi_datatype_t *origin_datatype, ompi_osc_rdma_peer_t *peer,
775+
ompi_datatype_t *origin_datatype, ompi_osc_rdma_peer_t *peer,
776776
OPAL_PTRDIFF_TYPE target_disp, int target_count,
777-
struct ompi_datatype_t *target_datatype, ompi_osc_rdma_request_t *request)
777+
ompi_datatype_t *target_datatype, ompi_osc_rdma_request_t *request)
778778
{
779779
ompi_osc_rdma_module_t *module = sync->module;
780780
mca_btl_base_registration_handle_t *target_handle;
@@ -807,9 +807,9 @@ static inline int ompi_osc_rdma_put_w_req (ompi_osc_rdma_sync_t *sync, const voi
807807
ompi_osc_rdma_put_contig, false);
808808
}
809809

810-
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,
810+
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,
811811
ompi_osc_rdma_peer_t *peer, OPAL_PTRDIFF_TYPE source_disp, int source_count,
812-
struct ompi_datatype_t *source_datatype, ompi_osc_rdma_request_t *request)
812+
ompi_datatype_t *source_datatype, ompi_osc_rdma_request_t *request)
813813
{
814814
ompi_osc_rdma_module_t *module = sync->module;
815815
mca_btl_base_registration_handle_t *source_handle;
@@ -841,9 +841,9 @@ static inline int ompi_osc_rdma_get_w_req (ompi_osc_rdma_sync_t *sync, void *ori
841841
source_handle, source_count, source_datatype, request,
842842
module->selected_btl->btl_get_limit, ompi_osc_rdma_get_contig, true);
843843
}
844-
int ompi_osc_rdma_put (const void *origin_addr, int origin_count, struct ompi_datatype_t *origin_datatype,
844+
int ompi_osc_rdma_put (const void *origin_addr, int origin_count, ompi_datatype_t *origin_datatype,
845845
int target_rank, OPAL_PTRDIFF_TYPE target_disp, int target_count,
846-
struct ompi_datatype_t *target_datatype, ompi_win_t *win)
846+
ompi_datatype_t *target_datatype, ompi_win_t *win)
847847
{
848848
ompi_osc_rdma_module_t *module = GET_MODULE(win);
849849
ompi_osc_rdma_peer_t *peer;
@@ -862,10 +862,10 @@ int ompi_osc_rdma_put (const void *origin_addr, int origin_count, struct ompi_da
862862
target_count, target_datatype, NULL);
863863
}
864864

865-
int ompi_osc_rdma_rput (const void *origin_addr, int origin_count, struct ompi_datatype_t *origin_datatype,
865+
int ompi_osc_rdma_rput (const void *origin_addr, int origin_count, ompi_datatype_t *origin_datatype,
866866
int target_rank, OPAL_PTRDIFF_TYPE target_disp, int target_count,
867-
struct ompi_datatype_t *target_datatype, struct ompi_win_t *win,
868-
struct ompi_request_t **request)
867+
ompi_datatype_t *target_datatype, ompi_win_t *win,
868+
ompi_request_t **request)
869869
{
870870
ompi_osc_rdma_module_t *module = GET_MODULE(win);
871871
ompi_osc_rdma_peer_t *peer;
@@ -897,9 +897,9 @@ int ompi_osc_rdma_rput (const void *origin_addr, int origin_count, struct ompi_d
897897
return OMPI_SUCCESS;
898898
}
899899

900-
int ompi_osc_rdma_get (void *origin_addr, int origin_count, struct ompi_datatype_t *origin_datatype,
900+
int ompi_osc_rdma_get (void *origin_addr, int origin_count, ompi_datatype_t *origin_datatype,
901901
int source_rank, OPAL_PTRDIFF_TYPE source_disp, int source_count,
902-
struct ompi_datatype_t *source_datatype, struct ompi_win_t *win)
902+
ompi_datatype_t *source_datatype, ompi_win_t *win)
903903
{
904904
ompi_osc_rdma_module_t *module = GET_MODULE(win);
905905
ompi_osc_rdma_peer_t *peer;
@@ -918,10 +918,10 @@ int ompi_osc_rdma_get (void *origin_addr, int origin_count, struct ompi_datatype
918918
source_disp, source_count, source_datatype, NULL);
919919
}
920920

921-
int ompi_osc_rdma_rget (void *origin_addr, int origin_count, struct ompi_datatype_t *origin_datatype,
921+
int ompi_osc_rdma_rget (void *origin_addr, int origin_count, ompi_datatype_t *origin_datatype,
922922
int source_rank, OPAL_PTRDIFF_TYPE source_disp, int source_count,
923-
struct ompi_datatype_t *source_datatype, struct ompi_win_t *win,
924-
struct ompi_request_t **request)
923+
ompi_datatype_t *source_datatype, ompi_win_t *win,
924+
ompi_request_t **request)
925925
{
926926
ompi_osc_rdma_module_t *module = GET_MODULE(win);
927927
ompi_osc_rdma_peer_t *peer;

‎ompi/mca/osc/rdma/osc_rdma_comm.h

+10-10
Original file line numberDiff line numberDiff line change
@@ -96,23 +96,23 @@ static inline int osc_rdma_get_remote_segment (ompi_osc_rdma_module_t *module, o
9696

9797
/* prototypes for implementations of MPI RMA window functions. these will be called from the
9898
* mpi interface (ompi/mpi/c) */
99-
int ompi_osc_rdma_put (const void *origin_addr, int origin_count, struct ompi_datatype_t *origin_dt,
99+
int ompi_osc_rdma_put (const void *origin_addr, int origin_count, ompi_datatype_t *origin_dt,
100100
int target, OPAL_PTRDIFF_TYPE target_disp, int target_count,
101-
struct ompi_datatype_t *target_dt, struct ompi_win_t *win);
101+
ompi_datatype_t *target_dt, ompi_win_t *win);
102102

103-
int ompi_osc_rdma_get (void *origin_addr, int origin_count, struct ompi_datatype_t *origin_dt,
103+
int ompi_osc_rdma_get (void *origin_addr, int origin_count, ompi_datatype_t *origin_dt,
104104
int target, OPAL_PTRDIFF_TYPE target_disp, int target_count,
105-
struct ompi_datatype_t *target_dt, struct ompi_win_t *win);
105+
ompi_datatype_t *target_dt, ompi_win_t *win);
106106

107-
int ompi_osc_rdma_rput (const void *origin_addr, int origin_count, struct ompi_datatype_t *origin_dt,
107+
int ompi_osc_rdma_rput (const void *origin_addr, int origin_count, ompi_datatype_t *origin_dt,
108108
int target, OPAL_PTRDIFF_TYPE target_disp, int target_count,
109-
struct ompi_datatype_t *target_dt, struct ompi_win_t *win,
110-
struct ompi_request_t **request);
109+
ompi_datatype_t *target_dt, ompi_win_t *win,
110+
ompi_request_t **request);
111111

112-
int ompi_osc_rdma_rget (void *origin_addr, int origin_count, struct ompi_datatype_t *origin_dt,
112+
int ompi_osc_rdma_rget (void *origin_addr, int origin_count, ompi_datatype_t *origin_dt,
113113
int target, OPAL_PTRDIFF_TYPE target_disp, int target_count,
114-
struct ompi_datatype_t *target_dt, struct ompi_win_t *win,
115-
struct ompi_request_t **request);
114+
ompi_datatype_t *target_dt, ompi_win_t *win,
115+
ompi_request_t **request);
116116

117117
/**
118118
* @brief read data from a remote memory region (blocking)

0 commit comments

Comments
 (0)