October 2014 - commits - mailman3.cels.anl.gov

[mpich] MPICH primary repository branch, master, updated. v3.2a1-48-g25e7fa9
by noreply＠mpich.org 06 Oct '14

06 Oct '14

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via 25e7fa93c9bc0bf8544478f3ff0d3db6de1def18 (commit) from 97ef2eeb57ef3dbad0d33da9b5e33d1735805505 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/25e7fa93c9bc0bf8544478f3ff0d3db6d… commit 25e7fa93c9bc0bf8544478f3ff0d3db6de1def18 Author: Sameh Sharkawi <sssharka(a)us.ibm.com> Date: Mon Oct 6 15:25:59 2014 -0400 pamid: Collectives using process topology segv if FCA on The non-blocking neighborhood collectives were not set up correctly in pamid. ibm (D199919) Signed-off-by: Su Huang <suhuang(a)us.ibm.com> diff --git a/src/mpid/pamid/src/comm/mpid_selectcolls.c b/src/mpid/pamid/src/comm/mpid_selectcolls.c index 3f4e247..a8c3f81 100644 --- a/src/mpid/pamid/src/comm/mpid_selectcolls.c +++ b/src/mpid/pamid/src/comm/mpid_selectcolls.c @@ -714,6 +714,11 @@ void MPIDI_Comm_coll_query(MPID_Comm *comm) comm->coll_fns->Neighbor_alltoall = MPIR_Neighbor_alltoall_default; comm->coll_fns->Neighbor_alltoallv = MPIR_Neighbor_alltoallv_default; comm->coll_fns->Neighbor_alltoallw = MPIR_Neighbor_alltoallw_default; + comm->coll_fns->Ineighbor_allgather = MPIR_Ineighbor_allgather_default; + comm->coll_fns->Ineighbor_allgatherv = MPIR_Ineighbor_allgatherv_default; + comm->coll_fns->Ineighbor_alltoall = MPIR_Ineighbor_alltoall_default; + comm->coll_fns->Ineighbor_alltoallv = MPIR_Ineighbor_alltoallv_default; + comm->coll_fns->Ineighbor_alltoallw = MPIR_Ineighbor_alltoallw_default; /* MPI-3 Support, optimized collectives hooked in */ comm->coll_fns->Ibarrier_req = MPIDO_Ibarrier; ----------------------------------------------------------------------- Summary of changes: src/mpid/pamid/src/comm/mpid_selectcolls.c | 5 +++++ 1 files changed, 5 insertions(+), 0 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.2a1-47-g97ef2ee
by noreply＠mpich.org 06 Oct '14

06 Oct '14

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via 97ef2eeb57ef3dbad0d33da9b5e33d1735805505 (commit) from 322bc78b3db3b14f7e3d3dca5b78dea6d5f48143 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/97ef2eeb57ef3dbad0d33da9b5e33d173… commit 97ef2eeb57ef3dbad0d33da9b5e33d1735805505 Author: Pavan Balaji <balaji(a)anl.gov> Date: Sun Oct 5 22:54:39 2014 -0500 Bug-fix: Expose sufficient memory as public. We were only exposing a part of the allocated buffer as public memory, but attempting to use all of it remotely. Thanks to Charles Archer @ Intel for reporting the error and contributing the fix. Signed-off-by: Min Si <msi(a)il.is.s.u-tokyo.ac.jp> diff --git a/test/mpi/rma/lockcontention3.c b/test/mpi/rma/lockcontention3.c index d70e26c..015eed6 100644 --- a/test/mpi/rma/lockcontention3.c +++ b/test/mpi/rma/lockcontention3.c @@ -123,7 +123,7 @@ int main( int argc, char *argv[] ) srcbuf = malloc(RMA_SIZE*sizeof(*srcbuf)); assert(srcbuf); - MPI_Win_create( rmabuffer, bufsize, sizeof(int), MPI_INFO_NULL, + MPI_Win_create( rmabuffer, bufsize * sizeof(int), sizeof(int), MPI_INFO_NULL, MPI_COMM_WORLD, &win ); /* Run a sequence of tests */ ----------------------------------------------------------------------- Summary of changes: test/mpi/rma/lockcontention3.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.2a1-46-g322bc78
by noreply＠mpich.org 05 Oct '14

05 Oct '14

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via 322bc78b3db3b14f7e3d3dca5b78dea6d5f48143 (commit) from 60cbd3c540485392669ad07547ca00ffa00c91c1 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/322bc78b3db3b14f7e3d3dca5b78dea6d… commit 322bc78b3db3b14f7e3d3dca5b78dea6d5f48143 Author: Wesley Bland <wbland(a)anl.gov> Date: Sun Oct 5 20:54:34 2014 -0500 Revert "Turn on some of the FT tests that are now passing" This reverts commit 32f3a924162eeaf9fb322cbb0ef70497c70d22d9. diff --git a/test/mpi/ft/testlist b/test/mpi/ft/testlist index 054dfd0..f26bd47 100644 --- a/test/mpi/ft/testlist +++ b/test/mpi/ft/testlist @@ -1,12 +1,12 @@ -die 4 mpiexecarg=-disable-auto-cleanup timeLimit=10 strict=false resultTest=TestStatusNoErrors +die 4 mpiexecarg=-disable-auto-cleanup timeLimit=10 strict=false resultTest=TestStatusNoErrors xfail=ticket1945 abort 2 mpiexecarg=-disable-auto-cleanup timeLimit=10 strict=false xfail=ticket1537 -sendalive 4 mpiexecarg=-disable-auto-cleanup timeLimit=10 strict=false resultTest=TestStatusNoErrors -isendalive 3 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false -multi_isendalive 4 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false +sendalive 4 mpiexecarg=-disable-auto-cleanup timeLimit=10 strict=false resultTest=TestStatusNoErrors xfail=ticket1945 +isendalive 3 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false xfail=ticket1945 +multi_isendalive 4 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false xfail=ticket1945 senddead 2 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false timeLimit=10 xfail=ticket1945 -recvdead 2 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false timeLimit=10 +recvdead 2 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false timeLimit=10 xfail=ticket1945 isenddead 2 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false timeLimit=10 xfail=ticket1945 -irecvdead 2 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false timeLimit=10 +irecvdead 2 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false timeLimit=10 xfail=ticket1945 barrier 4 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false timeLimit=10 xfail=ticket1945 gather 4 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false timeLimit=10 xfail=ticket1945 reduce 4 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false timeLimit=10 xfail=ticket1945 ----------------------------------------------------------------------- Summary of changes: test/mpi/ft/testlist | 12 ++++++------ 1 files changed, 6 insertions(+), 6 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.2a1-45-g60cbd3c
by noreply＠mpich.org 04 Oct '14

04 Oct '14

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via 60cbd3c540485392669ad07547ca00ffa00c91c1 (commit) from 051449e7b5421b006556ba4e8c6f3300b6efed06 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/60cbd3c540485392669ad07547ca00ffa… commit 60cbd3c540485392669ad07547ca00ffa00c91c1 Author: Ken Raffenetti <raffenet(a)mcs.anl.gov> Date: Fri Oct 3 12:18:37 2014 -0500 portals4: invalidate ME handle before request dequeue Invalidate the handle of an ME that has been matched and handled in the portals4 netmod *before* dequeuing the request. The prevents potential double unlinking failures. Signed-off-by: Antonio J. Pena <apenya(a)mcs.anl.gov> diff --git a/src/mpid/ch3/channels/nemesis/netmod/portals4/ptl_recv.c b/src/mpid/ch3/channels/nemesis/netmod/portals4/ptl_recv.c index ea94581..005b678 100644 --- a/src/mpid/ch3/channels/nemesis/netmod/portals4/ptl_recv.c +++ b/src/mpid/ch3/channels/nemesis/netmod/portals4/ptl_recv.c @@ -16,6 +16,10 @@ static void dequeue_req(const ptl_event_t *e) MPID_Request *const rreq = e->user_ptr; int s_len, r_len; + /* At this point we know the ME is unlinked. Invalidate the handle to + prevent further accesses, e.g. an attempted cancel. */ + REQ_PTL(rreq)->me = PTL_INVALID_HANDLE; + found = MPIDI_CH3U_Recvq_DP(rreq); MPIU_Assert(found); @@ -23,10 +27,6 @@ static void dequeue_req(const ptl_event_t *e) rreq->status.MPI_SOURCE = NPTL_MATCH_GET_RANK(e->match_bits); rreq->status.MPI_TAG = NPTL_MATCH_GET_TAG(e->match_bits); - /* At this point we know the ME is unlinked. Invalidate the handle to - prevent further accesses, e.g. an attempted cancel. */ - REQ_PTL(rreq)->me = PTL_INVALID_HANDLE; - MPID_Datatype_get_size_macro(rreq->dev.datatype, r_len); r_len *= rreq->dev.user_count; ----------------------------------------------------------------------- Summary of changes: .../channels/nemesis/netmod/portals4/ptl_recv.c | 8 ++++---- 1 files changed, 4 insertions(+), 4 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.2a1-44-g051449e
by noreply＠mpich.org 03 Oct '14

03 Oct '14

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via 051449e7b5421b006556ba4e8c6f3300b6efed06 (commit) from 818b6a61a2f4653dc12da9c14c46ab866ba2a02b (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/051449e7b5421b006556ba4e8c6f3300b… commit 051449e7b5421b006556ba4e8c6f3300b6efed06 Author: Igor Ivanov <Igor.Ivanov(a)itseez.com> Date: Fri Sep 5 18:18:49 2014 +0300 mpi/comm: Fix MPI_Intercomm_merge Previous code is based on a trick to get true context id using invalid communicator. There is a call of MPIR_Get_contextid() for new communicator that is not completelly built (no comm_commit and comm_hook calls). netmod/mxm uses comm_hook and can not work with this trick. This changes allow to avoid call of invalid communicator using temporary (intermediate) communicator. Signed-off-by: Igor Ivanov <Igor.Ivanov(a)itseez.com> Signed-off-by: Pavan Balaji <balaji(a)anl.gov> diff --git a/src/mpi/comm/intercomm_merge.c b/src/mpi/comm/intercomm_merge.c index 90b7c15..e12d991 100644 --- a/src/mpi/comm/intercomm_merge.c +++ b/src/mpi/comm/intercomm_merge.c @@ -20,6 +20,62 @@ int MPI_Intercomm_merge(MPI_Comm intercomm, int high, MPI_Comm *newintracomm) __ #endif /* -- End Profiling Symbol Block */ +/* These functions help implement the merge procedure */ +static int MPIR_Intercomm_merge_create_and_map_vcrt(MPID_Comm *comm_ptr, int local_high, MPID_Comm *new_intracomm_ptr); + + +/* This function creates VCRT for new communicator + * basing on VCRT of existing communicator. + */ + +#undef FUNCNAME +#define FUNCNAME MPIR_Intercomm_merge_create_and_map_vcrt +#undef FCNAME +#define FCNAME MPIDI_QUOTE(FUNCNAME) +static int MPIR_Intercomm_merge_create_and_map_vcrt(MPID_Comm *comm_ptr, int local_high, MPID_Comm *new_intracomm_ptr) +{ + int mpi_errno = MPI_SUCCESS; + int i, j; + + /* Now we know which group comes first. Build the new vcr + from the existing vcrs */ + mpi_errno = MPID_VCRT_Create( comm_ptr->local_size + comm_ptr->remote_size, &new_intracomm_ptr->vcrt ); + if (mpi_errno) MPIU_ERR_POP(mpi_errno); + + mpi_errno = MPID_VCRT_Get_ptr( new_intracomm_ptr->vcrt, &new_intracomm_ptr->vcr ); + if (mpi_errno) MPIU_ERR_POP(mpi_errno); + + if (local_high) { + /* remote group first */ + j = 0; + for (i = 0; i < comm_ptr->remote_size; i++) { + mpi_errno = MPID_VCR_Dup( comm_ptr->vcr[i], &new_intracomm_ptr->vcr[j++] ); + if (mpi_errno) MPIU_ERR_POP(mpi_errno); + } + for (i = 0; i < comm_ptr->local_size; i++) { + if (i == comm_ptr->rank) new_intracomm_ptr->rank = j; + mpi_errno = MPID_VCR_Dup( comm_ptr->local_vcr[i], &new_intracomm_ptr->vcr[j++] ); + if (mpi_errno) MPIU_ERR_POP(mpi_errno); + } + } + else { + /* local group first */ + j = 0; + for (i = 0; i < comm_ptr->local_size; i++) { + if (i == comm_ptr->rank) new_intracomm_ptr->rank = j; + mpi_errno = MPID_VCR_Dup( comm_ptr->local_vcr[i], &new_intracomm_ptr->vcr[j++] ); + if (mpi_errno) MPIU_ERR_POP(mpi_errno); + } + for (i = 0; i < comm_ptr->remote_size; i++) { + mpi_errno = MPID_VCR_Dup( comm_ptr->vcr[i], &new_intracomm_ptr->vcr[j++] ); + if (mpi_errno) MPIU_ERR_POP(mpi_errno); + } + } + +fn_fail: + return mpi_errno; +} + /* Define MPICH_MPI_FROM_PMPI if weak symbols are not supported to build the MPI routines */ #ifndef MPICH_MPI_FROM_PMPI @@ -33,7 +89,7 @@ int MPI_Intercomm_merge(MPI_Comm intercomm, int high, MPI_Comm *newintracomm) __ int MPIR_Intercomm_merge_impl(MPID_Comm *comm_ptr, int high, MPID_Comm **new_intracomm_ptr) { int mpi_errno = MPI_SUCCESS; - int local_high, remote_high, i, j, new_size; + int local_high, remote_high, new_size; MPIR_Context_id_t new_context_id; int errflag = FALSE; MPID_MPI_STATE_DECL(MPID_STATE_MPIR_INTERCOMM_MERGE_IMPL); @@ -118,44 +174,17 @@ int MPIR_Intercomm_merge_impl(MPID_Comm *comm_ptr, int high, MPID_Comm **new_int /* Now we know which group comes first. Build the new vcr from the existing vcrs */ - mpi_errno = MPID_VCRT_Create( new_size, &(*new_intracomm_ptr)->vcrt ); - if (mpi_errno) MPIU_ERR_POP(mpi_errno); - - mpi_errno = MPID_VCRT_Get_ptr( (*new_intracomm_ptr)->vcrt, &(*new_intracomm_ptr)->vcr ); + mpi_errno = MPIR_Intercomm_merge_create_and_map_vcrt(comm_ptr, local_high, (*new_intracomm_ptr)); if (mpi_errno) MPIU_ERR_POP(mpi_errno); - - if (local_high) { - /* remote group first */ - j = 0; - for (i = 0; i < comm_ptr->remote_size; i++) { - mpi_errno = MPID_VCR_Dup( comm_ptr->vcr[i], &(*new_intracomm_ptr)->vcr[j++] ); - if (mpi_errno) MPIU_ERR_POP(mpi_errno); - } - for (i = 0; i < comm_ptr->local_size; i++) { - if (i == comm_ptr->rank) (*new_intracomm_ptr)->rank = j; - mpi_errno = MPID_VCR_Dup( comm_ptr->local_vcr[i], &(*new_intracomm_ptr)->vcr[j++] ); - if (mpi_errno) MPIU_ERR_POP(mpi_errno); - } - } - else { - /* local group first */ - j = 0; - for (i = 0; i < comm_ptr->local_size; i++) { - if (i == comm_ptr->rank) (*new_intracomm_ptr)->rank = j; - mpi_errno = MPID_VCR_Dup( comm_ptr->local_vcr[i], &(*new_intracomm_ptr)->vcr[j++] ); - if (mpi_errno) MPIU_ERR_POP(mpi_errno); - } - for (i = 0; i < comm_ptr->remote_size; i++) { - mpi_errno = MPID_VCR_Dup( comm_ptr->vcr[i], &(*new_intracomm_ptr)->vcr[j++] ); - if (mpi_errno) MPIU_ERR_POP(mpi_errno); - } - } /* We've setup a temporary context id, based on the context id used by the intercomm. This allows us to perform the allreduce operations within the context id algorithm, since we already have a valid (almost - see comm_create_hook) communicator. */ + mpi_errno = MPIR_Comm_commit((*new_intracomm_ptr)); + if (mpi_errno) MPIU_ERR_POP(mpi_errno); + /* printf( "About to get context id \n" ); fflush( stdout ); */ /* In the multi-threaded case, MPIR_Get_contextid assumes that the calling routine already holds the single criticial section */ @@ -164,9 +193,24 @@ int MPIR_Intercomm_merge_impl(MPID_Comm *comm_ptr, int high, MPID_Comm **new_int if (mpi_errno) MPIU_ERR_POP(mpi_errno); MPIU_Assert(new_context_id != 0); + /* We release this communicator that was involved just to + * get valid context id and create true one + */ + mpi_errno = MPIR_Comm_release( (*new_intracomm_ptr),0); + if (mpi_errno) MPIU_ERR_POP(mpi_errno); + + mpi_errno = MPIR_Comm_create( new_intracomm_ptr ); + if (mpi_errno) MPIU_ERR_POP(mpi_errno); + + (*new_intracomm_ptr)->remote_size = (*new_intracomm_ptr)->local_size = new_size; + (*new_intracomm_ptr)->rank = -1; + (*new_intracomm_ptr)->comm_kind = MPID_INTRACOMM; (*new_intracomm_ptr)->context_id = new_context_id; (*new_intracomm_ptr)->recvcontext_id = new_context_id; + mpi_errno = MPIR_Intercomm_merge_create_and_map_vcrt(comm_ptr, local_high, (*new_intracomm_ptr)); + if (mpi_errno) MPIU_ERR_POP(mpi_errno); + mpi_errno = MPIR_Comm_commit((*new_intracomm_ptr)); if (mpi_errno) MPIU_ERR_POP(mpi_errno); ----------------------------------------------------------------------- Summary of changes: src/mpi/comm/intercomm_merge.c | 108 ++++++++++++++++++++++++++++------------ 1 files changed, 76 insertions(+), 32 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.2a1-43-g818b6a6
by noreply＠mpich.org 03 Oct '14

03 Oct '14

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via 818b6a61a2f4653dc12da9c14c46ab866ba2a02b (commit) from e4fb945f8c0ebff106d3c5758420c9311274e148 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/818b6a61a2f4653dc12da9c14c46ab866… commit 818b6a61a2f4653dc12da9c14c46ab866ba2a02b Author: halim amer <halim.amer(a)gmail.com> Date: Fri Sep 5 14:43:31 2014 -0500 Added an example for checking on which CPU a process is running on. This example does not rely on getcpu() or sched_getcpu() which are not available on old kernels/glibc. Signed-off-by: Ken Raffenetti <raffenet(a)mcs.anl.gov> diff --git a/src/pm/hydra/examples/print_process_bind.c b/src/pm/hydra/examples/print_process_bind.c new file mode 100644 index 0000000..7bc37e6 --- /dev/null +++ b/src/pm/hydra/examples/print_process_bind.c @@ -0,0 +1,66 @@ +/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil ; -*- */ +/* + * + * (C) 2014 by Argonne National Laboratory. + * See COPYRIGHT in top-level directory. + */ +#define PATH_MAX 100 + +#include "mpi.h" +#include <stdio.h> +#include <sys/types.h> +#include <unistd.h> +#include <string.h> + +/* This example can be used to check on which node+CPU + * each process is running on. This can serve to confirm + * The correctness of a Process-CPU binding policy for instance. + * + * To get the CPU Id on which a process is running, + * one can use the getcpu() or sched_getcpu(). + * However, getcpu() is only supported by Linux kernels + * >= 2.6.19 for x86_64 and i386 architectures. Also, + * sched_getcpu() is only available since glibc 2.6. + * + * This example implements an alternative, for systems + * not satisfying these requirements, by checking the column + * 39 of /proc/$pid/stat. + * + */ + +int main(int argc, char **argv) +{ + int rank, size, namelen; + FILE *fp; + char path[PATH_MAX]; + char processor_name[MPI_MAX_PROCESSOR_NAME]; + + MPI_Init(&argc, &argv); + MPI_Comm_rank(MPI_COMM_WORLD, &rank); + + MPI_Get_processor_name(processor_name, &namelen); + + char shell_cmd[100]; + sprintf(shell_cmd, "cat /proc/%d/stat | awk '{print $39}'", getpid()); + + if (rank == 0) { + printf("----------------------\n"); + printf(" PROCESS/CPU BINDING\n"); + printf("----------------------\n"); + } + + fp = popen(shell_cmd, "r"); + + while (fgets(path, PATH_MAX, fp) != NULL) { + printf("%s[%d]: running on CPU %s", processor_name, rank, path); + } + + pclose(fp); + + fflush(stdout); + + + MPI_Finalize(); + + return 0; +} ----------------------------------------------------------------------- Summary of changes: src/pm/hydra/examples/print_process_bind.c | 66 ++++++++++++++++++++++++++++ 1 files changed, 66 insertions(+), 0 deletions(-) create mode 100644 src/pm/hydra/examples/print_process_bind.c hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.2a1-42-ge4fb945
by noreply＠mpich.org 03 Oct '14

03 Oct '14

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via e4fb945f8c0ebff106d3c5758420c9311274e148 (commit) from 32f3a924162eeaf9fb322cbb0ef70497c70d22d9 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/e4fb945f8c0ebff106d3c5758420c9311… commit e4fb945f8c0ebff106d3c5758420c9311274e148 Author: Igor Ivanov <Igor.Ivanov(a)itseez.com> Date: Fri Oct 3 12:50:19 2014 +0300 netmode/mxm: Fix issue in anysource match - Fix issue in anysource match; - Improve debug output; Signed-off-by: Igor Ivanov <Igor.Ivanov(a)itseez.com> Signed-off-by: Pavan Balaji <balaji(a)anl.gov> diff --git a/src/mpid/ch3/channels/nemesis/netmod/mxm/mxm_cancel.c b/src/mpid/ch3/channels/nemesis/netmod/mxm/mxm_cancel.c index 202d149..ec96b7f 100644 --- a/src/mpid/ch3/channels/nemesis/netmod/mxm/mxm_cancel.c +++ b/src/mpid/ch3/channels/nemesis/netmod/mxm/mxm_cancel.c @@ -69,7 +69,7 @@ int MPID_nem_mxm_cancel_recv(MPIDI_VC_t * vc, MPID_Request * req) _dbg_mxm_out_req(req); fn_exit: - /* This function returns sero in case request is canceled + /* This function returns zero in case request is canceled * and nonzero otherwise */ return (!MPIR_STATUS_GET_CANCEL_BIT(req->status)); diff --git a/src/mpid/ch3/channels/nemesis/netmod/mxm/mxm_poll.c b/src/mpid/ch3/channels/nemesis/netmod/mxm/mxm_poll.c index 0defcbb..ba7686e 100644 --- a/src/mpid/ch3/channels/nemesis/netmod/mxm/mxm_poll.c +++ b/src/mpid/ch3/channels/nemesis/netmod/mxm/mxm_poll.c @@ -125,11 +125,8 @@ int MPID_nem_mxm_anysource_matched(MPID_Request * req) req_area = REQ_BASE(req); ret = mxm_req_cancel_recv(&req_area->mxm_req->item.recv); - if ((MXM_OK == ret) || (MXM_ERR_NO_PROGRESS == ret)) { - MPID_Segment_free(req->dev.segment_ptr); - } - else { - _mxm_req_wait(&req_area->mxm_req->item.base); + mxm_req_wait(&req_area->mxm_req->item.base); + if ((MXM_OK != ret) && (MXM_ERR_NO_PROGRESS != ret)) { matched = TRUE; } @@ -173,7 +170,7 @@ int MPID_nem_mxm_recv(MPIDI_VC_t * vc, MPID_Request * rreq) rreq->ch.noncontig = FALSE; _dbg_mxm_output(5, - "Recv ========> Getting USER msg for req %p (context %d rank %d tag %d size %d) \n", + "Recv ========> Getting USER msg for req %p (context %d from %d tag %d size %d) \n", rreq, context_id, rreq->dev.match.parts.rank, tag, data_sz); vc_area = VC_BASE(vc); @@ -244,7 +241,6 @@ static int _mxm_handle_rreq(MPID_Request * req) vc_area = VC_BASE(req->ch.vc); req_area = REQ_BASE(req); - _dbg_mxm_output(5, "========> Completing RECV req %p status %d\n", req, req->status.MPI_ERROR); _dbg_mxm_out_buf(req_area->iov_buf[0].ptr, (req_area->iov_buf[0].length > 16 ? 16 : req_area->iov_buf[0].length)); @@ -297,8 +293,6 @@ static int _mxm_handle_rreq(MPID_Request * req) iov = MPIU_Malloc(n_iov * sizeof(*iov)); MPIU_Assert(iov); - n_iov = req_area->iov_count; - iov_buf = req_area->iov_buf; for (index = 0; index < n_iov; index++) { iov[index].MPID_IOV_BUF = iov_buf[index].ptr; iov[index].MPID_IOV_LEN = iov_buf[index].length; @@ -360,6 +354,10 @@ static void _mxm_recv_completion_cb(void *context) list_enqueue(&mxm_obj->free_queue, &req_area->mxm_req->queue); } + _dbg_mxm_output(5, "========> %s RECV req %p status %d\n", + (MPIR_STATUS_GET_CANCEL_BIT(req->status) ? "Canceling" : "Completing"), + req, req->status.MPI_ERROR); + if (likely(!MPIR_STATUS_GET_CANCEL_BIT(req->status))) { _mxm_handle_rreq(req); } @@ -411,12 +409,7 @@ static int _mxm_irecv(MPID_nem_mxm_ep_t * ep, MPID_nem_mxm_req_area * req, int i ret = mxm_req_recv(mxm_rreq); if (MXM_OK != ret) { - if (ep) { - list_enqueue(&ep->free_queue, &req->mxm_req->queue); - } - else { - list_enqueue(&mxm_obj->free_queue, &req->mxm_req->queue); - } + list_enqueue(free_queue, &req->mxm_req->queue); mpi_errno = MPI_ERR_OTHER; goto fn_fail; } diff --git a/src/mpid/ch3/channels/nemesis/netmod/mxm/mxm_send.c b/src/mpid/ch3/channels/nemesis/netmod/mxm/mxm_send.c index f91edf3..b817459 100644 --- a/src/mpid/ch3/channels/nemesis/netmod/mxm/mxm_send.c +++ b/src/mpid/ch3/channels/nemesis/netmod/mxm/mxm_send.c @@ -248,8 +248,8 @@ int MPID_nem_mxm_send(MPIDI_VC_t * vc, const void *buf, int count, MPI_Datatype sreq->ch.noncontig = FALSE; _dbg_mxm_output(5, - "Send ========> Sending USER msg for req %p (context %d rank %d tag %d size %d) \n", - sreq, comm->context_id + context_offset, comm->rank, tag, data_sz); + "Send ========> Sending USER msg for req %p (context %d to %d tag %d size %d) \n", + sreq, comm->context_id + context_offset, rank, tag, data_sz); vc_area = VC_BASE(vc); req_area = REQ_BASE(sreq); @@ -350,8 +350,8 @@ int MPID_nem_mxm_ssend(MPIDI_VC_t * vc, const void *buf, int count, MPI_Datatype sreq->ch.noncontig = FALSE; _dbg_mxm_output(5, - "sSend ========> Sending USER msg for req %p (context %d rank %d tag %d size %d) \n", - sreq, comm->context_id + context_offset, comm->rank, tag, data_sz); + "sSend ========> Sending USER msg for req %p (context %d to %d tag %d size %d) \n", + sreq, comm->context_id + context_offset, rank, tag, data_sz); vc_area = VC_BASE(vc); req_area = REQ_BASE(sreq); @@ -452,8 +452,8 @@ int MPID_nem_mxm_isend(MPIDI_VC_t * vc, const void *buf, int count, MPI_Datatype sreq->ch.noncontig = FALSE; _dbg_mxm_output(5, - "iSend ========> Sending USER msg for req %p (context %d rank %d tag %d size %d) \n", - sreq, comm->context_id + context_offset, comm->rank, tag, data_sz); + "iSend ========> Sending USER msg for req %p (context %d to %d tag %d size %d) \n", + sreq, comm->context_id + context_offset, rank, tag, data_sz); vc_area = VC_BASE(vc); req_area = REQ_BASE(sreq); @@ -555,8 +555,8 @@ int MPID_nem_mxm_issend(MPIDI_VC_t * vc, const void *buf, int count, MPI_Datatyp sreq->ch.noncontig = FALSE; _dbg_mxm_output(5, - "isSend ========> Sending USER msg for req %p (context %d rank %d tag %d size %d) \n", - sreq, comm->context_id + context_offset, comm->rank, tag, data_sz); + "isSend ========> Sending USER msg for req %p (context %d to %d tag %d size %d) \n", + sreq, comm->context_id + context_offset, rank, tag, data_sz); vc_area = VC_BASE(vc); req_area = REQ_BASE(sreq); @@ -629,7 +629,6 @@ static int _mxm_handle_sreq(MPID_Request * req) vc_area = VC_BASE(req->ch.vc); req_area = REQ_BASE(req); - _dbg_mxm_output(5, "========> Completing SEND req %p status %d\n", req, req->status.MPI_ERROR); _dbg_mxm_out_buf(req_area->iov_buf[0].ptr, (req_area->iov_buf[0].length > 16 ? 16 : req_area->iov_buf[0].length)); @@ -679,6 +678,10 @@ static void _mxm_send_completion_cb(void *context) list_enqueue(&vc_area->mxm_ep->free_queue, &req_area->mxm_req->queue); + _dbg_mxm_output(5, "========> %s SEND req %p status %d\n", + (MPIR_STATUS_GET_CANCEL_BIT(req->status) ? "Canceling" : "Completing"), + req, req->status.MPI_ERROR); + if (likely(!MPIR_STATUS_GET_CANCEL_BIT(req->status))) { _mxm_handle_sreq(req); } @@ -750,7 +753,7 @@ static int _mxm_isend(MPID_nem_mxm_ep_t * ep, MPID_nem_mxm_req_area * req, ret = mxm_req_send(mxm_sreq); if (MXM_OK != ret) { - list_enqueue(&ep->free_queue, &req->mxm_req->queue); + list_enqueue(free_queue, &req->mxm_req->queue); mpi_errno = MPI_ERR_OTHER; goto fn_fail; } ----------------------------------------------------------------------- Summary of changes: .../ch3/channels/nemesis/netmod/mxm/mxm_cancel.c | 2 +- .../ch3/channels/nemesis/netmod/mxm/mxm_poll.c | 23 +++++++------------- .../ch3/channels/nemesis/netmod/mxm/mxm_send.c | 23 +++++++++++-------- 3 files changed, 22 insertions(+), 26 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.2a1-41-g32f3a92
by noreply＠mpich.org 03 Oct '14

03 Oct '14

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via 32f3a924162eeaf9fb322cbb0ef70497c70d22d9 (commit) from b338d7a68703c410ca3eb30f484769636bbc52a2 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/32f3a924162eeaf9fb322cbb0ef70497c… commit 32f3a924162eeaf9fb322cbb0ef70497c70d22d9 Author: Wesley Bland <wbland(a)anl.gov> Date: Fri Oct 3 08:58:03 2014 -0500 Turn on some of the FT tests that are now passing Signed-off-by: Rob Latham <robl(a)mcs.anl.gov> diff --git a/test/mpi/ft/testlist b/test/mpi/ft/testlist index f26bd47..054dfd0 100644 --- a/test/mpi/ft/testlist +++ b/test/mpi/ft/testlist @@ -1,12 +1,12 @@ -die 4 mpiexecarg=-disable-auto-cleanup timeLimit=10 strict=false resultTest=TestStatusNoErrors xfail=ticket1945 +die 4 mpiexecarg=-disable-auto-cleanup timeLimit=10 strict=false resultTest=TestStatusNoErrors abort 2 mpiexecarg=-disable-auto-cleanup timeLimit=10 strict=false xfail=ticket1537 -sendalive 4 mpiexecarg=-disable-auto-cleanup timeLimit=10 strict=false resultTest=TestStatusNoErrors xfail=ticket1945 -isendalive 3 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false xfail=ticket1945 -multi_isendalive 4 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false xfail=ticket1945 +sendalive 4 mpiexecarg=-disable-auto-cleanup timeLimit=10 strict=false resultTest=TestStatusNoErrors +isendalive 3 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false +multi_isendalive 4 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false senddead 2 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false timeLimit=10 xfail=ticket1945 -recvdead 2 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false timeLimit=10 xfail=ticket1945 +recvdead 2 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false timeLimit=10 isenddead 2 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false timeLimit=10 xfail=ticket1945 -irecvdead 2 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false timeLimit=10 xfail=ticket1945 +irecvdead 2 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false timeLimit=10 barrier 4 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false timeLimit=10 xfail=ticket1945 gather 4 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false timeLimit=10 xfail=ticket1945 reduce 4 mpiexecarg=-disable-auto-cleanup resultTest=TestStatusNoErrors strict=false timeLimit=10 xfail=ticket1945 ----------------------------------------------------------------------- Summary of changes: test/mpi/ft/testlist | 12 ++++++------ 1 files changed, 6 insertions(+), 6 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.2a1-40-gb338d7a
by noreply＠mpich.org 02 Oct '14

02 Oct '14

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via b338d7a68703c410ca3eb30f484769636bbc52a2 (commit) from 3a4d8eb15a3db9912f62e2883e2a3c72f0ee88e1 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/b338d7a68703c410ca3eb30f484769636… commit b338d7a68703c410ca3eb30f484769636bbc52a2 Author: William Gropp <wgropp(a)illinois.edu> Date: Wed Oct 1 13:37:05 2014 -0500 Create doctext index and suppress location Signed-off-by: Junchao Zhang <jczhang(a)mcs.anl.gov> diff --git a/Makefile.am b/Makefile.am index 4bbe965..e95edee 100644 --- a/Makefile.am +++ b/Makefile.am @@ -376,23 +376,30 @@ mandoc_path3=$(abs_top_builddir)/man/man3 htmldoc_path1=$(abs_top_builddir)/www/www1 htmldoc_path3=$(abs_top_builddir)/www/www3 doctext_docnotes=$(abs_top_srcdir)/maint/docnotes +# Provide an easily replaced url root for the generated index file. +# You can override this with URL desired in the index file generated by doctext. +# You can ignore this if you don't use mapnames or tohtml to add links +# to the MPI manual pages to documents. +htmldoc_root3="--your-url-here--" # Implicit suffix rules to cause the docs to be built. These are a bit of a # hack, but using the implicit rules like this instead of a shell for loop # permits make -jN to work. .c.man-phony: $(doctextman_verbose)$(DOCTEXT) -man -mpath $(mandoc_path3) -ext 3 \ - -heading MPI -quotefmt $(doctext_docnotes) $< + -heading MPI -quotefmt -nolocation $(doctext_docnotes) $< .c.html-phony: $(doctexthtml_verbose)$(DOCTEXT) -html -mpath $(htmldoc_path3) \ - -heading MPI -quotefmt $(doctext_docnotes) $< + -heading MPI -quotefmt -nolocation \ + -index $(htmldoc_path3)/mpi.cit -indexdir $(htmldoc_root3) \ + $(doctext_docnotes) $< .txt.man1-phony: $(doctextman_verbose)$(DOCTEXT) -man -mpath $(mandoc_path1) -ext 1 \ - -heading MPI -quotefmt $(doctext_docnotes) $< + -heading MPI -quotefmt -nolocation $(doctext_docnotes) $< .txt.html1-phony: $(doctexthtml_verbose)$(DOCTEXT) -html -mpath $(htmldoc_path1) \ - -heading MPI -quotefmt $(doctext_docnotes) $< + -heading MPI -quotefmt -nolocation $(doctext_docnotes) $< # use mandoc-local target to force directory creation before running DOCTEXT mandoc: ----------------------------------------------------------------------- Summary of changes: Makefile.am | 15 +++++++++++---- 1 files changed, 11 insertions(+), 4 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.2a1-39-g3a4d8eb
by noreply＠mpich.org 01 Oct '14

01 Oct '14

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via 3a4d8eb15a3db9912f62e2883e2a3c72f0ee88e1 (commit) via def275e90056f7af6c9566390c3820193df6c36b (commit) via 5bf646a352c0867aa79530da0ac856f100faf05f (commit) via c673fec42e015e9f714374dcb5d58528925663a1 (commit) via d6996640a0e50e932b105064fa026c53633cf88c (commit) from eda105b029572ebf7f4008f6de2e1afb9ccf6fc2 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/3a4d8eb15a3db9912f62e2883e2a3c72f… commit 3a4d8eb15a3db9912f62e2883e2a3c72f0ee88e1 Author: Xin Zhao <xinzhao3(a)illinois.edu> Date: Wed Oct 1 09:04:29 2014 -0500 Decrease iteration number in manyrma2 test. This test sometimes causes TIMEOUT in Jenkins TCP test. Signed-off-by: Pavan Balaji <balaji(a)anl.gov> diff --git a/test/mpi/rma/manyrma2.c b/test/mpi/rma/manyrma2.c index eb7f860..6773919 100644 --- a/test/mpi/rma/manyrma2.c +++ b/test/mpi/rma/manyrma2.c @@ -17,7 +17,7 @@ #define MAX_COUNT 65536*4/16 #define MAX_RMA_SIZE 2 /* 16 in manyrma performance test */ -#define MAX_RUNS 10 +#define MAX_RUNS 8 #define MAX_ITER_TIME 5.0 /* seconds */ typedef enum { SYNC_NONE = 0, http://git.mpich.org/mpich.git/commitdiff/def275e90056f7af6c9566390c3820193… commit def275e90056f7af6c9566390c3820193df6c36b Author: Xin Zhao <xinzhao3(a)illinois.edu> Date: Tue Sep 30 17:17:33 2014 -0500 Bug-fix: correct SHM barrier in MPI_Win_fence. Here we move the SHM barrier in MPI_Win_fence after the completion of coming-in RMA operations, so that following coming-in RMA operations on SHM in the next FENCE epoch will see the updated value. Signed-off-by: Pavan Balaji <balaji(a)anl.gov> diff --git a/src/mpid/ch3/src/ch3u_rma_sync.c b/src/mpid/ch3/src/ch3u_rma_sync.c index 8e1353c..4cbf25c 100644 --- a/src/mpid/ch3/src/ch3u_rma_sync.c +++ b/src/mpid/ch3/src/ch3u_rma_sync.c @@ -1170,28 +1170,6 @@ int MPIDI_Win_fence(int assert, MPID_Win *win_ptr) MPIR_T_PVAR_TIMER_END(RMA, rma_winfence_clearlock); } - if (!(assert & MPI_MODE_NOSUCCEED) && - (assert & MPI_MODE_NOPRECEDE || win_ptr->fence_issued == 0)) { - - /* In the FENCE that opens an epoch but does not close an epoch, - if SHM is allocated, perform a barrier among processes on the - same node, to prevent one process modifying another process's - memory before that process starts an epoch. */ - - if (win_ptr->shm_allocated == TRUE) { - MPID_Comm *node_comm_ptr = win_ptr->comm_ptr->node_comm; - - /* Ensure ordering of load/store operations. */ - OPA_read_write_barrier(); - - mpi_errno = MPIR_Barrier_impl(node_comm_ptr, &errflag); - if (mpi_errno) {goto fn_fail;} - - /* Ensure ordering of load/store operations. */ - OPA_read_write_barrier(); - } - } - /* Note that the NOPRECEDE and NOSUCCEED must be specified by all processes in the window's group if any specify it */ if (assert & MPI_MODE_NOPRECEDE) @@ -1201,7 +1179,7 @@ int MPIDI_Win_fence(int assert, MPID_Win *win_ptr) mpi_errno, MPI_ERR_RMA_SYNC, "**rmasync"); win_ptr->fence_issued = (assert & MPI_MODE_NOSUCCEED) ? 0 : 1; - goto fn_exit; + goto shm_barrier; } if (win_ptr->fence_issued == 0) @@ -1407,6 +1385,28 @@ int MPIDI_Win_fence(int assert, MPID_Win *win_ptr) win_ptr->epoch_state = MPIDI_EPOCH_NONE; } + shm_barrier: + if (!(assert & MPI_MODE_NOSUCCEED)) { + /* In a FENCE without MPI_MODE_NOSUCCEED (which means this FENCE + might start a new Active epoch), if SHM is allocated, perform + a barrier among processes on the same node, to prevent one + process modifying another process's memory before that process + starts an epoch. */ + + if (win_ptr->shm_allocated == TRUE) { + MPID_Comm *node_comm_ptr = win_ptr->comm_ptr->node_comm; + + /* Ensure ordering of load/store operations. */ + OPA_read_write_barrier(); + + mpi_errno = MPIR_Barrier_impl(node_comm_ptr, &errflag); + if (mpi_errno) {goto fn_fail;} + + /* Ensure ordering of load/store operations. */ + OPA_read_write_barrier(); + } + } + fn_exit: MPIU_CHKLMEM_FREEALL(); MPIDI_RMA_FUNC_EXIT(MPID_STATE_MPIDI_WIN_FENCE); http://git.mpich.org/mpich.git/commitdiff/5bf646a352c0867aa79530da0ac856f10… commit 5bf646a352c0867aa79530da0ac856f100faf05f Author: Xin Zhao <xinzhao3(a)illinois.edu> Date: Tue Sep 30 17:03:23 2014 -0500 Modify variable name my_counter to at_completion_counter. at_completion_counter is used to indicate if all Active Target operations have completed on this target. Signed-off-by: Pavan Balaji <balaji(a)anl.gov> diff --git a/src/mpid/ch3/include/mpidpre.h b/src/mpid/ch3/include/mpidpre.h index 76ef00f..4cf9d32 100644 --- a/src/mpid/ch3/include/mpidpre.h +++ b/src/mpid/ch3/include/mpidpre.h @@ -262,7 +262,7 @@ struct MPIDI_Win_target_state { }; #define MPIDI_DEV_WIN_DECL \ - volatile int my_counter; /* completion counter for operations \ + volatile int at_completion_counter; /* completion counter for operations \ targeting this window */ \ void **base_addrs; /* array of base addresses of the windows of \ all processes */ \ diff --git a/src/mpid/ch3/src/ch3u_handle_recv_req.c b/src/mpid/ch3/src/ch3u_handle_recv_req.c index f1232dc..262cf71 100644 --- a/src/mpid/ch3/src/ch3u_handle_recv_req.c +++ b/src/mpid/ch3/src/ch3u_handle_recv_req.c @@ -132,7 +132,7 @@ int MPIDI_CH3_ReqHandler_PutAccumRespComplete( MPIDI_VC_t *vc, /* here we increment the Active Target counter to guarantee the GET-like operation are completed when counter reaches zero. */ - win_ptr->my_counter++; + win_ptr->at_completion_counter++; iov[0].MPID_IOV_BUF = (MPID_IOV_BUF_CAST) get_accum_resp_pkt; iov[0].MPID_IOV_LEN = sizeof(*get_accum_resp_pkt); @@ -332,8 +332,8 @@ int MPIDI_CH3_ReqHandler_GetAccumRespComplete( MPIDI_VC_t *vc, /* here we decrement the Active Target counter to guarantee the GET-like operation are completed when counter reaches zero. */ - win_ptr->my_counter--; - MPIU_Assert(win_ptr->my_counter >= 0); + win_ptr->at_completion_counter--; + MPIU_Assert(win_ptr->at_completion_counter >= 0); MPIDI_CH3U_Request_complete(rreq); *complete = TRUE; @@ -545,7 +545,7 @@ int MPIDI_CH3_ReqHandler_FOPComplete( MPIDI_VC_t *vc, /* here we increment the Active Target counter to guarantee the GET-like operation are completed when counter reaches zero. */ - win_ptr->my_counter++; + win_ptr->at_completion_counter++; MPIDI_CH3U_SRBuf_alloc(resp_req, len); MPIU_ERR_CHKANDJUMP(resp_req->dev.tmpbuf_sz < len, mpi_errno, MPI_ERR_OTHER, "**nomemreq"); @@ -582,7 +582,7 @@ int MPIDI_CH3_ReqHandler_FOPComplete( MPIDI_VC_t *vc, /* here we increment the Active Target counter to guarantee the GET-like operation are completed when counter reaches zero. */ - win_ptr->my_counter++; + win_ptr->at_completion_counter++; MPID_Request_release(resp_req); goto finish_up; @@ -1239,7 +1239,7 @@ static int do_simple_get(MPID_Win *win_ptr, MPIDI_Win_lock_queue *lock_queue) /* here we increment the Active Target counter to guarantee the GET-like operation are completed when counter reaches zero. */ - win_ptr->my_counter++; + win_ptr->at_completion_counter++; MPIDI_Pkt_init(get_resp_pkt, MPIDI_CH3_PKT_GET_RESP); get_resp_pkt->request_handle = lock_queue->pt_single_op->request_handle; diff --git a/src/mpid/ch3/src/ch3u_handle_send_req.c b/src/mpid/ch3/src/ch3u_handle_send_req.c index 78d740c..fd732bb 100644 --- a/src/mpid/ch3/src/ch3u_handle_send_req.c +++ b/src/mpid/ch3/src/ch3u_handle_send_req.c @@ -57,8 +57,8 @@ int MPIDI_CH3_ReqHandler_GetSendRespComplete( MPIDI_VC_t *vc ATTRIBUTE((unused)) /* here we decrement the Active Target counter to guarantee the GET-like operation are completed when counter reaches zero. */ - win_ptr->my_counter--; - MPIU_Assert(win_ptr->my_counter >= 0); + win_ptr->at_completion_counter--; + MPIU_Assert(win_ptr->at_completion_counter >= 0); /* mark data transfer as complete and decrement CC */ MPIDI_CH3U_Request_complete(sreq); diff --git a/src/mpid/ch3/src/ch3u_rma_sync.c b/src/mpid/ch3/src/ch3u_rma_sync.c index ea64175..8e1353c 100644 --- a/src/mpid/ch3/src/ch3u_rma_sync.c +++ b/src/mpid/ch3/src/ch3u_rma_sync.c @@ -1269,7 +1269,7 @@ int MPIDI_Win_fence(int assert, MPID_Win *win_ptr) RMA ops on its window */ /* first initialize the completion counter. */ - win_ptr->my_counter += comm_size; + win_ptr->at_completion_counter += comm_size; mpi_errno = MPIR_Reduce_scatter_block_impl(MPI_IN_PLACE, rma_target_proc, 1, MPI_INT, MPI_SUM, comm_ptr, &errflag); @@ -1286,8 +1286,8 @@ int MPIDI_Win_fence(int assert, MPID_Win *win_ptr) /* Set the completion counter */ /* FIXME: MT: this needs to be done atomically because other procs have the address and could decrement it. */ - win_ptr->my_counter -= comm_size; - win_ptr->my_counter += rma_target_proc[0]; + win_ptr->at_completion_counter -= comm_size; + win_ptr->at_completion_counter += rma_target_proc[0]; MPIR_T_PVAR_TIMER_START(RMA, rma_winfence_issue); MPIR_T_PVAR_COUNTER_INC(RMA, rma_winfence_issue_aux, total_op_count); @@ -1380,11 +1380,11 @@ int MPIDI_Win_fence(int assert, MPID_Win *win_ptr) finish_up: /* wait for all operations from other processes to finish */ - if (win_ptr->my_counter) + if (win_ptr->at_completion_counter) { MPIR_T_PVAR_TIMER_START(RMA, rma_winfence_wait); MPID_Progress_start(&progress_state); - while (win_ptr->my_counter) + while (win_ptr->at_completion_counter) { mpi_errno = MPID_Progress_wait(&progress_state); /* --BEGIN ERROR HANDLING-- */ @@ -2292,7 +2292,7 @@ int MPIDI_Win_post(MPID_Group *post_grp_ptr, int assert, MPID_Win *win_ptr) } /* initialize the completion counter */ - win_ptr->my_counter += post_grp_size; + win_ptr->at_completion_counter += post_grp_size; if ((assert & MPI_MODE_NOCHECK) == 0) { @@ -2748,7 +2748,7 @@ int MPIDI_Win_complete(MPID_Win *win_ptr) dst = ranks_in_win_grp[i]; if (dst == rank) { /* FIXME: MT: this has to be done atomically */ - win_ptr->my_counter -= 1; + win_ptr->at_completion_counter -= 1; } else if (nops_to_proc[dst] == 0) { @@ -2842,13 +2842,13 @@ int MPIDI_Win_wait(MPID_Win *win_ptr) win_ptr->epoch_state = MPIDI_EPOCH_NONE; /* wait for all operations from other processes to finish */ - if (win_ptr->my_counter) + if (win_ptr->at_completion_counter) { MPID_Progress_state progress_state; MPIR_T_PVAR_TIMER_START(RMA, rma_winwait_wait); MPID_Progress_start(&progress_state); - while (win_ptr->my_counter) + while (win_ptr->at_completion_counter) { mpi_errno = MPID_Progress_wait(&progress_state); /* --BEGIN ERROR HANDLING-- */ @@ -2900,7 +2900,7 @@ int MPIDI_Win_test(MPID_Win *win_ptr, int *flag) MPIU_ERR_POP(mpi_errno); } - *flag = (win_ptr->my_counter) ? 0 : 1; + *flag = (win_ptr->at_completion_counter) ? 0 : 1; if (*flag) { /* Track access epoch state */ @@ -4603,7 +4603,7 @@ int MPIDI_CH3_PktHandler_Get( MPIDI_VC_t *vc, MPIDI_CH3_Pkt_t *pkt, /* here we increment the Active Target counter to guarantee the GET-like operation are completed when counter reaches zero. */ - win_ptr->my_counter++; + win_ptr->at_completion_counter++; if (MPIR_DATATYPE_IS_PREDEFINED(get_pkt->datatype)) { @@ -5012,7 +5012,7 @@ int MPIDI_CH3_PktHandler_CAS( MPIDI_VC_t *vc, MPIDI_CH3_Pkt_t *pkt, /* here we increment the Active Target counter to guarantee the GET-like operation are completed when counter reaches zero. */ - win_ptr->my_counter++; + win_ptr->at_completion_counter++; MPID_Request_release(req); goto fn_exit; @@ -5533,7 +5533,7 @@ int MPIDI_CH3_PktHandler_LockGetUnlock( MPIDI_VC_t *vc, MPIDI_CH3_Pkt_t *pkt, /* here we increment the Active Target counter to guarantee the GET-like operation are completed when counter reaches zero. */ - win_ptr->my_counter++; + win_ptr->at_completion_counter++; MPIDI_Pkt_init(get_resp_pkt, MPIDI_CH3_PKT_GET_RESP); get_resp_pkt->request_handle = lock_get_unlock_pkt->request_handle; @@ -6157,11 +6157,11 @@ int MPIDI_CH3_Finish_rma_op_target(MPIDI_VC_t *vc, MPID_Win *win_ptr, int is_rma if (flags & MPIDI_CH3_PKT_FLAG_RMA_AT_COMPLETE) { MPIU_Assert(win_ptr->current_lock_type == MPID_LOCK_NONE); - win_ptr->my_counter -= 1; - MPIU_Assert(win_ptr->my_counter >= 0); + win_ptr->at_completion_counter -= 1; + MPIU_Assert(win_ptr->at_completion_counter >= 0); /* Signal the local process when the op counter reaches 0. */ - if (win_ptr->my_counter == 0) + if (win_ptr->at_completion_counter == 0) MPIDI_CH3_Progress_signal_completion(); } diff --git a/src/mpid/ch3/src/mpid_rma.c b/src/mpid/ch3/src/mpid_rma.c index c112db0..685b665 100644 --- a/src/mpid/ch3/src/mpid_rma.c +++ b/src/mpid/ch3/src/mpid_rma.c @@ -285,7 +285,7 @@ static int win_init(MPI_Aint size, int disp_unit, int create_flavor, int model, (*win_ptr)->start_assert = 0; (*win_ptr)->comm_ptr = win_comm_ptr; - (*win_ptr)->my_counter = 0; + (*win_ptr)->at_completion_counter = 0; /* (*win_ptr)->base_addrs[] is set by caller; */ /* (*win_ptr)->sizes[] is set by caller; */ /* (*win_ptr)->disp_units[] is set by caller; */ http://git.mpich.org/mpich.git/commitdiff/c673fec42e015e9f714374dcb5d585289… commit c673fec42e015e9f714374dcb5d58528925663a1 Author: Xin Zhao <xinzhao3(a)illinois.edu> Date: Tue Sep 30 16:57:18 2014 -0500 Bug-fix: correct decrementing counter in GET-like operations. For GET-like operations, We should increment the Active Target counter when the process of sending back data is not completed immediately on target and a response request is created. We should decrement the counter when the process of sending back data is completed on target side. Signed-off-by: Pavan Balaji <balaji(a)anl.gov> diff --git a/src/mpid/ch3/src/ch3u_handle_recv_req.c b/src/mpid/ch3/src/ch3u_handle_recv_req.c index 4579050..f1232dc 100644 --- a/src/mpid/ch3/src/ch3u_handle_recv_req.c +++ b/src/mpid/ch3/src/ch3u_handle_recv_req.c @@ -89,6 +89,8 @@ int MPIDI_CH3_ReqHandler_PutAccumRespComplete( MPIDI_VC_t *vc, MPIDI_FUNC_ENTER(MPID_STATE_MPIDI_CH3_REQHANDLER_PUTACCUMRESPCOMPLETE); + MPID_Win_get_ptr(rreq->dev.target_win_handle, win_ptr); + /* Perform get in get-accumulate */ if (rreq->dev.resp_request_handle != MPI_REQUEST_NULL) { MPI_Aint type_size; @@ -128,6 +130,10 @@ int MPIDI_CH3_ReqHandler_PutAccumRespComplete( MPIDI_VC_t *vc, resp_req->dev.target_win_handle = rreq->dev.target_win_handle; resp_req->dev.flags = rreq->dev.flags; + /* here we increment the Active Target counter to guarantee the GET-like + operation are completed when counter reaches zero. */ + win_ptr->my_counter++; + iov[0].MPID_IOV_BUF = (MPID_IOV_BUF_CAST) get_accum_resp_pkt; iov[0].MPID_IOV_LEN = sizeof(*get_accum_resp_pkt); @@ -146,8 +152,6 @@ int MPIDI_CH3_ReqHandler_PutAccumRespComplete( MPIDI_VC_t *vc, get_acc_flag = 1; } - MPID_Win_get_ptr(rreq->dev.target_win_handle, win_ptr); - if (MPIDI_Request_get_type(rreq) == MPIDI_REQUEST_TYPE_ACCUM_RESP) { if (win_ptr->shm_allocated == TRUE) @@ -168,13 +172,6 @@ int MPIDI_CH3_ReqHandler_PutAccumRespComplete( MPIDI_VC_t *vc, if (mpi_errno) { MPIU_ERR_POP(mpi_errno); } } - if (get_acc_flag) { - /* here we decrement the Active Target counter to guarantee the GET-like - operation are completed when counter reaches zero. */ - win_ptr->my_counter--; - MPIU_Assert(win_ptr->my_counter >= 0); - } - /* mark data transfer as complete and decrement CC */ MPIDI_CH3U_Request_complete(rreq); *complete = TRUE; @@ -333,6 +330,11 @@ int MPIDI_CH3_ReqHandler_GetAccumRespComplete( MPIDI_VC_t *vc, MPI_WIN_NULL); if (mpi_errno) { MPIU_ERR_POP(mpi_errno); } + /* here we decrement the Active Target counter to guarantee the GET-like + operation are completed when counter reaches zero. */ + win_ptr->my_counter--; + MPIU_Assert(win_ptr->my_counter >= 0); + MPIDI_CH3U_Request_complete(rreq); *complete = TRUE; fn_exit: @@ -525,6 +527,8 @@ int MPIDI_CH3_ReqHandler_FOPComplete( MPIDI_VC_t *vc, MPIDI_Pkt_init(fop_resp_pkt, MPIDI_CH3_PKT_FOP_RESP); fop_resp_pkt->request_handle = rreq->dev.request_handle; + MPID_Win_get_ptr(rreq->dev.target_win_handle, win_ptr); + /* Copy original data into the send buffer. If data will fit in the header, use that. Otherwise allocate a temporary buffer. */ if (len <= sizeof(fop_resp_pkt->data)) { @@ -539,13 +543,15 @@ int MPIDI_CH3_ReqHandler_FOPComplete( MPIDI_VC_t *vc, resp_req->dev.flags = rreq->dev.flags; resp_req->dev.OnDataAvail = MPIDI_CH3_ReqHandler_GetAccumRespComplete; + /* here we increment the Active Target counter to guarantee the GET-like + operation are completed when counter reaches zero. */ + win_ptr->my_counter++; + MPIDI_CH3U_SRBuf_alloc(resp_req, len); MPIU_ERR_CHKANDJUMP(resp_req->dev.tmpbuf_sz < len, mpi_errno, MPI_ERR_OTHER, "**nomemreq"); MPIU_Memcpy( resp_req->dev.tmpbuf, rreq->dev.real_user_buf, len ); } - MPID_Win_get_ptr(rreq->dev.target_win_handle, win_ptr); - /* Apply the op */ if (rreq->dev.op != MPI_NO_OP) { uop = MPIR_OP_HDL_TO_FN(rreq->dev.op); @@ -573,6 +579,11 @@ int MPIDI_CH3_ReqHandler_FOPComplete( MPIDI_VC_t *vc, resp_req->dev.target_win_handle = rreq->dev.target_win_handle; resp_req->dev.flags = rreq->dev.flags; resp_req->dev.OnDataAvail = MPIDI_CH3_ReqHandler_GetAccumRespComplete; + + /* here we increment the Active Target counter to guarantee the GET-like + operation are completed when counter reaches zero. */ + win_ptr->my_counter++; + MPID_Request_release(resp_req); goto finish_up; } @@ -607,11 +618,6 @@ int MPIDI_CH3_ReqHandler_FOPComplete( MPIDI_VC_t *vc, MPIU_Free(rreq->dev.user_buf); } - /* here we decrement the Active Target counter to guarantee the GET-like - operation are completed when counter reaches zero. */ - win_ptr->my_counter--; - MPIU_Assert(win_ptr->my_counter >= 0); - *complete = 1; fn_exit: diff --git a/src/mpid/ch3/src/ch3u_rma_sync.c b/src/mpid/ch3/src/ch3u_rma_sync.c index 1258760..ea64175 100644 --- a/src/mpid/ch3/src/ch3u_rma_sync.c +++ b/src/mpid/ch3/src/ch3u_rma_sync.c @@ -4738,12 +4738,6 @@ int MPIDI_CH3_PktHandler_Accumulate( MPIDI_VC_t *vc, MPIDI_CH3_Pkt_t *pkt, MPIU_Object_set_ref(req, 1); *rreqp = req; - if (accum_pkt->type == MPIDI_CH3_PKT_GET_ACCUM) { - /* here we increment the Active Target counter to guarantee the GET-like - operation are completed when counter reaches zero. */ - win_ptr->my_counter++; - } - req->dev.user_count = accum_pkt->count; req->dev.op = accum_pkt->op; req->dev.real_user_buf = accum_pkt->addr; @@ -5015,6 +5009,11 @@ int MPIDI_CH3_PktHandler_CAS( MPIDI_VC_t *vc, MPIDI_CH3_Pkt_t *pkt, req->dev.target_win_handle = cas_pkt->target_win_handle; req->dev.flags = cas_pkt->flags; req->dev.OnDataAvail = MPIDI_CH3_ReqHandler_GetAccumRespComplete; + + /* here we increment the Active Target counter to guarantee the GET-like + operation are completed when counter reaches zero. */ + win_ptr->my_counter++; + MPID_Request_release(req); goto fn_exit; } @@ -5106,10 +5105,6 @@ int MPIDI_CH3_PktHandler_FOP( MPIDI_VC_t *vc, MPIDI_CH3_Pkt_t *pkt, MPIU_Object_set_ref(req, 1); /* Ref is held by progress engine */ *rreqp = NULL; - /* here we increment the Active Target counter to guarantee the GET-like - operation are completed when counter reaches zero. */ - win_ptr->my_counter++; - req->dev.user_buf = NULL; /* will be set later */ req->dev.user_count = 1; req->dev.datatype = fop_pkt->datatype; http://git.mpich.org/mpich.git/commitdiff/d6996640a0e50e932b105064fa026c536… commit d6996640a0e50e932b105064fa026c53633cf88c Author: Xin Zhao <xinzhao3(a)illinois.edu> Date: Mon Sep 29 15:36:05 2014 -0500 Bug-fix: call RMA finish function when RMA ops finishes. In the original implementation, for GACC/FOP/CAS, the function MPIDI_CH3_Finish_rma_op_target (includes operations that should be performed on target when that operation finishes on target) is not called when that operation real finishes, but is called after starting send back data. Here we fix it to make the function called after sending process on target is completed. Signed-off-by: Pavan Balaji <balaji(a)anl.gov> diff --git a/src/mpid/ch3/src/ch3u_handle_recv_req.c b/src/mpid/ch3/src/ch3u_handle_recv_req.c index df5583a..4579050 100644 --- a/src/mpid/ch3/src/ch3u_handle_recv_req.c +++ b/src/mpid/ch3/src/ch3u_handle_recv_req.c @@ -125,6 +125,8 @@ int MPIDI_CH3_ReqHandler_PutAccumRespComplete( MPIDI_VC_t *vc, resp_req->dev.OnFinal = MPIDI_CH3_ReqHandler_GetAccumRespComplete; resp_req->dev.OnDataAvail = MPIDI_CH3_ReqHandler_GetAccumRespComplete; + resp_req->dev.target_win_handle = rreq->dev.target_win_handle; + resp_req->dev.flags = rreq->dev.flags; iov[0].MPID_IOV_BUF = (MPID_IOV_BUF_CAST) get_accum_resp_pkt; iov[0].MPID_IOV_LEN = sizeof(*get_accum_resp_pkt); @@ -160,9 +162,11 @@ int MPIDI_CH3_ReqHandler_PutAccumRespComplete( MPIDI_VC_t *vc, } } + if (!get_acc_flag) { mpi_errno = MPIDI_CH3_Finish_rma_op_target(vc, win_ptr, TRUE, rreq->dev.flags, rreq->dev.source_win_handle); if (mpi_errno) { MPIU_ERR_POP(mpi_errno); } + } if (get_acc_flag) { /* here we decrement the Active Target counter to guarantee the GET-like @@ -317,11 +321,18 @@ int MPIDI_CH3_ReqHandler_GetAccumRespComplete( MPIDI_VC_t *vc, int *complete ) { int mpi_errno = MPI_SUCCESS; + MPID_Win *win_ptr; MPIDI_STATE_DECL(MPID_STATE_MPIDI_CH3_REQHANDLER_GETACCUMRESPCOMPLETE); MPIDI_FUNC_ENTER(MPID_STATE_MPIDI_CH3_REQHANDLER_GETACCUMRESPCOMPLETE); MPIU_Free(rreq->dev.user_buf); + MPID_Win_get_ptr(rreq->dev.target_win_handle, win_ptr); + + mpi_errno = MPIDI_CH3_Finish_rma_op_target(vc, win_ptr, TRUE, rreq->dev.flags, + MPI_WIN_NULL); + if (mpi_errno) { MPIU_ERR_POP(mpi_errno); } + MPIDI_CH3U_Request_complete(rreq); *complete = TRUE; fn_exit: @@ -524,6 +535,10 @@ int MPIDI_CH3_ReqHandler_FOPComplete( MPIDI_VC_t *vc, MPIU_ERR_CHKANDJUMP(resp_req == NULL, mpi_errno, MPI_ERR_OTHER, "**nomemreq"); MPIU_Object_set_ref(resp_req, 1); + resp_req->dev.target_win_handle = rreq->dev.target_win_handle; + resp_req->dev.flags = rreq->dev.flags; + resp_req->dev.OnDataAvail = MPIDI_CH3_ReqHandler_GetAccumRespComplete; + MPIDI_CH3U_SRBuf_alloc(resp_req, len); MPIU_ERR_CHKANDJUMP(resp_req->dev.tmpbuf_sz < len, mpi_errno, MPI_ERR_OTHER, "**nomemreq"); MPIU_Memcpy( resp_req->dev.tmpbuf, rreq->dev.real_user_buf, len ); @@ -552,8 +567,25 @@ int MPIDI_CH3_ReqHandler_FOPComplete( MPIDI_VC_t *vc, MPIU_ERR_CHKANDJUMP(mpi_errno != MPI_SUCCESS, mpi_errno, MPI_ERR_OTHER, "**ch3|rmamsg"); if (resp_req != NULL) { - MPID_Request_release(resp_req); + if (!MPID_Request_is_complete(resp_req)) { + /* sending process is not completed, set proper OnDataAvail + (it is initialized to NULL by lower layer) */ + resp_req->dev.target_win_handle = rreq->dev.target_win_handle; + resp_req->dev.flags = rreq->dev.flags; + resp_req->dev.OnDataAvail = MPIDI_CH3_ReqHandler_GetAccumRespComplete; + MPID_Request_release(resp_req); + goto finish_up; + } + else { + MPID_Request_release(resp_req); + } } + + /* There are additional steps to take if this is a passive + target RMA or the last operation from the source */ + mpi_errno = MPIDI_CH3_Finish_rma_op_target(vc, win_ptr, TRUE, rreq->dev.flags, + rreq->dev.source_win_handle); + if (mpi_errno) { MPIU_ERR_POP(mpi_errno); } } else { MPID_IOV iov[MPID_IOV_LIMIT]; @@ -569,18 +601,12 @@ int MPIDI_CH3_ReqHandler_FOPComplete( MPIDI_VC_t *vc, MPIU_ERR_CHKANDJUMP(mpi_errno != MPI_SUCCESS, mpi_errno, MPI_ERR_OTHER, "**ch3|rmamsg"); } + finish_up: /* Free temporary buffer allocated in PktHandler_FOP */ if (len > sizeof(int) * MPIDI_RMA_FOP_IMMED_INTS && rreq->dev.op != MPI_NO_OP) { MPIU_Free(rreq->dev.user_buf); } - /* There are additional steps to take if this is a passive - target RMA or the last operation from the source */ - - mpi_errno = MPIDI_CH3_Finish_rma_op_target(vc, win_ptr, TRUE, rreq->dev.flags, - rreq->dev.source_win_handle); - if (mpi_errno) { MPIU_ERR_POP(mpi_errno); } - /* here we decrement the Active Target counter to guarantee the GET-like operation are completed when counter reaches zero. */ win_ptr->my_counter--; diff --git a/src/mpid/ch3/src/ch3u_rma_sync.c b/src/mpid/ch3/src/ch3u_rma_sync.c index 0eeae75..1258760 100644 --- a/src/mpid/ch3/src/ch3u_rma_sync.c +++ b/src/mpid/ch3/src/ch3u_rma_sync.c @@ -5009,7 +5009,17 @@ int MPIDI_CH3_PktHandler_CAS( MPIDI_VC_t *vc, MPIDI_CH3_Pkt_t *pkt, MPIU_ERR_CHKANDJUMP(mpi_errno != MPI_SUCCESS, mpi_errno, MPI_ERR_OTHER, "**ch3|rmamsg"); if (req != NULL) { - MPID_Request_release(req); + if (!MPID_Request_is_complete(req)) { + /* sending process is not completed, set proper OnDataAvail + (it is initialized to NULL by lower layer) */ + req->dev.target_win_handle = cas_pkt->target_win_handle; + req->dev.flags = cas_pkt->flags; + req->dev.OnDataAvail = MPIDI_CH3_ReqHandler_GetAccumRespComplete; + MPID_Request_release(req); + goto fn_exit; + } + else + MPID_Request_release(req); } ----------------------------------------------------------------------- Summary of changes: src/mpid/ch3/include/mpidpre.h | 2 +- src/mpid/ch3/src/ch3u_handle_recv_req.c | 80 +++++++++++++++++------- src/mpid/ch3/src/ch3u_handle_send_req.c | 4 +- src/mpid/ch3/src/ch3u_rma_sync.c | 103 ++++++++++++++++--------------- src/mpid/ch3/src/mpid_rma.c | 2 +- test/mpi/rma/manyrma2.c | 2 +- 6 files changed, 115 insertions(+), 78 deletions(-) hooks/post-receive -- MPICH primary repository

1 0