February 2013 - commits - mailman3.cels.anl.gov

[mpich] MPICH primary repository branch, master, updated. v3.0.2-19-g72d3470
by noreply＠mpich.org 17 Feb '13

17 Feb '13

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 72d347070b52402039af3db2428a49089d2de35e (commit) from 3003efd864588017c4e48935ec89040db7e8f5c6 (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/72d347070b52402039af3db2428a49089… commit 72d347070b52402039af3db2428a49089d2de35e Author: Pavan Balaji <balaji(a)mcs.anl.gov> Date: Sun Feb 17 17:56:28 2013 -0600 Added support for Cobalt to the CHANGES file. No reviewer. diff --git a/CHANGES b/CHANGES index 946a654..744cb61 100644 --- a/CHANGES +++ b/CHANGES @@ -1,4 +1,16 @@ =============================================================================== + Changes in 3.0.3 +=============================================================================== + + # PM/PMI: Added support for the Cobalt process manager. + + # Several other minor bug fixes, memory leak fixes, and code cleanup. + A full list of changes is available at the following link: + + http://git.mpich.org/mpich.git/shortlog/v3.0.2..v3.0.3 + + +=============================================================================== Changes in 3.0.2 =============================================================================== ----------------------------------------------------------------------- Summary of changes: CHANGES | 12 ++++++++++++ 1 files changed, 12 insertions(+), 0 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

Re: [mpich-commits] [mpi-all-commits] [mpich] MPICH primary repository branch, master, updated. v3.0.2-18-g3003efd
by Dave Goodell 15 Feb '13

15 Feb '13

Perhaps also add a CHANGES entry indicating our "new" Cobalt support? -Dave On Feb 15, 2013, at 8:00 AM CST, mysql vizuser wrote: > 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 3003efd864588017c4e48935ec89040db7e8f5c6 (commit) > from 9a7b024b4ff2b1b758b6199d8e5f0297e35415cc (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/3003efd864588017c4e48935ec89040db… > > commit 3003efd864588017c4e48935ec89040db7e8f5c6 > Author: Pavan Balaji <balaji(a)mcs.anl.gov> > Date: Thu Feb 14 09:15:47 2013 -0600 > > Bug-fix: Allow configure to check for cobalt as well. > > No reviewer. > > diff --git a/src/pm/hydra/configure.ac b/src/pm/hydra/configure.ac > index 5865976..1398dcc 100644 > --- a/src/pm/hydra/configure.ac > +++ b/src/pm/hydra/configure.ac > @@ -164,9 +164,9 @@ fi > ######################################################################### > AC_ARG_WITH(hydra-bss, > [AC_HELP_STRING([--with-hydra-bss=name], > - [Bootstrap Server (user,ssh,rsh,fork,slurm,ll,lsf,sge,pbs,manual,persist)])], > + [Bootstrap Server (user,ssh,rsh,fork,slurm,ll,lsf,sge,pbs,cobalt,manual,persist)])], > [ hydra_bss=$withval ], > - [ hydra_bss="user,ssh,rsh,fork,slurm,ll,lsf,sge,pbs,manual,persist" ]) > + [ hydra_bss="user,ssh,rsh,fork,slurm,ll,lsf,sge,pbs,cobalt,manual,persist" ]) > AC_MSG_CHECKING(bootstrap server) > AC_MSG_RESULT($hydra_bss) > hydra_bss_names="`echo $hydra_bss | sed -e 's/:/ /g' -e 's/,/ /g'`" > @@ -223,6 +223,10 @@ for hydra_bss_name in ${hydra_bss_names}; do > fi > available_rmks="$available_rmks pbs" > ;; > + cobalt) > + hydra_bss_external=true > + available_rmks="$available_rmks cobalt" > + ;; > manual) > hydra_bss_external=true > available_launchers="$available_launchers manual" > diff --git a/src/pm/hydra/tools/bootstrap/external/Makefile.mk b/src/pm/hydra/tools/bootstrap/external/Makefile.mk > index bbd6119..91d5ee1 100644 > --- a/src/pm/hydra/tools/bootstrap/external/Makefile.mk > +++ b/src/pm/hydra/tools/bootstrap/external/Makefile.mk > @@ -49,6 +49,7 @@ libhydra_la_SOURCES += $(top_srcdir)/tools/bootstrap/external/external_common.c > $(top_srcdir)/tools/bootstrap/external/pbs_init.c \ > $(top_srcdir)/tools/bootstrap/external/pbs_query_native_int.c \ > $(top_srcdir)/tools/bootstrap/external/pbs_query_node_list.c \ > + $(top_srcdir)/tools/bootstrap/external/cobalt_init.c \ > $(top_srcdir)/tools/bootstrap/external/cobalt_query_native_int.c \ > $(top_srcdir)/tools/bootstrap/external/cobalt_query_node_list.c > > > ----------------------------------------------------------------------- > > Summary of changes: > src/pm/hydra/configure.ac | 8 ++++++-- > src/pm/hydra/tools/bootstrap/external/Makefile.mk | 1 + > 2 files changed, 7 insertions(+), 2 deletions(-) > > > hooks/post-receive > -- > MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.2-18-g3003efd
by noreply＠mpich.org 15 Feb '13

15 Feb '13

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 3003efd864588017c4e48935ec89040db7e8f5c6 (commit) from 9a7b024b4ff2b1b758b6199d8e5f0297e35415cc (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/3003efd864588017c4e48935ec89040db… commit 3003efd864588017c4e48935ec89040db7e8f5c6 Author: Pavan Balaji <balaji(a)mcs.anl.gov> Date: Thu Feb 14 09:15:47 2013 -0600 Bug-fix: Allow configure to check for cobalt as well. No reviewer. diff --git a/src/pm/hydra/configure.ac b/src/pm/hydra/configure.ac index 5865976..1398dcc 100644 --- a/src/pm/hydra/configure.ac +++ b/src/pm/hydra/configure.ac @@ -164,9 +164,9 @@ fi ######################################################################### AC_ARG_WITH(hydra-bss, [AC_HELP_STRING([--with-hydra-bss=name], - [Bootstrap Server (user,ssh,rsh,fork,slurm,ll,lsf,sge,pbs,manual,persist)])], + [Bootstrap Server (user,ssh,rsh,fork,slurm,ll,lsf,sge,pbs,cobalt,manual,persist)])], [ hydra_bss=$withval ], - [ hydra_bss="user,ssh,rsh,fork,slurm,ll,lsf,sge,pbs,manual,persist" ]) + [ hydra_bss="user,ssh,rsh,fork,slurm,ll,lsf,sge,pbs,cobalt,manual,persist" ]) AC_MSG_CHECKING(bootstrap server) AC_MSG_RESULT($hydra_bss) hydra_bss_names="`echo $hydra_bss | sed -e 's/:/ /g' -e 's/,/ /g'`" @@ -223,6 +223,10 @@ for hydra_bss_name in ${hydra_bss_names}; do fi available_rmks="$available_rmks pbs" ;; + cobalt) + hydra_bss_external=true + available_rmks="$available_rmks cobalt" + ;; manual) hydra_bss_external=true available_launchers="$available_launchers manual" diff --git a/src/pm/hydra/tools/bootstrap/external/Makefile.mk b/src/pm/hydra/tools/bootstrap/external/Makefile.mk index bbd6119..91d5ee1 100644 --- a/src/pm/hydra/tools/bootstrap/external/Makefile.mk +++ b/src/pm/hydra/tools/bootstrap/external/Makefile.mk @@ -49,6 +49,7 @@ libhydra_la_SOURCES += $(top_srcdir)/tools/bootstrap/external/external_common.c $(top_srcdir)/tools/bootstrap/external/pbs_init.c \ $(top_srcdir)/tools/bootstrap/external/pbs_query_native_int.c \ $(top_srcdir)/tools/bootstrap/external/pbs_query_node_list.c \ + $(top_srcdir)/tools/bootstrap/external/cobalt_init.c \ $(top_srcdir)/tools/bootstrap/external/cobalt_query_native_int.c \ $(top_srcdir)/tools/bootstrap/external/cobalt_query_node_list.c ----------------------------------------------------------------------- Summary of changes: src/pm/hydra/configure.ac | 8 ++++++-- src/pm/hydra/tools/bootstrap/external/Makefile.mk | 1 + 2 files changed, 7 insertions(+), 2 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.2-17-g9a7b024
by noreply＠mpich.org 08 Feb '13

08 Feb '13

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 9a7b024b4ff2b1b758b6199d8e5f0297e35415cc (commit) from 902bb9e890329482cf084663b25ee9dc45a96403 (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/9a7b024b4ff2b1b758b6199d8e5f0297e… commit 9a7b024b4ff2b1b758b6199d8e5f0297e35415cc Author: James Dinan <dinan(a)mcs.anl.gov> Date: Thu Feb 7 16:48:06 2013 -0600 Fix incorrect timing reporting in llist bench The linked list benchmarks were reportin the sum of the timings across all nodes, rather than the maximum. Reviewer: none diff --git a/test/mpi/rma/linked_list_bench_lock_all.c b/test/mpi/rma/linked_list_bench_lock_all.c index 9e7c5c0..90df021 100644 --- a/test/mpi/rma/linked_list_bench_lock_all.c +++ b/test/mpi/rma/linked_list_bench_lock_all.c @@ -242,7 +242,7 @@ int main(int argc, char **argv) { if (print_perf) { double max_time; - MPI_Reduce(&time, &max_time, 1, MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD); + MPI_Reduce(&time, &max_time, 1, MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD); if (procid == 0) { printf("Total time = %0.2f sec, elem/sec = %0.2f, sec/elem = %0.2f usec\n", max_time, NUM_ELEMS/max_time, max_time/NUM_ELEMS*1.0e6); diff --git a/test/mpi/rma/linked_list_bench_lock_excl.c b/test/mpi/rma/linked_list_bench_lock_excl.c index 275360f..7fe5240 100644 --- a/test/mpi/rma/linked_list_bench_lock_excl.c +++ b/test/mpi/rma/linked_list_bench_lock_excl.c @@ -245,7 +245,7 @@ int main(int argc, char **argv) { if (print_perf) { double max_time; - MPI_Reduce(&time, &max_time, 1, MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD); + MPI_Reduce(&time, &max_time, 1, MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD); if (procid == 0) { printf("Total time = %0.2f sec, elem/sec = %0.2f, sec/elem = %0.2f usec\n", max_time, NUM_ELEMS/max_time, max_time/NUM_ELEMS*1.0e6); diff --git a/test/mpi/rma/linked_list_bench_lock_shr.c b/test/mpi/rma/linked_list_bench_lock_shr.c index c96cf33..2ddb791 100644 --- a/test/mpi/rma/linked_list_bench_lock_shr.c +++ b/test/mpi/rma/linked_list_bench_lock_shr.c @@ -234,7 +234,7 @@ int main(int argc, char **argv) { if (print_perf) { double max_time; - MPI_Reduce(&time, &max_time, 1, MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD); + MPI_Reduce(&time, &max_time, 1, MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD); if (procid == 0) { printf("Total time = %0.2f sec, elem/sec = %0.2f, sec/elem = %0.2f usec\n", max_time, NUM_ELEMS/max_time, max_time/NUM_ELEMS*1.0e6); ----------------------------------------------------------------------- Summary of changes: test/mpi/rma/linked_list_bench_lock_all.c | 2 +- test/mpi/rma/linked_list_bench_lock_excl.c | 2 +- test/mpi/rma/linked_list_bench_lock_shr.c | 2 +- 3 files changed, 3 insertions(+), 3 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.2-16-g902bb9e
by noreply＠mpich.org 07 Feb '13

07 Feb '13

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 902bb9e890329482cf084663b25ee9dc45a96403 (commit) from 0a452a24cf55b8a648965ca1f5a504955880c354 (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/902bb9e890329482cf084663b25ee9dc4… commit 902bb9e890329482cf084663b25ee9dc45a96403 Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Thu Feb 7 15:31:04 2013 -0600 clmake: filter "install-sh" too Caused problems for the "mpich-osx" jenkins build. No reviewer. diff --git a/maint/clmake.in b/maint/clmake.in index 4a18f18..3891735 100644 --- a/maint/clmake.in +++ b/maint/clmake.in @@ -57,6 +57,7 @@ $rootSrcDir = ""; "building profiling interface in directory", "/usr/bin/install", "/.*bin/install", + ".*confdb/install-sh", "Copying Upshot", "Cleaning directory", "[*][*][*][*] Making .*\.\.\.\.", ----------------------------------------------------------------------- Summary of changes: maint/clmake.in | 1 + 1 files changed, 1 insertions(+), 0 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.2-15-g0a452a2
by noreply＠mpich.org 07 Feb '13

07 Feb '13

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 0a452a24cf55b8a648965ca1f5a504955880c354 (commit) from 5ffbf15451930ce35b5e08678d8b267ceb6a5fe4 (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/0a452a24cf55b8a648965ca1f5a504955… commit 0a452a24cf55b8a648965ca1f5a504955880c354 Author: William Gropp <wgropp(a)illinois.edu> Date: Wed Feb 6 07:31:15 2013 -0600 Added support for passing the length of a Fortran CHARACTER as size_t instead of int. Added a configure option to set; note that despite documentation to the contrary, it looks like some compilers still pass the length as an int (the common practice for decades), so changing to size_t must be made with extreme caution diff --git a/configure.ac b/configure.ac index f17eb74..e3cd063 100644 --- a/configure.ac +++ b/configure.ac @@ -1882,6 +1882,66 @@ information to configure with the FCFLAGS environment variable.]) AC_DEFINE(HAVE_MPI_F_INIT_WORKS_WITH_C,1,[Define if the Fortran init code for MPI works from C programs without special libraries]) fi + # + # Some Fortran compilers now pass CHARACTER length as a size_t instead + # of as an int. This is hard to test for, since the data is passed by + # value and for characters less than about 2GB long, the correct + # value will be used. In this case, we must use an approach similar to + # the one used by libtool for shared library options - look at the + # compiler name or vendor. + # Known compilers that use size_t instead of int: + # Intel Fortran + # gfortran + # Add others as they become known + AC_ARG_ENABLE(f77characterlen, + AC_HELP_STRING([--enable-f77characterlen], + [Select between int and size_t for the length of a Fortran CHARACTER, depending on the F77 compiler. If --enable-f77characterlen=size_t is given, force the use of size_t. This is used for passing Fortran CHARACTER data between C and Fortran, and is provided for experts. Note that the documentation provided by compiler vendors on the calling convention may not be accurate.]),,enable_f77characterlen=no) + + # Set the default + f77_uses_int_for_str=default + + case "$enable_f77characterlen" in + yes|no) + ;; + size_t) + f77_uses_int_for_str=no + enable_f77characterlen=yes + ;; + int) + f77_uses_int_for_str=yes + enable_f77characterlen=yes + ;; + *) + AC_MSG_ERROR([Invalid value provided for --enable-f77characterlen]) + ;; + esac + + # If we might change the size (enable) and we haven't set the choice, + # attempt to determine it from the compiler name. Risky, but we haven't + # found a reliable way to do this with test codes. + if test "$enable_f77characterlen" = "yes" -a \ + "$f77_uses_int_for_str" = "default" ; then + f77_uses_int_for_str=yes + f77Basename=`basename $F77` + case $f77Basename in + ifort*) + f77_uses_int_for_str=no + ;; + gfortran*) + f77_uses_int_for_str=no + ;; + esac + fi + # This test is disabled for now. Despite information in documentation + # on gfortran, it appears to pass lengths as int, at least in some + # builds (it used movl when tested in 2/2013). Tests that failed + # included infotestf.f, in a call to mpi_info_get. + # Leave this as a place holder until a proper test can be determined. + if test "$enable_f77characterlen" = "yes" -a \ + "$f77_uses_int_for_str" = "no" ; then + AC_DEFINE(USE_FORT_STR_LEN_SIZET,1,[Define if the length of a CHARACTER*(*) string in Fortran should be passed as size_t instead of int] ) + fi + fi # FC requires F77 as well. If the user disabled f77, do not run the diff --git a/src/binding/f77/mpi_fortimpl.h b/src/binding/f77/mpi_fortimpl.h index f8eb11f..0adfa11 100644 --- a/src/binding/f77/mpi_fortimpl.h +++ b/src/binding/f77/mpi_fortimpl.h @@ -22,16 +22,35 @@ * an MPI_Fint instead of an int. In that case, we will need an additional * case. */ +/* + * Many Fortran compilers at some point changed from passing the length of + * a CHARACTER*(N) in an int to using a size_t. This definition allows + * us to select either size_t or int. Determining which one the compiler + * is using may require reading the documentation or examining the generated + * code. + * + * The default is left as "int" because that is the correct legacy choice. + * Configure may need to check the compiler vendor and version to decide + * whether to select size_t. And despite documentation to the contrary, + * in our experiments, gfortran used int instead of size_t, which was + * verified by inspecting the assembly code. + */ +#ifdef USE_FORT_STR_LEN_SIZET +#define FORT_SIZE_INT size_t +#else +#define FORT_SIZE_INT int +#endif + #ifdef USE_FORT_MIXED_STR_LEN -#define FORT_MIXED_LEN_DECL , int +#define FORT_MIXED_LEN_DECL , FORT_SIZE_INT #define FORT_END_LEN_DECL -#define FORT_MIXED_LEN(a) , int a +#define FORT_MIXED_LEN(a) , FORT_SIZE_INT a #define FORT_END_LEN(a) #else #define FORT_MIXED_LEN_DECL -#define FORT_END_LEN_DECL , int +#define FORT_END_LEN_DECL , FORT_SIZE_INT #define FORT_MIXED_LEN(a) -#define FORT_END_LEN(a) , int a +#define FORT_END_LEN(a) , FORT_SIZE_INT a #endif /* ------------------------------------------------------------------------- */ ----------------------------------------------------------------------- Summary of changes: configure.ac | 60 ++++++++++++++++++++++++++++++++++++++++ src/binding/f77/mpi_fortimpl.h | 27 +++++++++++++++-- 2 files changed, 83 insertions(+), 4 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.2-14-g5ffbf15
by noreply＠mpich.org 06 Feb '13

06 Feb '13

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 5ffbf15451930ce35b5e08678d8b267ceb6a5fe4 (commit) from eed007281d4eca1e8bd662022a91526fb789b6bc (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/5ffbf15451930ce35b5e08678d8b267ce… commit 5ffbf15451930ce35b5e08678d8b267ceb6a5fe4 Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Wed Feb 6 22:27:59 2013 -0600 clmake: use `\W` instead of `\b` Patterns that end in non-word chars (such as `:`) will not match a `\b` atom when followed by a non-word like ` ` or `\n`. No reviewer. diff --git a/maint/clmake.in b/maint/clmake.in index 5532c47..4a18f18 100644 --- a/maint/clmake.in +++ b/maint/clmake.in @@ -188,7 +188,7 @@ while ($line = <>) { else { $is_command = 0; foreach $cmdname (@commands) { - if ($line =~ /^\s*$cmdname\b/) { + if ($line =~ /^\s*$cmdname\W/) { $is_command = 1; $cmdline = $line; last; ----------------------------------------------------------------------- Summary of changes: maint/clmake.in | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.2-13-geed0072
by noreply＠mpich.org 06 Feb '13

06 Feb '13

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 eed007281d4eca1e8bd662022a91526fb789b6bc (commit) via 748c0ae2a2abd424f81724d844c9c054b9126748 (commit) via e12f4293d0ab09f512b9fe091e96ce98b0532b6f (commit) from 9aa1c31b967e80344719fcb02dff7a9a05045b09 (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/eed007281d4eca1e8bd662022a91526fb… commit eed007281d4eca1e8bd662022a91526fb789b6bc Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Wed Feb 6 16:11:45 2013 -0600 ROMIO: fix buffer/datatype mismatch in Isend/Irecv Caught by the cool new type checking annotations in our `mpi.h` header that work with modern clang versions. No reviewer. diff --git a/src/mpi/romio/adio/common/ad_coll_exch_new.c b/src/mpi/romio/adio/common/ad_coll_exch_new.c index a2f32d7..10a441c 100644 --- a/src/mpi/romio/adio/common/ad_coll_exch_new.c +++ b/src/mpi/romio/adio/common/ad_coll_exch_new.c @@ -396,7 +396,7 @@ void ADIOI_Exch_file_views(int myrank, int nprocs, int file_ptr_type, INDICES, fd->comm, &recv_req_arr[j]); j++; MPI_Irecv(client_file_view_state_arr[i].flat_type_p->blocklens, - recv_count_arr[i].count, MPI_INT, i, + recv_count_arr[i].count, ADIO_OFFSET, i, BLOCK_LENS, fd->comm, &recv_req_arr[j]); j++; } @@ -412,7 +412,7 @@ void ADIOI_Exch_file_views(int myrank, int nprocs, int file_ptr_type, INDICES, fd->comm, &send_req_arr[j]); j++; MPI_Isend(flat_file_p->blocklens, - send_count_arr[i].count, MPI_INT, i, + send_count_arr[i].count, ADIO_OFFSET, i, BLOCK_LENS, fd->comm, &send_req_arr[j]); j++; } @@ -427,7 +427,7 @@ void ADIOI_Exch_file_views(int myrank, int nprocs, int file_ptr_type, &send_req_arr[j]); j++; MPI_Isend(flat_file_p->blocklens, - send_count_arr[i].count, MPI_INT, + send_count_arr[i].count, ADIO_OFFSET, fd->hints->ranklist[i], BLOCK_LENS, fd->comm, &send_req_arr[j]); j++; http://git.mpich.org/mpich.git/commitdiff/748c0ae2a2abd424f81724d844c9c054b… commit 748c0ae2a2abd424f81724d844c9c054b9126748 Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Wed Feb 6 16:10:19 2013 -0600 hydra: squash loop warning from modern clang `-Wempty-body` will cause this, and is enabled by default in modern clang when configuring MPICH with `--enable-strict`. No reviewer. diff --git a/src/pm/hydra/ui/mpich/mpiexec.c b/src/pm/hydra/ui/mpich/mpiexec.c index c238ce0..e9f01e1 100644 --- a/src/pm/hydra/ui/mpich/mpiexec.c +++ b/src/pm/hydra/ui/mpich/mpiexec.c @@ -101,7 +101,8 @@ static HYD_status qsort_node_list(void) int count, i; HYD_status status = HYD_SUCCESS; - for (count = 0, node = HYD_server_info.node_list; node; node = node->next, count++); + for (count = 0, node = HYD_server_info.node_list; node; node = node->next, count++) + /* skip */; HYDU_MALLOC(node_list, struct HYD_node **, count * sizeof(struct HYD_node *), status); for (i = 0, node = HYD_server_info.node_list; node; node = node->next, i++) http://git.mpich.org/mpich.git/commitdiff/e12f4293d0ab09f512b9fe091e96ce98b… commit e12f4293d0ab09f512b9fe091e96ce98b0532b6f Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Mon Feb 4 23:57:16 2013 -0600 fix Fortran MPI_WEIGHTS_EMPTY initialization We were not communicating the MPI_WEIGHTS_EMPTY common block value to C via the `mpirinitc_` routine. Also fix a minor warning about the C struct static initializer when `HAVE_C_MULTI_ATTR_ALIAS` is defined, as it usually is on Linux. The existing `dgraph_unwgtf` test does not catch this case. AFAICS, there is no black-box way to test this behavior using only the MPI interface provided that the MPI library correctly implements `MPI_UNWEIGHTED` handling, since the MPI implementation is forbidden to inspect the pointer value other than to check whether it is `MPI_UNWEIGHTED`. The best idea I've come up with so far is to create a sanity check routine that we can call from Fortran that ensures all Fortran-C interoperability constants are correctly recognized, but even that's a pretty weak test, since it must also be updated whenever a new Fortran common block is added. This will have to do for now... Follow-up to [d6ceef9]. Reviewed-by: dinan diff --git a/src/binding/f77/setbot.c.in b/src/binding/f77/setbot.c.in index 720f123..da4a91c 100644 --- a/src/binding/f77/setbot.c.in +++ b/src/binding/f77/setbot.c.in @@ -17,12 +17,15 @@ #define mpirinitc2_ mpirinitc2 #endif /* These functions are called from Fortran so only need prototypes in - this file. Note that the last argument is a character array, so + this file. Note that the second-to-last argument is a character array, so we need to include the elements of the Fortran character "dope vector". */ -FORT_DLL_SPEC void FORT_CALL mpirinitc_( void *, void *, void *, void *, void *, - void *, void * FORT_MIXED_LEN_DECL - FORT_END_LEN_DECL ); +FORT_DLL_SPEC void FORT_CALL mpirinitc_( void *si, void *ssi, + void *bt, void *ip, + void *uw, void *ecsi, + void *asn FORT_MIXED_LEN(d1), + void *we + FORT_END_LEN(d1)); FORT_DLL_SPEC void FORT_CALL mpirinitc2_( char * FORT_MIXED_LEN_DECL FORT_END_LEN_DECL ); @@ -67,8 +70,9 @@ void *MPIR_F_MPI_WEIGHTS_EMPTY = 0; FORT_DLL_SPEC void FORT_CALL mpirinitc_( void *si, void *ssi, void *bt, void *ip, void *uw, void *ecsi, - void *asn FORT_MIXED_LEN(d1) - FORT_END_LEN(d1) ) + void *asn FORT_MIXED_LEN(d1), + void *we + FORT_END_LEN(d1)) { MPI_F_STATUS_IGNORE = (MPI_Fint *) si; MPI_F_STATUSES_IGNORE = (MPI_Fint *) ssi; @@ -77,6 +81,7 @@ FORT_DLL_SPEC void FORT_CALL mpirinitc_( void *si, void *ssi, MPIR_F_MPI_UNWEIGHTED = uw; MPI_F_ERRCODES_IGNORE = (MPI_Fint *)ecsi; MPI_F_ARGVS_NULL = asn; + MPIR_F_MPI_WEIGHTS_EMPTY = we; } /* Initialize the Fortran ARGV_NULL to a blank. Using this routine avoids potential problems with string manipulation routines that @@ -201,7 +206,7 @@ struct mpif_cmblk9_t_ { MPI_Fint MPIF_WEIGHTS_EMPTY; }; typedef struct mpif_cmblk9_t_ mpif_cmblk9_t; -mpif_cmblk9_t mpifcmb9r @CMB_1INT_ALIGNMENT@ = {{0}}; +mpif_cmblk9_t mpifcmb9r @CMB_1INT_ALIGNMENT@ = {0}; extern mpif_cmblk9_t _CMPIFCMB9 __attribute__ ((alias("mpifcmb9r"))); extern mpif_cmblk9_t MPIFCMB9 __attribute__ ((alias("mpifcmb9r"))); extern mpif_cmblk9_t MPIFCMB9_ __attribute__ ((alias("mpifcmb9r"))); diff --git a/src/binding/f77/setbotf.f.in b/src/binding/f77/setbotf.f.in index 3c5398f..3053acf 100644 --- a/src/binding/f77/setbotf.f.in +++ b/src/binding/f77/setbotf.f.in @@ -21,12 +21,13 @@ C ARGVS_NULL, ARGV_NULL common /MPIFCMB6/ ecsi common /MPIFCMB7/ asn, pads_a common /MPIFCMB8/ an, pads_b + common /MPIFCMB9/ we save /MPIFCMB1/,/MPIFCMB2/ save /MPIFCMB3/,/MPIFCMB4/,/MPIFCMB5/,/MPIFCMB6/ - save /MPIFCMB7/,/MPIFCMB8/ + save /MPIFCMB7/,/MPIFCMB8/,/MPIFCMB9/ C MPI_ARGVS_NULL C (Fortran requires character data in a separate common block) - call mpirinitc(si, ssi, bt, ip, uw, ecsi, asn) + call mpirinitc(si, ssi, bt, ip, uw, ecsi, asn, we) call mpirinitc2(an) return end ----------------------------------------------------------------------- Summary of changes: src/binding/f77/setbot.c.in | 19 ++++++++++++------- src/binding/f77/setbotf.f.in | 5 +++-- src/mpi/romio/adio/common/ad_coll_exch_new.c | 6 +++--- src/pm/hydra/ui/mpich/mpiexec.c | 3 ++- 4 files changed, 20 insertions(+), 13 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.2-10-g9aa1c31
by noreply＠mpich.org 06 Feb '13

06 Feb '13

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 9aa1c31b967e80344719fcb02dff7a9a05045b09 (commit) from 3593b5c0130a485bfc0da8c383146c6f46e50f70 (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/9aa1c31b967e80344719fcb02dff7a9a0… commit 9aa1c31b967e80344719fcb02dff7a9a05045b09 Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Wed Feb 6 14:04:16 2013 -0600 clmake: more patterns and a regex fix Handle the case where GNU coreutils are installed in some odd `bin` directory, such as `gnubin`. Also fix a bad regex pattern (`[^\w]` means match a single character that is not `\` and is not `w`, rather than the opposite of `\w`, which is `\W`) by using the `\b` "boundary" atom. No reviewer. diff --git a/maint/clmake.in b/maint/clmake.in index 726675c..5532c47 100644 --- a/maint/clmake.in +++ b/maint/clmake.in @@ -39,6 +39,7 @@ $rootSrcDir = ""; "[A-Za-z0-9_\/-]*\/install-sh", "/usr/bin/ar", "mkdir", "/bin/mkdir", + "/.*bin/mkdir", "/usr/ucb/ld", "mpicc", "mpicxx", "compiling ROMIO in", @@ -55,11 +56,15 @@ $rootSrcDir = ""; "PATH=\"\\\$PATH.*\\sldconfig\\s", "building profiling interface in directory", "/usr/bin/install", + "/.*bin/install", "Copying Upshot", "Cleaning directory", "[*][*][*][*] Making .*\.\.\.\.", "Making .* in .*", - "Making upshot", "\\s*\$", ); + "Making upshot", "\\s*\$", + "ld: warning: directory not found for option '-L\\S*src/mpl'", + "ld: warning: directory not found for option '-L\\S*src/openpa/src'", +); # OtherNoise is an array of patterns that describe blocks of # text that we'd like to skip. The initial use it to handle @@ -183,7 +188,7 @@ while ($line = <>) { else { $is_command = 0; foreach $cmdname (@commands) { - if ($line =~ /^\s*$cmdname[^\w]/) { + if ($line =~ /^\s*$cmdname\b/) { $is_command = 1; $cmdline = $line; last; ----------------------------------------------------------------------- Summary of changes: maint/clmake.in | 9 +++++++-- 1 files changed, 7 insertions(+), 2 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.2-9-g3593b5c
by noreply＠mpich.org 06 Feb '13

06 Feb '13

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 3593b5c0130a485bfc0da8c383146c6f46e50f70 (commit) via 460cf5ca35bba3ec72979e7d2a3c189bd4a45853 (commit) via fbd9559322c9918e3973151df222c3d6d4261f10 (commit) from 2bdb181f2ff8607c3cfa6f7d306e94f604dc94ba (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/3593b5c0130a485bfc0da8c383146c6f4… commit 3593b5c0130a485bfc0da8c383146c6f46e50f70 Author: James Dinan <dinan(a)mcs.anl.gov> Date: Wed Feb 6 09:50:21 2013 -0600 PERF: Overlap req/wait for lock-all on shr men win Send out all lock requests before waiting for lock acquisition. This should improve the performance of lock-all operations on shared memory windows. Reviewer: goodell diff --git a/src/mpid/ch3/src/ch3u_rma_sync.c b/src/mpid/ch3/src/ch3u_rma_sync.c index 908064e..4a5ef1b 100644 --- a/src/mpid/ch3/src/ch3u_rma_sync.c +++ b/src/mpid/ch3/src/ch3u_rma_sync.c @@ -2376,11 +2376,15 @@ int MPIDI_Win_lock_all(int assert, MPID_Win *win_ptr) mpi_errno = MPIDI_CH3I_Send_lock_msg(i, MPI_LOCK_SHARED, win_ptr); if (mpi_errno) { MPIU_ERR_POP(mpi_errno); } + } + + for (i = 0; i < MPIR_Comm_size(win_ptr->comm_ptr); i++) { + /* Local process is already locked */ + if (i == win_ptr->myrank) continue; mpi_errno = MPIDI_CH3I_Wait_for_lock_granted(win_ptr, i); if (mpi_errno) { MPIU_ERR_POP(mpi_errno); } } - } fn_exit: http://git.mpich.org/mpich.git/commitdiff/460cf5ca35bba3ec72979e7d2a3c189bd… commit 460cf5ca35bba3ec72979e7d2a3c189bd4a45853 Author: James Dinan <dinan(a)mcs.anl.gov> Date: Tue Feb 5 10:01:42 2013 -0600 PERF: Flush should ignore inactive targets This fix enables flush to ignore non-local targets that have not been targeted by an RMA operation. Prior to this fix, a call to flush-all would result in acquiring the lock at all targets, which is extremely inefficient. Reviewer: goodell diff --git a/src/mpid/ch3/src/ch3u_rma_sync.c b/src/mpid/ch3/src/ch3u_rma_sync.c index 126c91f..908064e 100644 --- a/src/mpid/ch3/src/ch3u_rma_sync.c +++ b/src/mpid/ch3/src/ch3u_rma_sync.c @@ -2136,6 +2136,7 @@ int MPIDI_Win_flush(int rank, MPID_Win *win_ptr) { int mpi_errno = MPI_SUCCESS; int wait_for_rma_done_pkt = 0; + MPIDI_RMA_Op_t *rma_op; MPIDI_STATE_DECL(MPID_STATE_MPIDI_WIN_FLUSH); MPIDI_RMA_FUNC_ENTER(MPID_STATE_MPIDI_WIN_FLUSH); @@ -2174,6 +2175,16 @@ int MPIDI_Win_flush(int rank, MPID_Win *win_ptr) goto fn_exit; } + rma_op = MPIDI_CH3I_RMA_Ops_head(&win_ptr->targets[rank].rma_ops_list); + + /* If there is no activity at this target (e.g. lock-all was called, but we + * haven't communicated with this target), don't do anything. */ + if (win_ptr->targets[rank].remote_lock_state == MPIDI_CH3_WIN_LOCK_CALLED + && rma_op == NULL) + { + goto fn_exit; + } + /* MT: If another thread is performing a flush, wait for them to finish. */ if (win_ptr->targets[rank].remote_lock_state == MPIDI_CH3_WIN_LOCK_FLUSH) { http://git.mpich.org/mpich.git/commitdiff/fbd9559322c9918e3973151df222c3d6d… commit fbd9559322c9918e3973151df222c3d6d4261f10 Author: James Dinan <dinan(a)mcs.anl.gov> Date: Thu Nov 15 11:10:48 2012 -0600 Eliminate enqueueing of lock op in RMA ops list Prior to this patch, a lock entry was enqueued in the RMA ops list when Win_lock was called. This patch adds a new state tracking mechanism, which we use to record the synchronization state with respect to each RMA target. This new mechanism absorbs tracking of lock operation and the lock state at the target. It significantly simplifies the RMA synchronization and ops list processing. Reviewer: goodell diff --git a/src/mpid/ch3/include/mpidpre.h b/src/mpid/ch3/include/mpidpre.h index 366bf1f..de73708 100644 --- a/src/mpid/ch3/include/mpidpre.h +++ b/src/mpid/ch3/include/mpidpre.h @@ -202,6 +202,7 @@ typedef struct MPIDI_VC * MPID_VCR; enum MPIDI_CH3_Lock_states { MPIDI_CH3_WIN_LOCK_NONE = 0, + MPIDI_CH3_WIN_LOCK_CALLED, MPIDI_CH3_WIN_LOCK_REQUESTED, MPIDI_CH3_WIN_LOCK_GRANTED, MPIDI_CH3_WIN_LOCK_FLUSH @@ -250,6 +251,8 @@ struct MPIDI_Win_target_state { (shared/exclusive) of the target process for passive target RMA. Valid whenever state != NONE. */ + int remote_lock_assert; /* Assertion value provided in the call + to Lock */ }; #define MPIDI_DEV_WIN_DECL \ diff --git a/src/mpid/ch3/include/mpidrma.h b/src/mpid/ch3/include/mpidrma.h index f0f7fa1..ad8789d 100644 --- a/src/mpid/ch3/include/mpidrma.h +++ b/src/mpid/ch3/include/mpidrma.h @@ -12,7 +12,7 @@ typedef enum MPIDI_RMA_Op_type { MPIDI_RMA_PUT = 23, MPIDI_RMA_GET = 24, MPIDI_RMA_ACCUMULATE = 25, - MPIDI_RMA_LOCK = 26, + /* REMOVED: MPIDI_RMA_LOCK = 26, */ MPIDI_RMA_ACC_CONTIG = 27, MPIDI_RMA_GET_ACCUMULATE = 28, MPIDI_RMA_COMPARE_AND_SWAP = 29, @@ -70,7 +70,6 @@ typedef struct MPIDI_RMA_Op { int target_count; MPI_Datatype target_datatype; MPI_Op op; /* for accumulate */ - int lock_type; /* for win_lock */ /* Used to complete operations */ struct MPID_Request *request; MPIDI_RMA_dtype_info dtype_info; diff --git a/src/mpid/ch3/src/ch3u_rma_sync.c b/src/mpid/ch3/src/ch3u_rma_sync.c index 6ce857a..126c91f 100644 --- a/src/mpid/ch3/src/ch3u_rma_sync.c +++ b/src/mpid/ch3/src/ch3u_rma_sync.c @@ -1849,7 +1849,7 @@ int MPIDI_Win_test(MPID_Win *win_ptr, int *flag) int MPIDI_Win_lock(int lock_type, int dest, int assert, MPID_Win *win_ptr) { int mpi_errno = MPI_SUCCESS; - MPIDI_RMA_Op_t *rma_op = NULL; + struct MPIDI_Win_target_state *target_state; MPIDI_STATE_DECL(MPID_STATE_MPIDI_WIN_LOCK); MPIDI_RMA_FUNC_ENTER(MPID_STATE_MPIDI_WIN_LOCK); @@ -1868,14 +1868,11 @@ int MPIDI_Win_lock(int lock_type, int dest, int assert, MPID_Win *win_ptr) win_ptr->epoch_state != MPIDI_EPOCH_LOCK, mpi_errno, MPI_ERR_RMA_SYNC, "**rmasync"); -#ifdef HAVE_ERROR_CHECKING - rma_op = MPIDI_CH3I_RMA_Ops_head(&win_ptr->targets[dest].rma_ops_list); + target_state = &win_ptr->targets[dest]; - /* Check if a lock has already been issued (in the ops list or already issued) */ - MPIU_ERR_CHKANDJUMP((rma_op != NULL && rma_op->type == MPIDI_RMA_LOCK) || - win_ptr->targets[dest].remote_lock_state != MPIDI_CH3_WIN_LOCK_NONE, + /* Check if a lock has already been issued */ + MPIU_ERR_CHKANDJUMP(target_state->remote_lock_state != MPIDI_CH3_WIN_LOCK_NONE, mpi_errno, MPI_ERR_RMA_SYNC, "**rmasync"); -#endif /* HAVE_ERROR_CHECKING */ /* Track access epoch state */ if (win_ptr->epoch_state != MPIDI_EPOCH_LOCK_ALL) { @@ -1883,6 +1880,10 @@ int MPIDI_Win_lock(int lock_type, int dest, int assert, MPID_Win *win_ptr) win_ptr->epoch_state = MPIDI_EPOCH_LOCK; } + target_state->remote_lock_state = MPIDI_CH3_WIN_LOCK_CALLED; + target_state->remote_lock_mode = lock_type; + target_state->remote_lock_assert = assert; + if (dest == win_ptr->myrank) { /* The target is this process itself. We must block until the lock * is acquired. Once it is acquired, local puts, gets, accumulates @@ -1903,28 +1904,11 @@ int MPIDI_Win_lock(int lock_type, int dest, int assert, MPID_Win *win_ptr) mpi_errno = MPIDI_CH3I_Wait_for_lock_granted(win_ptr, dest); if (mpi_errno) { MPIU_ERR_POP(mpi_errno); } } - else { + else if (MPIR_PARAM_RMA_LOCK_IMMED) { /* TODO: Make this mode of operation available through an assert argument or info key. */ - if (MPIR_PARAM_RMA_LOCK_IMMED) { - mpi_errno = MPIDI_CH3I_Send_lock_msg(dest, lock_type, win_ptr); - MPIU_ERR_CHKANDJUMP(mpi_errno != MPI_SUCCESS, mpi_errno, MPI_ERR_OTHER, "**winRMAmessage"); - } - else { - MPIDI_RMA_Op_t *new_ptr = NULL; - - /* target is some other process. add the lock request to rma_ops_list */ - MPIU_INSTR_DURATION_START(rmaqueue_alloc); - mpi_errno = MPIDI_CH3I_RMA_Ops_alloc_tail(&win_ptr->targets[dest].rma_ops_list, &new_ptr); - MPIU_INSTR_DURATION_END(rmaqueue_alloc); - if (mpi_errno) { MPIU_ERR_POP(mpi_errno); } - - MPIU_INSTR_DURATION_START(rmaqueue_set); - new_ptr->type = MPIDI_RMA_LOCK; - new_ptr->target_rank = dest; - new_ptr->lock_type = lock_type; - MPIU_INSTR_DURATION_END(rmaqueue_set); - } + mpi_errno = MPIDI_CH3I_Send_lock_msg(dest, lock_type, win_ptr); + MPIU_ERR_CHKANDJUMP(mpi_errno != MPI_SUCCESS, mpi_errno, MPI_ERR_OTHER, "**winRMAmessage"); } fn_exit: @@ -1943,20 +1927,21 @@ int MPIDI_Win_lock(int lock_type, int dest, int assert, MPID_Win *win_ptr) int MPIDI_Win_unlock(int dest, MPID_Win *win_ptr) { int mpi_errno=MPI_SUCCESS; - int single_op_opt, type_size; - MPIDI_RMA_Op_t *rma_op, *curr_op; - MPID_Comm *comm_ptr; - MPIDI_VC_t * vc; - int wait_for_rma_done_pkt = 0, predefined; + int single_op_opt = 0; + MPIDI_RMA_Op_t *rma_op; + int wait_for_rma_done_pkt = 0; MPIDI_STATE_DECL(MPID_STATE_MPIDI_WIN_UNLOCK); MPIDI_RMA_FUNC_ENTER(MPID_STATE_MPIDI_WIN_UNLOCK); if (dest == MPI_PROC_NULL) goto fn_exit; - + MPIU_ERR_CHKANDJUMP(win_ptr->epoch_state != MPIDI_EPOCH_LOCK && win_ptr->epoch_state != MPIDI_EPOCH_LOCK_ALL, mpi_errno, MPI_ERR_RMA_SYNC, "**rmasync"); + MPIU_ERR_CHKANDJUMP(win_ptr->targets[dest].remote_lock_state == MPIDI_CH3_WIN_LOCK_NONE, + mpi_errno, MPI_ERR_RMA_SYNC, "**rmasync"); + /* Track access epoch state */ if (win_ptr->epoch_state == MPIDI_EPOCH_LOCK) { win_ptr->epoch_count--; @@ -1984,54 +1969,39 @@ int MPIDI_Win_unlock(int dest, MPID_Win *win_ptr) goto fn_exit; } - comm_ptr = win_ptr->comm_ptr; - rma_op = MPIDI_CH3I_RMA_Ops_head(&win_ptr->targets[dest].rma_ops_list); - - /* win_lock was not called. return error */ - if ( win_ptr->targets[dest].remote_lock_state == MPIDI_CH3_WIN_LOCK_NONE && - ( rma_op == NULL || rma_op->type != MPIDI_RMA_LOCK ) && - win_ptr->epoch_state != MPIDI_EPOCH_LOCK_ALL ) + + /* Lock was called, but the lock was not requested and there are no ops to + * perform. Do nothing and return. */ + if (rma_op == NULL && + win_ptr->targets[dest].remote_lock_state == MPIDI_CH3_WIN_LOCK_CALLED) { - MPIU_ERR_SETANDJUMP(mpi_errno, MPI_ERR_RMA_SYNC, "**rmasync"); - } - - if (rma_op && rma_op->target_rank != dest) { - /* The target rank is different from the one passed to win_lock! */ - MPIU_ERR_SETANDJUMP2(mpi_errno,MPI_ERR_OTHER,"**winunlockrank", - "**winunlockrank %d %d", dest, rma_op->target_rank); + win_ptr->targets[dest].remote_lock_state = MPIDI_CH3_WIN_LOCK_NONE; + goto fn_exit; } - /* Only win_lock+unlock called, no put/get/acc. If we haven't requested the - lock, we can do nothing and return. */ - if (rma_op && rma_op->type == MPIDI_RMA_LOCK && rma_op->next == NULL && - win_ptr->targets[dest].remote_lock_state == MPIDI_CH3_WIN_LOCK_NONE) { - MPIDI_CH3I_RMA_Ops_free(&win_ptr->targets[dest].rma_ops_list); - goto fn_exit; - } - - single_op_opt = 0; - - MPIDI_Comm_get_vc_set_active(comm_ptr, dest, &vc); - /* TODO: MPI-3: Add lock->cas/fop/gacc->unlock optimization. */ /* TODO: MPI-3: Add lock_all->op optimization. */ /* LOCK-OP-UNLOCK Optimization -- This optimization can't be used if we have already requested the lock. */ if ( MPIR_PARAM_RMA_MERGE_LOCK_OP_UNLOCK && - rma_op && rma_op->type == MPIDI_RMA_LOCK && - rma_op->next->next == NULL && - rma_op->next->type != MPIDI_RMA_COMPARE_AND_SWAP && - rma_op->next->type != MPIDI_RMA_FETCH_AND_OP && - rma_op->next->type != MPIDI_RMA_GET_ACCUMULATE ) { + win_ptr->targets[dest].remote_lock_state == MPIDI_CH3_WIN_LOCK_CALLED && + rma_op && rma_op->next == NULL /* There is only one op */ && + rma_op->type != MPIDI_RMA_COMPARE_AND_SWAP && + rma_op->type != MPIDI_RMA_FETCH_AND_OP && + rma_op->type != MPIDI_RMA_GET_ACCUMULATE ) + { /* Single put, get, or accumulate between the lock and unlock. If it * is of small size and predefined datatype at the target, we * do an optimization where the lock and the RMA operation are * sent in a single packet. Otherwise, we send a separate lock * request first. */ + int type_size, predefined; + MPIDI_VC_t *vc; + MPIDI_RMA_Op_t *curr_op = rma_op; + + MPIDI_Comm_get_vc_set_active(win_ptr->comm_ptr, dest, &vc); - curr_op = rma_op->next; - MPID_Datatype_get_size_macro(curr_op->origin_datatype, type_size); MPIDI_CH3I_DATATYPE_IS_PREDEFINED(curr_op->target_datatype, predefined); @@ -2059,15 +2029,8 @@ int MPIDI_Win_unlock(int dest, MPID_Win *win_ptr) /* Send a lock packet over to the target. wait for the lock_granted reply. Then do all the RMA ops. */ - if (win_ptr->targets[dest].remote_lock_state == MPIDI_CH3_WIN_LOCK_NONE) { - int lock_type; - - if (win_ptr->epoch_state == MPIDI_EPOCH_LOCK_ALL) - lock_type = MPI_LOCK_SHARED; - else - lock_type = rma_op->lock_type; - - mpi_errno = MPIDI_CH3I_Send_lock_msg(dest, lock_type, win_ptr); + if (win_ptr->targets[dest].remote_lock_state == MPIDI_CH3_WIN_LOCK_CALLED) { + mpi_errno = MPIDI_CH3I_Send_lock_msg(dest, win_ptr->targets[dest].remote_lock_mode, win_ptr); if (mpi_errno) { MPIU_ERR_POP(mpi_errno); } } @@ -2112,8 +2075,9 @@ int MPIDI_Win_unlock(int dest, MPID_Win *win_ptr) win_ptr->targets[dest].remote_lock_state = MPIDI_CH3_WIN_LOCK_NONE; } - fn_exit: MPIU_Assert(MPIDI_CH3I_RMA_Ops_isempty(&win_ptr->targets[dest].rma_ops_list)); + + fn_exit: MPIDI_RMA_FUNC_EXIT(MPID_STATE_MPIDI_WIN_UNLOCK); return mpi_errno; /* --BEGIN ERROR HANDLING-- */ @@ -2180,6 +2144,10 @@ int MPIDI_Win_flush(int rank, MPID_Win *win_ptr) win_ptr->epoch_state != MPIDI_EPOCH_LOCK_ALL, mpi_errno, MPI_ERR_RMA_SYNC, "**rmasync"); + /* Check if win_lock was called */ + MPIU_ERR_CHKANDJUMP(win_ptr->targets[rank].remote_lock_state == MPIDI_CH3_WIN_LOCK_NONE, + mpi_errno, MPI_ERR_RMA_SYNC, "**rmasync"); + /* Local flush: ops are performed immediately on the local process */ if (rank == win_ptr->comm_ptr->rank) { MPIU_Assert(win_ptr->targets[rank].remote_lock_state == MPIDI_CH3_WIN_LOCK_GRANTED); @@ -2228,23 +2196,8 @@ int MPIDI_Win_flush(int rank, MPID_Win *win_ptr) /* Send a lock packet over to the target, wait for the lock_granted reply, and perform the RMA ops. */ - if (win_ptr->targets[rank].remote_lock_state == MPIDI_CH3_WIN_LOCK_NONE) { - int lock_type; - if (win_ptr->epoch_state == MPIDI_EPOCH_LOCK_ALL) { - lock_type = MPI_LOCK_SHARED; - } - else { - MPIDI_RMA_Op_t *head = MPIDI_CH3I_RMA_Ops_head(&win_ptr->targets[rank].rma_ops_list); - - /* Ensure that win_lock is waiting at the head of the ops list */ - MPIU_ERR_CHKANDJUMP(MPIDI_CH3I_RMA_Ops_isempty(&win_ptr->targets[rank].rma_ops_list) || - head->type != MPIDI_RMA_LOCK || head->target_rank != rank, - mpi_errno, MPI_ERR_RMA_SYNC, "**rmasync"); - - lock_type = head->lock_type; - } - - mpi_errno = MPIDI_CH3I_Send_lock_msg(rank, lock_type, win_ptr); + if (win_ptr->targets[rank].remote_lock_state == MPIDI_CH3_WIN_LOCK_CALLED) { + mpi_errno = MPIDI_CH3I_Send_lock_msg(rank, win_ptr->targets[rank].remote_lock_mode, win_ptr); if (mpi_errno) { MPIU_ERR_POP(mpi_errno); } } @@ -2386,6 +2339,16 @@ int MPIDI_Win_lock_all(int assert, MPID_Win *win_ptr) /* Track access epoch state */ win_ptr->epoch_state = MPIDI_EPOCH_LOCK_ALL; + /* Set the target's lock state to "called" for all targets */ + /* FIXME: Don't use this O(p) approach */ + for (i = 0; i < MPIR_Comm_size(win_ptr->comm_ptr); i++) { + MPIU_Assert(win_ptr->targets[i].remote_lock_state == MPIDI_CH3_WIN_LOCK_NONE); + + win_ptr->targets[i].remote_lock_state = MPIDI_CH3_WIN_LOCK_CALLED; + win_ptr->targets[i].remote_lock_mode = MPI_LOCK_SHARED; + win_ptr->targets[i].remote_lock_assert = assert; + } + /* Immediately lock the local process for load/store access */ mpi_errno = MPIDI_CH3I_Acquire_local_lock(win_ptr, MPI_LOCK_SHARED); if (mpi_errno != MPI_SUCCESS) { MPIU_ERR_POP(mpi_errno); } @@ -2530,19 +2493,12 @@ static int MPIDI_CH3I_Do_passive_target_rma(MPID_Win *win_ptr, int target_rank, *wait_for_rma_done_pkt = 0; } else { - MPIDI_RMA_Op_t *head = MPIDI_CH3I_RMA_Ops_head(&win_ptr->targets[target_rank].rma_ops_list); - /* go through the list and move the first get operation (if there is one) to the end. Note that the first operation may be a lock, so we can skip it */ - if (head->type == MPIDI_RMA_LOCK) { - curr_ptr = head->next; - } else { - curr_ptr = head; - } - *wait_for_rma_done_pkt = 1; + curr_ptr = MPIDI_CH3I_RMA_Ops_head(&win_ptr->targets[target_rank].rma_ops_list); while (curr_ptr != NULL) { if (curr_ptr->type == MPIDI_RMA_GET) { @@ -2562,13 +2518,6 @@ static int MPIDI_CH3I_Do_passive_target_rma(MPID_Win *win_ptr, int target_rank, curr_ptr = MPIDI_CH3I_RMA_Ops_head(&win_ptr->targets[target_rank].rma_ops_list); - /* Remove the lock operation if it's still on the head of the list */ - if (!MPIDI_CH3I_RMA_Ops_isempty(&win_ptr->targets[target_rank].rma_ops_list) && - MPIDI_CH3I_RMA_Ops_head(&win_ptr->targets[target_rank].rma_ops_list)->type == MPIDI_RMA_LOCK) - { - MPIDI_CH3I_RMA_Ops_free_and_next(&win_ptr->targets[target_rank].rma_ops_list, &curr_ptr); - } - nops = 0; while (curr_ptr != NULL) { nops++; @@ -2667,7 +2616,7 @@ static int MPIDI_CH3I_Send_lock_msg(int dest, int lock_type, MPID_Win *win_ptr) MPIDI_STATE_DECL(MPID_STATE_MPIDI_SEND_LOCK_MSG); MPIDI_RMA_FUNC_ENTER(MPID_STATE_MPIDI_SEND_LOCK_MSG); - MPIU_Assert(win_ptr->targets[dest].remote_lock_state == MPIDI_CH3_WIN_LOCK_NONE); + MPIU_Assert(win_ptr->targets[dest].remote_lock_state == MPIDI_CH3_WIN_LOCK_CALLED); MPIDI_Comm_get_vc_set_active(win_ptr->comm_ptr, dest, &vc); @@ -2756,6 +2705,9 @@ static int MPIDI_CH3I_Wait_for_lock_granted(MPID_Win *win_ptr, int target_rank) * packet sets the remote_lock_state flag to GRANTED. */ + MPIU_Assert(win_ptr->targets[target_rank].remote_lock_state == MPIDI_CH3_WIN_LOCK_REQUESTED || + win_ptr->targets[target_rank].remote_lock_state == MPIDI_CH3_WIN_LOCK_GRANTED); + /* poke the progress engine until remote_lock_state flag is set to GRANTED */ if (win_ptr->targets[target_rank].remote_lock_state != MPIDI_CH3_WIN_LOCK_GRANTED) { MPID_Progress_state progress_state; @@ -2896,9 +2848,9 @@ static int MPIDI_CH3I_Send_lock_put_or_acc(MPID_Win *win_ptr, int target_rank) MPIDI_RMA_FUNC_ENTER(MPID_STATE_MPIDI_CH3I_SEND_LOCK_PUT_OR_ACC); - lock_type = MPIDI_CH3I_RMA_Ops_head(&win_ptr->targets[target_rank].rma_ops_list)->lock_type; + lock_type = win_ptr->targets[target_rank].remote_lock_mode; - rma_op = MPIDI_CH3I_RMA_Ops_head(&win_ptr->targets[target_rank].rma_ops_list)->next; + rma_op = MPIDI_CH3I_RMA_Ops_head(&win_ptr->targets[target_rank].rma_ops_list); win_ptr->pt_rma_puts_accs[rma_op->target_rank]++; @@ -3093,9 +3045,9 @@ static int MPIDI_CH3I_Send_lock_get(MPID_Win *win_ptr, int target_rank) MPIDI_RMA_FUNC_ENTER(MPID_STATE_MPIDI_CH3I_SEND_LOCK_GET); - lock_type = MPIDI_CH3I_RMA_Ops_head(&win_ptr->targets[target_rank].rma_ops_list)->lock_type; + lock_type = win_ptr->targets[target_rank].remote_lock_mode; - rma_op = MPIDI_CH3I_RMA_Ops_head(&win_ptr->targets[target_rank].rma_ops_list)->next; + rma_op = MPIDI_CH3I_RMA_Ops_head(&win_ptr->targets[target_rank].rma_ops_list); /* create a request, store the origin buf, cnt, datatype in it, and pass a handle to it in the get packet. When the get ----------------------------------------------------------------------- Summary of changes: src/mpid/ch3/include/mpidpre.h | 3 + src/mpid/ch3/include/mpidrma.h | 3 +- src/mpid/ch3/src/ch3u_rma_sync.c | 195 ++++++++++++++++---------------------- 3 files changed, 85 insertions(+), 116 deletions(-) hooks/post-receive -- MPICH primary repository

1 0