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

[mpich] MPICH primary repository branch, master, updated. v3.0.1-81-gd3e1b75
by noreply＠mpich.org 25 Jan '13

25 Jan '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 d3e1b7542d877c438054b9e40e3719bc7d3b86c5 (commit) via 412672dbfcfa8dc2eac1192e474d04201644d0d1 (commit) via 44d08f8a49fa103d2db80f84e6e3ac6f4391b5c8 (commit) via e56307ab731333a06b9aa6bb7ea060142eb7efff (commit) via 7f528a332290e730b397932779ffe571b20c4f5c (commit) via a4d103b7ece57a6aae7bac77e58927177a3b2a49 (commit) via 46eacfc76cde82fa9f5bf16975de58a1c9798487 (commit) via 2cfb60178370094f4df8ba43e535f987ff405bb3 (commit) via d6b69b468128c0c356db2d31e8fd4ca0414c3ea4 (commit) from 6a954d01628b1443336bee7e197d91d56690d8f6 (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/d3e1b7542d877c438054b9e40e3719bc7… commit d3e1b7542d877c438054b9e40e3719bc7d3b86c5 Merge: 6a954d0 412672d Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Fri Jan 25 18:20:22 2013 -0600 Merge branch 'nemesis-mpit' Adds Ralf's changes to nemesis that add MPI_T performance variables. I left his commits mostly alone and added several cleanup commits on top of them. http://git.mpich.org/mpich.git/commitdiff/412672dbfcfa8dc2eac1192e474d04201… commit 412672dbfcfa8dc2eac1192e474d04201644d0d1 Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Tue Jan 22 18:15:39 2013 -0600 always define stat vars This makes the stats code more robust to persnickety compilers that might not appreciate references to extern variables that don't really exist. Reviewed-by: rgunter diff --git a/src/mpid/ch3/channels/nemesis/src/mpid_nem_init.c b/src/mpid/ch3/channels/nemesis/src/mpid_nem_init.c index c8cbe64..19df691 100644 --- a/src/mpid/ch3/channels/nemesis/src/mpid_nem_init.c +++ b/src/mpid/ch3/channels/nemesis/src/mpid_nem_init.c @@ -37,13 +37,13 @@ static int get_local_procs(MPIDI_PG_t *pg, int our_pg_rank, int *num_local_p, char *MPID_nem_asymm_base_addr = 0; +/* used by mpid_nem_inline.h and mpid_nem_finalize.c */ +uint64_t *MPID_nem_fbox_fall_back_to_queue_count = NULL; + #if ENABLE_NEM_STATISTICS /* MPIT support */ MPIR_T_SIMPLE_HANDLE_CREATOR(fbox_count_creator, uint64_t, MPID_nem_mem_region.num_local) -/* used by mpid_nem_inline.h and mpid_nem_finalize.c */ -uint64_t *MPID_nem_fbox_fall_back_to_queue_count = NULL; - #undef FUNCNAME #define FUNCNAME MPID_nem_init_stats #undef FCNAME diff --git a/src/mpid/ch3/src/ch3u_recvq.c b/src/mpid/ch3/src/ch3u_recvq.c index 01ce6bb..01e3dcb 100644 --- a/src/mpid/ch3/src/ch3u_recvq.c +++ b/src/mpid/ch3/src/ch3u_recvq.c @@ -59,7 +59,9 @@ MPID_Request ** const MPID_Recvq_posted_head_ptr = &recvq_posted_head; MPID_Request ** const MPID_Recvq_unexpected_head_ptr = &recvq_unexpected_head; #endif -#if ENABLE_RECVQ_STATISTICS +/* Always define these, since our macros will always generate references to + * them. Every compiler we have tested optimizes these "if(0){...}" code paths + * away, even at "-O0". */ static unsigned int posted_qlen = 0; static unsigned int unexpected_qlen = 0; static MPI_Aint posted_recvq_match_attempts = 0; @@ -67,15 +69,6 @@ static MPI_Aint unexpected_recvq_match_attempts = 0; static double time_failed_matching_postedq = 0.0; static double time_matching_unexpectedq = 0.0; uint64_t MPIDI_CH3I_unexpected_recvq_buffer_size = 0; /* used in ch3u_eager.c and ch3u_handle_recv_pkt.c */ -#else -extern unsigned int posted_qlen; -extern unsigned int unexpected_qlen; -extern MPI_Aint posted_recvq_match_attempts; -extern MPI_Aint unexpected_recvq_match_attempts; -extern double time_failed_matching_postedq; -extern double time_matching_unexpectedq; -extern uint64_t MPIDI_CH3I_unexpected_recvq_buffer_size; -#endif /* ENABLE_RECVQ_STATISTICS */ /* If the MPIDI_Message_match structure fits into a pointer size, we http://git.mpich.org/mpich.git/commitdiff/44d08f8a49fa103d2db80f84e6e3ac6f4… commit 44d08f8a49fa103d2db80f84e6e3ac6f4391b5c8 Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Tue Jan 22 17:51:35 2013 -0600 fixups for new MPI_T tests Add the copyright header, limit printing to rank 0, and print a "finished" message to make it clearer that the test completed and didn't just crash. Reviewed-by: rgunter diff --git a/test/mpi/manual/mpi_t/nem_fbox_fallback_to_queue_count.c b/test/mpi/manual/mpi_t/nem_fbox_fallback_to_queue_count.c index 8fea096..f045212 100644 --- a/test/mpi/manual/mpi_t/nem_fbox_fallback_to_queue_count.c +++ b/test/mpi/manual/mpi_t/nem_fbox_fallback_to_queue_count.c @@ -1,3 +1,16 @@ +/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil ; -*- */ +/* + * + * (C) 2013 by Argonne National Laboratory. + * See COPYRIGHT in top-level directory. + */ + +/* This test checks that the nemesis code correctly exposes statistics related + * to "fbox" handling. It also attempts to verify that it accurately maintains + * these statistics. + * + * Originally written by Ralf Gunter Correa Carvalho. */ + #include <mpi.h> #include <assert.h> #include <string.h> @@ -160,12 +173,15 @@ int main(int argc, char *argv[]) MPI_Datatype dtype; MPI_T_enum enumtype; - printf("MPIT pvar test: nem_fbox_fall_back_to_queue_count\n"); fflush(stdout); - MPI_Init(NULL, NULL); MPI_Comm_size(MPI_COMM_WORLD, &size); MPI_Comm_rank(MPI_COMM_WORLD, &rank); + if (rank == 0) { + printf("MPIT pvar test: nem_fbox_fall_back_to_queue_count\n"); + fflush(stdout); + } + /* Ensure we're using exactly two ranks. */ assert(size == 2); @@ -203,6 +219,11 @@ int main(int argc, char *argv[]) MPI_T_pvar_handle_free(session, &fbox_handle); MPI_T_pvar_session_free(&session); + if (rank == 0) { + printf("finished\n"); + fflush(stdout); + } + TRY(MPI_T_finalize()); MPI_Finalize(); diff --git a/test/mpi/manual/mpi_t/unexpected_recvq_buffer_size.c b/test/mpi/manual/mpi_t/unexpected_recvq_buffer_size.c index de896cc..4753e02 100644 --- a/test/mpi/manual/mpi_t/unexpected_recvq_buffer_size.c +++ b/test/mpi/manual/mpi_t/unexpected_recvq_buffer_size.c @@ -1,3 +1,15 @@ +/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil ; -*- */ +/* + * + * (C) 2013 by Argonne National Laboratory. + * See COPYRIGHT in top-level directory. + */ + +/* This test checks that the nemesis code correctly exposes statistics related + * to unexpected receive queue buffer/message sizes. + * + * Originally written by Ralf Gunter Correa Carvalho. */ + #include <mpi.h> #include <assert.h> #include <string.h> @@ -91,12 +103,15 @@ int main(int argc, char *argv[]) MPI_Datatype dtype; MPI_T_enum enumtype; - printf("MPIT pvar test: unexpected_recvq_buffer_size\n"); fflush(stdout); - MPI_Init(NULL, NULL); MPI_Comm_size(MPI_COMM_WORLD, &size); MPI_Comm_rank(MPI_COMM_WORLD, &rank); + if (rank == 0) { + printf("MPIT pvar test: unexpected_recvq_buffer_size\n"); + fflush(stdout); + } + /* Ensure we're using exactly two ranks. */ /* Future tests (using collectives) might need this because of the MPI_Barrier */ assert(size == 2); @@ -131,6 +146,11 @@ int main(int argc, char *argv[]) MPI_T_pvar_handle_free(session, &uqsize_handle); MPI_T_pvar_session_free(&session); + if (rank == 0) { + printf("finished\n"); + fflush(stdout); + } + TRY(MPI_T_finalize()); MPI_Finalize(); http://git.mpich.org/mpich.git/commitdiff/e56307ab731333a06b9aa6bb7ea060142… commit e56307ab731333a06b9aa6bb7ea060142eb7efff Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Tue Jan 22 14:34:57 2013 -0600 move new MPI_T tests to manual/mpi_t dir Reviewed-by: rgunter diff --git a/test/mpi/configure.ac b/test/mpi/configure.ac index dd403ba..f0da35b 100644 --- a/test/mpi/configure.ac +++ b/test/mpi/configure.ac @@ -1409,6 +1409,7 @@ AC_OUTPUT(maint/testmerge \ errors/f90/Makefile \ manual/Makefile \ manual/manyconnect \ + manual/mpi_t/Makefile \ perf/Makefile \ testlist \ cxx/testlist \ diff --git a/test/mpi/manual/mpi_t/Makefile.am b/test/mpi/manual/mpi_t/Makefile.am index 3f84934..24110f8 100644 --- a/test/mpi/manual/mpi_t/Makefile.am +++ b/test/mpi/manual/mpi_t/Makefile.am @@ -7,5 +7,9 @@ include $(top_srcdir)/Makefile.mtest -noinst_PROGRAMS = mpit_test mpit_test2 +# these tests are specific to the MPICH ch3:nemesis implementation of nemesis +nemesis_tests = nem_fbox_fallback_to_queue_count \ + unexpected_recvq_buffer_size +noinst_PROGRAMS = mpit_test mpit_test2 \ + $(nemesis_tests) diff --git a/test/mpi_t/nem_fbox_fallback_to_queue_count.c b/test/mpi/manual/mpi_t/nem_fbox_fallback_to_queue_count.c similarity index 100% rename from test/mpi_t/nem_fbox_fallback_to_queue_count.c rename to test/mpi/manual/mpi_t/nem_fbox_fallback_to_queue_count.c diff --git a/test/mpi_t/unexpected_recvq_buffer_size.c b/test/mpi/manual/mpi_t/unexpected_recvq_buffer_size.c similarity index 100% rename from test/mpi_t/unexpected_recvq_buffer_size.c rename to test/mpi/manual/mpi_t/unexpected_recvq_buffer_size.c http://git.mpich.org/mpich.git/commitdiff/7f528a332290e730b397932779ffe571b… commit 7f528a332290e730b397932779ffe571b20c4f5c Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Tue Jan 22 17:20:06 2013 -0600 fixup for new pvar build system code Reviewed-by: rgunter diff --git a/configure.ac b/configure.ac index d97df82..f17eb74 100644 --- a/configure.ac +++ b/configure.ac @@ -372,15 +372,15 @@ AC_ARG_ENABLE(g, all - All of the above choices ],,enable_g=none) -AC_ARG_ENABLE(mpit_pvars, -[--enable-mpit-pvars=list - Selectively enable MPIT performance variables. +AC_ARG_ENABLE([mpit_pvars], +[--enable-mpit-pvars=list - Selectively enable MPI_T performance variables. list is a comma-separated list of variable names, including none - No performance info recorded recvq - All message queue-related nem - All nemesis-related all - All variables above -],,enable_mpit_pvars=none) +],[],[enable_mpit_pvars=all]) dnl We may want to force MPI_Aint to be the same size as MPI_Offset, dnl particularly on 32 bit systems with large (64 bit) file systems. @@ -401,6 +401,7 @@ AC_ARG_ENABLE(fast, nochkmsg - No error checking, i.e. --disable-error-checking notiming - No timing collection, i.e. --disable-timing. ndebug - Appends -DNDEBUG to MPICHLIB_CFLAGS. + nompit - Equivalent to "--disable-mpit-pvars". all|yes - "defopt", "nochkmsg", "notiming" and "ndebug" are enabled when --enable-fast is specified without any option. none - None of above options, i.e. --disable-fast. Note that @@ -818,6 +819,9 @@ for option in $enable_fast ; do ndebug) enable_append_ndebug=yes ;; + nompit) + enable_mpit_pvars=no + ;; # # [BRT] removed the reseting of enable_g so that --with-enable=dbg,meminit # can be specified with -enable-fast. This change was largely made for the @@ -1726,32 +1730,23 @@ PAC_COMPILER_SHLIB_FLAGS([CC],[$cc_shlib_conf]) AC_SUBST_FILE([cc_shlib_conf]) # --------------------------------------------------------------------------- -# Support for MPIT performance variables +# Support for MPI_T performance variables # enable-mpit-pvars # strip off multiple options, separated by commas save_IFS="$IFS" IFS="," for var in $enable_mpit_pvars ; do - case "$var" in - nem) - enable_nem_pvars=yes - ;; - recvq) - enable_recvq_pvars=yes - ;; - all|yes) - enable_nem_pvars=yes - enable_recvq_pvars=yes - ;; - no|none) - ;; - *) - IFS=$save_IFS - AC_MSG_WARN([Unknown value $option for enable-mpit-pvars]) - IFS="," - ;; - esac + AS_CASE(["$var"], + [nem],[enable_nem_pvars=yes], + [recvq],[enable_recvq_pvars=yes], + [all|yes], + [enable_nem_pvars=yes + enable_recvq_pvars=yes], + [no|none],[], + [IFS=$save_IFS + AC_MSG_WARN([Unknown value ($option) for enable-mpit-pvars]) + IFS=","]) done IFS="$save_IFS" @@ -1761,7 +1756,7 @@ else status_nem_pvars=0 fi AC_DEFINE_UNQUOTED(ENABLE_NEM_STATISTICS,$status_nem_pvars, - [Define to 1 to enable nemesis-related MPIT performance variables]) + [Define to 1 to enable nemesis-related MPI_T performance variables]) if test -n "$enable_recvq_pvars" ; then status_recvq_pvars=1 @@ -1769,7 +1764,7 @@ else status_recvq_pvars=0 fi AC_DEFINE_UNQUOTED(ENABLE_RECVQ_STATISTICS,$status_recvq_pvars, - [Define to 1 to enable message receive queue-related MPIT performance variables]) + [Define to 1 to enable message receive queue-related MPI_T performance variables]) # --------------------------------------------------------------------------- # Support for the language bindings: Fortran 77, Fortran 90, and C++ http://git.mpich.org/mpich.git/commitdiff/a4d103b7ece57a6aae7bac77e58927177… commit a4d103b7ece57a6aae7bac77e58927177a3b2a49 Author: Ralf Gunter <ralfgunter(a)gmail.com> Date: Fri Dec 7 09:31:58 2012 -0600 Add configure args to selectively enable MPIT pvars. Users can now pass --enable-mpit-pvars to configure to statically enable/disable MPIT performance variables so as to limit performance impacts only to those variables they need. Current valid arguments are as follows: none - No performance info recorded recvq - All message queue-related variables nem - All nemesis-related variables all - All variables above A macro framework is added to facilitate the addition of new variables (see MPIR_T_GATE on mpid_common_statistics.h). Also add a check to the MPIT nemesis test to ensure that the variables being tested have been registered with the runtime. diff --git a/configure.ac b/configure.ac index a183a57..d97df82 100644 --- a/configure.ac +++ b/configure.ac @@ -372,6 +372,16 @@ AC_ARG_ENABLE(g, all - All of the above choices ],,enable_g=none) +AC_ARG_ENABLE(mpit_pvars, +[--enable-mpit-pvars=list - Selectively enable MPIT performance variables. + list is a comma-separated list of variable names, + including + none - No performance info recorded + recvq - All message queue-related + nem - All nemesis-related + all - All variables above +],,enable_mpit_pvars=none) + dnl We may want to force MPI_Aint to be the same size as MPI_Offset, dnl particularly on 32 bit systems with large (64 bit) file systems. AC_ARG_WITH(aint-size, @@ -1716,6 +1726,52 @@ PAC_COMPILER_SHLIB_FLAGS([CC],[$cc_shlib_conf]) AC_SUBST_FILE([cc_shlib_conf]) # --------------------------------------------------------------------------- +# Support for MPIT performance variables + +# enable-mpit-pvars +# strip off multiple options, separated by commas +save_IFS="$IFS" +IFS="," +for var in $enable_mpit_pvars ; do + case "$var" in + nem) + enable_nem_pvars=yes + ;; + recvq) + enable_recvq_pvars=yes + ;; + all|yes) + enable_nem_pvars=yes + enable_recvq_pvars=yes + ;; + no|none) + ;; + *) + IFS=$save_IFS + AC_MSG_WARN([Unknown value $option for enable-mpit-pvars]) + IFS="," + ;; + esac +done +IFS="$save_IFS" + +if test -n "$enable_nem_pvars" ; then + status_nem_pvars=1 +else + status_nem_pvars=0 +fi +AC_DEFINE_UNQUOTED(ENABLE_NEM_STATISTICS,$status_nem_pvars, + [Define to 1 to enable nemesis-related MPIT performance variables]) + +if test -n "$enable_recvq_pvars" ; then + status_recvq_pvars=1 +else + status_recvq_pvars=0 +fi +AC_DEFINE_UNQUOTED(ENABLE_RECVQ_STATISTICS,$status_recvq_pvars, + [Define to 1 to enable message receive queue-related MPIT performance variables]) + +# --------------------------------------------------------------------------- # Support for the language bindings: Fortran 77, Fortran 90, and C++ # # First, we handle the case of no explicit enable/disable option. In that diff --git a/src/mpid/ch3/channels/nemesis/include/mpid_nem_inline.h b/src/mpid/ch3/channels/nemesis/include/mpid_nem_inline.h index e2bff14..6d215a1 100644 --- a/src/mpid/ch3/channels/nemesis/include/mpid_nem_inline.h +++ b/src/mpid/ch3/channels/nemesis/include/mpid_nem_inline.h @@ -83,7 +83,7 @@ MPID_nem_mpich_send_header (void* buf, int size, MPIDI_VC_t *vc, int *again) goto return_success; } usequeue_l: - MPIR_T_INC(MPID_nem_fbox_fall_back_to_queue_count[MPID_nem_mem_region.local_ranks[vc->lpid]]); + MPIR_T_INC(NEM_STATISTICS, MPID_nem_fbox_fall_back_to_queue_count[MPID_nem_mem_region.local_ranks[vc->lpid]]); #endif /*USE_FASTBOX */ @@ -315,7 +315,7 @@ MPID_nem_mpich_sendv_header (MPID_IOV **iov, int *n_iov, MPIDI_VC_t *vc, int *ag goto return_success; } usequeue_l: - MPIR_T_INC(MPID_nem_fbox_fall_back_to_queue_count[MPID_nem_mem_region.local_ranks[vc->lpid]]); + MPIR_T_INC(NEM_STATISTICS, MPID_nem_fbox_fall_back_to_queue_count[MPID_nem_mem_region.local_ranks[vc->lpid]]); #endif /*USE_FASTBOX */ @@ -469,7 +469,7 @@ MPID_nem_mpich_send_seg_header (MPID_Segment *segment, MPIDI_msg_sz_t *segment_f } } usequeue_l: - MPIR_T_INC(MPID_nem_fbox_fall_back_to_queue_count[MPID_nem_mem_region.local_ranks[vc->lpid]]); + MPIR_T_INC(NEM_STATISTICS, MPID_nem_fbox_fall_back_to_queue_count[MPID_nem_mem_region.local_ranks[vc->lpid]]); #endif /*USE_FASTBOX */ diff --git a/src/mpid/ch3/channels/nemesis/include/mpidi_nem_statistics.h b/src/mpid/ch3/channels/nemesis/include/mpidi_nem_statistics.h index 0e942f6..bdf6a0a 100644 --- a/src/mpid/ch3/channels/nemesis/include/mpidi_nem_statistics.h +++ b/src/mpid/ch3/channels/nemesis/include/mpidi_nem_statistics.h @@ -7,12 +7,9 @@ #ifndef _MPIDI_NEM_STATISTICS_H_ #define _MPIDI_NEM_STATISTICS_H_ -#define ENABLE_STATISTICS 1 #include "mpidi_common_statistics.h" -#define ENABLE_NEM_STATISTICS 1 - -/* from mpid_nem_init.c */ +/* Do not protect by enable macro - see mpidi_common_statistics.h */ extern uint64_t *MPID_nem_fbox_fall_back_to_queue_count; #endif /* _MPIDI_NEM_STATISTICS_H_ */ diff --git a/src/mpid/ch3/channels/nemesis/src/mpid_nem_init.c b/src/mpid/ch3/channels/nemesis/src/mpid_nem_init.c index 7bf7952..c8cbe64 100644 --- a/src/mpid/ch3/channels/nemesis/src/mpid_nem_init.c +++ b/src/mpid/ch3/channels/nemesis/src/mpid_nem_init.c @@ -37,6 +37,7 @@ static int get_local_procs(MPIDI_PG_t *pg, int our_pg_rank, int *num_local_p, char *MPID_nem_asymm_base_addr = 0; +#if ENABLE_NEM_STATISTICS /* MPIT support */ MPIR_T_SIMPLE_HANDLE_CREATOR(fbox_count_creator, uint64_t, MPID_nem_mem_region.num_local) @@ -75,6 +76,7 @@ static int MPID_nem_init_stats(int n_local_ranks) fn_fail: return mpi_errno; } +#endif /* ENABLE_NEM_STATISTICS */ #undef FUNCNAME #define FUNCNAME MPID_nem_init diff --git a/src/mpid/ch3/include/mpidi_common_statistics.h b/src/mpid/ch3/include/mpidi_common_statistics.h index d4c1d2e..d430bca 100644 --- a/src/mpid/ch3/include/mpidi_common_statistics.h +++ b/src/mpid/ch3/include/mpidi_common_statistics.h @@ -7,16 +7,34 @@ #ifndef _MPIDI_COMMON_STATISTICS_H_ #define _MPIDI_COMMON_STATISTICS_H_ -#if ENABLE_STATISTICS +/* Statically decides whether or not to perform 'action'. + * + * This has the unfortunate side-effect that the compiler will complain about + * undeclared variables in the 'action' parameter. + * Thankfully, since by linking time they will be long gone (due to the compiler + * optimizing away the whole macro), a simple 'extern' declaration suffices. + * This is why some statistics headers declare their variables as extern + * outside of the macro scope. + */ +#define MPIR_T_GATE(PVAR_CLASS, action) \ + do { \ + if (ENABLE_##PVAR_CLASS) { \ + action; \ + } \ + } while(0) + /* TODO add some code here and probably elsewhere to make these show up in the * MPI_T_pvar_ interface */ +#define MPIR_T_INC_impl(x) (++(x)) +#define MPIR_T_DEC_impl(x) (--(x)) -#define MPIR_T_INC(x) (++(x)) -#define MPIR_T_DEC(x) (--(x)) +#define MPIR_T_INC(PVAR_CLASS, x) MPIR_T_GATE(PVAR_CLASS, MPIR_T_INC_impl(x)) +#define MPIR_T_DEC(PVAR_CLASS, x) MPIR_T_GATE(PVAR_CLASS, MPIR_T_DEC_impl(x)) -#define MPIR_T_START_TIMER(start) MPID_Wtime(&start) -#define MPIR_T_END_TIMER(start, mpit_variable) \ + +#define MPIR_T_START_TIMER_impl(start) MPID_Wtime(&start) +#define MPIR_T_END_TIMER_impl(start, mpit_variable) \ do { \ MPID_Time_t end; \ double temp_delta = 0.0; \ @@ -25,21 +43,16 @@ (mpit_variable) += temp_delta; \ } while(0) -#define MPIR_T_SUBTRACT(x, y) ((x) -= (y)) -#define MPIR_T_ADD(x, y) ((x) += (y)) - -#else +#define MPIR_T_START_TIMER(PVAR_CLASS, start) MPIR_T_GATE(PVAR_CLASS, MPIR_T_START_TIMER_impl(start)) +#define MPIR_T_END_TIMER(PVAR_CLASS, start, pvar) MPIR_T_GATE(PVAR_CLASS, MPIR_T_END_TIMER_impl(start, pvar)) -#define MPIR_T_INC(x) -#define MPIR_T_DEC(x) -#define MPIR_T_START_TIMER(start) -#define MPIR_T_END_TIMER(start, mpit_variable) +#define MPIR_T_SUBTRACT_impl(x, y) ((x) -= (y)) +#define MPIR_T_ADD_impl(x, y) ((x) += (y)) -#define MPIR_T_SUBTRACT(x, y) -#define MPIR_T_ADD(x, y) +#define MPIR_T_SUBTRACT(PVAR_CLASS, x, y) MPIR_T_GATE(PVAR_CLASS, MPIR_T_SUBTRACT_impl(x, y)) +#define MPIR_T_ADD(PVAR_CLASS, x, y) MPIR_T_GATE(PVAR_CLASS, MPIR_T_ADD_impl(x, y)) -#endif /* ENABLE_STATISTICS */ #define MPIR_T_SIMPLE_HANDLE_CREATOR(TAG, TYPE, COUNT) \ static int TAG(void *obj_handle, \ @@ -51,7 +64,7 @@ static int TAG(void *obj_handle, handle->handle_state = handle->info->var_state; \ handle->bytes = sizeof(TYPE); \ \ - *countp = (COUNT); \ + *countp = (COUNT); \ return MPI_SUCCESS; \ } diff --git a/src/mpid/ch3/include/mpidi_recvq_statistics.h b/src/mpid/ch3/include/mpidi_recvq_statistics.h index 53a8da3..d0109a2 100644 --- a/src/mpid/ch3/include/mpidi_recvq_statistics.h +++ b/src/mpid/ch3/include/mpidi_recvq_statistics.h @@ -7,11 +7,10 @@ #ifndef _MPIDI_RECVQ_STATISTICS_H_ #define _MPIDI_RECVQ_STATISTICS_H_ -#define ENABLE_STATISTICS 1 #include "mpidi_common_statistics.h" -#define ENABLE_RECVQ_STATISTICS 1 - -extern uint64_t MPIDI_CH3I_unexpected_recvq_buffer_size; /* from ch3u_recvq.c */ +/* Do not protect by enable macro - see mpidi_common_statistics.h */ +/* from ch3u_recvq.c */ +extern uint64_t MPIDI_CH3I_unexpected_recvq_buffer_size; #endif /* _MPIDI_RECVQ_STATISTICS_H_ */ diff --git a/src/mpid/ch3/src/ch3u_eager.c b/src/mpid/ch3/src/ch3u_eager.c index 64310c2..2ca4e78 100644 --- a/src/mpid/ch3/src/ch3u_eager.c +++ b/src/mpid/ch3/src/ch3u_eager.c @@ -436,7 +436,7 @@ int MPIDI_CH3_PktHandler_EagerShortSend( MPIDI_VC_t *vc, MPIDI_CH3_Pkt_t *pkt, a buffer that we've allocated). */ /* printf( "Allocating into tmp\n" ); fflush(stdout); */ recv_data_sz = rreq->dev.recv_data_sz; - MPIR_T_ADD(MPIDI_CH3I_unexpected_recvq_buffer_size, recv_data_sz); + MPIR_T_ADD(RECVQ_STATISTICS, MPIDI_CH3I_unexpected_recvq_buffer_size, recv_data_sz); rreq->dev.tmpbuf = MPIU_Malloc(recv_data_sz); if (!rreq->dev.tmpbuf) { MPIU_ERR_SETANDJUMP(mpi_errno,MPI_ERR_OTHER,"**nomem"); diff --git a/src/mpid/ch3/src/ch3u_handle_recv_pkt.c b/src/mpid/ch3/src/ch3u_handle_recv_pkt.c index 7f3f020..8ea67a8 100644 --- a/src/mpid/ch3/src/ch3u_handle_recv_pkt.c +++ b/src/mpid/ch3/src/ch3u_handle_recv_pkt.c @@ -274,7 +274,7 @@ int MPIDI_CH3U_Receive_data_unexpected(MPID_Request * rreq, char *buf, MPIDI_msg } if (MPIDI_Request_get_msg_type(rreq) == MPIDI_REQUEST_EAGER_MSG) - MPIR_T_ADD(MPIDI_CH3I_unexpected_recvq_buffer_size, rreq->dev.tmpbuf_sz); + MPIR_T_ADD(RECVQ_STATISTICS, MPIDI_CH3I_unexpected_recvq_buffer_size, rreq->dev.tmpbuf_sz); rreq->dev.OnDataAvail = MPIDI_CH3_ReqHandler_UnpackUEBufComplete; diff --git a/src/mpid/ch3/src/ch3u_recvq.c b/src/mpid/ch3/src/ch3u_recvq.c index 6b3b823..01ce6bb 100644 --- a/src/mpid/ch3/src/ch3u_recvq.c +++ b/src/mpid/ch3/src/ch3u_recvq.c @@ -67,7 +67,16 @@ static MPI_Aint unexpected_recvq_match_attempts = 0; static double time_failed_matching_postedq = 0.0; static double time_matching_unexpectedq = 0.0; uint64_t MPIDI_CH3I_unexpected_recvq_buffer_size = 0; /* used in ch3u_eager.c and ch3u_handle_recv_pkt.c */ -#endif /* defined(ENABLE_RECVQ_STATISTICS) */ +#else +extern unsigned int posted_qlen; +extern unsigned int unexpected_qlen; +extern MPI_Aint posted_recvq_match_attempts; +extern MPI_Aint unexpected_recvq_match_attempts; +extern double time_failed_matching_postedq; +extern double time_matching_unexpectedq; +extern uint64_t MPIDI_CH3I_unexpected_recvq_buffer_size; +#endif /* ENABLE_RECVQ_STATISTICS */ + /* If the MPIDI_Message_match structure fits into a pointer size, we * can directly work on it */ @@ -114,7 +123,7 @@ MPIR_T_SIMPLE_HANDLE_CREATOR(simple_double_creator, double, 1) int MPIDI_CH3U_Recvq_init(void) { int mpi_errno = MPI_SUCCESS; -#ifdef ENABLE_RECVQ_STATISTICS +#if ENABLE_RECVQ_STATISTICS int idx = -1; mpi_errno = MPIR_T_pvar_add("posted_recvq_length", MPI_T_VERBOSITY_USER_DETAIL, @@ -282,14 +291,14 @@ int MPIDI_CH3U_Recvq_FU(int source, int tag, int context_id, MPI_Status *s) match.parts.rank = source; if (tag != MPI_ANY_TAG && source != MPI_ANY_SOURCE) { - MPIR_T_START_TIMER(timer_start); + MPIR_T_START_TIMER(RECVQ_STATISTICS, timer_start); while (rreq != NULL) { - MPIR_T_INC(unexpected_recvq_match_attempts); + MPIR_T_INC(RECVQ_STATISTICS, unexpected_recvq_match_attempts); if (MATCH_WITH_NO_MASK(rreq->dev.match, match)) break; rreq = rreq->dev.next; } - MPIR_T_END_TIMER(timer_start, time_matching_unexpectedq); + MPIR_T_END_TIMER(RECVQ_STATISTICS, timer_start, time_matching_unexpectedq); } else { mask.parts.context_id = mask.parts.rank = mask.parts.tag = ~0; @@ -298,14 +307,14 @@ int MPIDI_CH3U_Recvq_FU(int source, int tag, int context_id, MPI_Status *s) if (source == MPI_ANY_SOURCE) match.parts.rank = mask.parts.rank = 0; - MPIR_T_START_TIMER(timer_start); + MPIR_T_START_TIMER(RECVQ_STATISTICS, timer_start); while (rreq != NULL) { - MPIR_T_INC(unexpected_recvq_match_attempts); + MPIR_T_INC(RECVQ_STATISTICS, unexpected_recvq_match_attempts); if (MATCH_WITH_LEFT_MASK(rreq->dev.match, match, mask)) break; rreq = rreq->dev.next; } - MPIR_T_END_TIMER(timer_start, time_matching_unexpectedq); + MPIR_T_END_TIMER(RECVQ_STATISTICS, timer_start, time_matching_unexpectedq); } /* Save the information about the request before releasing the @@ -365,14 +374,17 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU(MPI_Request sreq_id, /* FIXME: Why doesn't this exit after it finds the first match? */ cur_rreq = recvq_unexpected_head; while (cur_rreq != NULL) { - MPIR_T_START_TIMER(timer_start); - if (cur_rreq->dev.sender_req_id == sreq_id && - MPIR_T_INC(unexpected_recvq_match_attempts) && - (MATCH_WITH_NO_MASK(cur_rreq->dev.match, *match))) { - matching_prev_rreq = prev_rreq; - matching_cur_rreq = cur_rreq; + MPIR_T_START_TIMER(RECVQ_STATISTICS, timer_start); + + if (cur_rreq->dev.sender_req_id == sreq_id) { + MPIR_T_INC(RECVQ_STATISTICS, unexpected_recvq_match_attempts); + + if (MATCH_WITH_NO_MASK(cur_rreq->dev.match, *match)) { + matching_prev_rreq = prev_rreq; + matching_cur_rreq = cur_rreq; + } } - MPIR_T_END_TIMER(timer_start, time_matching_unexpectedq); + MPIR_T_END_TIMER(RECVQ_STATISTICS, timer_start, time_matching_unexpectedq); prev_rreq = cur_rreq; cur_rreq = cur_rreq->dev.next; @@ -390,10 +402,10 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU(MPI_Request sreq_id, recvq_unexpected_tail = matching_prev_rreq; } - MPIR_T_DEC(unexpected_qlen); + MPIR_T_DEC(RECVQ_STATISTICS, unexpected_qlen); rreq = matching_cur_rreq; - MPIR_T_SUBTRACT(MPIDI_CH3I_unexpected_recvq_buffer_size, rreq->dev.tmpbuf_sz); + MPIR_T_SUBTRACT(RECVQ_STATISTICS, MPIDI_CH3I_unexpected_recvq_buffer_size, rreq->dev.tmpbuf_sz); } else { rreq = NULL; @@ -425,7 +437,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_matchonly(int source, int tag, int context_i MPIU_THREAD_CS_ASSERT_HELD(MSGQUEUE); /* Store how much time is spent traversing the queue */ - MPIR_T_START_TIMER(timer_start); + MPIR_T_START_TIMER(RECVQ_STATISTICS, timer_start); /* Optimize this loop for an empty unexpected receive queue */ rreq = recvq_unexpected_head; @@ -438,7 +450,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_matchonly(int source, int tag, int context_i if (tag != MPI_ANY_TAG && source != MPI_ANY_SOURCE) { do { - MPIR_T_INC(unexpected_recvq_match_attempts); + MPIR_T_INC(RECVQ_STATISTICS, unexpected_recvq_match_attempts); if (MATCH_WITH_NO_MASK(rreq->dev.match, match)) { if (prev_rreq != NULL) { prev_rreq->dev.next = rreq->dev.next; @@ -450,8 +462,8 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_matchonly(int source, int tag, int context_i if (rreq->dev.next == NULL) { recvq_unexpected_tail = prev_rreq; } - MPIR_T_DEC(unexpected_qlen); - MPIR_T_SUBTRACT(MPIDI_CH3I_unexpected_recvq_buffer_size, rreq->dev.tmpbuf_sz); + MPIR_T_DEC(RECVQ_STATISTICS, unexpected_qlen); + MPIR_T_SUBTRACT(RECVQ_STATISTICS, MPIDI_CH3I_unexpected_recvq_buffer_size, rreq->dev.tmpbuf_sz); rreq->comm = comm; MPIR_Comm_add_ref(comm); @@ -472,7 +484,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_matchonly(int source, int tag, int context_i match.parts.rank = mask.parts.rank = 0; do { - MPIR_T_INC(unexpected_recvq_match_attempts); + MPIR_T_INC(RECVQ_STATISTICS, unexpected_recvq_match_attempts); if (MATCH_WITH_LEFT_MASK(rreq->dev.match, match, mask)) { if (prev_rreq != NULL) { prev_rreq->dev.next = rreq->dev.next; @@ -483,8 +495,8 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_matchonly(int source, int tag, int context_i if (rreq->dev.next == NULL) { recvq_unexpected_tail = prev_rreq; } - MPIR_T_DEC(unexpected_qlen); - MPIR_T_SUBTRACT(MPIDI_CH3I_unexpected_recvq_buffer_size, rreq->dev.tmpbuf_sz); + MPIR_T_DEC(RECVQ_STATISTICS, unexpected_qlen); + MPIR_T_SUBTRACT(RECVQ_STATISTICS, MPIDI_CH3I_unexpected_recvq_buffer_size, rreq->dev.tmpbuf_sz); rreq->comm = comm; MPIR_Comm_add_ref(comm); @@ -500,7 +512,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_matchonly(int source, int tag, int context_i } lock_exit: - MPIR_T_END_TIMER(timer_start, time_matching_unexpectedq); + MPIR_T_END_TIMER(RECVQ_STATISTICS, timer_start, time_matching_unexpectedq); *foundp = found; @@ -542,7 +554,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_or_AEP(int source, int tag, MPIU_THREAD_CS_ASSERT_HELD(MSGQUEUE); /* Store how much time is spent traversing the queue */ - MPIR_T_START_TIMER(timer_start); + MPIR_T_START_TIMER(RECVQ_STATISTICS, timer_start); /* Optimize this loop for an empty unexpected receive queue */ rreq = recvq_unexpected_head; @@ -555,7 +567,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_or_AEP(int source, int tag, if (tag != MPI_ANY_TAG && source != MPI_ANY_SOURCE) { do { - MPIR_T_INC(unexpected_recvq_match_attempts); + MPIR_T_INC(RECVQ_STATISTICS, unexpected_recvq_match_attempts); if (MATCH_WITH_NO_MASK(rreq->dev.match, match)) { if (prev_rreq != NULL) { prev_rreq->dev.next = rreq->dev.next; @@ -567,10 +579,10 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_or_AEP(int source, int tag, if (rreq->dev.next == NULL) { recvq_unexpected_tail = prev_rreq; } - MPIR_T_DEC(unexpected_qlen); + MPIR_T_DEC(RECVQ_STATISTICS, unexpected_qlen); if (MPIDI_Request_get_msg_type(rreq) == MPIDI_REQUEST_EAGER_MSG) - MPIR_T_SUBTRACT(MPIDI_CH3I_unexpected_recvq_buffer_size, rreq->dev.tmpbuf_sz); + MPIR_T_SUBTRACT(RECVQ_STATISTICS, MPIDI_CH3I_unexpected_recvq_buffer_size, rreq->dev.tmpbuf_sz); rreq->comm = comm; MPIR_Comm_add_ref(comm); @@ -592,7 +604,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_or_AEP(int source, int tag, match.parts.rank = mask.parts.rank = 0; do { - MPIR_T_INC(unexpected_recvq_match_attempts); + MPIR_T_INC(RECVQ_STATISTICS, unexpected_recvq_match_attempts); if (MATCH_WITH_LEFT_MASK(rreq->dev.match, match, mask)) { if (prev_rreq != NULL) { prev_rreq->dev.next = rreq->dev.next; @@ -603,10 +615,10 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_or_AEP(int source, int tag, if (rreq->dev.next == NULL) { recvq_unexpected_tail = prev_rreq; } - MPIR_T_DEC(unexpected_qlen); + MPIR_T_DEC(RECVQ_STATISTICS, unexpected_qlen); if (MPIDI_Request_get_msg_type(rreq) == MPIDI_REQUEST_EAGER_MSG) - MPIR_T_SUBTRACT(MPIDI_CH3I_unexpected_recvq_buffer_size, rreq->dev.tmpbuf_sz); + MPIR_T_SUBTRACT(RECVQ_STATISTICS, MPIDI_CH3I_unexpected_recvq_buffer_size, rreq->dev.tmpbuf_sz); rreq->comm = comm; MPIR_Comm_add_ref(comm); @@ -621,7 +633,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_or_AEP(int source, int tag, } while (rreq); } } - MPIR_T_END_TIMER(timer_start, time_matching_unexpectedq); + MPIR_T_END_TIMER(RECVQ_STATISTICS, timer_start, time_matching_unexpectedq); /* A matching request was not found in the unexpected queue, so we need to allocate a new request and add it to the posted queue */ @@ -679,7 +691,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_or_AEP(int source, int tag, recvq_posted_head = rreq; } recvq_posted_tail = rreq; - MPIR_T_INC(posted_qlen); + MPIR_T_INC(RECVQ_STATISTICS, posted_qlen); MPIDI_POSTED_RECV_ENQUEUE_HOOK(rreq); } @@ -688,7 +700,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_or_AEP(int source, int tag, /* If a match was not found, the timer was stopped after the traversal */ if (found) - MPIR_T_END_TIMER(timer_start, time_matching_unexpectedq); + MPIR_T_END_TIMER(RECVQ_STATISTICS, timer_start, time_matching_unexpectedq); MPIDI_FUNC_EXIT(MPID_STATE_MPIDI_CH3U_RECVQ_FDU_OR_AEP); return rreq; @@ -723,7 +735,7 @@ int MPIDI_CH3U_Recvq_DP(MPID_Request * rreq) /* MT FIXME is this right? or should the caller do this? */ MPIU_THREAD_CS_ENTER(MSGQUEUE,); - MPIR_T_START_TIMER(timer_start); + MPIR_T_START_TIMER(RECVQ_STATISTICS, timer_start); cur_rreq = recvq_posted_head; while (cur_rreq != NULL) { if (cur_rreq == rreq) { @@ -736,7 +748,7 @@ int MPIDI_CH3U_Recvq_DP(MPID_Request * rreq) if (cur_rreq->dev.next == NULL) { recvq_posted_tail = prev_rreq; } - MPIR_T_DEC(posted_qlen); + MPIR_T_DEC(RECVQ_STATISTICS, posted_qlen); /* Notify channel that rreq has been dequeued and check if it has already matched rreq, fail if so */ dequeue_failed = MPIDI_POSTED_RECV_DEQUEUE_HOOK(rreq); @@ -749,7 +761,7 @@ int MPIDI_CH3U_Recvq_DP(MPID_Request * rreq) cur_rreq = cur_rreq->dev.next; } if (!found) - MPIR_T_END_TIMER(timer_start, time_failed_matching_postedq); + MPIR_T_END_TIMER(RECVQ_STATISTICS, timer_start, time_failed_matching_postedq); MPIU_THREAD_CS_EXIT(MSGQUEUE,); @@ -799,9 +811,9 @@ MPID_Request * MPIDI_CH3U_Recvq_FDP_or_AEU(MPIDI_Message_match * match, rreq = recvq_posted_head; - MPIR_T_START_TIMER(timer_start); + MPIR_T_START_TIMER(RECVQ_STATISTICS, timer_start); while (rreq != NULL) { - MPIR_T_INC(posted_recvq_match_attempts); + MPIR_T_INC(RECVQ_STATISTICS, posted_recvq_match_attempts); if (MATCH_WITH_LEFT_RIGHT_MASK(rreq->dev.match, *match, rreq->dev.mask)) { if (prev_rreq != NULL) { prev_rreq->dev.next = rreq->dev.next; @@ -812,7 +824,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDP_or_AEU(MPIDI_Message_match * match, if (rreq->dev.next == NULL) { recvq_posted_tail = prev_rreq; } - MPIR_T_DEC(posted_qlen); + MPIR_T_DEC(RECVQ_STATISTICS, posted_qlen); /* give channel a chance to match the request, try again if so */ channel_matched = MPIDI_POSTED_RECV_DEQUEUE_HOOK(rreq); @@ -825,7 +837,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDP_or_AEU(MPIDI_Message_match * match, prev_rreq = rreq; rreq = rreq->dev.next; } - MPIR_T_END_TIMER(timer_start, time_failed_matching_postedq); + MPIR_T_END_TIMER(RECVQ_STATISTICS, timer_start, time_failed_matching_postedq); /* A matching request was not found in the posted queue, so we need to allocate a new request and add it to the unexpected queue */ @@ -844,7 +856,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDP_or_AEU(MPIDI_Message_match * match, recvq_unexpected_head = rreq; } recvq_unexpected_tail = rreq; - MPIR_T_INC(unexpected_qlen); + MPIR_T_INC(RECVQ_STATISTICS, unexpected_qlen); } found = FALSE; @@ -891,7 +903,7 @@ static inline void dequeue_and_set_error(MPID_Request **req, MPID_Request *prev if (recvq_posted_tail == *req) recvq_posted_tail = prev_req; - MPIR_T_DEC(posted_qlen); + MPIR_T_DEC(RECVQ_STATISTICS, posted_qlen); /* set error and complete */ (*req)->status.MPI_ERROR = *error; diff --git a/test/mpi_t/nem_fbox_fallback_to_queue_count.c b/test/mpi_t/nem_fbox_fallback_to_queue_count.c index e4a0530..8fea096 100644 --- a/test/mpi_t/nem_fbox_fallback_to_queue_count.c +++ b/test/mpi_t/nem_fbox_fallback_to_queue_count.c @@ -154,7 +154,8 @@ void recv_first_test() int main(int argc, char *argv[]) { int i, size, num, name_len, desc_len, count, verb, thread_support; - int varclass, bind, readonly, continuous, atomic, fbox_idx; + int varclass, bind, readonly, continuous, atomic; + int fbox_idx = -1; char name[STR_LEN], desc[STR_LEN]; MPI_Datatype dtype; MPI_T_enum enumtype; @@ -183,6 +184,9 @@ int main(int argc, char *argv[]) fbox_idx = i; } + /* Ensure variable was registered by the runtime */ + assert(fbox_idx != -1); + /* Initialize MPIT session & variable handle. */ MPI_T_pvar_session_create(&session); MPI_T_pvar_handle_alloc(session, fbox_idx, NULL, &fbox_handle, &count); http://git.mpich.org/mpich.git/commitdiff/46eacfc76cde82fa9f5bf16975de58a1c… commit 46eacfc76cde82fa9f5bf16975de58a1c9798487 Author: Ralf Gunter <ralfgunter(a)gmail.com> Date: Thu Nov 29 16:31:48 2012 -0600 Add MPIT variables and corresponding tests. The following MPIT performance variables are added: - nem_fbox_fall_back_to_queue_count; - time_failed_matching_postedq; - time_matching_unexpectedq; - unexpected_recvq_buffer_size. Some common MPIT-related routines, handle creation functions and enable/disable switches are gathered in a common file. Two test cases are furnished, but not yet integrated with 'make check'. diff --git a/src/mpid/ch3/channels/nemesis/include/mpid_nem_inline.h b/src/mpid/ch3/channels/nemesis/include/mpid_nem_inline.h index ee71cc6..e2bff14 100644 --- a/src/mpid/ch3/channels/nemesis/include/mpid_nem_inline.h +++ b/src/mpid/ch3/channels/nemesis/include/mpid_nem_inline.h @@ -11,6 +11,7 @@ #include "my_papi_defs.h" #include "mpiiov.h" +#include "mpidi_nem_statistics.h" extern int MPID_nem_lmt_shm_pending; extern MPID_nem_cell_ptr_t MPID_nem_prefetched_cell; @@ -82,6 +83,8 @@ MPID_nem_mpich_send_header (void* buf, int size, MPIDI_VC_t *vc, int *again) goto return_success; } usequeue_l: + MPIR_T_INC(MPID_nem_fbox_fall_back_to_queue_count[MPID_nem_mem_region.local_ranks[vc->lpid]]); + #endif /*USE_FASTBOX */ #ifdef PREFETCH_CELL @@ -312,6 +315,7 @@ MPID_nem_mpich_sendv_header (MPID_IOV **iov, int *n_iov, MPIDI_VC_t *vc, int *ag goto return_success; } usequeue_l: + MPIR_T_INC(MPID_nem_fbox_fall_back_to_queue_count[MPID_nem_mem_region.local_ranks[vc->lpid]]); #endif /*USE_FASTBOX */ @@ -465,6 +469,7 @@ MPID_nem_mpich_send_seg_header (MPID_Segment *segment, MPIDI_msg_sz_t *segment_f } } usequeue_l: + MPIR_T_INC(MPID_nem_fbox_fall_back_to_queue_count[MPID_nem_mem_region.local_ranks[vc->lpid]]); #endif /*USE_FASTBOX */ diff --git a/src/mpid/ch3/channels/nemesis/include/mpidi_nem_statistics.h b/src/mpid/ch3/channels/nemesis/include/mpidi_nem_statistics.h new file mode 100644 index 0000000..0e942f6 --- /dev/null +++ b/src/mpid/ch3/channels/nemesis/include/mpidi_nem_statistics.h @@ -0,0 +1,18 @@ +/* -*- Mode: c-basic-offset:4 ; indent-tabs-mode:nil ; -*- */ +/* + * (C) 2013 by Argonne National Laboratory. + * See COPYRIGHT in top-level directory. + */ + +#ifndef _MPIDI_NEM_STATISTICS_H_ +#define _MPIDI_NEM_STATISTICS_H_ + +#define ENABLE_STATISTICS 1 +#include "mpidi_common_statistics.h" + +#define ENABLE_NEM_STATISTICS 1 + +/* from mpid_nem_init.c */ +extern uint64_t *MPID_nem_fbox_fall_back_to_queue_count; + +#endif /* _MPIDI_NEM_STATISTICS_H_ */ diff --git a/src/mpid/ch3/channels/nemesis/src/mpid_nem_finalize.c b/src/mpid/ch3/channels/nemesis/src/mpid_nem_finalize.c index baed188..f5cc1a5 100644 --- a/src/mpid/ch3/channels/nemesis/src/mpid_nem_finalize.c +++ b/src/mpid/ch3/channels/nemesis/src/mpid_nem_finalize.c @@ -10,6 +10,8 @@ #include "pmi.h" #endif +#include "mpidi_nem_statistics.h" + #undef FUNCNAME #define FUNCNAME MPID_nem_finalize #undef FCNAME @@ -54,6 +56,10 @@ int MPID_nem_finalize(void) my_papi_close(); #endif /*PAPI_MONITOR */ +#if ENABLE_NEM_STATISTICS + MPIU_Free(MPID_nem_fbox_fall_back_to_queue_count); +#endif + fn_exit: MPIDI_FUNC_EXIT(MPID_STATE_MPID_NEM_FINALIZE); return mpi_errno; diff --git a/src/mpid/ch3/channels/nemesis/src/mpid_nem_init.c b/src/mpid/ch3/channels/nemesis/src/mpid_nem_init.c index 947da3b..7bf7952 100644 --- a/src/mpid/ch3/channels/nemesis/src/mpid_nem_init.c +++ b/src/mpid/ch3/channels/nemesis/src/mpid_nem_init.c @@ -8,6 +8,7 @@ #include "mpid_nem_impl.h" #include "mpid_nem_nets.h" #include <errno.h> +#include "mpidi_nem_statistics.h" /* constants for configure time selection of local LMT implementations */ #define MPID_NEM_LOCAL_LMT_NONE 0 @@ -36,6 +37,45 @@ static int get_local_procs(MPIDI_PG_t *pg, int our_pg_rank, int *num_local_p, char *MPID_nem_asymm_base_addr = 0; +/* MPIT support */ +MPIR_T_SIMPLE_HANDLE_CREATOR(fbox_count_creator, uint64_t, MPID_nem_mem_region.num_local) + +/* used by mpid_nem_inline.h and mpid_nem_finalize.c */ +uint64_t *MPID_nem_fbox_fall_back_to_queue_count = NULL; + +#undef FUNCNAME +#define FUNCNAME MPID_nem_init_stats +#undef FCNAME +#define FCNAME MPIDI_QUOTE(FUNCNAME) +static int MPID_nem_init_stats(int n_local_ranks) +{ + int mpi_errno = MPI_SUCCESS; + int idx = -1; + + MPID_nem_fbox_fall_back_to_queue_count = MPIU_Calloc(n_local_ranks, sizeof(uint64_t)); + + mpi_errno = MPIR_T_pvar_add("nem_fbox_fall_back_to_queue_count", + MPI_T_VERBOSITY_USER_DETAIL, + MPI_T_PVAR_CLASS_COUNTER, + MPI_AINT, + MPI_T_ENUM_NULL, + "array counting how many times nemesis had to fall back to " + "the regular queue when sending messages between pairs of " + "local processes", + MPI_T_BIND_NO_OBJECT, + /*readonly=*/ FALSE, + /*continuous=*/ TRUE, + /*atomic=*/ FALSE, + MPIR_T_PVAR_IMPL_SIMPLE, + /*var_state=*/ MPID_nem_fbox_fall_back_to_queue_count, + &fbox_count_creator, + &idx); + if (mpi_errno) MPIU_ERR_POP(mpi_errno); + +fn_fail: + return mpi_errno; +} + #undef FUNCNAME #define FUNCNAME MPID_nem_init #undef FCNAME @@ -341,6 +381,10 @@ MPID_nem_init(int pg_rank, MPIDI_PG_t *pg_p, int has_parent ATTRIBUTE((unused))) my_papi_start( pg_rank ); #endif /*PAPI_MONITOR */ +#if ENABLE_NEM_STATISTICS + MPID_nem_init_stats(num_local); +#endif + MPIU_CHKPMEM_COMMIT(); fn_exit: return mpi_errno; @@ -349,6 +393,7 @@ MPID_nem_init(int pg_rank, MPIDI_PG_t *pg_p, int has_parent ATTRIBUTE((unused))) MPIU_CHKPMEM_REAP(); goto fn_exit; /* --END ERROR HANDLING-- */ + } /* MPID_nem_vc_init initialize nemesis' part of the vc */ diff --git a/src/mpid/ch3/include/mpidi_common_statistics.h b/src/mpid/ch3/include/mpidi_common_statistics.h new file mode 100644 index 0000000..d4c1d2e --- /dev/null +++ b/src/mpid/ch3/include/mpidi_common_statistics.h @@ -0,0 +1,58 @@ +/* -*- Mode: c-basic-offset:4 ; indent-tabs-mode:nil ; -*- */ +/* + * (C) 2013 by Argonne National Laboratory. + * See COPYRIGHT in top-level directory. + */ + +#ifndef _MPIDI_COMMON_STATISTICS_H_ +#define _MPIDI_COMMON_STATISTICS_H_ + +#if ENABLE_STATISTICS + +/* TODO add some code here and probably elsewhere to make these show up in the + * MPI_T_pvar_ interface */ + +#define MPIR_T_INC(x) (++(x)) +#define MPIR_T_DEC(x) (--(x)) + +#define MPIR_T_START_TIMER(start) MPID_Wtime(&start) +#define MPIR_T_END_TIMER(start, mpit_variable) \ + do { \ + MPID_Time_t end; \ + double temp_delta = 0.0; \ + MPID_Wtime(&(end)); \ + MPID_Wtime_diff(&(start), &end, &temp_delta); \ + (mpit_variable) += temp_delta; \ + } while(0) + +#define MPIR_T_SUBTRACT(x, y) ((x) -= (y)) +#define MPIR_T_ADD(x, y) ((x) += (y)) + +#else + +#define MPIR_T_INC(x) +#define MPIR_T_DEC(x) + +#define MPIR_T_START_TIMER(start) +#define MPIR_T_END_TIMER(start, mpit_variable) + +#define MPIR_T_SUBTRACT(x, y) +#define MPIR_T_ADD(x, y) + +#endif /* ENABLE_STATISTICS */ + +#define MPIR_T_SIMPLE_HANDLE_CREATOR(TAG, TYPE, COUNT) \ +static int TAG(void *obj_handle, \ + struct MPIR_T_pvar_handle *handle, \ + int *countp) \ +{ \ + /* the IMPL_SIMPLE code reads/writes "bytes" bytes from the location given \ + * by the "handle_state" pointer */ \ + handle->handle_state = handle->info->var_state; \ + handle->bytes = sizeof(TYPE); \ + \ + *countp = (COUNT); \ + return MPI_SUCCESS; \ +} + +#endif /* _MPIDI_COMMON_STATISTICS_H_ */ diff --git a/src/mpid/ch3/include/mpidi_recvq_statistics.h b/src/mpid/ch3/include/mpidi_recvq_statistics.h new file mode 100644 index 0000000..53a8da3 --- /dev/null +++ b/src/mpid/ch3/include/mpidi_recvq_statistics.h @@ -0,0 +1,17 @@ +/* -*- Mode: c-basic-offset:4 ; indent-tabs-mode:nil ; -*- */ +/* + * (C) 2013 by Argonne National Laboratory. + * See COPYRIGHT in top-level directory. + */ + +#ifndef _MPIDI_RECVQ_STATISTICS_H_ +#define _MPIDI_RECVQ_STATISTICS_H_ + +#define ENABLE_STATISTICS 1 +#include "mpidi_common_statistics.h" + +#define ENABLE_RECVQ_STATISTICS 1 + +extern uint64_t MPIDI_CH3I_unexpected_recvq_buffer_size; /* from ch3u_recvq.c */ + +#endif /* _MPIDI_RECVQ_STATISTICS_H_ */ diff --git a/src/mpid/ch3/src/ch3u_eager.c b/src/mpid/ch3/src/ch3u_eager.c index 4800e23..64310c2 100644 --- a/src/mpid/ch3/src/ch3u_eager.c +++ b/src/mpid/ch3/src/ch3u_eager.c @@ -5,6 +5,7 @@ */ #include "mpidimpl.h" +#include "mpidi_recvq_statistics.h" /* * Send an eager message. To optimize for the important, short contiguous @@ -435,6 +436,7 @@ int MPIDI_CH3_PktHandler_EagerShortSend( MPIDI_VC_t *vc, MPIDI_CH3_Pkt_t *pkt, a buffer that we've allocated). */ /* printf( "Allocating into tmp\n" ); fflush(stdout); */ recv_data_sz = rreq->dev.recv_data_sz; + MPIR_T_ADD(MPIDI_CH3I_unexpected_recvq_buffer_size, recv_data_sz); rreq->dev.tmpbuf = MPIU_Malloc(recv_data_sz); if (!rreq->dev.tmpbuf) { MPIU_ERR_SETANDJUMP(mpi_errno,MPI_ERR_OTHER,"**nomem"); diff --git a/src/mpid/ch3/src/ch3u_handle_recv_pkt.c b/src/mpid/ch3/src/ch3u_handle_recv_pkt.c index 7828606..7f3f020 100644 --- a/src/mpid/ch3/src/ch3u_handle_recv_pkt.c +++ b/src/mpid/ch3/src/ch3u_handle_recv_pkt.c @@ -6,6 +6,7 @@ #include "mpidimpl.h" #include "mpidrma.h" +#include "mpidi_recvq_statistics.h" /* * This file contains the dispatch routine called by the ch3 progress @@ -272,6 +273,9 @@ int MPIDI_CH3U_Receive_data_unexpected(MPID_Request * rreq, char *buf, MPIDI_msg *complete = FALSE; } + if (MPIDI_Request_get_msg_type(rreq) == MPIDI_REQUEST_EAGER_MSG) + MPIR_T_ADD(MPIDI_CH3I_unexpected_recvq_buffer_size, rreq->dev.tmpbuf_sz); + rreq->dev.OnDataAvail = MPIDI_CH3_ReqHandler_UnpackUEBufComplete; fn_fail: diff --git a/src/mpid/ch3/src/ch3u_recvq.c b/src/mpid/ch3/src/ch3u_recvq.c index de99de1..6b3b823 100644 --- a/src/mpid/ch3/src/ch3u_recvq.c +++ b/src/mpid/ch3/src/ch3u_recvq.c @@ -5,6 +5,7 @@ */ #include "mpidimpl.h" +#include "mpidi_recvq_statistics.h" /* MPIDI_POSTED_RECV_ENQUEUE_HOOK(req): Notifies channel that req has been enqueued on the posted recv queue. Returns void. */ @@ -58,24 +59,14 @@ MPID_Request ** const MPID_Recvq_posted_head_ptr = &recvq_posted_head; MPID_Request ** const MPID_Recvq_unexpected_head_ptr = &recvq_unexpected_head; #endif -/* TODO decide control this independently via configure or with the existing - * --enable-timing option (#ifdef COLLECT_STATS) */ -#define ENABLE_RECVQ_STATISTICS 1 -#ifdef ENABLE_RECVQ_STATISTICS +#if ENABLE_RECVQ_STATISTICS static unsigned int posted_qlen = 0; static unsigned int unexpected_qlen = 0; static MPI_Aint posted_recvq_match_attempts = 0; static MPI_Aint unexpected_recvq_match_attempts = 0; -/* TODO add some code here and probably elsewhere to make these show up in the - * MPI_T_pvar_ interface */ -#define MPIR_T_INC(x) (++(x)) -#define MPIR_T_DEC(x) (--(x)) - -#else - -#define MPIR_T_INC(x) -#define MPIR_T_DEC(x) - +static double time_failed_matching_postedq = 0.0; +static double time_matching_unexpectedq = 0.0; +uint64_t MPIDI_CH3I_unexpected_recvq_buffer_size = 0; /* used in ch3u_eager.c and ch3u_handle_recv_pkt.c */ #endif /* defined(ENABLE_RECVQ_STATISTICS) */ /* If the MPIDI_Message_match structure fits into a pointer size, we @@ -110,34 +101,11 @@ static MPI_Aint unexpected_recvq_match_attempts = 0; (((match1).parts.tag & (mask).parts.tag) == ((match2).parts.tag & (mask).parts.tag)) && \ ((match1).parts.context_id == (match2).parts.context_id))) -/* will be invoked to populate the custom parts of pvar_handle objects */ -static int simple_uint_creator(void *obj_handle, - struct MPIR_T_pvar_handle *handle, - int *countp) -{ - /* the IMPL_SIMPLE code reads/writes "bytes" bytes from the location given - * by the "handle_state" pointer */ - handle->handle_state = handle->info->var_state; - handle->bytes = sizeof(unsigned int); - - /* a single unsigned int should be read/written */ - *countp = 1; - return MPI_SUCCESS; -} -static int simple_aint_creator(void *obj_handle, - struct MPIR_T_pvar_handle *handle, - int *countp) -{ - /* the IMPL_SIMPLE code reads/writes "bytes" bytes from the location given - * by the "handle_state" pointer */ - handle->handle_state = handle->info->var_state; - handle->bytes = sizeof(MPI_Aint); - - /* a single Aint should be read/written */ - *countp = 1; - return MPI_SUCCESS; -} +/* will be invoked to populate the custom parts of pvar_handle objects */ +MPIR_T_SIMPLE_HANDLE_CREATOR(simple_aint_creator, MPI_Aint, 1) +MPIR_T_SIMPLE_HANDLE_CREATOR(simple_uint_creator, unsigned int, 1) +MPIR_T_SIMPLE_HANDLE_CREATOR(simple_double_creator, double, 1) #undef FUNCNAME #define FUNCNAME MPIDI_CH3U_Recvq_init @@ -213,6 +181,58 @@ int MPIDI_CH3U_Recvq_init(void) &simple_aint_creator, &idx); if (mpi_errno) MPIU_ERR_POP(mpi_errno); + + /* time spent unsuccessfully trying to match incoming message with posted receives */ + mpi_errno = MPIR_T_pvar_add("time_failed_matching_postedq", + MPI_T_VERBOSITY_USER_DETAIL, + MPI_T_PVAR_CLASS_TIMER, + MPI_DOUBLE, + MPI_T_ENUM_NULL, + "total time spent on unsuccessful search passes on the posted receives queue", + MPI_T_BIND_NO_OBJECT, + /*readonly=*/FALSE, + /*continuous=*/TRUE, + /*atomic=*/FALSE, + MPIR_T_PVAR_IMPL_SIMPLE, + /*var_state=*/&time_failed_matching_postedq, + &simple_double_creator, + &idx); + if (mpi_errno) MPIU_ERR_POP(mpi_errno); + + /* time spent trying to match a posted receive with messages in the unexpected queue */ + mpi_errno = MPIR_T_pvar_add("time_matching_unexpectedq", + MPI_T_VERBOSITY_USER_DETAIL, + MPI_T_PVAR_CLASS_TIMER, + MPI_DOUBLE, + MPI_T_ENUM_NULL, + "total time spent on search passes on the unexpected receive queue", + MPI_T_BIND_NO_OBJECT, + /*readonly=*/FALSE, + /*continuous=*/TRUE, + /*atomic=*/FALSE, + MPIR_T_PVAR_IMPL_SIMPLE, + /*var_state=*/&time_matching_unexpectedq, + &simple_double_creator, + &idx); + if (mpi_errno) MPIU_ERR_POP(mpi_errno); + + /* allocated buffer size in the unexpected receive queue */ + mpi_errno = MPIR_T_pvar_add("unexpected_recvq_buffer_size", + MPI_T_VERBOSITY_USER_DETAIL, + MPI_T_PVAR_CLASS_LEVEL, + MPI_AINT, + MPI_T_ENUM_NULL, + "total buffer size allocated in the unexpected receive queue", + MPI_T_BIND_NO_OBJECT, + /*readonly=*/TRUE, + /*continuous=*/TRUE, + /*atomic=*/FALSE, + MPIR_T_PVAR_IMPL_SIMPLE, + /*var_state=*/&MPIDI_CH3I_unexpected_recvq_buffer_size, + &simple_aint_creator, + &idx); + if (mpi_errno) MPIU_ERR_POP(mpi_errno); + #endif fn_fail: return mpi_errno; @@ -246,6 +266,7 @@ fn_fail: int MPIDI_CH3U_Recvq_FU(int source, int tag, int context_id, MPI_Status *s) { MPID_Request * rreq; + MPID_Time_t timer_start; int found = 0; MPIDI_Message_match match, mask; MPIDI_STATE_DECL(MPID_STATE_MPIDI_CH3U_RECVQ_FU); @@ -261,12 +282,14 @@ int MPIDI_CH3U_Recvq_FU(int source, int tag, int context_id, MPI_Status *s) match.parts.rank = source; if (tag != MPI_ANY_TAG && source != MPI_ANY_SOURCE) { + MPIR_T_START_TIMER(timer_start); while (rreq != NULL) { MPIR_T_INC(unexpected_recvq_match_attempts); if (MATCH_WITH_NO_MASK(rreq->dev.match, match)) break; rreq = rreq->dev.next; } + MPIR_T_END_TIMER(timer_start, time_matching_unexpectedq); } else { mask.parts.context_id = mask.parts.rank = mask.parts.tag = ~0; @@ -275,12 +298,14 @@ int MPIDI_CH3U_Recvq_FU(int source, int tag, int context_id, MPI_Status *s) if (source == MPI_ANY_SOURCE) match.parts.rank = mask.parts.rank = 0; + MPIR_T_START_TIMER(timer_start); while (rreq != NULL) { MPIR_T_INC(unexpected_recvq_match_attempts); if (MATCH_WITH_LEFT_MASK(rreq->dev.match, match, mask)) break; rreq = rreq->dev.next; } + MPIR_T_END_TIMER(timer_start, time_matching_unexpectedq); } /* Save the information about the request before releasing the @@ -319,6 +344,7 @@ int MPIDI_CH3U_Recvq_FU(int source, int tag, int context_id, MPI_Status *s) MPID_Request * MPIDI_CH3U_Recvq_FDU(MPI_Request sreq_id, MPIDI_Message_match * match) { + MPID_Time_t timer_start; MPID_Request * rreq; MPID_Request * prev_rreq; MPID_Request * cur_rreq; @@ -339,12 +365,15 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU(MPI_Request sreq_id, /* FIXME: Why doesn't this exit after it finds the first match? */ cur_rreq = recvq_unexpected_head; while (cur_rreq != NULL) { + MPIR_T_START_TIMER(timer_start); if (cur_rreq->dev.sender_req_id == sreq_id && MPIR_T_INC(unexpected_recvq_match_attempts) && (MATCH_WITH_NO_MASK(cur_rreq->dev.match, *match))) { matching_prev_rreq = prev_rreq; matching_cur_rreq = cur_rreq; } + MPIR_T_END_TIMER(timer_start, time_matching_unexpectedq); + prev_rreq = cur_rreq; cur_rreq = cur_rreq->dev.next; } @@ -363,6 +392,8 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU(MPI_Request sreq_id, MPIR_T_DEC(unexpected_qlen); rreq = matching_cur_rreq; + + MPIR_T_SUBTRACT(MPIDI_CH3I_unexpected_recvq_buffer_size, rreq->dev.tmpbuf_sz); } else { rreq = NULL; @@ -382,6 +413,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU(MPI_Request sreq_id, #define FCNAME MPIDI_QUOTE(FUNCNAME) MPID_Request * MPIDI_CH3U_Recvq_FDU_matchonly(int source, int tag, int context_id, MPID_Comm *comm, int *foundp) { + MPID_Time_t timer_start; int found = FALSE; MPID_Request *rreq, *prev_rreq; MPIDI_Message_match match; @@ -392,6 +424,9 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_matchonly(int source, int tag, int context_i MPIU_THREAD_CS_ASSERT_HELD(MSGQUEUE); + /* Store how much time is spent traversing the queue */ + MPIR_T_START_TIMER(timer_start); + /* Optimize this loop for an empty unexpected receive queue */ rreq = recvq_unexpected_head; if (rreq) { @@ -416,6 +451,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_matchonly(int source, int tag, int context_i recvq_unexpected_tail = prev_rreq; } MPIR_T_DEC(unexpected_qlen); + MPIR_T_SUBTRACT(MPIDI_CH3I_unexpected_recvq_buffer_size, rreq->dev.tmpbuf_sz); rreq->comm = comm; MPIR_Comm_add_ref(comm); @@ -448,6 +484,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_matchonly(int source, int tag, int context_i recvq_unexpected_tail = prev_rreq; } MPIR_T_DEC(unexpected_qlen); + MPIR_T_SUBTRACT(MPIDI_CH3I_unexpected_recvq_buffer_size, rreq->dev.tmpbuf_sz); rreq->comm = comm; MPIR_Comm_add_ref(comm); @@ -463,6 +500,8 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_matchonly(int source, int tag, int context_i } lock_exit: + MPIR_T_END_TIMER(timer_start, time_matching_unexpectedq); + *foundp = found; MPIDI_FUNC_EXIT(MPID_STATE_MPIDI_CH3U_RECVQ_FDU_MATCHONLY); @@ -491,6 +530,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_or_AEP(int source, int tag, int context_id, MPID_Comm *comm, void *user_buf, int user_count, MPI_Datatype datatype, int * foundp) { + MPID_Time_t timer_start; int found; MPID_Request *rreq, *prev_rreq; MPIDI_Message_match match; @@ -501,6 +541,9 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_or_AEP(int source, int tag, MPIU_THREAD_CS_ASSERT_HELD(MSGQUEUE); + /* Store how much time is spent traversing the queue */ + MPIR_T_START_TIMER(timer_start); + /* Optimize this loop for an empty unexpected receive queue */ rreq = recvq_unexpected_head; if (rreq) { @@ -526,6 +569,9 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_or_AEP(int source, int tag, } MPIR_T_DEC(unexpected_qlen); + if (MPIDI_Request_get_msg_type(rreq) == MPIDI_REQUEST_EAGER_MSG) + MPIR_T_SUBTRACT(MPIDI_CH3I_unexpected_recvq_buffer_size, rreq->dev.tmpbuf_sz); + rreq->comm = comm; MPIR_Comm_add_ref(comm); rreq->dev.user_buf = user_buf; @@ -559,6 +605,9 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_or_AEP(int source, int tag, } MPIR_T_DEC(unexpected_qlen); + if (MPIDI_Request_get_msg_type(rreq) == MPIDI_REQUEST_EAGER_MSG) + MPIR_T_SUBTRACT(MPIDI_CH3I_unexpected_recvq_buffer_size, rreq->dev.tmpbuf_sz); + rreq->comm = comm; MPIR_Comm_add_ref(comm); rreq->dev.user_buf = user_buf; @@ -572,7 +621,8 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_or_AEP(int source, int tag, } while (rreq); } } - + MPIR_T_END_TIMER(timer_start, time_matching_unexpectedq); + /* A matching request was not found in the unexpected queue, so we need to allocate a new request and add it to the posted queue */ { @@ -634,8 +684,11 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_or_AEP(int source, int tag, } lock_exit: - *foundp = found; + + /* If a match was not found, the timer was stopped after the traversal */ + if (found) + MPIR_T_END_TIMER(timer_start, time_matching_unexpectedq); MPIDI_FUNC_EXIT(MPID_STATE_MPIDI_CH3U_RECVQ_FDU_OR_AEP); return rreq; @@ -657,6 +710,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDU_or_AEP(int source, int tag, int MPIDI_CH3U_Recvq_DP(MPID_Request * rreq) { int found; + MPID_Time_t timer_start; MPID_Request * cur_rreq; MPID_Request * prev_rreq; int dequeue_failed; @@ -669,6 +723,7 @@ int MPIDI_CH3U_Recvq_DP(MPID_Request * rreq) /* MT FIXME is this right? or should the caller do this? */ MPIU_THREAD_CS_ENTER(MSGQUEUE,); + MPIR_T_START_TIMER(timer_start); cur_rreq = recvq_posted_head; while (cur_rreq != NULL) { if (cur_rreq == rreq) { @@ -693,6 +748,8 @@ int MPIDI_CH3U_Recvq_DP(MPID_Request * rreq) prev_rreq = cur_rreq; cur_rreq = cur_rreq->dev.next; } + if (!found) + MPIR_T_END_TIMER(timer_start, time_failed_matching_postedq); MPIU_THREAD_CS_EXIT(MSGQUEUE,); @@ -726,6 +783,7 @@ int MPIDI_CH3U_Recvq_DP(MPID_Request * rreq) MPID_Request * MPIDI_CH3U_Recvq_FDP_or_AEU(MPIDI_Message_match * match, int * foundp) { + MPID_Time_t timer_start; int found; MPID_Request * rreq; MPID_Request * prev_rreq; @@ -741,6 +799,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDP_or_AEU(MPIDI_Message_match * match, rreq = recvq_posted_head; + MPIR_T_START_TIMER(timer_start); while (rreq != NULL) { MPIR_T_INC(posted_recvq_match_attempts); if (MATCH_WITH_LEFT_RIGHT_MASK(rreq->dev.match, *match, rreq->dev.mask)) { @@ -766,6 +825,7 @@ MPID_Request * MPIDI_CH3U_Recvq_FDP_or_AEU(MPIDI_Message_match * match, prev_rreq = rreq; rreq = rreq->dev.next; } + MPIR_T_END_TIMER(timer_start, time_failed_matching_postedq); /* A matching request was not found in the posted queue, so we need to allocate a new request and add it to the unexpected queue */ diff --git a/test/mpi_t/nem_fbox_fallback_to_queue_count.c b/test/mpi_t/nem_fbox_fallback_to_queue_count.c new file mode 100644 index 0000000..e4a0530 --- /dev/null +++ b/test/mpi_t/nem_fbox_fallback_to_queue_count.c @@ -0,0 +1,206 @@ +#include <mpi.h> +#include <assert.h> +#include <string.h> +#include <stdlib.h> +#include <stdint.h> +#include <stdio.h> + +#define TRY(func) \ + do { \ + err = (func); \ + if (err != MPI_SUCCESS) \ + MPI_Abort(MPI_COMM_WORLD, err); \ + } while(0) + +#define STR_LEN 100 +#define BUF_COUNT 10 + +uint64_t null_fbox[2] = { 0 }; +int err, rank; +MPI_T_pvar_session session; +MPI_T_pvar_handle fbox_handle; + +/* Check that we can successfuly write to the variable. */ +void blank_test() +{ + int i; + uint64_t temp[2] = { -1 }; + + temp[0] = 0x1234; temp[1] = 0xABCD; + TRY(MPI_T_pvar_write(session, fbox_handle, temp)); + + temp[0] = 0xCD34; temp[1] = 0x12AB; + TRY(MPI_T_pvar_read(session, fbox_handle, temp)); + assert(temp[0] == 0x1234); assert(temp[1] == 0xABCD); +} + +/* Nemesis' fastbox falls back to regular queues when more than one message + * is yet to be delivered. + * Here, the sender posts all sends before the receiver has a chance to + * acknowledge any of them; this should force the sender to fall_back to the + * queue every time. */ +void send_first_test() +{ + uint64_t nem_fbox_fall_back_to_queue_count[2] = {-1}; + + /* Reset the fbox variable. */ + MPI_T_pvar_write(session, fbox_handle, null_fbox); + + if (rank == 0) { + char send_buf[BUF_COUNT] = { 0x12 }; + + /* Check that the variable has been correctly initialized. */ + TRY(MPI_T_pvar_read(session, fbox_handle, nem_fbox_fall_back_to_queue_count)); + assert(nem_fbox_fall_back_to_queue_count[1] == 0); + + MPI_Send(send_buf, BUF_COUNT, MPI_CHAR, 1, 0, MPI_COMM_WORLD); + TRY(MPI_T_pvar_read(session, fbox_handle, nem_fbox_fall_back_to_queue_count)); + assert(nem_fbox_fall_back_to_queue_count[1] == 0); + + MPI_Send(send_buf, BUF_COUNT, MPI_CHAR, 1, 0, MPI_COMM_WORLD); + TRY(MPI_T_pvar_read(session, fbox_handle, nem_fbox_fall_back_to_queue_count)); + assert(nem_fbox_fall_back_to_queue_count[1] == 1); + + MPI_Send(send_buf, BUF_COUNT, MPI_CHAR, 1, 0, MPI_COMM_WORLD); + TRY(MPI_T_pvar_read(session, fbox_handle, nem_fbox_fall_back_to_queue_count)); + assert(nem_fbox_fall_back_to_queue_count[1] == 2); + + MPI_Send(send_buf, BUF_COUNT, MPI_CHAR, 1, 0, MPI_COMM_WORLD); + TRY(MPI_T_pvar_read(session, fbox_handle, nem_fbox_fall_back_to_queue_count)); + assert(nem_fbox_fall_back_to_queue_count[1] == 3); + + /* Make sure we've posted the sends before the receiver gets a chance + * to receive them. + * FIXME: Ideally this should use a barrier, but that uses messages + * internally and hence will sometimes screw up the asserts above. + */ + sleep(1); + + } else if (rank == 1) { + char recv_buf[BUF_COUNT]; + MPI_Status status; + + sleep(1); /* see above */ + + MPI_Recv(recv_buf, BUF_COUNT, MPI_CHAR, 0, 0, MPI_COMM_WORLD, &status); + MPI_Recv(recv_buf, BUF_COUNT, MPI_CHAR, 0, 0, MPI_COMM_WORLD, &status); + MPI_Recv(recv_buf, BUF_COUNT, MPI_CHAR, 0, 0, MPI_COMM_WORLD, &status); + MPI_Recv(recv_buf, BUF_COUNT, MPI_CHAR, 0, 0, MPI_COMM_WORLD, &status); + } + + MPI_Barrier(MPI_COMM_WORLD); /* ensure we've finished this test before + * moving on to the next */ +} + +/* By posting receives ahead of time, messages should be taken out of the + * fastbox as soon as they are delivered. Hence, the counter should remain 0 + * throughout. + * FIXME: This doesn't quite work yet, in part because of the barrier (which + * also uses messages). May want to 'sleep' between sends as a + * workaround. + */ +void recv_first_test() +{ + uint64_t nem_fbox_fall_back_to_queue_count[2] = {-1}; + + /* Reset the fbox variable. */ + MPI_T_pvar_write(session, fbox_handle, null_fbox); + + if (rank == 0) { + char send_buf[BUF_COUNT] = { 0x12 }; + + MPI_Barrier(MPI_COMM_WORLD); /* see below */ + + /* Check that the variable has been correctly initialized. */ + TRY(MPI_T_pvar_read(session, fbox_handle, nem_fbox_fall_back_to_queue_count)); + assert(nem_fbox_fall_back_to_queue_count[1] == 0); + + MPI_Send(send_buf, BUF_COUNT, MPI_CHAR, 1, 0, MPI_COMM_WORLD); + TRY(MPI_T_pvar_read(session, fbox_handle, nem_fbox_fall_back_to_queue_count)); + assert(nem_fbox_fall_back_to_queue_count[1] == 0); + + MPI_Send(send_buf, BUF_COUNT, MPI_CHAR, 1, 0, MPI_COMM_WORLD); + TRY(MPI_T_pvar_read(session, fbox_handle, nem_fbox_fall_back_to_queue_count)); + assert(nem_fbox_fall_back_to_queue_count[1] == 0); + + MPI_Send(send_buf, BUF_COUNT, MPI_CHAR, 1, 0, MPI_COMM_WORLD); + TRY(MPI_T_pvar_read(session, fbox_handle, nem_fbox_fall_back_to_queue_count)); + assert(nem_fbox_fall_back_to_queue_count[1] == 0); + + MPI_Send(send_buf, BUF_COUNT, MPI_CHAR, 1, 0, MPI_COMM_WORLD); + TRY(MPI_T_pvar_read(session, fbox_handle, nem_fbox_fall_back_to_queue_count)); + assert(nem_fbox_fall_back_to_queue_count[1] == 0); + } else if (rank == 1) { + char recv_buf[BUF_COUNT]; + MPI_Request reqs[4]; + + MPI_Irecv(recv_buf, BUF_COUNT, MPI_CHAR, 0, 0, MPI_COMM_WORLD, &reqs[0]); + MPI_Irecv(recv_buf, BUF_COUNT, MPI_CHAR, 0, 0, MPI_COMM_WORLD, &reqs[1]); + MPI_Irecv(recv_buf, BUF_COUNT, MPI_CHAR, 0, 0, MPI_COMM_WORLD, &reqs[2]); + MPI_Irecv(recv_buf, BUF_COUNT, MPI_CHAR, 0, 0, MPI_COMM_WORLD, &reqs[3]); + + MPI_Barrier(MPI_COMM_WORLD); /* make sure we've posted the receives + * before the sender gets a chance + * to send them */ + + MPI_Status status[4]; + MPI_Waitall(4, reqs, status); + } + + MPI_Barrier(MPI_COMM_WORLD); /* ensure we've finished this test before + * moving on to the next */ +} + +int main(int argc, char *argv[]) +{ + int i, size, num, name_len, desc_len, count, verb, thread_support; + int varclass, bind, readonly, continuous, atomic, fbox_idx; + char name[STR_LEN], desc[STR_LEN]; + MPI_Datatype dtype; + MPI_T_enum enumtype; + + printf("MPIT pvar test: nem_fbox_fall_back_to_queue_count\n"); fflush(stdout); + + MPI_Init(NULL, NULL); + MPI_Comm_size(MPI_COMM_WORLD, &size); + MPI_Comm_rank(MPI_COMM_WORLD, &rank); + + /* Ensure we're using exactly two ranks. */ + assert(size == 2); + + /* Standard MPIT initialization. */ + TRY(MPI_T_init_thread(MPI_THREAD_SINGLE, &thread_support)); + TRY(MPI_T_pvar_get_num(&num)); + + /* Locate desired MPIT variable. */ + for (i = 0; i < num; i++) { + name_len = desc_len = STR_LEN; + TRY(MPI_T_pvar_get_info(i, name, &name_len, &verb, &varclass, &dtype, + &enumtype, desc, &desc_len, &bind, &readonly, + &continuous, &atomic)); + + if (strcmp(name, "nem_fbox_fall_back_to_queue_count") == 0) + fbox_idx = i; + } + + /* Initialize MPIT session & variable handle. */ + MPI_T_pvar_session_create(&session); + MPI_T_pvar_handle_alloc(session, fbox_idx, NULL, &fbox_handle, &count); + + /* Ensure the variable is of the correct size. */ + assert(count == 2); + + /* Run a batch of tests. */ + blank_test(); + send_first_test(); + /* recv_first_test(); */ + + /* Cleanup. */ + MPI_T_pvar_handle_free(session, &fbox_handle); + MPI_T_pvar_session_free(&session); + + TRY(MPI_T_finalize()); + MPI_Finalize(); + + return 0; +} diff --git a/test/mpi_t/unexpected_recvq_buffer_size.c b/test/mpi_t/unexpected_recvq_buffer_size.c new file mode 100644 index 0000000..de896cc --- /dev/null +++ b/test/mpi_t/unexpected_recvq_buffer_size.c @@ -0,0 +1,138 @@ +#include <mpi.h> +#include <assert.h> +#include <string.h> +#include <stdio.h> + +#define TRY(func) \ + do { \ + err = (func); \ + if (err != MPI_SUCCESS) \ + MPI_Abort(MPI_COMM_WORLD, err); \ + } while(0) + +#define EAGER_SIZE 10 +#define RNDV_SIZE 100000 +#define STR_LEN 100 + +int err, rank; +MPI_T_pvar_session session; +MPI_T_pvar_handle uqsize_handle; + +/* The first receive will block waiting for the last send, since messages from + * a given rank are received in order. */ +void reversed_tags_test() +{ + size_t unexpected_recvq_buffer_size; + + if (rank == 0) { + int send_buf[EAGER_SIZE] = { 0x1234 }; + + MPI_Send(send_buf, EAGER_SIZE, MPI_INT, 1, 0xA, MPI_COMM_WORLD); + MPI_Send(send_buf, EAGER_SIZE, MPI_INT, 1, 0xB, MPI_COMM_WORLD); + MPI_Send(send_buf, EAGER_SIZE, MPI_INT, 1, 0xC, MPI_COMM_WORLD); + MPI_Send(send_buf, EAGER_SIZE, MPI_INT, 1, 0xD, MPI_COMM_WORLD); + } else if (rank == 1) { + int recv_buf[EAGER_SIZE]; + MPI_Status status; + + MPI_Recv(recv_buf, EAGER_SIZE, MPI_INT, 0, 0xD, MPI_COMM_WORLD, &status); + TRY(MPI_T_pvar_read(session, uqsize_handle, &unexpected_recvq_buffer_size)); + assert(unexpected_recvq_buffer_size == 3*EAGER_SIZE*sizeof(int)); + + MPI_Recv(recv_buf, EAGER_SIZE, MPI_INT, 0, 0xC, MPI_COMM_WORLD, &status); + TRY(MPI_T_pvar_read(session, uqsize_handle, &unexpected_recvq_buffer_size)); + assert(unexpected_recvq_buffer_size == 2*EAGER_SIZE*sizeof(int)); + + MPI_Recv(recv_buf, EAGER_SIZE, MPI_INT, 0, 0xB, MPI_COMM_WORLD, &status); + TRY(MPI_T_pvar_read(session, uqsize_handle, &unexpected_recvq_buffer_size)); + assert(unexpected_recvq_buffer_size == 1*EAGER_SIZE*sizeof(int)); + + MPI_Recv(recv_buf, EAGER_SIZE, MPI_INT, 0, 0xA, MPI_COMM_WORLD, &status); + TRY(MPI_T_pvar_read(session, uqsize_handle, &unexpected_recvq_buffer_size)); + assert(unexpected_recvq_buffer_size == 0*EAGER_SIZE*sizeof(int)); + } + + MPI_Barrier(MPI_COMM_WORLD); /* make sure this test is over before going to the next one */ +} + +/* Rendezvous-based messages will never be unexpected (except for the initial RTS, + * which has an empty buffer anyhow). + */ +void rndv_test() +{ + size_t unexpected_recvq_buffer_size; + + if (rank == 0) { + int send_buf[RNDV_SIZE] = { 0x5678 }; + + MPI_Send(send_buf, RNDV_SIZE, MPI_INT, 1, 0, MPI_COMM_WORLD); + MPI_Send(send_buf, RNDV_SIZE, MPI_INT, 1, 0, MPI_COMM_WORLD); + } else if (rank == 1) { + int recv_buf[RNDV_SIZE]; + MPI_Status status; + + MPI_Recv(recv_buf, RNDV_SIZE, MPI_INT, 0, 0, MPI_COMM_WORLD, &status); + TRY(MPI_T_pvar_read(session, uqsize_handle, &unexpected_recvq_buffer_size)); + assert(unexpected_recvq_buffer_size == 0); + + MPI_Recv(recv_buf, RNDV_SIZE, MPI_INT, 0, 0, MPI_COMM_WORLD, &status); + TRY(MPI_T_pvar_read(session, uqsize_handle, &unexpected_recvq_buffer_size)); + assert(unexpected_recvq_buffer_size == 0); + } + + MPI_Barrier(MPI_COMM_WORLD); /* make sure this test is over before going to the next one */ +} + +int main(int argc, char *argv[]) +{ + int i, size, num, name_len, desc_len, verb, thread_support; + int varclass, bind, readonly, continuous, atomic, uqsize_idx, count; + char name[STR_LEN], desc[STR_LEN]; + MPI_Datatype dtype; + MPI_T_enum enumtype; + + printf("MPIT pvar test: unexpected_recvq_buffer_size\n"); fflush(stdout); + + MPI_Init(NULL, NULL); + MPI_Comm_size(MPI_COMM_WORLD, &size); + MPI_Comm_rank(MPI_COMM_WORLD, &rank); + + /* Ensure we're using exactly two ranks. */ + /* Future tests (using collectives) might need this because of the MPI_Barrier */ + assert(size == 2); + + /* Standard MPIT initialization. */ + TRY(MPI_T_init_thread(MPI_THREAD_SINGLE, &thread_support)); + TRY(MPI_T_pvar_get_num(&num)); + + /* Locate desired MPIT variable. */ + for (i = 0; i < num; i++) { + name_len = desc_len = STR_LEN; + TRY(MPI_T_pvar_get_info(i, name, &name_len, &verb, &varclass, &dtype, + &enumtype, desc, &desc_len, &bind, &readonly, + &continuous, &atomic)); + + if (strcmp(name, "unexpected_recvq_buffer_size") == 0) + uqsize_idx = i; + } + + /* Initialize MPIT session & variable handle. */ + MPI_T_pvar_session_create(&session); + MPI_T_pvar_handle_alloc(session, uqsize_idx, NULL, &uqsize_handle, &count); + + /* Ensure the variable is of the correct size. */ + assert(count == 1); + + /* Run a batch of tests. */ + reversed_tags_test(); + rndv_test(); + + /* Cleanup. */ + MPI_T_pvar_handle_free(session, &uqsize_handle); + MPI_T_pvar_session_free(&session); + + TRY(MPI_T_finalize()); + MPI_Finalize(); + + return 0; +} http://git.mpich.org/mpich.git/commitdiff/2cfb60178370094f4df8ba43e535f987f… commit 2cfb60178370094f4df8ba43e535f987ff405bb3 Author: Ralf Gunter <ralfgunter(a)gmail.com> Date: Thu Nov 29 14:25:05 2012 -0600 MPI_T_pvar_read/write now respect the handle creator's count variable diff --git a/src/mpi_t/pvar_handle_alloc.c b/src/mpi_t/pvar_handle_alloc.c index e26316f..e61ce49 100644 --- a/src/mpi_t/pvar_handle_alloc.c +++ b/src/mpi_t/pvar_handle_alloc.c @@ -61,6 +61,9 @@ int MPIR_T_pvar_handle_alloc_impl(MPI_T_pvar_session session, int pvar_index, vo mpi_errno = info->create_fn(obj_handle, (*handle), count); if (mpi_errno) MPIU_ERR_POP(mpi_errno); + /* some creator functions may not assign this */ + (*handle)->count = *count; + MPL_DL_APPEND(session->hlist, *handle); MPIU_CHKPMEM_COMMIT(); diff --git a/src/mpi_t/pvar_read.c b/src/mpi_t/pvar_read.c index aaf9e8a..34d7041 100644 --- a/src/mpi_t/pvar_read.c +++ b/src/mpi_t/pvar_read.c @@ -35,7 +35,7 @@ int MPIR_T_pvar_read_impl(MPI_T_pvar_session session, MPI_T_pvar_handle handle, /* the extra indirection through "info" might be too costly for some tools, * consider moving this value to or caching it in the handle itself */ if (likely(handle->info->impl_kind == MPIR_T_PVAR_IMPL_SIMPLE)) { - MPIU_Memcpy(buf, handle->handle_state, handle->bytes); + MPIU_Memcpy(buf, handle->handle_state, handle->count * handle->bytes); } else { MPIU_Assertp(FALSE); /* _IMPL_CB not yet implemented */ diff --git a/src/mpi_t/pvar_write.c b/src/mpi_t/pvar_write.c index 2c7781c..26f4ecb 100644 --- a/src/mpi_t/pvar_write.c +++ b/src/mpi_t/pvar_write.c @@ -37,7 +37,7 @@ int MPIR_T_pvar_write_impl(MPI_T_pvar_session session, MPI_T_pvar_handle handle, /* the extra indirection through "info" might be too costly for some tools, * consider moving this value to or caching it in the handle itself */ if (likely(handle->info->impl_kind == MPIR_T_PVAR_IMPL_SIMPLE)) { - MPIU_Memcpy(handle->handle_state, buf, handle->bytes); + MPIU_Memcpy(handle->handle_state, buf, handle->count * handle->bytes); } else { MPIU_Assertp(FALSE); /* _IMPL_CB not yet implemented */ http://git.mpich.org/mpich.git/commitdiff/d6b69b468128c0c356db2d31e8fd4ca04… commit d6b69b468128c0c356db2d31e8fd4ca0414c3ea4 Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Tue Jan 22 14:38:31 2013 -0600 move MPI_T tests to their own subdir in "manual" Reviewed-by: rgunter diff --git a/test/mpi/manual/Makefile.am b/test/mpi/manual/Makefile.am index d466feb..4ba8d9e 100644 --- a/test/mpi/manual/Makefile.am +++ b/test/mpi/manual/Makefile.am @@ -7,9 +7,10 @@ include $(top_srcdir)/Makefile.mtest +SUBDIRS = mpi_t + noinst_PROGRAMS = singjoin testconnect testconnectserial dimsbalanced \ - spawntest_master spawntest_child segfault \ - mpit_test mpit_test2 + spawntest_master spawntest_child segfault # testconnectserial would like MPICHLIBSTR to be defined as the installation # directory of the MPI library. This definition is not required. diff --git a/test/mpi/manual/mpi_t/Makefile.am b/test/mpi/manual/mpi_t/Makefile.am new file mode 100644 index 0000000..3f84934 --- /dev/null +++ b/test/mpi/manual/mpi_t/Makefile.am @@ -0,0 +1,11 @@ +## -*- Mode: Makefile; -*- +## vim: set ft=automake : +## +## (C) 2013 by Argonne National Laboratory. +## See COPYRIGHT in top-level directory. +## + +include $(top_srcdir)/Makefile.mtest + +noinst_PROGRAMS = mpit_test mpit_test2 + diff --git a/test/mpi/manual/mpit_test.c b/test/mpi/manual/mpi_t/mpit_test.c similarity index 100% rename from test/mpi/manual/mpit_test.c rename to test/mpi/manual/mpi_t/mpit_test.c diff --git a/test/mpi/manual/mpit_test2.c b/test/mpi/manual/mpi_t/mpit_test2.c similarity index 100% rename from test/mpi/manual/mpit_test2.c rename to test/mpi/manual/mpi_t/mpit_test2.c ----------------------------------------------------------------------- Summary of changes: configure.ac | 51 +++++ src/mpi_t/pvar_handle_alloc.c | 3 + src/mpi_t/pvar_read.c | 2 +- src/mpi_t/pvar_write.c | 2 +- .../ch3/channels/nemesis/include/mpid_nem_inline.h | 5 + .../nemesis/include/mpidi_nem_statistics.h | 15 ++ .../ch3/channels/nemesis/src/mpid_nem_finalize.c | 6 + src/mpid/ch3/channels/nemesis/src/mpid_nem_init.c | 47 ++++ src/mpid/ch3/include/mpidi_common_statistics.h | 71 ++++++ src/mpid/ch3/include/mpidi_recvq_statistics.h | 16 ++ src/mpid/ch3/src/ch3u_eager.c | 2 + src/mpid/ch3/src/ch3u_handle_recv_pkt.c | 4 + src/mpid/ch3/src/ch3u_recvq.c | 197 +++++++++++------ test/mpi/configure.ac | 1 + test/mpi/manual/Makefile.am | 5 +- test/mpi/manual/mpi_t/Makefile.am | 15 ++ test/mpi/manual/{ => mpi_t}/mpit_test.c | 0 test/mpi/manual/{ => mpi_t}/mpit_test2.c | 0 .../mpi_t/nem_fbox_fallback_to_queue_count.c | 231 ++++++++++++++++++++ .../manual/mpi_t/unexpected_recvq_buffer_size.c | 158 +++++++++++++ 20 files changed, 761 insertions(+), 70 deletions(-) create mode 100644 src/mpid/ch3/channels/nemesis/include/mpidi_nem_statistics.h create mode 100644 src/mpid/ch3/include/mpidi_common_statistics.h create mode 100644 src/mpid/ch3/include/mpidi_recvq_statistics.h create mode 100644 test/mpi/manual/mpi_t/Makefile.am rename test/mpi/manual/{ => mpi_t}/mpit_test.c (100%) rename test/mpi/manual/{ => mpi_t}/mpit_test2.c (100%) create mode 100644 test/mpi/manual/mpi_t/nem_fbox_fallback_to_queue_count.c create mode 100644 test/mpi/manual/mpi_t/unexpected_recvq_buffer_size.c hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.1-72-g6a954d0
by noreply＠mpich.org 25 Jan '13

25 Jan '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 6a954d01628b1443336bee7e197d91d56690d8f6 (commit) via 4aba817fda95da257258befed3fa34167ee44507 (commit) via b187ee20f4167555ed752601ba56f62878c02fe4 (commit) via 237d91bf62040f46f32ecf5f65a475c578bd6b9d (commit) from 39ed84ae13cbed02f561144a0b509c052339377d (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/6a954d01628b1443336bee7e197d91d56… commit 6a954d01628b1443336bee7e197d91d56690d8f6 Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Fri Jan 25 16:50:49 2013 -0600 add `xfail=ticketXYZ` annotations to failing tests These five tests are currently failing on a regular basis with MPICH. All need to be dealt with at some point, but we cannot fix them right now and they make it hard to spot new failures in the tests. references #1767, #1788 Reviewed-by: dinan diff --git a/test/mpi/datatype/testlist b/test/mpi/datatype/testlist index ccef43d..755b25b 100644 --- a/test/mpi/datatype/testlist +++ b/test/mpi/datatype/testlist @@ -48,5 +48,5 @@ get-elements 1 hindexed_block 1 mpiversion=3.0 hindexed_block_contents 1 mpiversion=3.0 longdouble 1 -large-count 1 mpiversion=3.0 +large-count 1 mpiversion=3.0 xfail=ticket1767 cxx-types 1 mpiversion=3.0 diff --git a/test/mpi/perf/testlist b/test/mpi/perf/testlist index 503729f..b81ea53 100644 --- a/test/mpi/perf/testlist +++ b/test/mpi/perf/testlist @@ -1,10 +1,10 @@ transp-datatype 2 sendrecvl 2 -twovec 1 +twovec 1 xfail=ticket1788 dtpack 1 -nestvec 1 -nestvec2 1 -indexperf 1 +nestvec 1 xfail=ticket1788 +nestvec2 1 xfail=ticket1788 +indexperf 1 xfail=ticket1788 non_zero_root 4 timer 1 # The commcreatep test looks at how communicator creation scales with group http://git.mpich.org/mpich.git/commitdiff/4aba817fda95da257258befed3fa34167… commit 4aba817fda95da257258befed3fa34167ee44507 Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Fri Jan 25 14:51:17 2013 -0600 TAP: add "xfail" support This is used to indicate tests that are correct, but functionality that is expected to fail in the current version of the MPI implementation. This is expressed in `testlist` files with an argument string like `xfail=ticketXYZ` or `xfail=REASON` after the program name. Unfortunately, spaces, quotes, and `#` characters are not permitted in the format at this time, due to the primitive nature of the testfile parser code. If strict MPI testing is requested (by configuring the test suite with `--enable-strictmpi`) then the `xfail=BLAH` arguments are ignored under the assumption that they are only accurate for the current version of the MPICH implementation. No xfail support has been added to the XML output at this time. Reviewed-by: dinan diff --git a/test/mpi/runtests.in b/test/mpi/runtests.in index 3046cc8..56b6e39 100644 --- a/test/mpi/runtests.in +++ b/test/mpi/runtests.in @@ -362,6 +362,7 @@ sub RunList { my $requiresMPIX = ""; my $progEnv = ""; my $mpiVersion = ""; + my $xfail = ""; if ($#args >= 1) { $np = $args[1]; } # Process the key=value arguments for (my $i=2; $i <= $#args; $i++) { @@ -395,6 +396,12 @@ sub RunList { elsif ($key eq "mpix") { $requiresMPIX = $value } + elsif ($key eq "xfail") { + if ($value eq "") { + print STDERR "\"xfail=\" requires an argument\n"; + } + $xfail = $value; + } else { print STDERR "Unrecognized key $key in $listfileSource\n"; } @@ -441,6 +448,13 @@ sub RunList { next; } + if (lc($testIsStrict) eq "true") { + # Strict MPI testing was requested, so assume that a non-MPICH MPI + # implementation is being tested and the "xfail" implementation + # assumptions do not hold. + $xfail = ''; + } + if (lc($requiresMPIX) eq "true" && lc($MPIHasMPIX) eq "no") { unless (-d $programname) { SkippedTest($programname, $np, $workdir, "tests MPIX extensions, MPIX testing disabled"); @@ -455,19 +469,19 @@ sub RunList { } else { $total_run++; - if (&BuildMPIProgram( $programname ) == 0) { + if (&BuildMPIProgram( $programname, $xfail ) == 0) { if ($batchRun == 1) { &AddMPIProgram( $programname, $np, $ResultTest, $InitForRun, $timeLimit, $progArgs, - $progEnv, $mpiexecArgs ); + $progEnv, $mpiexecArgs, $xfail ); } else { &RunMPIProgram( $programname, $np, $ResultTest, $InitForRun, $timeLimit, $progArgs, - $progEnv, $mpiexecArgs ); + $progEnv, $mpiexecArgs, $xfail ); } } - else { + elsif ($xfail ne '') { # We expected to run this program, so failure to build # is an error $found_error = 1; @@ -520,7 +534,7 @@ sub ProcessImplicitList { if ($programname eq "runtests") { next; } # Ignore self if (-x $programname) { $total_run++; - &RunMPIProgram( $programname, $np_default, "", "", "", "", "", "" ); + &RunMPIProgram( $programname, $np_default, "", "", "", "", "", "", "" ); } } close PGMS; @@ -540,8 +554,8 @@ sub ProcessImplicitList { if (! -s $programname) { next; } $programname =~ s/\.c//; $total_run++; - if (&BuildMPIProgram( $programname ) == 0) { - &RunMPIProgram( $programname, $np_default, "", "", "", "", "", "" ); + if (&BuildMPIProgram( $programname, "") == 0) { + &RunMPIProgram( $programname, $np_default, "", "", "", "", "", "", "" ); } else { # We expected to run this program, so failure to build @@ -562,7 +576,7 @@ sub ProcessImplicitList { # If the 3rd arg is not present, the a default that simply checks that the # return status is 0 and that the output is " No Errors" is used. sub RunMPIProgram { - my ($programname,$np,$ResultTest,$InitForTest,$timeLimit,$progArgs,$progEnv,$mpiexecArgs) = @_; + my ($programname,$np,$ResultTest,$InitForTest,$timeLimit,$progArgs,$progEnv,$mpiexecArgs,$xfail) = @_; my $found_error = 0; my $found_noerror = 0; my $inline = ""; @@ -655,10 +669,10 @@ sub RunMPIProgram { } } if ($found_error) { - &RunTestFailed( $programname, $np, $curdir, $inline ); + &RunTestFailed( $programname, $np, $curdir, $inline, $xfail ); } else { - &RunTestPassed( $programname, $np, $curdir ); + &RunTestPassed( $programname, $np, $curdir, $xfail ); } &RunPostMsg( $programname, $np, $curdir ); } @@ -666,7 +680,7 @@ sub RunMPIProgram { # This version simply writes the mpiexec command out, with the output going # into a file, and recording the output status of the run. sub AddMPIProgram { - my ($programname,$np,$ResultTest,$InitForTest,$timeLimit,$progArgs,$progEnv,$mpiexecArgs) = @_; + my ($programname,$np,$ResultTest,$InitForTest,$timeLimit,$progArgs,$progEnv,$mpiexecArgs, $xfail) = @_; if (! -x $programname) { print STDERR "Could not find $programname!"; @@ -734,7 +748,8 @@ sub AddMPIProgram { # # Return value is 0 on success, non zero on failure sub BuildMPIProgram { - my $programname = $_[0]; + my $programname = shift; + my $xfail = shift; my $rc = 0; if ($verbose) { print STDERR "making $programname\n"; } if (! -x $programname) { $remove_this_pgm = 1; } @@ -752,7 +767,7 @@ sub BuildMPIProgram { # in the summary file (which is otherwise written by the # RunMPIProgram step) &RunPreMsg( $programname, $np, $curdir ); - &RunTestFailed( $programname, $np, $curdir, "Failed to build $programname; $output" ); + &RunTestFailed( $programname, $np, $curdir, "Failed to build $programname; $output", $xfail ); &RunPostMsg( $programname, $np, $curdir ); } return $rc; @@ -966,12 +981,16 @@ sub RunPostMsg { } } sub RunTestPassed { - my ($programname, $np, $workdir) = @_; + my ($programname, $np, $workdir, $xfail) = @_; if ($xmloutput) { print XMLOUT "<STATUS>pass</STATUS>$newline"; } if ($tapoutput) { - print TAPOUT "ok ${total_run} - $workdir/$programname $np\n"; + my $xfailstr = ''; + if ($xfail ne '') { + $xfailstr = " # TODO $xfail"; + } + print TAPOUT "ok ${total_run} - $workdir/$programname ${np}${xfailstr}\n"; } } sub RunTestFailed { @@ -979,6 +998,7 @@ sub RunTestFailed { my $np = shift; my $workdir = shift; my $output = shift; + my $xfail = shift; if ($xmloutput) { my $xout = $output; @@ -994,7 +1014,11 @@ sub RunTestFailed { } if ($tapoutput) { - print TAPOUT "not ok ${total_run} - $workdir/$programname $np\n"; + my $xfailstr = ''; + if ($xfail ne '') { + $xfailstr = " # TODO $xfail"; + } + print TAPOUT "not ok ${total_run} - $workdir/$programname ${np}${xfailstr}\n"; print TAPOUT " ---\n"; print TAPOUT " Directory: $workdir\n"; print TAPOUT " File: $programname\n"; http://git.mpich.org/mpich.git/commitdiff/b187ee20f4167555ed752601ba56f6287… commit b187ee20f4167555ed752601ba56f62878c02fe4 Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Fri Jan 25 14:50:14 2013 -0600 runtests: permit arbitrary capitalization in args At Jim's request... Reviewed-by: dinan diff --git a/test/mpi/runtests.in b/test/mpi/runtests.in index 7eec47f..3046cc8 100644 --- a/test/mpi/runtests.in +++ b/test/mpi/runtests.in @@ -434,14 +434,14 @@ sub RunList { } # Check whether strict is required by MPI but not by the # test (use strict=false for tests that use non-standard extensions) - if ($requiresStrict eq "false" && $testIsStrict eq "true") { + if (lc($requiresStrict) eq "false" && lc($testIsStrict) eq "true") { unless (-d $programname) { SkippedTest($programname, $np, $workdir, "non-strict test, strict MPI mode requested"); } next; } - if ($requiresMPIX eq "true" && $MPIHasMPIX eq "no") { + if (lc($requiresMPIX) eq "true" && lc($MPIHasMPIX) eq "no") { unless (-d $programname) { SkippedTest($programname, $np, $workdir, "tests MPIX extensions, MPIX testing disabled"); } http://git.mpich.org/mpich.git/commitdiff/237d91bf62040f46f32ecf5f65a475c57… commit 237d91bf62040f46f32ecf5f65a475c578bd6b9d Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Fri Jan 25 13:21:06 2013 -0600 TAP: add "# SKIP" directive support This way tests that were skipped because of strictness or version constraints will show up in the summary.tap output file. Reviewed-by: dinan diff --git a/test/mpi/runtests.in b/test/mpi/runtests.in index e71a978..7eec47f 100644 --- a/test/mpi/runtests.in +++ b/test/mpi/runtests.in @@ -47,7 +47,8 @@ $testIsStrict = "@MPI_IS_STRICT@"; $MPIhasMPIX = "@MPI_HAS_MPIX@"; $np_arg = "-n"; # Name of argument to specify the number of processes $err_count = 0; # Number of programs that failed. -$total_count = 0; # Number of programs tested +$total_run = 0; # Number of programs tested +$total_seen = 0; # Number of programs considered for testing $np_default = 2; # Default number of processes to use $np_max = -1; # Maximum number of processes to use (overrides any # value in the test list files. -1 is Infinity @@ -251,7 +252,7 @@ if ($xmloutput && $closeXMLOutput) { } if ($tapoutput) { - print TAPOUT "1..$total_count"; + print TAPOUT "1..$total_seen"; close TAPOUT; } @@ -262,13 +263,13 @@ if ($batchRun) { } else { if ($err_count) { - print "$err_count tests failed out of $total_count\n"; + print "$err_count tests failed out of $total_run\n"; if ($xmloutput) { print "Details in $xmlfullfile\n"; } } else { - print " All $total_count tests passed!\n"; + print " All $total_run tests passed!\n"; } if ($tapoutput) { print "TAP formatted results in $tapfullfile\n"; @@ -400,33 +401,60 @@ sub RunList { } } + # skip empty lines + if ($programname eq "") { next; } + + if ($np eq "") { $np = $np_default; } + if ($np_max > 0 && $np > $np_max) { $np = $np_max; } + + # allows us to accurately output TAP test numbers without disturbing the + # original totals that have traditionally been reported + # + # These "unless" blocks are ugly, but permit us to honor skipping + # criteria for directories as well without counting directories as tests + # in our XML/TAP output. + unless (-d $programname) { + $total_seen++; + } + # If a minimum MPI version is specified, check against the # available MPI. If the version is unknown, we ignore this # test (thus, all tests will be run). if ($mpiVersion ne "" && $MPIMajorVersion ne "unknown" && $MPIMinorVersion ne "unknown") { my ($majorReq,$minorReq) = split(/\./,$mpiVersion); - if ($majorReq > $MPIMajorVersion) { next; } - if ($majorReq == $MPIMajorVersion && - $minorReq > $MPIMinorVersion) { next; } + if ($majorReq > $MPIMajorVersion or + ($majorReq == $MPIMajorVersion && $minorReq > $MPIMinorVersion)) + { + unless (-d $programname) { + SkippedTest($programname, $np, $workdir, "requires MPI version $mpiVersion"); + } + next; + } } # Check whether strict is required by MPI but not by the # test (use strict=false for tests that use non-standard extensions) - if ($requiresStrict eq "false" && $testIsStrict eq "true") { next; } + if ($requiresStrict eq "false" && $testIsStrict eq "true") { + unless (-d $programname) { + SkippedTest($programname, $np, $workdir, "non-strict test, strict MPI mode requested"); + } + next; + } - if ($requiresMPIX eq "true" && $MPIHasMPIX eq "no") { next; } + if ($requiresMPIX eq "true" && $MPIHasMPIX eq "no") { + unless (-d $programname) { + SkippedTest($programname, $np, $workdir, "tests MPIX extensions, MPIX testing disabled"); + } + next; + } - if ($np eq "") { $np = $np_default; } - if ($np_max > 0 && $np > $np_max) { $np = $np_max; } - # skip empty lines - if ($programname eq "") { next; } if (-d $programname) { # If a directory, go into the that directory and # look for a new list file &ProcessDir( $programname, $listfile ); } else { - $total_count++; + $total_run++; if (&BuildMPIProgram( $programname ) == 0) { if ($batchRun == 1) { &AddMPIProgram( $programname, $np, $ResultTest, @@ -491,7 +519,7 @@ sub ProcessImplicitList { if (-d $programname) { next; } # Ignore directories if ($programname eq "runtests") { next; } # Ignore self if (-x $programname) { - $total_count++; + $total_run++; &RunMPIProgram( $programname, $np_default, "", "", "", "", "", "" ); } } @@ -511,7 +539,7 @@ sub ProcessImplicitList { # Skip messages from ls about no files if (! -s $programname) { next; } $programname =~ s/\.c//; - $total_count++; + $total_run++; if (&BuildMPIProgram( $programname ) == 0) { &RunMPIProgram( $programname, $np_default, "", "", "", "", "", "" ); } @@ -943,7 +971,7 @@ sub RunTestPassed { print XMLOUT "<STATUS>pass</STATUS>$newline"; } if ($tapoutput) { - print TAPOUT "ok ${total_count} - $workdir/$programname $np\n"; + print TAPOUT "ok ${total_run} - $workdir/$programname $np\n"; } } sub RunTestFailed { @@ -966,7 +994,7 @@ sub RunTestFailed { } if ($tapoutput) { - print TAPOUT "not ok ${total_count} - $workdir/$programname $np\n"; + print TAPOUT "not ok ${total_run} - $workdir/$programname $np\n"; print TAPOUT " ---\n"; print TAPOUT " Directory: $workdir\n"; print TAPOUT " File: $programname\n"; @@ -996,6 +1024,20 @@ sub RunTestFailed { } } } + +sub SkippedTest { + my $programname = shift; + my $np = shift; + my $workdir = shift; + my $reason = shift; + + # simply omit from the XML output + + if ($tapoutput) { + print TAPOUT "ok ${total_seen} - $workdir/$programname $np # SKIP $reason\n"; + } +} + # ---------------------------------------------------------------------------- # Alternate init routines sub InitQuickTimeout { ----------------------------------------------------------------------- Summary of changes: test/mpi/datatype/testlist | 2 +- test/mpi/perf/testlist | 8 ++-- test/mpi/runtests.in | 130 +++++++++++++++++++++++++++++++++----------- 3 files changed, 103 insertions(+), 37 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.1-68-g39ed84a
by noreply＠mpich.org 24 Jan '13

24 Jan '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 39ed84ae13cbed02f561144a0b509c052339377d (commit) from 29393baef505112cf45a79a89a0c587290d4bdeb (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 ----------------------------------------------------------------- ----------------------------------------------------------------------- Summary of changes: test/mpi/runtests.in | 4 ++-- 1 files changed, 2 insertions(+), 2 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

Re: [mpich-commits] [mpi-all-commits] [mpich] MPICH primary repository branch, master, updated. v3.0.1-67-g29393ba
by Dave Goodell 24 Jan '13

24 Jan '13

Whoops, that should be #1787… On Jan 24, 2013, at 2:54 PM 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 29393baef505112cf45a79a89a0c587290d4bdeb (commit) > from 0739939ec85d44d05f00c84c6804715c35c94c78 (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 ----------------------------------------------------------------- > commit 29393baef505112cf45a79a89a0c587290d4bdeb > Author: Dave Goodell <goodell(a)mcs.anl.gov> > Date: Thu Jan 24 14:38:28 2013 -0600 > > tt#1786: set MPID_MAX_THREAD_LEVEL in ch3:sock correctly > > fixes #1786 > > No reviewer. > > ----------------------------------------------------------------------- > > Summary of changes: > src/mpid/ch3/channels/nemesis/mpichprereq | 4 ---- > src/mpid/ch3/channels/sock/mpichprereq | 5 ----- > src/mpid/ch3/channels/sock/subconfigure.m4 | 2 ++ > 3 files changed, 2 insertions(+), 9 deletions(-) > delete mode 100644 src/mpid/ch3/channels/nemesis/mpichprereq > delete mode 100644 src/mpid/ch3/channels/sock/mpichprereq > > > hooks/post-receive > -- > MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.1-67-g29393ba
by noreply＠mpich.org 24 Jan '13

24 Jan '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 29393baef505112cf45a79a89a0c587290d4bdeb (commit) from 0739939ec85d44d05f00c84c6804715c35c94c78 (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 ----------------------------------------------------------------- commit 29393baef505112cf45a79a89a0c587290d4bdeb Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Thu Jan 24 14:38:28 2013 -0600 tt#1786: set MPID_MAX_THREAD_LEVEL in ch3:sock correctly fixes #1786 No reviewer. ----------------------------------------------------------------------- Summary of changes: src/mpid/ch3/channels/nemesis/mpichprereq | 4 ---- src/mpid/ch3/channels/sock/mpichprereq | 5 ----- src/mpid/ch3/channels/sock/subconfigure.m4 | 2 ++ 3 files changed, 2 insertions(+), 9 deletions(-) delete mode 100644 src/mpid/ch3/channels/nemesis/mpichprereq delete mode 100644 src/mpid/ch3/channels/sock/mpichprereq hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.1-66-g0739939
by noreply＠mpich.org 24 Jan '13

24 Jan '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 0739939ec85d44d05f00c84c6804715c35c94c78 (commit) from 99b86a31577ff24934f99e7ce58c0d501ff429e1 (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 ----------------------------------------------------------------- commit 0739939ec85d44d05f00c84c6804715c35c94c78 Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Thu Jan 24 14:00:31 2013 -0600 tt#1785: add a symbol for configure testing Should allow PETSc to decide whether or not they can safely use nonblocking collectives with ch3:sock. Reviewed-by: buntinas ----------------------------------------------------------------------- Summary of changes: src/mpid/ch3/channels/sock/src/ch3_progress.c | 5 +++++ 1 files changed, 5 insertions(+), 0 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.1-65-g99b86a3
by noreply＠mpich.org 24 Jan '13

24 Jan '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 99b86a31577ff24934f99e7ce58c0d501ff429e1 (commit) from fc206c72fbdfd5d881a5adb7748eaabf063cb77b (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 ----------------------------------------------------------------- commit 99b86a31577ff24934f99e7ce58c0d501ff429e1 Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Thu Jan 24 12:58:03 2013 -0600 tt#1785: NBC progress issues with ch3:sock We made progress for MPI_Wait and friends, but not MPI_Test/MPI_Iprobe/etc. Thanks to Jed Brown for letting us know about this issue. fixes #1785 Reviewed-by: buntinas ----------------------------------------------------------------------- Summary of changes: src/mpid/ch3/channels/sock/src/ch3_progress.c | 5 ++++ test/mpi/coll/Makefile.am | 1 + test/mpi/coll/ibarrier.c | 33 +++++++++++++++++++++++++ test/mpi/coll/testlist | 4 +++ 4 files changed, 43 insertions(+), 0 deletions(-) create mode 100644 test/mpi/coll/ibarrier.c hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, portals-work, created. v3.0.1-63-g608e384
by noreply＠mpich.org 24 Jan '13

24 Jan '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, portals-work has been created at 608e3845d35feb8fc2735ca94f54a12f67c798bd (commit) - Log ----------------------------------------------------------------- commit 608e3845d35feb8fc2735ca94f54a12f67c798bd Author: Darius Buntinas <buntinas(a)mcs.anl.gov> Date: Wed Jan 23 17:15:03 2013 -0600 Various Portals netmod fixes o Added debug messages o Fixed bug where the lower-bound for contig datatypes was ignored o Added missing MEUnlink for truncated messages ----------------------------------------------------------------------- hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.1-64-gfc206c7
by noreply＠mpich.org 24 Jan '13

24 Jan '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 fc206c72fbdfd5d881a5adb7748eaabf063cb77b (commit) via 20545d5fe83d4e31b8fa5552836c64b730f6a100 (commit) from ffc5d6d5ecefc1d7917686cd6022e69ee81e1123 (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 ----------------------------------------------------------------- commit fc206c72fbdfd5d881a5adb7748eaabf063cb77b Author: James Dinan <dinan(a)mcs.anl.gov> Date: Thu Jan 24 09:55:20 2013 -0600 Fix poor error handling in RMA code commit 20545d5fe83d4e31b8fa5552836c64b730f6a100 Author: James Dinan <dinan(a)mcs.anl.gov> Date: Thu Jan 24 09:54:01 2013 -0600 Fix error handling bug for struct datatypes Reviewer: goodell ----------------------------------------------------------------------- Summary of changes: src/mpi/datatype/type_create_struct.c | 2 +- src/mpid/ch3/src/ch3u_handle_recv_req.c | 6 ++++-- 2 files changed, 5 insertions(+), 3 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.1-62-gffc5d6d
by noreply＠mpich.org 23 Jan '13

23 Jan '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 ffc5d6d5ecefc1d7917686cd6022e69ee81e1123 (commit) from ba027ca6db3509fffbe8b8a9afaa150618d9a9ea (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 ----------------------------------------------------------------- commit ffc5d6d5ecefc1d7917686cd6022e69ee81e1123 Author: Darius Buntinas <buntinas(a)mcs.anl.gov> Date: Wed Jan 23 16:47:05 2013 -0600 added DBG message for memcopy ----------------------------------------------------------------------- Summary of changes: .../channels/nemesis/include/mpid_nem_memdefs.h | 1 + 1 files changed, 1 insertions(+), 0 deletions(-) hooks/post-receive -- MPICH primary repository

1 0