4 hours agoMerge pull request #3940 from raffenet/dtpools-confdb master
Ken Raffenetti [Tue, 23 Jul 2019 13:23:00 +0000]
Merge pull request #3940 from raffenet/dtpools-confdb

dtpools: Use private copy of confdb

Approved-by: Hui Zhou <>

4 hours agoMerge pull request #3860 from raffenet/attr-check
Ken Raffenetti [Tue, 23 Jul 2019 13:22:44 +0000]
Merge pull request #3860 from raffenet/attr-check

mpi/attr: Remove problematic error check

Approved-by: Hui Zhou <>

21 hours agoMerge pull request #3935 from tarudoodi/fix-coll-testing
Ken Raffenetti [Mon, 22 Jul 2019 20:35:25 +0000]
Merge pull request #3935 from tarudoodi/fix-coll-testing

test: use correct cvar for testing

Approved-by: Ken Raffenetti <>

21 hours agodtpools: Use private copy of confdb
Ken Raffenetti [Mon, 22 Jul 2019 20:20:28 +0000]
dtpools: Use private copy of confdb

Avoid using confdb from the testsuite, which could cause issues when
building from tarball. Fixes pmodels/mpich#3939.

22 hours agoMerge pull request #3936 from hajimefu/klocwork-18172-pub
Hui Zhou [Mon, 22 Jul 2019 18:47:20 +0000]
Merge pull request #3936 from hajimefu/klocwork-18172-pub

mpir: Assert indirect sizes are positive

Approved-by: Hui Zhou <>
Approved-by: Ken Raffenetti <>

27 hours agoconfig: Move ptr alignment check to confdb
Ken Raffenetti [Mon, 22 Jul 2019 14:26:01 +0000]
config: Move ptr alignment check to confdb

This check is no longer in use by MPICH, but we will retain the logic
in confdb for possible later use.

27 hours agompi/attr: Remove problematic error check
Ken Raffenetti [Wed, 19 Jun 2019 14:34:40 +0000]
mpi/attr: Remove problematic error check

An attempt was made to catch users providing the wrong pointer type
when retrieving an attribute value. However, the check could cause
working codes to crash on platforms without strict alignment
requirements (e.g. x86_64). Remove this check, as it is not required.

Reported-by: Nick Radcliffe <>

3 days agompir: Assert indirect sizes are positive
Hajime Fujita [Wed, 17 Jul 2019 20:45:09 +0000]
mpir: Assert indirect sizes are positive

This patch adds assertions in MPIR_Handle_indirect_init to ensure
given sizes are positive. This will help not only static code
analyzer but also humans understanding requirements and behavior
of the function.

Fixes Klocwork 18172

3 days agoMerge pull request #3933 from raffenet/mtest-maxbufsize
Ken Raffenetti [Fri, 19 Jul 2019 17:41:22 +0000]
Merge pull request #3933 from raffenet/mtest-maxbufsize

mtest: Lower default max buffer size

Approved-by: Hui Zhou <>
Approved-by: Ken Raffenetti <>

4 days agotest: use correct cvar for testing
Taru Doodi [Thu, 18 Jul 2019 20:44:40 +0000]
test: use correct cvar for testing

5 days agodtpools: add sanity guard against erroneous maxbufsize
Hui Zhou [Tue, 16 Jul 2019 23:02:35 +0000]
dtpools: add sanity guard against erroneous maxbufsize

In the case user supply an erroneous maxbufsize, such as an
uninitialized variable or overflown integer, the DTP_obj_create will
enter a dead (or unnecessarily long) loop, which can be difficult to
debug. Add sanity guard against such issues.

5 days agomtest: Lower default max buffer size
Ken Raffenetti [Tue, 16 Jul 2019 20:51:59 +0000]
mtest: Lower default max buffer size

The previous value was negative when MPI_Aint is a 32-bit signed
integer. Use a new default of 1GB that should be safe on all platforms
we test on.

6 days agoMerge pull request #3929 from hajimefu/gl1784-namedrxctx-pub
Ken Raffenetti [Wed, 17 Jul 2019 14:54:57 +0000]
Merge pull request #3929 from hajimefu/gl1784-namedrxctx-pub

ch4/ofi: Use FI_NAMED_RX_CTX with SEP

Approved-by: Ken Raffenetti <>

7 days agoMerge pull request #3931 from raffenet/mpl-fdopen
Ken Raffenetti [Tue, 16 Jul 2019 14:03:53 +0000]
Merge pull request #3931 from raffenet/mpl-fdopen

mpl: Fix configure check

Approved-by: Hui Zhou <>

7 days agompl: Fix configure check
Ken Raffenetti [Mon, 15 Jul 2019 19:42:16 +0000]
mpl: Fix configure check

fdopen is declared in stdio.h. This check would always fail because it
was looking in the wrong header file.

8 days agoMerge pull request #3928 from hajimefu/gl1786-rm-pobj-pub
Ken Raffenetti [Mon, 15 Jul 2019 13:41:00 +0000]
Merge pull request #3928 from hajimefu/gl1786-rm-pobj-pub

ch4/ofi: Remove send_handler wrapper

Approved-by: Ken Raffenetti <>

10 days agoch4/ofi: Correct function state names
Hajime Fujita [Sat, 13 Jul 2019 04:14:08 +0000]
ch4/ofi: Correct function state names

10 days agoch4/ofi: Remove always-true param from MPIDI_OFI_do_control_send
Hajime Fujita [Sat, 13 Jul 2019 04:07:25 +0000]
ch4/ofi: Remove always-true param from MPIDI_OFI_do_control_send

`use_comm_table` turns out to be always true, no longer needed.

10 days agoch4/ofi: Remove lock parameters from AM functions
Hajime Fujita [Sat, 13 Jul 2019 03:56:04 +0000]
ch4/ofi: Remove lock parameters from AM functions

These parameters were used to specify a need for per-object locking.
Now that per-object lock has been removed from CH4, they are no longer

10 days agoch4/ofi: Use FI_NAMED_RX_CTX with SEP
Hajime Fujita [Wed, 27 Mar 2019 19:38:39 +0000]
ch4/ofi: Use FI_NAMED_RX_CTX with SEP

When scalable endpoint is enabled, FI_NAMED_RX_CTX is needed to
specify a destination receive context index.

10 days agoch4/ofi: Remove unused send_handler
Hajime Fujita [Wed, 27 Mar 2019 20:47:01 +0000]
ch4/ofi: Remove unused send_handler

10 days agoch4/ofi: Remove use of send_handler wrapper
Hajime Fujita [Wed, 27 Mar 2019 20:42:15 +0000]
ch4/ofi: Remove use of send_handler wrapper

MPIDI_OFI_send_handler was introduced to abstract CQ data availability,
but given that now we always assume CQ data support, we no longer
need this wrapper. This patch removes its usage.

10 days agoch4/ofi: Remove unused enum values
Hajime Fujita [Fri, 29 Mar 2019 21:00:35 +0000]
ch4/ofi: Remove unused enum values

10 days agoMerge pull request #3927 from tarudoodi/test-script-typo-fix
Hui Zhou [Fri, 12 Jul 2019 18:47:36 +0000]
Merge pull request #3927 from tarudoodi/test-script-typo-fix

test: fix typos in

Approved-by: Hui Zhou <>

11 days agotest: fix typos in
Taru Doodi [Fri, 12 Jul 2019 16:22:00 +0000]
test: fix typos in

11 days agoMerge pull request #3877 from yfguo/uninline-progress
Ken Raffenetti [Thu, 11 Jul 2019 19:15:38 +0000]
Merge pull request #3877 from yfguo/uninline-progress

ch4: uninline progress engine

Approved-by: Ken Reffenetti <>

12 days agoch4/stubnm: fix header inclusion order
Yanfei Guo [Wed, 26 Jun 2019 18:40:18 +0000]
ch4/stubnm: fix header inclusion order

12 days agoch4/stubnm: remove global.c.
Yanfei Guo [Wed, 26 Jun 2019 15:14:29 +0000]
ch4/stubnm: remove global.c.

The file only contains a function table which is a duplication of

12 days agoch4: uninline progress engine
Yanfei Guo [Fri, 22 Mar 2019 21:04:07 +0000]
ch4: uninline progress engine

Experiments has shown no performance impact after unlinine the entire
progress engine. This change significantly reduces the build size and

Co-authored-by: Hui Zhou <>

12 days agoMerge pull request #3918 from raffenet/tbsend
Ken Raffenetti [Thu, 11 Jul 2019 15:34:28 +0000]
Merge pull request #3918 from raffenet/tbsend

mpid: Remove unused tBsend interface

Approved-by: Giuseppe Congiu <>

12 days agompid: Remove unused tBsend interface
Ken Raffenetti [Tue, 9 Jul 2019 13:55:00 +0000]
mpid: Remove unused tBsend interface

According to an old comment, "this interface had the semantics of
MPI_Bsend, except that it returns the internal error code
MPID_WOULD_BLOCK if the message can''t be sent immediately (t is for
'try')." I cannot find an instance of this interface (or
MPID_WOULD_BLOCK) in pamid, ch3 or ch4, so just remove it.

12 days agoMerge pull request #3922 from raffenet/remove-retry2
Ken Raffenetti [Wed, 10 Jul 2019 18:44:40 +0000]
Merge pull request #3922 from raffenet/remove-retry2

ch4/ofi: Remove dubious macro

Approved-by: Hajime Fujita <>

13 days agoch4/ofi: Remove duplicate macro
Ken Raffenetti [Wed, 10 Jul 2019 16:24:37 +0000]
ch4/ofi: Remove duplicate macro

There is no difference between MPIDI_OFI_CALL and
MPIDI_OFI_CALL_NOLOCK. Convert usage and remove the duplicate.

13 days agoch4/ofi: Remove dubious macro
Ken Raffenetti [Wed, 10 Jul 2019 14:34:45 +0000]
ch4/ofi: Remove dubious macro

MPIDI_OFI_CALL_RETRY2 took a simple function as argument.  The rest of
the macro performed communication to be executed in a retry loop if
-FI_EAGAIN is returned. We already have MPIDI_OFI_CALL_RETRY. It is
unclear what value is added by taking a simple function call as a
macro argument instead of calling it normally, so just remove it.

13 days agoch4/ofi: Remove unused macro argument
Ken Raffenetti [Wed, 10 Jul 2019 15:07:08 +0000]
ch4/ofi: Remove unused macro argument

MPIDI_OFI_CALL_RETRY does not act on the LOCK argument
provided. Remove it and all related usages from the netmod code.

13 days agoMerge pull request #3919 from shawnccx/kloc_fix
Ken Raffenetti [Wed, 10 Jul 2019 14:30:17 +0000]
Merge pull request #3919 from shawnccx/kloc_fix

ch4/ofi: Add error check for MPIDI_OFI_INIT_CHUNK_CONTEXT

Approved-by: Ken Reffenetti <>

13 days agoMerge pull request #3914 from hajimefu/simplify-pcomm-pub
Ken Raffenetti [Tue, 9 Jul 2019 20:46:10 +0000]
Merge pull request #3914 from hajimefu/simplify-pcomm-pub

ch4: Fix persistent receive with MPI_ANY_SOURCE

Approved-by: Ken Raffenetti <>

2 weeks agoxfail: Update Fortran ssend test entries
Hajime Fujita [Wed, 3 Jul 2019 22:07:42 +0000]
xfail: Update Fortran ssend test entries

Mark Fortran ssend test entries to point to pmodels/mpich#3915

2 weeks agoxfail: Remove fortran psendf tests
Hajime Fujita [Wed, 3 Jul 2019 17:15:07 +0000]
xfail: Remove fortran psendf tests

2 weeks agomtest: Extend inactivereq to add ANY_SOURCE receive
Hajime Fujita [Tue, 16 Oct 2018 21:40:39 +0000]
mtest: Extend inactivereq to add ANY_SOURCE receive

This patch adds ANY_SOURCE receive handling test for inactive
persistent requests.

2 weeks agomtest: Enhance multi_psend_derived for ANY_SOURCE
Hajime Fujita [Tue, 16 Oct 2018 21:40:08 +0000]
mtest: Enhance multi_psend_derived for ANY_SOURCE

This patch extends multi_psend_derived to support
persistent receive with MPI_ANY_SOURCE.

2 weeks agoch4/generic: Use MPI_Aint for count
Hajime Fujita [Tue, 9 Jul 2019 16:25:06 +0000]
ch4/generic: Use MPI_Aint for count

2 weeks agoch4: Remove persistent API from NM/SHM/MPIDIG
Hajime Fujita [Wed, 3 Jul 2019 04:33:10 +0000]
ch4: Remove persistent API from NM/SHM/MPIDIG

Given that now persistent communication is handled at CH4,
related logics inside netmod/shmmod/MPIDIG are no longer used.

2 weeks agoch4: Refactor persistent communication handling
Hajime Fujita [Tue, 2 Jul 2019 21:14:08 +0000]
ch4: Refactor persistent communication handling

Current CH4 has a bug in persistent communication handling under
ch4-direct=auto mode, when MPI_Recv_init is called with
MPI_ANY_SOURCE as the source rank.

The issue is that current MPID_Recv_init implementation makes
two persistent requests from shmmod and netmod, but when MPID_Irecv
is called from MPID_Startall, MPID_Irecv again invokes shmmod and
netmod recv functions. So two difference places are handling
ANY_SOURCE logic twice.

There are two ways to fix the issue:
1) Create a special version of MPID_Irecv dedicated for persistent
2) Handle the persistent communication logic completely in CH4.
   At the time of MPID_Recv_init, CH4 does not treat ANY_SOURCE
   specially. When MPID_Irecv is called from MPID_Startall,
   MPID_Irecv handles ANY_SOURCE internally.

This patch implements 2). This means netmod/shmmod will lose a
capability to implement their own optimization logic for persistent
communication. It is considered okay for now, given every netmod/shmmod
ends up with calling MPIDIG to handle persistent communication.

1) was also considered and tried but it turned out to be a little bit
intrusive and required some CH4-level API change.
See for details.

This patch is based on Ken Raffenetti <>'s idea.

2 weeks agoch4/generic: Free NM/AM req on anysrc recv completion
Hajime Fujita [Tue, 9 Oct 2018 15:28:55 +0000]
ch4/generic: Free NM/AM req on anysrc recv completion

This patch fixes a request object leak when
* AM-only mode is used (or the netmod decides to use MPIDIG), and
* CH4 shared memory is turned on (`--enable-ch4-direct=auto`), and
* Receive from MPI_ANYSOURCE is performed, and
* Netmod receives the message

In case of anysource recieve, a request object generated by
the shmmod is always returned to the user. Thus, the user-level
request cleanup functions (e.g. MPI_Wait) will only free the
request object from shmmod. Therefore, netmod should decrement
its request object by itself, otherwise it will leak.

This logic follows what the OFI netmod is already doing (see

Signed-off-by: Ken Raffenetti <>

2 weeks agoch4/ofi: Release datatype on recv cancel
Hajime Fujita [Tue, 9 Oct 2018 15:15:41 +0000]
ch4/ofi: Release datatype on recv cancel

When a posted receive is canceled, and that receive takes a derived
datatype as an argument, we must release the refcount for the
datatype that was incremented at the time of receive posting,
otherwise it will lead to resource leak.

Signed-off-by: Ken Raffenetti <>

2 weeks agoMerge pull request #3917 from rzambre/without-num-queued-threads
Ken Raffenetti [Tue, 9 Jul 2019 15:33:53 +0000]
Merge pull request #3917 from rzambre/without-num-queued-threads

common: Remove counter that tracks number of threads waiting on a lock

Approved-by: Pavan Balaji <>

2 weeks agoch4/ofi: Add error check for MPIDI_OFI_INIT_CHUNK_CONTEXT
Chongxiao Cao [Mon, 8 Jul 2019 17:35:20 +0000]
ch4/ofi: Add error check for MPIDI_OFI_INIT_CHUNK_CONTEXT

Fix klocwork issue.

2 weeks agocommon: Remove counter that tracks number of threads waiting on a lock
Rohit Zambre [Thu, 9 May 2019 22:51:51 +0000]
common: Remove counter that tracks number of threads waiting on a lock

This counter adds overhead in both uncontended and contended cases.

2 weeks agoMerge pull request #3855 from hzhou/1906_mpix_h
Hui Zhou [Wed, 3 Jul 2019 22:08:46 +0000]
Merge pull request #3855 from hzhou/1906_mpix_h

bug: fix enable memory tracing and pmix together

Approved-by: Ken Reffenetti <>

2 weeks agoMerge pull request #3843 from raffenet/ucx-submodule
Ken Raffenetti [Wed, 3 Jul 2019 22:04:23 +0000]
Merge pull request #3843 from raffenet/ucx-submodule

ch4/ucx: Update submodule to v1.5.2

No reviewer.

2 weeks agoMerge pull request #3909 from raffenet/fix-intel-tls
Ken Raffenetti [Wed, 3 Jul 2019 21:23:02 +0000]
Merge pull request #3909 from raffenet/fix-intel-tls

init: Workaround Intel compiler TLS bug on macOS

Approved-by: Hui Zhou <>

2 weeks agoinit: Workaround Intel compiler TLS bug on macOS
Ken Raffenetti [Tue, 2 Jul 2019 19:59:23 +0000]
init: Workaround Intel compiler TLS bug on macOS

Work around a bug in the Intel compiler on macOS that does not treat
MPIR_Per_thread as thread-local storage when it is
instantiated. References to MPIR_Per_thread in other files would then
conflict with the non-TLS version. Add an explicit reference in
initthread.c to trick the compiler. Fixes pmodels/mpich#3437.

2 weeks agoMerge pull request #3913 from hzhou/1907_lonely_common
Hui Zhou [Wed, 3 Jul 2019 17:56:11 +0000]
Merge pull request #3913 from hzhou/1907_lonely_common

posix: lonely common symbol get lost during static linking

Approved-by: Wesley Bland <>
Approved-by: Ken Raffenetti <>

2 weeks agoposix: lonely common symbol get lost during static linking
Hui Zhou [Wed, 3 Jul 2019 12:12:13 +0000]
posix: lonely common symbol get lost during static linking

It only manifested on mac osx with --disable-shared.

2 weeks agoMerge pull request #3912 from hzhou/1907_test_coll_algos
Hui Zhou [Wed, 3 Jul 2019 11:56:55 +0000]
Merge pull request #3912 from hzhou/1907_test_coll_algos

test: fix typo in

Approved-by: Gengbin Zheng <>

2 weeks agotest: fix typo in
Hui Zhou [Tue, 2 Jul 2019 22:48:42 +0000]
test: fix typo in

2 weeks agoMerge pull request #3840 from hzhou/1906_segment
Hui Zhou [Tue, 2 Jul 2019 21:20:29 +0000]
Merge pull request #3840 from hzhou/1906_segment

datatypes: bugfix: update orig_block in segment_seek

Approved-by: Pavan Balaji <>
Approved-by: Yanfei Guo <>

3 weeks agoMerge pull request #3907 from zhenggb72/alltoallw-fix
Hui Zhou [Tue, 2 Jul 2019 16:20:41 +0000]
Merge pull request #3907 from zhenggb72/alltoallw-fix

coll:  fix typo for ialltoallw

Approved-by: Hui Zhou <>

3 weeks agocoll: fix typo
Gengbin Zheng [Tue, 2 Jul 2019 15:14:24 +0000]
coll:  fix typo

3 weeks agoMerge pull request #3904 from hzhou/1907_xfail_bcastlength
Hui Zhou [Tue, 2 Jul 2019 13:41:52 +0000]
Merge pull request #3904 from hzhou/1907_xfail_bcastlength

jenkins: xfail errors/coll/bcastlength for ucx am-only

Approved-by: Ken Raffenetti <>

3 weeks agojenkins: xfail errors/coll/bcastlength for ucx am-only
Hui Zhou [Mon, 1 Jul 2019 20:45:56 +0000]
jenkins: xfail errors/coll/bcastlength for ucx am-only

3 weeks agoMerge pull request #3898 from zhenggb72/sched-free
Hui Zhou [Mon, 1 Jul 2019 20:37:59 +0000]
Merge pull request #3898 from zhenggb72/sched-free

coll: Add MPIR_TSP_sched_free function for GenTran scheduler

Approved-by: Hui Zhou <>

3 weeks agoch4/ucx: Update submodule to v1.5.2
Ken Raffenetti [Mon, 10 Jun 2019 21:11:55 +0000]
ch4/ucx: Update submodule to v1.5.2

3 weeks agoMerge pull request #3879 from tarudoodi/add-algo-tests
Hui Zhou [Mon, 1 Jul 2019 19:30:07 +0000]
Merge pull request #3879 from tarudoodi/add-algo-tests

mtest: Add algo tests

Approved-by: Hui Zhou <>

3 weeks agomtest: Add tests for Alltoallv gentran inplace algo
Taru Doodi [Wed, 26 Jun 2019 14:22:07 +0000]
mtest: Add tests for Alltoallv gentran inplace algo

3 weeks agomtest: Add tests for Alltoallv gentran blocked algo
Taru Doodi [Wed, 26 Jun 2019 14:20:45 +0000]
mtest: Add tests for Alltoallv gentran blocked algo

3 weeks agomtest: Add tests for Alltoallw gentran inplace algorithm
Taru Doodi [Tue, 25 Jun 2019 22:02:39 +0000]
mtest: Add tests for Alltoallw gentran inplace algorithm

3 weeks agomtest: Add tests for Alltoallw gentran blocked algorithm
Taru Doodi [Tue, 25 Jun 2019 21:59:53 +0000]
mtest: Add tests for Alltoallw gentran blocked algorithm

3 weeks agocoll: Add MPIR_TSP_sched_free function for GenTran scheduler
Gengbin Zheng [Fri, 28 Jun 2019 14:57:17 +0000]
coll: Add MPIR_TSP_sched_free function for GenTran scheduler

Make free'ing of a schedule as a separate function as this will
be called from different places.

3 weeks agoMerge pull request #3900 from gcongiu/pr/mem/eager-fix
Giuseppe Congiu [Mon, 1 Jul 2019 14:52:55 +0000]
Merge pull request #3900 from gcongiu/pr/mem/eager-fix

ch4/shm: free local rank cache in fastbox allocation if error occurs

Approved-by: Ken Raffenetti <>

3 weeks agoMerge pull request #3901 from lnkgyv/fortran_cross_variables
Hui Zhou [Mon, 1 Jul 2019 14:50:26 +0000]
Merge pull request #3901 from lnkgyv/fortran_cross_variables

doc: add sample variable for cross compile

Approved-by: Hui Zhou <>

3 weeks agoMerge pull request #3899 from lnkgyv/configure_cross_fix
Hui Zhou [Mon, 1 Jul 2019 14:39:08 +0000]
Merge pull request #3899 from lnkgyv/configure_cross_fix

configure: fix 'confcross' placement.

Approved-by: Hui Zhou <>

3 weeks agodoc: add sample variable for cross compile
Yurij Goncharuk [Mon, 1 Jul 2019 10:12:41 +0000]
doc: add sample variable for cross compile

3 weeks agoch4/shm: free local rank cache in fastbox allocation if error occurs
Giuseppe Congiu [Mon, 1 Jul 2019 01:34:01 +0000]
ch4/shm: free local rank cache in fastbox allocation if error occurs

The fastbox first_poll_local_ranks cache should be freed if an error
happens. This patch changes the allocation from using MPL_malloc to
MPIR_CHKPMEM_MALLOC which manages errors and redirects to appropriate
error handling, i.e., MPIR_CHKPMEM_REAP.

3 weeks agoMerge pull request #3851 from gcongiu/testsuite-fix-disable-static-option
Giuseppe Congiu [Sat, 29 Jun 2019 01:30:54 +0000]
Merge pull request #3851 from gcongiu/testsuite-fix-disable-static-option

mpi/test: remove dtpools static library dependency

Approved-by: Pavan Balaji <>
Approved-by: Ken Raffenetti <>
Approved-by: Hui Zhou <>

3 weeks agotest/mpi: link dtpools to tests using libtool
Giuseppe Congiu [Sun, 16 Jun 2019 21:27:02 +0000]
test/mpi: link dtpools to tests using libtool

3 weeks agotest/mpi: convert dtpools into a convenience library
Giuseppe Congiu [Fri, 14 Jun 2019 23:47:06 +0000]
test/mpi: convert dtpools into a convenience library

3 weeks agoconfigure: fix 'confcross' placement.
Yurij Goncharuk [Fri, 28 Jun 2019 16:23:40 +0000]
configure: fix 'confcross' placement.

During cross compile export variables are source via 'confcross' file
which is located in the mpich root source directory. It's need to set
full path to file to be sourced where bash works in posix mode.

3 weeks agomtest: Add tests for gatherv gentran linear algorithm
Taru Doodi [Tue, 25 Jun 2019 19:50:17 +0000]
mtest: Add tests for gatherv gentran linear algorithm

3 weeks agoMerge pull request #3887 from tarudoodi/alltoall-scattered
Hui Zhou [Fri, 28 Jun 2019 15:40:00 +0000]
Merge pull request #3887 from tarudoodi/alltoall-scattered

coll: TSP Ialltoall scattered

Approved-by: Hui Zhou <>

3 weeks agoMerge pull request #3890 from zhenggb72/scatterv-linear
Hui Zhou [Fri, 28 Jun 2019 15:36:29 +0000]
Merge pull request #3890 from zhenggb72/scatterv-linear

iscatterv gentran-based linear algorithm

Approved-by: Hui Zhou <>

3 weeks agoMerge pull request #3895 from hzhou/1906_warnings
Hui Zhou [Fri, 28 Jun 2019 15:02:01 +0000]
Merge pull request #3895 from hzhou/1906_warnings

coll: warnings squash

Approved-by: Wesley Bland <>

3 weeks agomtest: Add tests for alltoall scattered algo
Taru Doodi [Wed, 26 Jun 2019 21:43:42 +0000]
mtest: Add tests for alltoall scattered algo

3 weeks agocoll: Implements alltoall scattered algorithm
Taru Doodi [Thu, 19 Jul 2018 15:30:21 +0000]
coll: Implements alltoall scattered algorithm

Co-authored-by: Akhil Langer <>

3 weeks agoofi: warning squash [-Wunused-variable]
Hui Zhou [Fri, 28 Jun 2019 01:44:09 +0000]
ofi: warning squash [-Wunused-variable]

3 weeks agoinit: -Wmissing-braces in global variable init
Hui Zhou [Fri, 28 Jun 2019 01:41:03 +0000]
init: -Wmissing-braces in global variable init

3 weeks agocoll: remove FUNCNAME added by last few commits
Hui Zhou [Fri, 28 Jun 2019 01:35:55 +0000]
coll: remove FUNCNAME added by last few commits

3 weeks agocoll: warnings squash
Hui Zhou [Fri, 28 Jun 2019 01:23:02 +0000]
coll: warnings squash

3 weeks agoMerge pull request #3888 from tarudoodi/allreduce-rabnsfnr-pr
Hui Zhou [Fri, 28 Jun 2019 00:37:36 +0000]
Merge pull request #3888 from tarudoodi/allreduce-rabnsfnr-pr

coll: TSP Iallreduce using recursive exchange reduce scatter and allgatherv

Approved-by: Hui Zhou <>

3 weeks agoMerge pull request #3893 from raffenet/opa-ignore
Hui Zhou [Thu, 27 Jun 2019 23:34:58 +0000]
Merge pull request #3893 from raffenet/opa-ignore

gitignore: Add opa's copy of confdb

Approved-by: Hui Zhou <>

3 weeks agoMerge pull request #3882 from hzhou/1906_cvar_group
Hui Zhou [Thu, 27 Jun 2019 22:16:37 +0000]
Merge pull request #3882 from hzhou/1906_cvar_group

cvar: add enum group

Approved-by: Gengbin Zheng <>
Approved-by: Rubasri Kalidas <>

3 weeks agomtest: Add tests for allreduce recexch reduce scatter-recexch allgatherv algo
Taru Doodi [Wed, 26 Jun 2019 22:27:33 +0000]
mtest: Add tests for allreduce recexch reduce scatter-recexch allgatherv algo

3 weeks agocoll: Add recexch reduce scatter - recexch allgatherv algorithm for allreduce
Rubasri Kalidas [Mon, 13 Aug 2018 14:34:57 +0000]
coll: Add recexch reduce scatter - recexch allgatherv algorithm for allreduce

This commit adds allreduce using reduce scatter
followed by allgatherv (recursive exchange). This algorithm
allows us to have 'K' number of processes participating
in each exchange.

3 weeks agocoll: Move step2 in reduce scatter recursive exchange to a separate function
Rubasri Kalidas [Mon, 13 Aug 2018 14:32:43 +0000]
coll: Move step2 in reduce scatter recursive exchange to a separate function

3 weeks agocoll: Extend reduce scatter recursive exchange to include distance halving
Rubasri Kalidas [Mon, 13 Aug 2018 14:26:36 +0000]
coll: Extend reduce scatter recursive exchange to include distance halving

3 weeks agogitignore: Add opa's copy of confdb
Ken Raffenetti [Thu, 27 Jun 2019 20:01:32 +0000]
gitignore: Add opa's copy of confdb

Hui Zhou [Thu, 27 Jun 2019 03:22:39 +0000]

Patch to work with enum group.

3 weeks agoMerge pull request #3713 from shawnccx/ch4_mqd_am_march_19
Yanfei Guo [Thu, 27 Jun 2019 18:37:48 +0000]
Merge pull request #3713 from shawnccx/ch4_mqd_am_march_19

ch4: Message Queue Dumping Support for CH4 active message

Approved-by: Ken Raffenetti <>

3 weeks agoMerge pull request #3874 from zhenggb72/alltoallw
Hui Zhou [Thu, 27 Jun 2019 18:26:50 +0000]
Merge pull request #3874 from zhenggb72/alltoallw

Alltoallw inplace and block algorithms

Approved-by: Hui Zhou <>