GitLab

We call memory barriers at proper places in RMA sync calls
as following, and remove unnecessary memory barriers:

(1) Win_fence: very beginning and very end.
(2) Win_post/Win_complete: very beginning.
(3) Win_start/Win_wait/Win_test: very end.
(4) Win_lock/Win_lock_all: very end.
(5) Win_unlock/Win_unlock_all: very beginning.
(6) Win_flush/Win_flush_local/Win_flush_all/Win_flush_local_all: very beginning.

About the reason of doing this, please refer to comments
at the beginning of src/mpid/ch3/src/ch3u_rma_sync.c.

No reviewer.

In ending RMA synchronization calls, we poke the
progress engine at last if we never poke it before.
Because some program execution depends on the
incoming events in progress engine, if we never
process them we may cause deadlock in the program.

No reviewer.

Originally free_ops_before_completion functions only
works with active target. Here we modify it to accomodate
passive target as well.

Also, everytime we trigger free_ops_before_completion,
we lose the chance to do real Win_flush_local operation
and must do a Win_flush instead. Here we transfer
Win_flush_local to Win_flush if disable_flush_local flag
is set, and unset that flag after the current flush
is fone.

No reviewer.

No reviewer.

Use int instead of size_t in RMA pkt header to reduce
packet size.

No reviewer.

No reviewer.

No reviewer.

In this patch we allow GET/GACC response packets to
piggyback some IMMED data, just like what we did
for PUT/GACC/FOP/CAS packets.

No reviewer.

Originally we only allows LOCK request to be piggybacked
with small RMA operations (all data can be fit in packet
header). This brings communication overhead for larger
operations since origin side needs to wait for the LOCK
ACK before it can transmit data to the target.

In this patch we add support of piggybacking LOCK with
RMA operations with arbitrary size. Note that (1) this
only works with basic datatypes; (2) if the LOCK cannot
be satisfied, we temporarily buffer this operation on
the target side.

No reviewer.

No reviewer.

No reviewer.

No reviewer.

accumulated_ops_cnt is used to track no. of accumulated
posted RMA operations between two synchronization calls,
so that we can decide when to poke progress engine based
on the current value of this counter.

Here we initialize it to zero in the BEGINNING synchronization
calls (Win_fence, Win_start, first Win_lock, Win_lock_all),
and correctly decrement it in the ENDING synchronization calls
(Win_fence, Win_complete, Win_unlock, Win_unlock_all,
Win_flush, Win_flush_local, Win_flush_all, Win_flush_local_all).
We also use a per-target counter to track single target case.

No reviewer.

No reviewer.

Arrange RMA packet definition and structures in
src/mpid/ch3/include/mpidpkt.h in the following
order:

1. RMA operation packets: PUT, GET, ACC, GACC, CAS, FOP
2. RMA operation response packets: GET_RESP, GACC_RESP, CAS_RESP, FOP_RESP
3. RMA control packets: LOCK, UNLOCK, FLUSH, DECR_AT_COUNTER
4. RMA control response packets: LOCK_ACK, FLUSH_ACK

No reviewer.

Arrange RMA sync functions in src/mpid/ch3/src/ch3u_rma_sync.c
in the following order:

Win_fence
Win_post
Win_start
Win_complete
Win_wait
Win_test
Win_lock
Win_unlock
Win_flush
Win_flush_local
Win_lock_all
Win_unlock_all
Win_flush_all
Win_flush_local_all
Win_sync

No reviewer.

artem.v.yalozo authored Oct 23, 2014 and Kenneth Raffenetti committed Dec 08, 2014

This patch provides the following fix wrt Windows conformance feature
(makes single code working on both platforms Linux and Windows):
    - RMA mutexes fix for Windows

Change-Id: Ib4f7b2ec8a07813f0ed35281a1d584637c84c0a9
Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

andrey.lobanov authored Oct 17, 2014 and Kenneth Raffenetti committed Dec 05, 2014

  - name publishing/lookup support (pmi v1 and v2)
  - job and node attrs (v2)

Change-Id: Id18d968da0d0bbf6e8cb2e7acffaf77d82a5e8b0
Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

Paul Coffman authored Dec 04, 2014 and Rob Latham committed Dec 05, 2014

The original fix for 'romio gpfs: select correct read buffer' was still
missing a critical piece for the last round to use the correct read
buffer, resulting in a correctness issue that was missed by IOR but
still found by the IBM PE test team.  The fix was to correctly toggle
the buffer after the last read.
Signed-off-by: Paul Coffman <pkcoff@us.ibm.com>
Signed-off-by: Rob Latham <robl@mcs.anl.gov>

No reviewer

During error code creation, the error class was erroneously modified by
applying ERROR_DYN_MASK when.  The dynamic bit is already set for
user-defined error classes, so this bug had no effect in all existing
MPICH tests.  However, when a predefined error class was passed during
error code creation, it would be incorrectly marked as dynamic,
resulting in an invalid result when the error class of a returned error
code was returned via MPI_Error_class.
Signed-off-by: Wesley Bland <wbland@anl.gov>

Kuleshov Aleksey authored Dec 01, 2014 and Sangmin Seo committed Dec 02, 2014

Type of the third argument for MPIR_Pack_size_impl should be a pointer
to MPI_Aint. This patch fixes the wrong usage of int pointer for
MPIR_Pack_size_impl in NEWMAD and MXM netmods.
Signed-off-by: Sangmin Seo <sseo@anl.gov>

Signed-off-by: Wesley Bland <wbland@anl.gov>

Also remove the function prototype declaration since it is not used
out-of-order.
Signed-off-by: Wesley Bland <wbland@anl.gov>

Signed-off-by: Wesley Bland <wbland@anl.gov>

The function to convert the group of failed procs to a bitarray was
incorrectly quiting early if one of the globally known failed processes
was not in the communciator being dealt with.
Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

Since pamid doesn't include any of the fault tolerance functions, it
should never say that a message is pending failure. We also can't call
abort in here since the function is usind at the MPI layer.
Signed-off-by: Paul Coffman <pkcoff@us.ibm.com>

If a fatal error occurs, pass the MPI error code to MPID_Abort. To ensure
non-zero exit status with dynamic error codes, we set the first available
dynamic error class to 1. #Refs 1537
Signed-off-by: Pavan Balaji <balaji@anl.gov>

Implement abort in the Hydra PMI server and modify simple PMI to send
an abort command. Previously, we just exited the calling process and
relied on the process manager to detect it and cleanup the rest of the
job. Refs #1537
Signed-off-by: Pavan Balaji <balaji@anl.gov>

We simply use PMI_Abort in both the sock and nemesis code. Remove
extra functions and constants that are not useful. Refs #1537
Signed-off-by: Pavan Balaji <balaji@anl.gov>

Paul Coffman authored Nov 24, 2014 and Rob Latham committed Nov 24, 2014

ROMIO GPFSMPIO_P2PCONTIG threaded read needs to toggle first read buffer

When using both the GPFSMPIO_P2PCONTIG and GPFSMPIO_PTHREADIO
optimizations there was a correctness bug when reading where for the
first round the read buffer did not toggle to the two-phase buffer for
the pthread reader, resulting in diseminating the data from the wrong
buffer.  The fix is to do the toggle after the first read.
Signed-off-by: Paul Coffman <pkcoff@us.ibm.com>
Signed-off-by: Rob Latham <robl@mcs.anl.gov>

It is possible that a request handler of RMA request is
called for the second time inside the first called request
handler on the same request.

Consider the following case: a req is queued up in Nemesis
SHM queue with ref count of 2: one is for request completion
and another is for dequeueing from SHM queue. The first
called req handler completed this request and decrement ref
count to 1. This request is still in the queue. However,
within this handler, we trigger the same req handler on the
same request again (for example making progress on SHM queue),
and the second called handler also tries to complete this
request, which leads to the wrong execution.

In this patch we check if request has already been completed
when entering the req handler, to prevent processing the same
request twice. We also move the function finish_op_on_target()
(where the same req handler can be triggered again)
after request completion routine, so that we can mark the
current request as completed before enter the same req handler
for the second time.

Fix #2204
Signed-off-by: Pavan Balaji <balaji@anl.gov>

William Gropp authored Nov 24, 2014 and Rob Latham committed Nov 24, 2014

Update the use of DOCTEXT to match the rest of MPICH, including adding
-nolocation (drop the location of the source file from the documentation)
and ensure that the mpi.cit file contains the I/O routines as well as
the others (this file can be used to add links to the man pages in
other documents).
Signed-off-by: Rob Latham <robl@mcs.anl.gov>

Charles J Archer authored Nov 17, 2014 and Pavan Balaji committed Nov 21, 2014

 * Implements a tag matching interface netmod over the OFIWG Scalable Fabric Interfaces (SFI)

Paul Coffman authored Nov 20, 2014 and Rob Latham committed Nov 21, 2014

At the end of MPIDI_Init_collsel_extension in the pami device init code
mpid_init.c there is logic to disable the optimized collectives based on
criteria that is invalid on BGQ but was nonetheless always evaluating to
true and disabling the optimized collectives on BGQ.  Compiler
directives were placed around the logic to avoid this code for the  BGQ
platform.
Signed-off-by: Paul Coffman <pkcoff@us.ibm.com>
Signed-off-by: Rob Latham <robl@mcs.anl.gov>

If user does not explicitly set alloc_shm to TRUE in win_create, we
should never detect SHM windows because of expensive overhead. However,
current code does not check this info flag. This patch fixed it.

Closes #2161
Signed-off-by: Xin Zhao <xinzhao3@illinois.edu>

Signed-off-by: Antonio J. Pena <apenya@mcs.anl.gov>