charm.git
9 years agoinit maxfiles to number of physnodes
Eric Bohm [Wed, 8 Feb 2012 18:53:37 +0000 (12:53 -0600)]
init maxfiles to number of physnodes

9 years agodescribes what the example is for
Eric Bohm [Wed, 8 Feb 2012 00:28:34 +0000 (18:28 -0600)]
describes what the example is for

9 years agoMerge branch 'charm' of charmgit:charm into charm
Eric Bohm [Wed, 8 Feb 2012 00:24:33 +0000 (18:24 -0600)]
Merge branch 'charm' of charmgit:charm into charm

9 years agoAdd map object for pupdisk to distribute them to one per node when possible
Eric Bohm [Wed, 8 Feb 2012 00:23:55 +0000 (18:23 -0600)]
Add map object for pupdisk to distribute them to one per node when possible

9 years agoMeshStreamer and NDMeshStreamer: default to a single buffer per destination
Lukasz Wesolowski [Tue, 7 Feb 2012 23:00:41 +0000 (17:00 -0600)]
MeshStreamer and NDMeshStreamer: default to a single buffer per destination
when first argument to constructor is too small or invalid.

9 years agoadded missing cmidirect.c
Yanhua Sun [Tue, 7 Feb 2012 22:41:38 +0000 (14:41 -0800)]
added missing cmidirect.c

9 years agoMerge branch 'charm' of charmgit:charm into charm
Yanhua Sun [Tue, 7 Feb 2012 21:20:39 +0000 (13:20 -0800)]
Merge branch 'charm' of charmgit:charm into charm

9 years agoadded CmiDirect in Gemini
Yanhua Sun [Tue, 7 Feb 2012 21:20:22 +0000 (13:20 -0800)]
added CmiDirect in Gemini

9 years agominor tweaks on print messages
Gengbin Zheng [Tue, 7 Feb 2012 20:23:21 +0000 (14:23 -0600)]
minor tweaks on print messages

9 years agoadded a new pattern for repeating: e.g 2X0-4
Gengbin Zheng [Tue, 7 Feb 2012 05:35:28 +0000 (23:35 -0600)]
added a new pattern for repeating:  e.g 2X0-4
For Cray, with +setcpuaffinity and no pemap and commap, it will compute a map for comm threads

9 years agominor tweak on print
Gengbin Zheng [Tue, 7 Feb 2012 04:49:40 +0000 (22:49 -0600)]
minor tweak on print

9 years agoworking version. needs small amount of placing tweaking to put
Eric Bohm [Tue, 7 Feb 2012 01:58:02 +0000 (19:58 -0600)]
working version.  needs small amount of placing tweaking to put
one element per node

9 years agofix an invalid pointer reference error whern npes = 1
Gengbin Zheng [Mon, 6 Feb 2012 15:57:31 +0000 (09:57 -0600)]
fix an invalid pointer reference error whern npes = 1

9 years agominor celan up code, allow using a link list instead of looping sendQ.
Gengbin Zheng [Mon, 6 Feb 2012 06:53:09 +0000 (00:53 -0600)]
minor celan up code, allow using a link list instead of looping sendQ.

9 years agonew env "CHARM_UGNI_MEMPOOL_SIZE"
Gengbin Zheng [Sat, 4 Feb 2012 19:33:30 +0000 (13:33 -0600)]
new env "CHARM_UGNI_MEMPOOL_SIZE"

9 years agoa function to read size from string with K,M,G.
Gengbin Zheng [Sat, 4 Feb 2012 19:32:06 +0000 (13:32 -0600)]
a function to read size from string with K,M,G.

9 years agopassing destNode to LrtsSendNetworkFunc.
Gengbin Zheng [Fri, 3 Feb 2012 23:11:27 +0000 (15:11 -0800)]
passing destNode to LrtsSendNetworkFunc.

9 years agoGPUManager: only print mempool diagnostics when debugging flag is on
Lukasz Wesolowski [Fri, 3 Feb 2012 22:34:26 +0000 (16:34 -0600)]
GPUManager: only print mempool diagnostics when debugging flag is on

9 years agoCmiAbort to LrtsAbort
Gengbin Zheng [Fri, 3 Feb 2012 07:38:36 +0000 (01:38 -0600)]
CmiAbort to LrtsAbort

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Fri, 3 Feb 2012 07:37:32 +0000 (01:37 -0600)]
Merge branch 'charm' of charmgit:charm into charm

9 years agoCmiAbort to LrtsAbort in machine LRTS
Gengbin Zheng [Fri, 3 Feb 2012 07:37:14 +0000 (01:37 -0600)]
CmiAbort to LrtsAbort in machine LRTS

9 years agoCmiAbort in LRTS change to LrtsAbort
Gengbin Zheng [Fri, 3 Feb 2012 07:34:13 +0000 (01:34 -0600)]
CmiAbort in LRTS change to LrtsAbort

9 years agochange number of iterations
Gengbin Zheng [Fri, 3 Feb 2012 06:30:54 +0000 (00:30 -0600)]
change number of iterations

9 years agomove CmiAbort to machine common code, and add a new LrtsAbort for LRTS.
Gengbin Zheng [Fri, 3 Feb 2012 06:26:47 +0000 (00:26 -0600)]
move CmiAbort to machine common code, and add a new LrtsAbort for LRTS.
clean up shmem and semaphore in pxshm version
add LrtsSendNetwork() which is a wrapper for LrtsSend() with pxshm support.

9 years agocode clean up
Yanhua Sun [Thu, 2 Feb 2012 23:53:44 +0000 (17:53 -0600)]
code clean up

9 years agoremove semphore file at exit
Gengbin Zheng [Thu, 2 Feb 2012 17:38:38 +0000 (11:38 -0600)]
remove semphore file at exit

9 years agoRevert "charmxi: print commit ID when erroring out" et al.
Phil Miller [Thu, 2 Feb 2012 15:32:04 +0000 (09:32 -0600)]
Revert "charmxi: print commit ID when erroring out" et al.

Rebuilding charmxi everytime the output of 'git describe' changes
means that all the headers get rebuilt, and then the whole
system. That makes development much less pleasant.

This reverts commit af295183fad45e9cb1eb8085afebd3f6d30915c9.
This reverts commit 8d3d3c41040aa4f16d2a438c28057e099c6af497.
This reverts commit d88c354dfdc9f42ae6b5d49421c41307e8bf6950.

9 years agofix a bug when running on 1 pe, due to commenting out size checking in AdvanceCommuni...
Gengbin Zheng [Thu, 2 Feb 2012 16:22:11 +0000 (08:22 -0800)]
fix a bug when running on 1 pe, due to commenting out size checking in AdvanceCommunication.

9 years agominor fix for sendbuf
Yanhua Sun [Thu, 2 Feb 2012 08:22:34 +0000 (02:22 -0600)]
minor fix for sendbuf

9 years agofix the assignment to source_addr
Gengbin Zheng [Thu, 2 Feb 2012 07:24:07 +0000 (01:24 -0600)]
fix the assignment to source_addr

9 years agominor change
Yanhua Sun [Thu, 2 Feb 2012 07:10:12 +0000 (01:10 -0600)]
minor change

9 years agomake sendbuf a linklist in non-smp
Yanhua Sun [Thu, 2 Feb 2012 06:52:16 +0000 (00:52 -0600)]
make sendbuf a linklist in non-smp

9 years agoadded top in pcqueue
Yanhua Sun [Thu, 2 Feb 2012 06:51:49 +0000 (00:51 -0600)]
added top in pcqueue

9 years agominor change, avoid repeatly opening xpmem_fd file.
Gengbin Zheng [Thu, 2 Feb 2012 06:42:54 +0000 (00:42 -0600)]
minor change, avoid repeatly opening xpmem_fd file.

9 years agoadd xpmem into machine common core.
Gengbin Zheng [Wed, 1 Feb 2012 07:41:49 +0000 (01:41 -0600)]
add xpmem into machine common core.

9 years agoan alternative to pxshm for intra-node communication, uses Linux kernel module xpmem
Gengbin Zheng [Wed, 1 Feb 2012 06:26:39 +0000 (22:26 -0800)]
an alternative to pxshm for intra-node communication, uses Linux kernel module xpmem

9 years agoNDMeshStreamer: bug fixes
Lukasz Wesolowski [Wed, 1 Feb 2012 06:11:48 +0000 (00:11 -0600)]
NDMeshStreamer: bug fixes

9 years agoMeshStreamer: checking in a new version of the code generalized
Lukasz Wesolowski [Wed, 1 Feb 2012 00:43:08 +0000 (18:43 -0600)]
MeshStreamer: checking in a new version of the code generalized
to N dimensions. Keeping the old version around at least until
this one is well tested.

9 years agocatch signals to clean up pxshm files
Gengbin Zheng [Tue, 31 Jan 2012 21:21:37 +0000 (15:21 -0600)]
catch signals to clean up pxshm files

9 years agoUsed the existing immediate msg queue to implement the function that a worker thread...
Chao Mei [Tue, 31 Jan 2012 20:59:43 +0000 (14:59 -0600)]
Used the existing immediate msg queue to implement the function that a worker thread has sth to notify the local comm thread.

9 years agominor change in coding style to avoid potential bug
Chao Mei [Tue, 31 Jan 2012 03:34:30 +0000 (21:34 -0600)]
minor change in coding style to avoid potential bug

9 years agoadded locks to ensure GNI_Call thread safe
Yanhua Sun [Tue, 31 Jan 2012 03:19:22 +0000 (21:19 -0600)]
added locks to ensure GNI_Call thread safe

9 years agoadded multiple worker thread sending
Yanhua Sun [Mon, 30 Jan 2012 23:12:41 +0000 (17:12 -0600)]
added multiple worker thread sending

9 years agoImplemented a functionality that allows user to designate some function to be execute...
Chao Mei [Mon, 30 Jan 2012 21:27:59 +0000 (15:27 -0600)]
Implemented a functionality that allows user to designate some function to be executed on the local comm thread in SMP mode. To use such functionality, define macro CMK_LEVERAGE_COMMTHREAD=1 and refer to machine-commthd-util.c for details. Note that this functionality is not supported in the standalone charm mode of net layer.

9 years agoMerge branch 'charm' of charmgit:charm into charm
Chao Mei [Mon, 30 Jan 2012 19:03:32 +0000 (13:03 -0600)]
Merge branch 'charm' of charmgit:charm into charm

9 years agoMinor changes to pingpong so that the program could measure intra-node latency in...
Chao Mei [Mon, 30 Jan 2012 19:02:50 +0000 (13:02 -0600)]
Minor changes to pingpong so that the program could measure intra-node latency in SMP mode.

9 years agocall the correct migrate constructor
Gengbin Zheng [Mon, 30 Jan 2012 18:22:31 +0000 (12:22 -0600)]
call the correct migrate constructor

9 years agoaddjust max hit
Yanhua Sun [Mon, 30 Jan 2012 05:06:15 +0000 (23:06 -0600)]
addjust max hit

9 years agofix a bug when hit send buffer max, the control msg is discarded mistakenly due to...
Gengbin Zheng [Mon, 30 Jan 2012 03:49:50 +0000 (21:49 -0600)]
fix a bug when hit send buffer max, the control msg is discarded mistakenly due to wrong status value.

9 years agoadded useSendQueue useRecvQueue in gni
Yanhua Sun [Mon, 30 Jan 2012 00:58:14 +0000 (16:58 -0800)]
added useSendQueue useRecvQueue in gni

9 years agoimplement big message pipelining.
Gengbin Zheng [Sun, 29 Jan 2012 18:03:08 +0000 (10:03 -0800)]
implement big message pipelining.

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Sat, 28 Jan 2012 05:01:22 +0000 (21:01 -0800)]
Merge branch 'charm' of charmgit:charm into charm

9 years agominor
Gengbin Zheng [Sat, 28 Jan 2012 05:01:02 +0000 (21:01 -0800)]
minor

9 years agochange smsg max size
Yanhua Sun [Sat, 28 Jan 2012 02:48:43 +0000 (18:48 -0800)]
change smsg max size

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Thu, 26 Jan 2012 05:55:17 +0000 (21:55 -0800)]
Merge branch 'charm' of charmgit:charm into charm

9 years agofixed pingpong size
Gengbin Zheng [Thu, 26 Jan 2012 05:52:38 +0000 (21:52 -0800)]
fixed pingpong size

9 years agoFix histogram row index calculations
Aaron Becker [Wed, 25 Jan 2012 18:53:11 +0000 (12:53 -0600)]
Fix histogram row index calculations

9 years agomove handleOneRecvedMsg after tracing
Gengbin Zheng [Wed, 25 Jan 2012 17:18:18 +0000 (09:18 -0800)]
move handleOneRecvedMsg after tracing

9 years agofixed timer problem in gemini
Yanhua Sun [Wed, 25 Jan 2012 03:29:58 +0000 (21:29 -0600)]
fixed timer problem in gemini

9 years agoMerge branch 'charm' of charmgit:charm into charm
Yanhua Sun [Wed, 25 Jan 2012 03:28:40 +0000 (21:28 -0600)]
Merge branch 'charm' of charmgit:charm into charm

9 years agoGemini uses its own timer instead of general one
Yanhua Sun [Wed, 25 Jan 2012 03:28:33 +0000 (21:28 -0600)]
Gemini uses its own timer instead of general one

9 years agoMeshStreamer: added support for incomplete meshes.
Lukasz Wesolowski [Tue, 24 Jan 2012 05:04:14 +0000 (23:04 -0600)]
MeshStreamer: added support for incomplete meshes.

9 years agomake ibverb work in standalone mode
Gengbin Zheng [Mon, 23 Jan 2012 19:22:47 +0000 (11:22 -0800)]
make ibverb work in standalone mode

9 years agonew histogram example for MSA
Aaron Becker [Fri, 20 Jan 2012 05:17:51 +0000 (23:17 -0600)]
new histogram example for MSA

9 years agofix CmiBarrier newly added to charmInit for bigsim
Gengbin Zheng [Thu, 19 Jan 2012 16:39:18 +0000 (10:39 -0600)]
fix CmiBarrier newly added to charmInit for bigsim

9 years agofixed a bug in ugni barrier
Yanhua Sun [Wed, 18 Jan 2012 21:07:28 +0000 (15:07 -0600)]
fixed a bug in ugni barrier

9 years agopup CkArray::stableLocations
Gengbin Zheng [Wed, 18 Jan 2012 19:24:15 +0000 (13:24 -0600)]
pup CkArray::stableLocations

9 years agodisable CMK_BROADCAST_USE_CMIREFERENCE when with SMP
Gengbin Zheng [Wed, 18 Jan 2012 05:43:53 +0000 (23:43 -0600)]
disable CMK_BROADCAST_USE_CMIREFERENCE when with SMP

9 years agowhen pxshm is enaled print some info to confirm
Gengbin Zheng [Wed, 18 Jan 2012 05:43:03 +0000 (23:43 -0600)]
when pxshm is enaled print some info to confirm

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Wed, 18 Jan 2012 04:03:30 +0000 (20:03 -0800)]
Merge branch 'charm' of charmgit:charm into charm

9 years agouse CmiSyncBroadcastAllAndFree
Gengbin Zheng [Wed, 18 Jan 2012 04:03:10 +0000 (20:03 -0800)]
use CmiSyncBroadcastAllAndFree

9 years agofix timer synchronized problem in ugni
Yanhua Sun [Tue, 17 Jan 2012 21:56:09 +0000 (13:56 -0800)]
fix timer synchronized problem in ugni

9 years agofix for compilation errors
Gengbin Zheng [Mon, 16 Jan 2012 23:24:32 +0000 (15:24 -0800)]
fix for compilation errors

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Mon, 16 Jan 2012 23:22:13 +0000 (15:22 -0800)]
Merge branch 'charm' of charmgit:charm into charm

9 years agowhen can use CMK_BROADCAST_USE_CMIREFERENCE, optimize ListSend functions to avoid...
Gengbin Zheng [Mon, 16 Jan 2012 23:21:39 +0000 (15:21 -0800)]
when can use CMK_BROADCAST_USE_CMIREFERENCE, optimize ListSend functions to avoid copying.

9 years agoMerge branch 'charm' of charmgit:charm into charm
Xiang Ni [Mon, 16 Jan 2012 18:01:15 +0000 (12:01 -0600)]
Merge branch 'charm' of charmgit:charm into charm

9 years agosupport CsdScheduleNodePoll
Xiang Ni [Mon, 16 Jan 2012 18:00:56 +0000 (12:00 -0600)]
support CsdScheduleNodePoll

9 years agooptimize the previous change, only need to make a copy, when refcount > 1, i.e. some...
Gengbin Zheng [Mon, 16 Jan 2012 17:51:10 +0000 (09:51 -0800)]
optimize the previous change, only need to make a copy, when refcount > 1, i.e. some outgoing messages are buffered.

9 years agoa function to get memory refcount.
Gengbin Zheng [Mon, 16 Jan 2012 17:50:19 +0000 (09:50 -0800)]
a function to get memory refcount.

9 years agosame fix for processProcBcastMsg
Gengbin Zheng [Mon, 16 Jan 2012 17:09:13 +0000 (09:09 -0800)]
same fix for processProcBcastMsg

9 years agowhen use refcounter in broadcast, it may create a subtle situation that a message...
Gengbin Zheng [Mon, 16 Jan 2012 17:02:19 +0000 (09:02 -0800)]
when use refcounter in broadcast, it may create a subtle situation that a message which is being sent, is executed by a local entry function and gets changed, therefore the message sent out later from the same buffer is incorrect.
When doing such broadcast, always make a copy of message when deliver to the processor itself.

9 years agowhen using CMK_BROADCAST_USE_CMIREFERENCE, make sure msg is copied once in case msg...
Gengbin Zheng [Mon, 16 Jan 2012 06:07:47 +0000 (22:07 -0800)]
when using CMK_BROADCAST_USE_CMIREFERENCE, make sure msg is copied once in case msg is on stack.

9 years agoindentation
Gengbin Zheng [Sun, 15 Jan 2012 06:50:06 +0000 (22:50 -0800)]
indentation

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Sun, 15 Jan 2012 06:18:45 +0000 (00:18 -0600)]
Merge branch 'charm' of charmgit:charm into charm

9 years agosimply and bug fix to comm thread tracing.
Gengbin Zheng [Sun, 15 Jan 2012 06:18:02 +0000 (00:18 -0600)]
simply and bug fix to comm thread tracing.
messages need to be changed (srcpe and eventID) when switching between worker thread and comm threads.

9 years agofix trace commthrad bug of receiving comm thread wrong
Yanhua Sun [Sun, 15 Jan 2012 06:12:11 +0000 (22:12 -0800)]
fix trace commthrad bug of receiving comm thread wrong

9 years agoclean up tracing comm thread
Gengbin Zheng [Sun, 15 Jan 2012 04:30:17 +0000 (20:30 -0800)]
clean up tracing comm thread
1. make TRACE_COMM)* macros to simplify code
2. fix tracing when converse messages are encountered
3. fix srcpe on comm thread to pretend that the msg is sent now from comm thread (need to be tested abd verify in Projections)

9 years agosanity check, disable CMK_SMP_TRACE_COMMTHREAD if not SMP.
Gengbin Zheng [Sat, 14 Jan 2012 23:16:10 +0000 (17:16 -0600)]
sanity check, disable CMK_SMP_TRACE_COMMTHREAD if not SMP.

9 years agoinitialize epIdx in envelope so that trace comm threads can work properly.
Gengbin Zheng [Sat, 14 Jan 2012 08:38:26 +0000 (00:38 -0800)]
initialize epIdx in envelope so that trace comm threads can work properly.

9 years agotwo more comm thread tracing functions need to be sure the msg is charm message
Gengbin Zheng [Sat, 14 Jan 2012 06:01:09 +0000 (00:01 -0600)]
two more comm thread tracing functions need to be sure the msg is charm message

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Sat, 14 Jan 2012 05:34:29 +0000 (23:34 -0600)]
Merge branch 'charm' of charmgit:charm into charm

9 years agofix a bug in traceSendMsgComm() which should only trace charm messages (no Converse...
Gengbin Zheng [Sat, 14 Jan 2012 05:33:49 +0000 (23:33 -0600)]
fix a bug in traceSendMsgComm() which should only trace charm messages (no Converse messages)

9 years agoadded gemini trace comm thread
Yanhua Sun [Sat, 14 Jan 2012 04:21:50 +0000 (20:21 -0800)]
added gemini trace comm thread

9 years agofix minor things for VC++
Gengbin Zheng [Fri, 13 Jan 2012 15:04:58 +0000 (09:04 -0600)]
fix minor things for VC++

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Fri, 13 Jan 2012 07:32:17 +0000 (23:32 -0800)]
Merge branch 'charm' of charmgit:charm into charm

9 years agoreduced numIter a bit and some other minor changes
Gengbin Zheng [Fri, 13 Jan 2012 07:29:05 +0000 (23:29 -0800)]
reduced numIter a bit and some other minor changes

9 years agominor change for using +comm_thread_only_recv
Gengbin Zheng [Fri, 13 Jan 2012 07:19:50 +0000 (23:19 -0800)]
minor change for using +comm_thread_only_recv

9 years agochange the implementation of ARMCI_Malloc_local to also support migration. (using...
Gengbin Zheng [Fri, 13 Jan 2012 05:28:42 +0000 (23:28 -0600)]
change the implementation of ARMCI_Malloc_local to also support migration. (using isomalloc)

9 years agoAllocating a new lock whenever a pool moves
Nikhil Jain [Fri, 13 Jan 2012 03:58:08 +0000 (21:58 -0600)]
Allocating a new lock whenever a pool moves

9 years agoJim's poolfix patch plus some indent fixes. Passes make test.
Eric Bohm [Thu, 12 Jan 2012 23:17:51 +0000 (17:17 -0600)]
Jim's poolfix patch plus some indent fixes.  Passes make test.