charm.git
13 years agomake portable to CkDirect implementations other than ibverbs
Eric Bohm [Mon, 7 Apr 2008 07:02:26 +0000 (07:02 +0000)]
make portable to CkDirect implementations other than ibverbs

13 years agoibverbs layer is cranky about field order in the userhandle.
Eric Bohm [Mon, 7 Apr 2008 06:39:27 +0000 (06:39 +0000)]
ibverbs layer is cranky about field order in the userhandle.
this fixes bug

13 years agousing sgemm now
Abhinav Bhatele [Sun, 6 Apr 2008 06:14:05 +0000 (06:14 +0000)]
using sgemm now

13 years agoremoved inlining from lookup functions because it breaks with -O3
Aaron Becker [Fri, 4 Apr 2008 23:27:26 +0000 (23:27 +0000)]
removed inlining from lookup functions because it breaks with -O3

13 years agoLatest fixes to edge_bisect. Serial looks to be working fine now.
Terry L. Wilmarth [Fri, 4 Apr 2008 23:09:14 +0000 (23:09 +0000)]
Latest fixes to edge_bisect.  Serial looks to be working fine now.

13 years agoimprove robustness in cmidirect implementation
Eric Bohm [Fri, 4 Apr 2008 05:38:56 +0000 (05:38 +0000)]
improve robustness in cmidirect implementation

13 years agofixed silly bug in copyEdgeAdjacency
Aaron Becker [Fri, 4 Apr 2008 02:18:29 +0000 (02:18 +0000)]
fixed silly bug in copyEdgeAdjacency

13 years ago*** empty log message ***
Terry L. Wilmarth [Thu, 3 Apr 2008 22:37:57 +0000 (22:37 +0000)]
*** empty log message ***

13 years agocomment
Gengbin Zheng [Thu, 3 Apr 2008 22:27:09 +0000 (22:27 +0000)]
comment

13 years ago*** empty log message ***
Terry L. Wilmarth [Thu, 3 Apr 2008 18:54:23 +0000 (18:54 +0000)]
*** empty log message ***

13 years agocorrected array access bug, count is now a double instead of an int (haven't thought...
Pritish Jetley [Thu, 3 Apr 2008 17:32:28 +0000 (17:32 +0000)]
corrected array access bug, count is now a double instead of an int (haven't thought about machine precision
issues yet). per iteration time is printed now.

13 years agotwo new cknodereduction class variables need to be pup'ed so that checkpoint/restart...
Gengbin Zheng [Thu, 3 Apr 2008 07:55:35 +0000 (07:55 +0000)]
two new cknodereduction class variables need to be pup'ed so that checkpoint/restart will work.

13 years agoadded a node barrier to avoid a race condition of overwriting readonlies.
Gengbin Zheng [Thu, 3 Apr 2008 07:25:27 +0000 (07:25 +0000)]
added a node barrier to avoid a race condition of overwriting readonlies.

13 years agosubtle bug fixed
Abhinav Bhatele [Thu, 3 Apr 2008 01:12:02 +0000 (01:12 +0000)]
subtle bug fixed

13 years agobg/p ckdirect update
Eric Bohm [Thu, 3 Apr 2008 00:30:17 +0000 (00:30 +0000)]
bg/p ckdirect update

13 years agoFixed a memory bug...
Terry L. Wilmarth [Wed, 2 Apr 2008 22:38:01 +0000 (22:38 +0000)]
Fixed a memory bug...

13 years agoAdds CMK_DIRECT as a conv-mach symbol which can be tested for CmiDirect support.
Eric Bohm [Wed, 2 Apr 2008 20:29:59 +0000 (20:29 +0000)]
Adds CMK_DIRECT as a conv-mach symbol which can be tested for CmiDirect support.

13 years agoImplements CmiDirect interface on Blue Gene/P.
Eric Bohm [Wed, 2 Apr 2008 20:28:33 +0000 (20:28 +0000)]
Implements CmiDirect interface on Blue Gene/P.
Adds CMK_DIRECT as a conv-mach symbol which can be tested for CmiDirect support.

13 years agoRemoved mallocs and frees in cmidirect, while calling cmidirect_ready
Sayantan Chakravorty [Wed, 2 Apr 2008 18:53:30 +0000 (18:53 +0000)]
Removed mallocs and frees in cmidirect, while calling cmidirect_ready
and after a message has been received

13 years ago*** empty log message ***
Terry L. Wilmarth [Wed, 2 Apr 2008 16:43:56 +0000 (16:43 +0000)]
*** empty log message ***

13 years ago*** empty log message ***
Abhinav Bhatele [Wed, 2 Apr 2008 03:40:16 +0000 (03:40 +0000)]
*** empty log message ***

13 years agoeverything fixed hopefully
Abhinav Bhatele [Wed, 2 Apr 2008 03:39:51 +0000 (03:39 +0000)]
everything fixed hopefully

13 years ago*** empty log message ***
Abhinav Bhatele [Wed, 2 Apr 2008 02:06:45 +0000 (02:06 +0000)]
*** empty log message ***

13 years agobug fixed
Abhinav Bhatele [Wed, 2 Apr 2008 00:13:26 +0000 (00:13 +0000)]
bug fixed

13 years agofirst version
Abhinav Bhatele [Tue, 1 Apr 2008 23:54:45 +0000 (23:54 +0000)]
first version

13 years agoEdge bisection code. Works on one serial element, but locking breaks on the second
Terry L. Wilmarth [Tue, 1 Apr 2008 21:34:25 +0000 (21:34 +0000)]
Edge bisection code.  Works on one serial element, but locking breaks on the second
bisection.

13 years agofirst version after move
Abhinav Bhatele [Tue, 1 Apr 2008 19:33:15 +0000 (19:33 +0000)]
first version after move

13 years agocode moved to examples
Abhinav Bhatele [Tue, 1 Apr 2008 19:31:49 +0000 (19:31 +0000)]
code moved to examples

13 years agofixed declaration for replaceadaptadjonedge
Aaron Becker [Tue, 1 Apr 2008 17:55:51 +0000 (17:55 +0000)]
fixed declaration for replaceadaptadjonedge

13 years agohave an ARR_CHECK version
Pritish Jetley [Tue, 1 Apr 2008 02:40:28 +0000 (02:40 +0000)]
have an ARR_CHECK version

13 years agofixed bug
Pritish Jetley [Tue, 1 Apr 2008 01:31:01 +0000 (01:31 +0000)]
fixed bug

13 years agofixed compilation problem when compiling charm with CMK_OPTIMIZE and the
Filippo Gioachin [Mon, 31 Mar 2008 23:52:57 +0000 (23:52 +0000)]
fixed compilation problem when compiling charm with CMK_OPTIMIZE and the
user program without CMK_OPTIMIZE

13 years agoicc on ranger's opterons hates -xT using -xO instead
Eric Bohm [Mon, 31 Mar 2008 22:33:48 +0000 (22:33 +0000)]
icc on ranger's opterons hates -xT using -xO instead

13 years agosilly but annoying bug fixes
Abhinav Bhatele [Mon, 31 Mar 2008 21:08:59 +0000 (21:08 +0000)]
silly but annoying bug fixes

13 years agoAdded functions to add to and clear edge adjacencies
Aaron Becker [Mon, 31 Mar 2008 21:08:03 +0000 (21:08 +0000)]
Added functions to add to and clear edge adjacencies

13 years agoPoll cmidirect q for 1 processor runs as well
Sayantan Chakravorty [Mon, 31 Mar 2008 20:31:54 +0000 (20:31 +0000)]
Poll cmidirect q for 1 processor runs as well

13 years agoremote messages removed
Abhinav Bhatele [Mon, 31 Mar 2008 19:10:06 +0000 (19:10 +0000)]
remote messages removed

13 years agoMoved bulk adaptivity attr handling code from FEM_add_element_local to BulkAdapt...
Aaron Becker [Mon, 31 Mar 2008 17:53:20 +0000 (17:53 +0000)]
Moved bulk adaptivity attr handling code from FEM_add_element_local to BulkAdapt::add_element

13 years agoNeeded newline at EOF.
Terry L. Wilmarth [Mon, 31 Mar 2008 13:58:52 +0000 (13:58 +0000)]
Needed newline at EOF.

13 years agobug fixes
Abhinav Bhatele [Sun, 30 Mar 2008 07:48:09 +0000 (07:48 +0000)]
bug fixes

13 years agofixed a reduction bug
Abhinav Bhatele [Sun, 30 Mar 2008 07:10:16 +0000 (07:10 +0000)]
fixed a reduction bug

13 years agoopen sts file only at the traceClose time instead of fopen at the beginning of the...
Gengbin Zheng [Sun, 30 Mar 2008 05:39:19 +0000 (05:39 +0000)]
open sts file only at the traceClose time instead of fopen at the beginning of the execution.

13 years agolook for bugs
Abhinav Bhatele [Sun, 30 Mar 2008 05:23:37 +0000 (05:23 +0000)]
look for bugs

13 years agofirst version of CkDirect matmul3d
Abhinav Bhatele [Sat, 29 Mar 2008 23:05:15 +0000 (23:05 +0000)]
first version of CkDirect matmul3d

13 years ago1d instead of 3d chare array
Pritish Jetley [Sat, 29 Mar 2008 21:24:26 +0000 (21:24 +0000)]
1d instead of 3d chare array

13 years agoMakefile with 2 targets, ckdirect and msg based
Sayantan Chakravorty [Sat, 29 Mar 2008 18:56:51 +0000 (18:56 +0000)]
Makefile with 2 targets, ckdirect and msg based

13 years agore-enabled eltsComp
Pritish Jetley [Sat, 29 Mar 2008 09:06:33 +0000 (09:06 +0000)]
re-enabled eltsComp

13 years agouse os malloc as default now
Gengbin Zheng [Fri, 28 Mar 2008 21:11:11 +0000 (21:11 +0000)]
use os malloc as default now

13 years agocode cleanup
Aaron Becker [Fri, 28 Mar 2008 20:18:21 +0000 (20:18 +0000)]
code cleanup

13 years agoremove c++ style comments in c.
Gengbin Zheng [Fri, 28 Mar 2008 20:02:47 +0000 (20:02 +0000)]
remove c++ style comments in c.

13 years agounnecessary copy removed and use of local pointers for chares on the same
Abhinav Bhatele [Fri, 28 Mar 2008 19:19:24 +0000 (19:19 +0000)]
unnecessary copy removed and use of local pointers for chares on the same
processor

13 years agoinclude conv-config.h for the int8 type definition.
Gengbin Zheng [Fri, 28 Mar 2008 18:48:59 +0000 (18:48 +0000)]
include conv-config.h for the int8 type definition.

13 years agodefine a new pup_int8.
Gengbin Zheng [Fri, 28 Mar 2008 17:48:02 +0000 (17:48 +0000)]
define a new pup_int8.

13 years agouse a new puper to pup int8 type.
Gengbin Zheng [Fri, 28 Mar 2008 17:47:22 +0000 (17:47 +0000)]
use a new puper to pup int8 type.

13 years agoa complete change to promot int to long to overcome the intmax limit.
Gengbin Zheng [Fri, 28 Mar 2008 16:23:41 +0000 (16:23 +0000)]
a complete change to promot int to long to overcome the intmax limit.

13 years agono eltsComp
Pritish Jetley [Fri, 28 Mar 2008 15:19:27 +0000 (15:19 +0000)]
no eltsComp

13 years agochanged the return type of CmiIsomallocLength to long.
Gengbin Zheng [Fri, 28 Mar 2008 14:08:00 +0000 (14:08 +0000)]
changed the return type of CmiIsomallocLength to long.
Something forgot to check in yesterday.

13 years agono redundant computation
Pritish Jetley [Fri, 28 Mar 2008 08:39:52 +0000 (08:39 +0000)]
no redundant computation

13 years agofixed bug. all elements being touched now.
Pritish Jetley [Fri, 28 Mar 2008 07:59:13 +0000 (07:59 +0000)]
fixed bug. all elements being touched now.
next item - remove redundant computations and replace with copies.

13 years ago*** empty log message ***
Pritish Jetley [Fri, 28 Mar 2008 05:50:35 +0000 (05:50 +0000)]
*** empty log message ***

13 years ago*** empty log message ***
Pritish Jetley [Fri, 28 Mar 2008 05:49:06 +0000 (05:49 +0000)]
*** empty log message ***

13 years agoadded eltsComp
Pritish Jetley [Fri, 28 Mar 2008 05:46:31 +0000 (05:46 +0000)]
added eltsComp

13 years agoiterations++ was in the wrong place
Pritish Jetley [Fri, 28 Mar 2008 05:28:37 +0000 (05:28 +0000)]
iterations++ was in the wrong place

13 years agoremove the constraint that the number of slots on each processor can not exceed the...
Gengbin Zheng [Fri, 28 Mar 2008 02:52:06 +0000 (02:52 +0000)]
remove the constraint that the number of slots on each processor can not exceed the intMax. This is done by prompting relevant int type to long.

13 years ago*** empty log message ***
Pritish Jetley [Thu, 27 Mar 2008 23:28:32 +0000 (23:28 +0000)]
*** empty log message ***

13 years agoAdded initialization and cleanup for bulk adjacency attributes in add/remove
Aaron Becker [Thu, 27 Mar 2008 23:04:17 +0000 (23:04 +0000)]
Added initialization and cleanup for bulk adjacency attributes in add/remove
element

13 years ago*** empty log message ***
Pritish Jetley [Thu, 27 Mar 2008 22:22:54 +0000 (22:22 +0000)]
*** empty log message ***

13 years agofixed a bug that when calling an operation on a remote processor (grab or free),...
Gengbin Zheng [Thu, 27 Mar 2008 22:00:08 +0000 (22:00 +0000)]
fixed a bug that when calling an operation on a remote processor (grab or free), it passed wrong nslots parameter.

13 years agofixed Filippo's check in. Get rid of preloading of conv-config.sh, which changed...
Gengbin Zheng [Thu, 27 Mar 2008 16:31:05 +0000 (16:31 +0000)]
fixed Filippo's check in. Get rid of preloading of conv-config.sh, which changed the value of $@ on some platform for example IBM SP.

13 years agofixed previous checkin errors
Filippo Gioachin [Thu, 27 Mar 2008 00:46:28 +0000 (00:46 +0000)]
fixed previous checkin errors

13 years agoInclude cmidirect.h and ckdirect.h in include directory of build
Sayantan Chakravorty [Wed, 26 Mar 2008 23:20:12 +0000 (23:20 +0000)]
Include cmidirect.h and ckdirect.h in include directory of build

13 years agoCmiDirect/CkDirect changes
Pritish Jetley [Wed, 26 Mar 2008 22:44:11 +0000 (22:44 +0000)]
CmiDirect/CkDirect changes

13 years ago*** empty log message ***
Pritish Jetley [Wed, 26 Mar 2008 22:25:57 +0000 (22:25 +0000)]
*** empty log message ***

13 years ago3d versions, with and without CkDirect
Pritish Jetley [Wed, 26 Mar 2008 22:23:00 +0000 (22:23 +0000)]
3d versions, with and without CkDirect

13 years agoforgot check for machines without python
Filippo Gioachin [Wed, 26 Mar 2008 21:15:01 +0000 (21:15 +0000)]
forgot check for machines without python

13 years ago*** empty log message ***
Filippo Gioachin [Wed, 26 Mar 2008 18:31:26 +0000 (18:31 +0000)]
*** empty log message ***

13 years ago3d stencil computation; no copying of send/recv buffers; quite a pain in the rear...
Pritish Jetley [Wed, 26 Mar 2008 07:08:58 +0000 (07:08 +0000)]
3d stencil computation; no copying of send/recv buffers; quite a pain in the rear to write,
even more so to understand.

13 years agoAdded flag "-debug" to charmc for usage with charmdebug.
Filippo Gioachin [Wed, 26 Mar 2008 03:33:19 +0000 (03:33 +0000)]
Added flag "-debug" to charmc for usage with charmdebug.
Some work done on CmiReduce to make it possible to have concurrent reductions.
Added memory tagging by chare ID in charmdebug.
Fixed breakpointing with charmdebug: now each entry point is defined by its sequential index in the EP table (before a string match was performed).
Added stepping: delivery of a single message in the queue. Also, all the system entry methods are currently always executed, and never get buffered by charmdebug (they can still breakpoint if the user so chooses).
Fixed race condition in python interface. To do that the interface itself has been modified, in particular with regard to high-level python. Python is now more flexible and it allows any chare collection to be "python" and receive requests. The replies are reduced to a PythonCCS mainchare which will handle the socket.
The python directory has now another module (charmdebug-python) for usage by charmdebug introspection code.

13 years agoAdded flag "-debug" to charmc for usage with charmdebug.
Filippo Gioachin [Wed, 26 Mar 2008 03:21:29 +0000 (03:21 +0000)]
Added flag "-debug" to charmc for usage with charmdebug.
Some work done on CmiReduce to make it possible to have concurrent reductions.
Added memory tagging by chare ID in charmdebug.
Fixed breakpointing with charmdebug: now each entry point is defined by its sequential index in the EP table (before a string match was performed).
Added stepping: delivery of a single message in the queue. Also, all the system entry methods are currently always executed, and never get buffered by charmdebug (they can still breakpoint if the user so chooses).
Fixed race condition in python interface. To do that the interface itself has been modified, in particular with regard to high-level python. Python is now more flexible and it allows any chare collection to be "python" and receive requests. The replies are reduced to a PythonCCS mainchare which will handle the socket.
The python directory has now another module (charmdebug-python) for usage by charmdebug introspection code.

13 years agoAdded copy adaptAdj functionality. Also if my previous commit didn't work,
Aaron Becker [Tue, 25 Mar 2008 23:16:53 +0000 (23:16 +0000)]
Added copy adaptAdj functionality. Also if my previous commit didn't work,
fixed edge adjacency creation bug

13 years agoclean up
Pritish Jetley [Mon, 24 Mar 2008 04:35:28 +0000 (04:35 +0000)]
clean up

13 years agofixed bugs
Pritish Jetley [Mon, 24 Mar 2008 04:34:26 +0000 (04:34 +0000)]
fixed bugs

13 years agobug fixes ... answers are now correct
Abhinav Bhatele [Mon, 24 Mar 2008 02:25:08 +0000 (02:25 +0000)]
bug fixes ... answers are now correct

13 years agoIncreased timeout value for ibverbs
Sayantan Chakravorty [Sun, 23 Mar 2008 21:36:24 +0000 (21:36 +0000)]
Increased timeout value for ibverbs

13 years agoChanged timeout value
Sayantan Chakravorty [Sun, 23 Mar 2008 21:14:40 +0000 (21:14 +0000)]
Changed timeout value

13 years agoanother bug :)
Abhinav Bhatele [Sun, 23 Mar 2008 07:56:07 +0000 (07:56 +0000)]
another bug :)

13 years agoa bug fixed
Abhinav Bhatele [Sun, 23 Mar 2008 07:50:30 +0000 (07:50 +0000)]
a bug fixed

13 years agocomm of C fixed
Abhinav Bhatele [Sun, 23 Mar 2008 07:07:36 +0000 (07:07 +0000)]
comm of C fixed

13 years agomemset
Pritish Jetley [Sun, 23 Mar 2008 06:29:21 +0000 (06:29 +0000)]
memset

13 years agocannon's algorithms
Pritish Jetley [Sun, 23 Mar 2008 06:07:08 +0000 (06:07 +0000)]
cannon's algorithms

13 years agoSMP machine.c with optimized spanning tree broadcast.
Sameer Kumar [Fri, 21 Mar 2008 14:35:26 +0000 (14:35 +0000)]
SMP machine.c with optimized spanning tree broadcast.

13 years agoremove pcqueue lock for AMD64.
Gengbin Zheng [Wed, 19 Mar 2008 22:40:47 +0000 (22:40 +0000)]
remove pcqueue lock for AMD64.
This is because AMD64 offers a slightly stronger memory-consistency model, in that it does not reorder a store ahead of a load. Probably we can do away without using lock and fence.
But THIS NEEDS MORE TESTING.

13 years agotopology code checked in
Abhinav Bhatele [Wed, 19 Mar 2008 19:25:10 +0000 (19:25 +0000)]
topology code checked in

13 years agoskip netfem for multicore version.
Gengbin Zheng [Wed, 19 Mar 2008 18:00:49 +0000 (18:00 +0000)]
skip netfem for multicore version.

13 years agofor multicore/smp, copy argc and argv after some of them which are specific to machin...
Gengbin Zheng [Wed, 19 Mar 2008 14:07:29 +0000 (14:07 +0000)]
for multicore/smp, copy argc and argv after some of them which are specific to machine.c are consumed.
This change is TENTATIVE, should be tested on ALL charm versions!

13 years agosecond version
Abhinav Bhatele [Wed, 19 Mar 2008 00:28:55 +0000 (00:28 +0000)]
second version

13 years agoModified adjacency dumper to put a | separator between the lists of
Terry L. Wilmarth [Tue, 18 Mar 2008 20:19:13 +0000 (20:19 +0000)]
Modified adjacency dumper to put a | separator between the lists of
elements that are adjacent to each edge for better readability.

13 years agoremoved static keyword from user handle definition
Pritish Jetley [Tue, 18 Mar 2008 20:15:27 +0000 (20:15 +0000)]
removed static keyword from user handle definition

13 years agolookupEdgeAdaptAdjacencies was getting the int table instead of the char
Terry L. Wilmarth [Tue, 18 Mar 2008 20:01:52 +0000 (20:01 +0000)]
lookupEdgeAdaptAdjacencies was getting the int table instead of the char
table where the data seems to have been stored.

13 years agoThere is some flaw in an assert in adapt_adj's dump adjacencies function
Terry L. Wilmarth [Tue, 18 Mar 2008 19:52:54 +0000 (19:52 +0000)]
There is some flaw in an assert in adapt_adj's dump adjacencies function
which I have remedied.