paboyle
5c0adf7bf2
Make clang happy with parenthesis
2017-02-16 23:51:33 +00:00
paboyle
bd600702cf
Vectorise the XYZT face gathering better.
...
Hard coded for simd_layout <= 2 in any given spread out direction; full generality is inconsistent
with efficiency.
2017-02-15 11:11:04 +00:00
paboyle
a48ee6f0f2
Don't use MPI3_leader any more. No real gain and complex
2017-02-07 01:31:24 -05:00
paboyle
73547cca66
MPI3 working i think
2017-02-07 01:30:02 -05:00
paboyle
123c673db7
Policy to control async or sync SendRecv
2017-02-07 01:24:54 -05:00
paboyle
61f82216e2
Communicator Policy, NodeCount distinct from Rank count
2017-02-07 01:22:53 -05:00
azusayamaguchi
f85b35314d
Fix a routine for single node processor coor from rank
2016-11-08 11:49:13 +00:00
azusayamaguchi
6e548a8ad5
Linux compile needed
2016-11-04 11:34:16 +00:00
paboyle
f41a230b32
Decrease mpi3l verbose
2016-11-02 19:54:03 +00:00
paboyle
757a928f9a
Improvement to use own SHM_OPEN call to avoid openmpi bug.
2016-11-02 12:37:46 +00:00
paboyle
32375aca65
Semaphore sleep/wake up on remote processes.
2016-11-02 09:27:20 +00:00
paboyle
bb94ddd0eb
Tidy up of mpi3; also some cleaning of the dslash controls.
2016-11-02 08:07:09 +00:00
paboyle
791cb050c8
Comms improvements
2016-11-01 11:35:43 +00:00
paboyle
09f66100d3
MPI 3 compile on non-linux
2016-10-25 06:01:12 +01:00
azusayamaguchi
d7d92af09d
Travis fail fix attempt
2016-10-25 01:45:53 +01:00
azusayamaguchi
d97a27f483
Verbose
2016-10-25 01:05:31 +01:00
azusayamaguchi
7c3363b91e
Compiles all comms targets
2016-10-25 00:04:17 +01:00
azusayamaguchi
b94478fa51
mpi, mpi3, shmem all compile.
...
mpi, mpi3 pass single node multi-rank
2016-10-24 23:45:31 +01:00
azusayamaguchi
b6a65059a2
Update to use shared memory to contain the stencil comms buffers
...
Tested on 2.1.1.1 1.2.1.1 4.1.1.1 1.4.1.1 2.2.1.1 subnode decompositions
2016-10-24 17:30:43 +01:00
azusayamaguchi
c190221fd3
Internal SHM comms in non-simd directions working
...
Need to fix simd directions
2016-10-22 18:14:27 +01:00
azusayamaguchi
910b8dd6a1
use simd type
2016-10-21 22:35:29 +01:00
azusayamaguchi
09fd5c43a7
Reasonably fast version
2016-10-21 15:17:39 +01:00
azusayamaguchi
fad96cf250
StencilBufs
2016-10-21 13:36:00 +01:00
azusayamaguchi
f331809c27
Use variable type for loop
2016-10-21 13:35:37 +01:00
paboyle
306160ad9a
bcopy threaded
2016-10-21 12:07:28 +01:00
paboyle
a762b1fb71
MPI3 working with a bounce through shared memory on my laptop.
...
Longer term plan: make the "u_comm_buf" in Stencil point to the shared region and avoid the
send between ranks on same node.
2016-10-21 09:03:26 +01:00
paboyle
b58adc6a4b
commVector
2016-10-20 17:00:15 +01:00
paboyle
5fe2b85cbd
MPI3 and shared memory support
2016-10-20 16:58:01 +01:00
paboyle
32bc7a6ab8
MPI back out of change that hangs
...
AVX2 for clang, gcc needs the -mfma flag.
2016-08-05 10:36:00 +01:00
paboyle
62601bb649
Bug fix
2016-07-08 20:46:29 +01:00
paboyle
ef97e32152
Adding persistent communicators
2016-07-08 17:16:08 +01:00
paboyle
680645f849
Merge branch 'release/v0.5.0'
2016-06-30 15:15:03 -07:00
Guido Cossu
5e02392f9c
Fixed compilation error for benchmark_dwf
...
Some parts were assuming floating point precision
2016-06-20 12:30:51 +01:00
Richard Rollins
86187d7cca
Removed write to stdout in constructor for MPI CartesianCommunicator
2016-06-14 15:34:20 +01:00
paboyle
d6b64f47d9
Uint64 sum for IO rates
2016-03-16 02:27:22 -07:00
paboyle
a359f7a9f5
Merge branch 'master' of https://github.com/paboyle/Grid
2016-03-11 16:07:07 -08:00
paboyle
b606deb3f0
Uint64 gsum
2016-03-11 16:06:54 -08:00
paboyle
090e7aa930
Merge remote-tracking branch 'origin/chulwoo-dec12-2015'
...
Merge Chulwoo's Lanczos related improvements.
Merge Nd!=4 fixes for pure gauge HMC from Evan.
2016-03-08 09:55:14 +00:00
paboyle
e55c35734b
Fix a nocompile
2016-03-03 20:33:28 +00:00
Peter Boyle
6aeaf6f568
Parallel IO worked on. I'm puzzled because I already thought I shook this out on MacOS + OpenMPI and then
...
turned up problems on the BlueWaters Cray.
Gets 75MB/s from home filesystem on parallel configuration read. Need to make the RNG IO parallel,
and also to look at aggregating bigger writes for the parallel write.
Not sure what the home filesystem is.
2016-02-21 08:03:21 -06:00
paboyle
a3fbabf404
Bug fix
2016-02-18 18:08:24 +00:00
Peter Boyle
41c2b09184
Shmem comms [NO MPI] target added. The dwf test runs and passes.
...
Not really shaken out to my satisfaction though as I want more tests done, so don't declare as working.
But committing my current while I try a few experimentals.
2016-02-14 14:24:38 -06:00
paboyle
294dbf1bf0
Compile on OpenMPI shmem
2016-02-11 23:45:51 +00:00
Peter Boyle
7f927a541c
Shmem related fixes for shmem compile
2016-02-11 07:37:39 -06:00
paboyle
e2f73e3ead
Updates for shmem
2016-02-10 16:50:32 -08:00
Jung
5c57d4f403
Merge branch 'master' of https://github.com/paboyle/Grid into scidac1_2
...
Conflicts:
lib/qcd/action/fermion/WilsonKernels.h
2016-01-11 11:36:45 -05:00
Jung
5924e5a562
Merge branch 'master' of https://github.com/paboyle/Grid into scidac1_2
...
Conflicts:
configure
lib/qcd/action/Actions.h
lib/qcd/action/fermion/WilsonKernels.h
2016-01-06 03:44:57 -05:00
paboyle
aae8bf31a7
Global edit adding copyright and license info to every source file.
2016-01-02 14:51:32 +00:00
Peter Boyle
dc814f30da
Binary IO file for generic Grid array parallel I/O.
...
Number of IO MPI tasks can be varied by selecting which
dimensions use parallel IO and which dimensions use Serial send to boss
I/O.
Thus can neck down from, say 1024 nodes = 4x4x8x8 to {1,8,32,64,128,256,1024} nodes
doing the I/O.
Interpolates nicely between ALL nodes write their data, a single boss per time-plane
in processor space [old UKQCD fortran code did this], and a single node doing all I/O.
Not sure I have the transfer sizes big enough and am not overly convinced fstream
is guaranteed to not give buffer inconsistencies unless I set streambuf size to zero.
Practically it has worked on 8 tasks, 2x1x2x2 writing /cloning NERSC configurations
on my MacOS + OpenMPI and Clang environment.
It is VERY easy to switch to pwrite at a later date, and also easy to send x-strips around from
each node in order to gather bigger chunks at the syscall level.
That would push us up to the circa 8x 18*4*8 == 4KB size write chunk, and by taking, say, x/y non
parallel we get to 16MB contiguous chunks written in multi 4KB transactions
per IOnode in 64^3 lattices for configuration I/O.
I suspect this is fine for system performance.
2015-08-26 13:40:29 +01:00
neo
48bf4878c1
Experimental support for ARM
2015-06-09 15:46:21 +09:00