portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-15 02:05:37 +00:00
Code Releases Activity
3,650 Commits 86 Branches 14 Tags 74 MiB
45df59720e
Commit Graph

116 Commits

Author SHA1 Message Date
Peter Boyle
fb8d4b2357 Lots of debug on performance Mobius 2016-12-08 17:28:28 +00:00
Guido Cossu
143c70e29f Debugged the threaded version. Cleaning up 2016-12-07 04:40:25 +00:00
Guido Cossu
b812d5e39c Added single threaded version of the derivative for the Ls vectorised DWF 2016-12-06 16:31:13 +00:00
Guido Cossu
ae9688e343 Reporting also the total mflops 2016-11-28 11:37:02 +00:00
Antonin Portelli
ca21003f01 Merge branch 'feature/fft-opt' into feature/feynman-rules 
# Conflicts:
#	lib/FFT.h
#	lib/qcd/action/fermion/WilsonFermion5D.h
#	tests/core/Test_fft.cc
 2016-10-26 18:44:47 +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
6a9eae6b6b Reporting improvements 2016-10-21 13:36:18 +01:00
Antonin Portelli
997fd882ff Merge branch 'develop' into feature/feynman-rules 
# Conflicts:
#	lib/Threads.h
#	lib/qcd/action/fermion/WilsonFermion.cc
#	lib/qcd/action/fermion/WilsonFermion.h
#	lib/qcd/utils/SUn.h
#	lib/simd/Grid_avx.h
#	lib/simd/Intel512common.h
 2016-10-19 18:35:18 +01:00
azusayamaguchi
81f2aeaece KNL streaming stores, and KNL performance coutners 2016-10-12 11:45:22 +01:00
paboyle
96f1d1b828 Debugged Domain wall and Overlap feynman rules (infinite Ls, finite mass). 2016-10-10 23:46:45 +01:00
Guido Cossu
2e453dfbf5 Added some instrumentation to benchmark the force computation 2016-10-06 17:52:45 +01:00
paboyle
4089984431 Timing hooks 2016-10-06 09:25:12 +01:00
paboyle
b6713ecb60 Momentum space rules for Overlap, DWF untested to date 2016-09-26 09:39:09 +01:00
paboyle
48fb1cdc11 Update domain 5d vectorised impl type, move the type over to 4d redblack with 
the dense OO inverse
 2016-07-14 23:54:35 +01:00
paboyle
6d58cb2a68 Enable reordering of the loops in the assembler for cache friendly. 
This gets in the way of L2 prefetching however. Do next next link in stencil
prefetching.
 2016-06-30 14:35:01 -07:00
paboyle
55f65b81b5 Improvements to the assembler interface that let us move chunks of the 
site and s loop into the kernels. This will save on function call overhead and
guarantee L2 prefetching strategy is right since OMP can't distribute the
sub-chunks of work.
 2016-06-09 01:12:36 -07:00
paboyle
8ac021de73 Added a test an fixed it for red black precon Ls innermost vectorised DWF 2016-06-07 13:16:56 -07:00
paboyle
53d06046b0 Compiling updates for KNL 2016-06-03 03:47:54 -07:00
paboyle
139cc5f1ae Large change with KNL preparation 2016-06-03 03:24:26 -07:00
paboyle
1e554350ac The threaded coms didn't agree with GCC. Suprised, and looks like GCC bug. 2016-04-29 16:49:18 -07:00
paboyle
9b6ab6db16 simd in 5th dimension support 2016-04-19 15:38:01 -07:00
paboyle
e8dddb1596 Adding extra benchmark 2016-04-06 10:32:54 +01:00
paboyle
e67fc2be18 Adding a trial for openmp overhead minimisation 2016-03-31 16:00:37 +01:00
paboyle
165bffc2e7 Avx512 changes for assembler kernels 2016-03-26 22:25:45 -06:00
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
61413565d0 Back off the inlined spin proj as not working 2016-03-02 07:03:09 -08:00
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
Peter Boyle
22422a84d9 Small problem in compressor fix 2016-02-17 19:03:09 -06:00
Peter Boyle
c9fadf97a5 Simplify the compressor interface again. 2016-02-17 18:16:45 -06:00
Peter Boyle
81395e85d1 Regressing to not overlap comms and compute becasue bluewaters, edison, and cori are so rubbish at it. 2016-02-16 13:56:44 -06:00
Peter Boyle
a0fc47c6f9 Cheaper implementation 2016-02-15 16:02:36 -06:00
paboyle
e2f73e3ead Updates for shmem 2016-02-10 16:50:32 -08:00
Jung
411ac49dd7 GparityWilsonTM typedef added. Not yet tested 
Conflicts:
	configure
	lib/qcd/action/fermion/WilsonKernels.h
 2016-01-25 01:36:28 -05:00
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
paboyle
fc6ad65751 Pushed the overlap comms tweaks 2016-01-11 06:34:22 -08:00
paboyle
dafc74020c Overlap comms compute improvements in hand op kernels, and better timing from Edison and Cori 2016-01-10 16:54:27 -08:00
paboyle
d19321dfde Overlap comms compute changes 2016-01-10 19:20:16 +00:00
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
c99d748da6 Timing reports in benchmarks now reflect the asynch comms thread statistics 2016-01-04 14:42:16 +00:00
paboyle
02452afd36 Optional overlap of comms with compute 2016-01-04 14:18:40 +00:00
paboyle
331768dcff Added overlap comms compute mode 2016-01-03 01:38:11 +00:00
paboyle
aae8bf31a7 Global edit adding copyright and license info to every source file. 2016-01-02 14:51:32 +00:00
paboyle
34a0fde2ad Fixes to fermion force terms after sign of gamma_mu (0...3) change. 
Thought I had already committed these.

Believe I have got the Gparity fermion force working.

* tests/Test_gpdwf_force.cc     -- correctly predicts dS for two flavour pseudofermion
                                   based on a small dt update of U field.

* tests/Test_hmc_EODWFRatio_Gparity.cc -- ran 1 trajectory on 8^4 with dH=0.21.

Need to accumulate a full plaquette log to believe fully which will take some hours of run time.
 2015-12-15 23:14:12 +00:00
Jung
f2b4edc090 Fixes for Gparity comparison with CPS (Instantiation, Gamma matrix convention) 2015-12-07 02:04:57 -05:00
paboyle
b2c02a6106 Runs fastst on cori 2015-11-28 16:58:16 -08:00
paboyle
e9ff25b06b Small threading change makes a difference on Cori. 2015-11-07 00:07:05 -08:00
paboyle
899ca41cb8 Merge branch 'master' of github.com:paboyle/Grid 
Conflicts:
	lib/qcd/action/fermion/WilsonFermion5D.cc
 2015-11-06 03:50:04 -08:00
paboyle
a2ff068e29 Asm and threading for many core 2015-11-06 03:47:14 -08:00
Peter Boyle
28022755ae Stencil class name global change to StencilImpl typedef 2015-11-06 05:30:17 -06:00
Peter Boyle
55cfc89459 * Finished the template/policy style introduction of gparity, except the gparity force terms. 
So valence sector looks ok.

FermionOperatorImpl.h provides the policy classes.

Expect HMC will introduce a smearing policy and a fermion representation change policy template
param. Will also probably need multi-precision work.

* HMC is running even-odd and non-checkerboarded (checked 4^4 wilson fermion/wilson gauge).

There appears to be a bug in the multi-level integrator -- <e-dH> passes with single level but
not with multi-level.

In any case there looks to be quite a bit to clean up.

This is the "const det" style implementation that is not appropriate  yet for clover since
it assumes that Mee is indept of the gauge fields. Easily fixed in future.
 2015-08-15 23:25:49 +01:00
Peter Boyle
7d3512ab21 Gparity valence test now working. 
Interface in FermionOperator will change a lot in future
 2015-08-14 00:01:04 +01:00
Peter Boyle
84a66476ab Rework/global edit to enforce type templating of fermion operators. 
Allows multi-precision work and paves the way for alternate BC's and such like
allowing for example G-parity which is important for K pipi programme.
In particular, can drive an extra flavour index into the fermion fields
using template types.
 2015-08-10 20:47:44 +01:00
Peter Boyle
2157a6919a Changes making force term test for DWF pass. 2015-08-01 22:06:07 +09:00
Peter Boyle
1d67d29183 Jackson smoothed chebyshev and (untested) completion of force terms 
for Cayley, Partial and Cont fraction dwf and overlap.
have even odd and unprec forces.
 2015-08-01 05:58:35 +09:00
Peter Boyle
4fe110bd07 Two flavour HMC for Wilson/Wilson is conserving energy. 
Still to check plaq and <e(-dH)>, but nevertheless this is
progress
 2015-07-29 17:53:39 +09:00
Peter Boyle
4cc2ef84d3 Committing incomplete work for parameter file I/O. 
MacroMagic.h is central. Guido and I plan to move
over to generating virtual (XML, JSON, YAML, text, binary) encoding
from macro based system.
 2015-07-27 18:32:28 +09:00
Peter Boyle
d9d4c5916a Elemental force term for Wilson dslash added and tests thereof passing. 
Now need to construct pseudofermion two flavour, ratio, one flavour, ratio
action fragments.
 2015-07-26 10:54:38 +09:00
Peter Boyle
d1afebf71e Sizable improvement in multigrid for unsquared. 
6000 matmuls CG unprec
2000 matmuls CG prec (4000 eo muls)
1050 matmuls PGCR on 16^3 x 32 x 8 m=.01

Substantial effort on timing and logging infrastructure
 2015-07-24 01:31:13 +09:00
Peter Boyle
03ca506a3d Big commit fixing nocompiles in defective C++11 compilers (gcc, icpc). stared getting to 
near the bleeding edge I guess
 2015-06-30 15:17:27 +01:00
Peter Boyle
98c817df1b big commit fixing nocompiles in defective C++11 compilers (gcc, icpc). stared getting to 
near the bleeding edge I guess
 2015-06-30 15:03:11 +01:00
Peter Boyle
a17684ebe2 Some small steps towards a multigrid 2015-06-22 12:49:44 +01:00
Azusa Yamaguchi
dc7c77e1d5 Merge branch 'master' of https://github.com/paboyle/Grid 2015-06-20 14:22:29 +01:00
Peter Boyle
b92060f511 5d OpDir direction interface refers to the 5d dims, not 4d to present a 
sensible and consistent external interface.
 2015-06-09 22:41:59 +01:00
Peter Boyle
1e5b015ee3 Some unary ops and coarse grid support 2015-06-09 10:26:19 +01:00
Peter Boyle
d6f1ddf99c Conjugate residual algorithm; some more unary functions 2015-06-08 12:04:59 +01:00
Peter Boyle
3845f267cb Domain wall fermions now invert ; have the basis set up for 
Tanh/Zolo * (Cayley/PartFrac/ContFrac) * (Mobius/Shamir/Wilson)
Approx        Representation               Kernel.

All are done with space-time taking part in checkerboarding, Ls uncheckerboarded

Have only so far tested the Domain Wall limit of mobius, and at that only checked
that it
i)  Inverts
ii) 5dim DW == Ls copies of 4dim D2
iii) MeeInv Mee == 1
iv) Meo+Mee+Moe+Moo == M unprec.
v) MpcDagMpc is hermitan
vi) Mdag is the adjoint of M between stochastic vectors.

That said, the RB schur solve, RB MpcDagMpc solve, Unprec solve
all converge and the true residual becomes small; so pretty good tests.
 2015-06-02 16:57:12 +01:00