portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2026-07-17 15:43:27 +01:00
Code Releases Activity
1,972 Commits 95 Branches 14 Tags
33d199a0ad3d1be0d583961911c2766d8f8874e4
Commit Graph
100 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
paboyle b1508e4124 Merge branch 'feature/mpi3' into develop 2016-10-25 06:06:36 +01:00
paboyle b820076b91 Merge branch 'develop' into feature/mpi3 2016-10-25 06:02:33 +01:00
paboyle 09f66100d3 MPI 3 compile on non-linux 2016-10-25 06:01:12 +01:00
paboyle 2c54a53d0a Compile verbose reduce 2016-10-21 12:12:14 +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 5b5925b8e5 Forgot to add 2016-10-20 17:09:40 +01:00
paboyle b58adc6a4b commVector 2016-10-20 17:00:15 +01:00
paboyle f9d5e95d72 allocator template typedefs moved to AlignedAllocator 2016-10-20 16:59:39 +01:00
paboyle 4f8e636a43 commVector 2016-10-20 16:59:16 +01:00
paboyle 9b39f35ae6 commVector different for SHMEM compat 2016-10-20 16:58:53 +01:00
paboyle 5fe2b85cbd MPI3 and shared memory support 2016-10-20 16:58:01 +01:00
paboyle c7cccaaa69 Comm vector for shmem 2016-10-20 16:57:31 +01:00
paboyle cbcfea466f MPI3 2016-10-20 16:57:14 +01:00
paboyle 4955672fc3 MPI3 2016-10-20 16:57:00 +01:00
paboyle 39f1c880b8 mpi3 2016-10-20 16:56:40 +01:00
paboyle 8c043da5b7 SHMEM and comms allocator made different 2016-10-20 16:56:05 +01:00
paboyle 3cbe974eb4 Layout 2016-10-20 16:55:21 +01:00
paboyle 7af9b87318 Cache face tables to improve performance. 
Extract merge now looking poor.
 2016-10-18 09:51:37 +01:00
paboyle 811ca45473 GNU clang hack for AVX512 since there are missing reduce intrinsics in Clang 3.9 and GCC-6 AVX512 support 2016-10-17 16:23:21 +01:00
paboyle bc1a4d40ba Faster integer handling avoid push_back 2016-10-17 16:16:44 +01:00
paboyle c8079e6621 Time the face gateher in x-dir more carefully 2016-10-13 22:28:50 +01:00
paboyle 2d4a45c758 Typecast pointer 2016-10-12 09:14:15 +01:00
paboyle 0f182f033b Drop macos with gcc 2016-10-11 22:29:06 +01:00
paboyle 7240d73184 Parallelise the x faces; fix the segv on KNL with comms 2016-10-11 22:21:07 +01:00
paboyle 42cd148f5e Base pointer for comms buffer under AVX512 assembly 2016-10-11 16:06:06 +01:00
paboyle 4089984431 Timing hooks 2016-10-06 09:25:12 +01:00
paboyle 8c89391c02 FFTW unresolved fixed when no fftw3.h 2016-08-24 16:41:47 +01:00
paboyle bfac5195b8 tidy up 2016-08-24 16:38:36 +01:00
paboyle a782ca3238 Merge branch 'feature/fft-flop-count' into develop 2016-08-24 15:06:17 +01:00
paboyle 744691097f Printing 2016-08-24 15:05:56 +01:00
paboyle ff6da364e8 FFT double and single precision gives good performance now in multithreaded code. 2016-08-24 15:05:00 +01:00
paboyle 88be3b39bb Merge branch 'develop' of https://github.com/paboyle/Grid into develop 2016-08-22 18:29:36 +01:00
paboyle 8a02824e08 Merge branch 'feature/FFT' into develop 2016-08-22 16:25:04 +01:00
paboyle 356e7940fd fftw can be switched off 2016-08-22 16:24:49 +01:00
paboyle 73ce476890 Include fftw headers 2016-08-22 16:24:21 +01:00
paboyle 29c4ef41de Adding a test for libfftw3 2016-08-22 16:21:01 +01:00
paboyle e423a09974 FFT improved and test_FFT passing under MPI 8 processes, 8^4 for LatticeComplexD and LatticeSpinMatrixD 2016-08-18 02:23:21 +01:00
paboyle 17097a93ec FFTW test ran over 4 mpi processes. 2016-08-17 01:33:55 +01:00
paboyle 94a6373a7f Merge branch 'feature/eigen-cleanup' into develop 2016-08-15 23:58:34 +01:00
paboyle 4ab7dbfd57 Instantiate 2016-08-15 23:00:40 +01:00
paboyle 90e70790f3 Feature for z-Mobius prep 2016-08-15 22:31:29 +01:00
paboyle fc25d2295c fftw download 2016-08-06 00:28:52 +01:00
paboyle 8dc2cfcedb Adding fftw header pulling 2016-08-06 00:28:28 +01:00
paboyle 5a68715be3 Richards sweep test 2016-08-05 10:51:57 +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 27f3ecc833 Merge branch 'feature/bugfix-ck-cj' into develop 2016-07-16 01:59:52 +01:00
paboyle f9e90eeb1f Sign error on the force for 4d fields fixed 2016-07-16 01:52:44 +01:00
paboyle fad5c675eb sign error on the 4d gparity force 2016-07-16 01:51:56 +01:00
paboyle 4908b77d46 Fixed conflicts. PLEASE avoid making wholesale cosmetic only changes, this created 
a HUGE amount of difficult to resolve and understand conflicts .

Wholesale formatting, reordering functions etc... in a central file like Tensor_class
or Grid_vector_types while others are also editing without making substantial functionality
changes creates pain.
 2016-07-15 20:59:07 +01:00
paboyle f4dd5062d7 Merge branch 'develop' of https://github.com/paboyle/Grid into develop 2016-07-15 19:26:06 +01:00
paboyle da34d75841 Merge branch 'feature/Ls-vectorised-actions' into develop 2016-07-15 19:09:47 +01:00
paboyle 980ff18956 Solving the instantiation no compile issue 2016-07-15 17:19:44 +01:00
paboyle 1a6c7204ac Disable instantiation; Use cache version instead 2016-07-15 00:34:39 +01:00
paboyle 49310fbab3 Done with red black change over 2016-07-15 00:08:43 +01:00
paboyle 6049d5ac47 Update 2016-07-15 00:08:32 +01:00
paboyle 35d0d35238 Updated file list 2016-07-15 00:02:53 +01:00
paboyle c0e878705e Updated file list 2016-07-15 00:02:39 +01:00
paboyle 5c0c8efb9e Updated file list 2016-07-15 00:02:11 +01:00
paboyle dfd714e1ef Multiple implementations for the 5d hopping terms, depending on cache friendly 
ops and/or the 5th direction being vectorised
All use 4d redblack.
 2016-07-15 00:00:09 +01:00
paboyle 79a8ca1a62 Rewrite for performance. Impl dependent instantiations give 
4d linalg impls of the 5d hopping terms (and inverse)
Cache friendly loop orderings of the above
Dense matrix stored and apply to the above

-- Switch to Ls vectorised, and use dense matrix approach for the MooeeInv
   and rotate/shift of the Mooee M5D routines.
 2016-07-14 23:58:15 +01:00
paboyle fb45eb2eb2 5d ls vec rename of impl class 2016-07-14 23:57:26 +01:00
paboyle a307274c96 Fermion impl rename for ls vectorised 5d approaches 2016-07-14 23:56:13 +01:00
paboyle 3f2c44a5fe Updating the class to 5d selection based on impl type 2016-07-14 23:55:26 +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 8a79e93cc2 Rename the 5d domain wall fermion vectorised Ls impl class 2016-07-14 23:53:00 +01:00
paboyle 3493b51879 Modest updates 2016-07-14 23:52:13 +01:00
paboyle de3e79d300 red black for Ls vectorised is 4d red black. Update accordingly now I've made this choice 2016-07-14 23:49:42 +01:00
paboyle dd62a61c5c Added broadcast and rotation of simd vectors 2016-07-14 23:49:00 +01:00
paboyle 8f47d0b5ab Rotation needed for hopping term in fifth dim with Ls vectorised fields 2016-07-14 23:45:36 +01:00
paboyle 42af132dab Fix for chris kellys request to peek poke on checkerboarded fields 2016-07-14 23:44:48 +01:00
paboyle 9db2c6525d updating benchmarks for red black 4d for Ls vectorised code 2016-07-14 23:44:02 +01:00
paboyle adbc7c1188 Adding files for multiple implementations (cache opt) and Ls vectorisation 
of the 5D cayley form chiral fermions for the 5d matrix. With Ls entirely
in the vector direction, s-hopping terms involve rotations.

The serial dependence of the LDU inversion for Mobius and 4d even odd
checkerboarding is removed by simply applying Ls^2 operations (vectorised
many ways) as a dense matrix operation.

This should give similar throughput but high flops (non-compulsory flops)
but enable use of the KNL cache friendly kernels throughout the code.

Ls is still constrained to be a multiple of Nsimd, which is as much as 8 for AVX512
with single precision.
 2016-07-14 22:59:21 +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 c667d9fdcc Trying to make compile clean on travis; seem to have a make -j 4 problem with fftw 2016-07-07 23:26:39 +01:00
paboyle 7dbb94bab2 Update 2016-07-07 22:51:37 +01:00
paboyle 236dcc820b typo fix 2016-07-07 22:46:11 +01:00
paboyle a42a441a6a Rename the reconfigure script to ./autogen.sh 2016-07-07 22:35:45 +01:00
paboyle a0676beeb1 Open up dependency on Eigen and FFTW 2016-07-07 22:31:07 +01:00
paboyle fc4a043663 Colors and banner clean up 2016-07-02 16:15:38 +01:00
paboyle 61ba50665e Merge branch 'hotfix/v0.5.1' into develop 2016-07-01 16:34:30 +01:00
paboyle bfe14000a9 Double compile fix 2016-07-01 16:33:51 +01:00
paboyle 1ceff48133 Merge branch 'release/v0.5.0' into develop 2016-06-30 15:15:59 -07:00
paboyle 680645f849 Merge branch 'release/v0.5.0' 2016-06-30 15:15:03 -07:00
paboyle 3fc6e03ad1 Version file 2016-06-30 14:44:09 -07:00
paboyle 2d6614f3a1 Merge branch 'feature/knl-cache-opt' into develop 2016-06-30 14:36:20 -07:00
paboyle 4e041b5103 Merge branch 'feature/knl-cache-opt' of https://github.com/paboyle/Grid into feature/knl-cache-opt 2016-06-30 14:36:08 -07:00
paboyle 712b9a3489 Asm only for avx512 2016-06-30 14:35:02 -07:00
paboyle bdaa5b1767 Updated to have perfect prefetching for the s-vectorised kernel with any cache blocking. 2016-06-30 14:35:02 -07:00
paboyle 8fcefc021a Improved the prefetching when using cache blocking codes 2016-06-30 14:35:02 -07:00
paboyle 1445189361 COntrol the prefetch strategy 2016-06-30 14:35:02 -07:00
paboyle 05c884a62a Prefetch change 2016-06-30 14:35:01 -07:00
paboyle a25bec87d9 Prefetch during save 2016-06-30 14:35:01 -07:00
paboyle 2d8bb4c594 Tweaks 2016-06-30 14:35:01 -07:00
paboyle 51cb2d4328 update file lists 2016-06-30 14:35:01 -07:00
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 c8b35d960c Merge branch 'develop' of https://github.com/paboyle/Grid into feature/knl-cache-opt 2016-06-30 14:30:49 -07:00
paboyle 532f41dd61 Asm only for avx512 2016-06-30 14:00:34 -07:00
paboyle 661b0ab45d Updated to have perfect prefetching for the s-vectorised kernel with any cache blocking. 2016-06-30 13:07:42 -07:00