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
8,153 Commits 95 Branches 14 Tags
feature/verify-checksum-host
Commit Graph
100 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
paboyle 7132a4fd28 Update 2025-12-02 23:22:42 +00:00
paboyle e8057d6b4a Updated for verbose on host vs. device side csum 2025-12-02 23:15:32 +00:00
paboyle 973584e039 Update on aurora 2025-12-02 22:24:54 +00:00
paboyle 4a0aaf0786 Fix issue with Aurora compilers 2025-11-21 21:41:13 +00:00
paboyle 9c3835524c Fix compile warn 2025-11-21 21:41:12 +00:00
paboyle 549351bb8a Stag verbose clean up 2025-11-20 18:22:57 +00:00
paboyle c54d87a472 Aurora compile fix for new compiler 2025-11-06 18:17:33 +00:00
paboyle 23581333e6 link cufft 2025-08-21 22:25:55 +01:00
paboyle e5fa3d887f Compile on CUDA 2025-08-21 22:10:27 +01:00
paboyle 583fa7bb0a FFTW guarded after CUDA adn HIP 2025-08-21 22:00:12 +01:00
paboyle a3420e6fa9 Update for grid view logging 2025-08-14 21:29:20 +00:00
paboyle 732836d9f8 Missed one 2025-08-14 20:25:54 +00:00
paboyle 87658f7b53 ASSERT tripped in Shuhei's branch 2025-08-14 20:08:54 +00:00
paboyle e7f51e5fb1 Timer pointers for hadrons compat. 
Reluctantly, this interface is silly to pass timers around.
 2025-08-11 21:11:36 +01:00
paboyle 5adf2657dd Updated to compile and run fast on CUDA 2025-08-10 00:00:13 +01:00
paboyle 82cfff2990 A2A meson field BLAS based momentum project 2025-08-07 15:51:15 +00:00
paboyle 4397b1c442 Debugged momentum projection for A2A Meson Field 2025-08-07 15:51:01 +00:00
paboyle 9e6a4a4737 Assertion updates to macros (mostly) with backtrace. 
WIlson flow to include options for DBW2, Iwasaki, Symanzik.
View logging for data assurance
 2025-08-07 15:48:38 +00:00
paboyle 41f344bbd3 Merge with Christoph GPT checksum debug 2025-07-15 03:06:09 +00:00
paboyle a77cd50b2f Update comms logging in Cshift 2025-07-11 14:36:10 +00:00
paboyle 73af020f98 improved 2025-06-27 06:08:54 +00:00
paboyle bffb83c46e std::cout<<GridLogMessage<<"Debug:"<<std::endl; 
std::cout<<GridLogMessage<<"  --dylib-map     : print dynamic library map, useful for interpreting signal backtraces "<<std::endl;
    std::cout<<GridLogMessage<<"  --heartbeat     : periodic itimer wakeup (interrupts stuck system calls!) "<<std::endl;
    std::cout<<GridLogMessage<<"  --signal-delay n : pause for n seconds after signal handling (useful to get ALL nodes in stuck state) "<<std::endl;
    std::cout<<GridLogMessage<<"  --debug-stdout  : print stdout from EVERY node to file Grid.stdout/err.rank "<<std::endl;
    std::cout<<GridLogMessage<<"  --debug-signals : catch sigsegv and print a blame report, handle SIGHUP with a backtrace to stderr"<<std::endl;
    std::cout<<GridLogMessage<<"  --debug-heartbeat : periodically report backtrace "<<std::endl;

--dylib-map : Grid prints its dylib regions
--heartbeat : itimer based / SIGALRM wake up which seems to make Aurora
more stable
--debug-heartbeat : periodically report to stderr where we are in code

Now have libunwind option (configure: --with-unwind=<prefix>) to give an
Asynch-Signal safe backtrace. Avoid glibc backtrace due to mallocs.
 2025-06-27 06:08:54 +00:00
paboyle 7031f37350 Use libunwind for backtrace as it is signal asynch safe 2025-06-27 06:08:54 +00:00
paboyle 829dd74cb2 Verbose change 2025-06-27 06:08:54 +00:00
paboyle 66e671985d P2P 2025-06-27 06:08:54 +00:00
paboyle 5afcbcf0f3 Cshift uses flight recorder 2025-06-27 06:08:54 +00:00
paboyle 9730579312 Simplify and verbose 2025-06-27 06:08:51 +00:00
paboyle bfae14d035 More flight logging 2025-06-27 06:07:34 +00:00
paboyle b78fc73d19 Better signal handler 2025-06-27 06:07:34 +00:00
paboyle d418f78352 Making running on Aurora more debuggable 2025-05-23 20:58:16 +00:00
paboyle 25163998a0 Makes SYCL compiler happy 2025-05-23 20:57:11 +00:00
paboyle bb317aba8d Lattice = for sycl 2025-05-13 12:50:58 +00:00
paboyle 644cc6647e JSON update 2025-05-13 12:50:58 +00:00
paboyle 72397ce23b SYCL interface change 2025-05-13 12:50:58 +00:00
paboyle 25ab9325e7 Use hostVector but remove construct resize 2025-03-11 15:02:32 +00:00
paboyle 19f9378b98 Should work on Aurora nowb 2025-03-11 13:50:43 +00:00
paboyle 1d22841811 Working on aurora, GPT issue turned up is fixed 2025-03-06 03:20:18 +00:00
paboyle c4d9aa1a21 Config command that makes GPT happier 2025-02-27 20:12:49 +00:00
paboyle 6ae809ed40 Print not liked on GPT compile 2025-02-27 20:12:49 +00:00
paboyle 438dfbdb83 Only throw if there is a pending list entry in CommsComplete 2025-02-25 16:57:27 +00:00
paboyle b2ce760cf4 Verbose issue with GPT 2025-02-25 16:55:23 +00:00
paboyle ba9bbe0221 Bounce MPI through host 2025-02-12 19:34:59 +00:00
paboyle 4c3dd82d84 CSHIFT with bounce throuhgh Host memory on MPI packets 2025-02-12 19:09:53 +00:00
paboyle 44e911b5b7 Comment change 2025-02-12 17:37:55 +00:00
paboyle a7a16df9d0 GET not put has kinder barrier sequence for NVLINK type access as when 
GET is done, I can use it without barrier. Moves a barrier to a nicer
place, overlapped with DtoH DMA
 2025-02-12 14:59:28 +00:00
paboyle 382e0abefd Was issueing a double fence -- the gather also fences 2025-02-12 14:57:28 +00:00
paboyle 6fdefe5b90 Barrier sequencing if doing "GET" not "PUT" is different. 
This is somewhat better timing for Barriers
 2025-02-12 14:55:20 +00:00
paboyle 4788dd8e2e More states in packet progression for GPU non aware MPI 2025-02-12 14:53:57 +00:00
paboyle 1cc5f221f3 GET not put ordering is better as I know when I've got all MY data 2025-02-12 14:53:05 +00:00
paboyle 93251bfba0 GET not put for better ordering in the downstream dependent kernels -- I 
know when I'm done, so we can move a barrier / handshake between ranks
intranode to a point off critical path
 2025-02-12 14:50:21 +00:00
paboyle 18b79508b8 New line better for pretty print 2025-02-12 14:49:48 +00:00
paboyle 4de5ed1613 Remove vector view. The std::vector will not inform Memory manager of 
deletion and so a stale entry could be left. It is not and should not be
used.
 2025-02-12 14:48:46 +00:00
paboyle 0baaddbe98 Pipeline mode commit on Aurora. 5+ TF/s on 16^3x32 per tile at 384 
nodes.
More concurrency/fine grained scheduling is possible.
 2025-02-04 19:27:26 +00:00
paboyle b50fb34e71 Perf on Aurora 2025-02-01 18:39:34 +00:00
paboyle de84d730ff Fastest run config on Aurora to date 2025-02-01 18:08:40 +00:00
paboyle c4fc972fec Merge branch 'feature/deprecate-uvm' into develop 2025-01-31 16:32:36 +00:00
paboyle 8cf809e231 Best results on Aurora so far 2025-01-31 16:14:45 +00:00
paboyle 94019a922e Significantly better performance on Aurora without using pipeline mode 2025-01-30 16:36:46 +00:00
paboyle d6b2727f86 Pipeline mode getting better -- 2 nodes @ 10TF/s per node on Aurora 2025-01-29 09:22:21 +00:00
paboyle 74a4f43946 Optional host buffer bounce for no CUDA aware MPI 2025-01-28 15:22:46 +00:00
paboyle 1caf8b0f86 Rename 2025-01-28 15:22:37 +00:00
paboyle 8fe429346f Dslash testing for reproduce 2024-11-11 23:11:11 +00:00
paboyle 5cc4f3241d Meson field test 2024-10-18 15:42:30 +00:00
paboyle a78a61d76f Update configure 2024-10-15 14:38:45 +00:00
paboyle 2eff3f34ed Alternate reduction; default to grids own but make a configure flag 
--enable-reduction=grid|mpi
 2024-10-15 14:36:06 +00:00
paboyle 03687c1d62 Final version of test, closer to original again 2024-10-15 14:35:17 +00:00
paboyle febfe4e77f Make my own reduction a configure flag 2024-10-15 14:32:35 +00:00
paboyle 4d1aa134b5 Use normal reduction, configure flag to force deterministic 2024-10-15 14:32:11 +00:00
paboyle 5ec879860a Odd rounding issue - bears looking into 2024-10-15 14:30:54 +00:00
paboyle b728af903c Fast axpy norm under CFLAG 2024-10-11 03:23:09 +00:00
paboyle 54f1999030 axpy_norm_fast -- wasn't using the determinstic MPI sum causing issues 2024-10-11 03:22:18 +00:00
paboyle fd58f0b669 Return ok 2024-10-11 03:21:21 +00:00
paboyle c5c67b706e cl::sycl -> SYCL 2024-10-10 22:04:12 +00:00
paboyle be7a543e2c Revert barriers -- these were not the problem 2024-10-10 22:03:29 +00:00
paboyle 68f112d576 New software moves cl::sycl 2024-10-10 22:03:04 +00:00
paboyle ec1395a304 Better flight logging 2024-10-10 22:01:57 +00:00
paboyle beb0e474ee Use deterministic own brand reduction 2024-10-10 22:01:24 +00:00
paboyle 2b5fdcbbc5 New software version 2024-10-10 21:59:02 +00:00
paboyle 295127d456 Deterministic homebrew reduction 2024-10-10 21:58:26 +00:00
paboyle 7dcfb13694 New software stack 2024-10-10 21:57:35 +00:00
paboyle 9fa8bd6438 Configure for AOT on Aurora latest software 2024-09-23 11:25:44 +00:00
paboyle 02c8178f16 Almost working on Aurora 2024-09-23 09:43:50 +00:00
paboyle e637fbacae Verbose remove 2024-09-23 09:42:43 +00:00
paboyle 066544281f Deprecate UVM 2024-09-17 13:34:27 +00:00
paboyle 11be10d2c0 Aurora testing 2024-09-10 18:11:52 +00:00
paboyle 160969a758 UVM tester, doesn't turn up anything 2024-09-10 18:09:42 +00:00
paboyle 622f78ebea SYCL updates -- operator = giving trouble on Aurora. 
SYCL reduction is failing intermittently with SVM interface - returns
zero, expect non-zero.
Think I need to remove ALL dependence on SVM.
 2024-09-04 13:53:48 +00:00
paboyle 89c0519f83 Repro test 2024-03-12 16:11:33 +00:00
paboyle 2704b82084 Merge branch 'develop' of https://github.com/paboyle/Grid into develop 2024-03-12 15:16:24 +00:00
paboyle cf8632bbac Britney test option 2024-03-12 15:15:35 +00:00
paboyle d224297972 PBS scripts 2024-03-12 15:15:16 +00:00
paboyle b812a7b4c6 Staggered launch script 2024-03-06 01:32:40 +00:00
paboyle 891a366f73 Repro CG script 2024-03-06 01:22:55 +00:00
paboyle 10116b3be8 Force device copyable and tell SYCL to shut it. 2024-03-06 01:13:27 +00:00
paboyle a46a0f0882 force device copyable and don't take crap from SYCL 2024-03-06 01:12:49 +00:00
paboyle a26a8a38f4 Merge branch 'develop' of https://github.com/paboyle/Grid into develop 2024-03-06 00:05:00 +00:00
paboyle 7435315d50 More blasted shell variables 2024-03-06 00:03:59 +00:00
paboyle 9b5f741e85 Reproducing CG can be more useful now 2024-03-06 00:03:16 +00:00
paboyle 517822fdd2 SPR HBM benchmarking right and also PVC batched GEMM 2024-03-06 00:02:27 +00:00
paboyle 1b93a9be88 Print out the hostname 2024-03-06 00:01:58 +00:00