portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-09 23:45:36 +00:00
Code Releases Activity

Perf results

Browse Source
This commit is contained in:
Peter Boyle 2021-09-16 01:06:28 +01:00
parent cc4a27b9e6
commit 145acf2919
4 changed files with 465 additions and 139 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

293
systems/Tursa/dwf.16node.perf Normal file
View File

@ -0,0 +1,293 @@
tu-c0r1n00 - 0 device=0 binding=--interleave=0,1
tu-c0r1n00 - 1 device=1 binding=--interleave=2,3
tu-c0r1n00 - 2 device=2 binding=--interleave=4,5
tu-c0r1n00 - 3 device=3 binding=--interleave=6,7
tu-c0r2n21 - 0 device=0 binding=--interleave=0,1
tu-c0r2n06 - 3 device=3 binding=--interleave=6,7
tu-c0r2n12 - 2 device=2 binding=--interleave=4,5
tu-c0r2n21 - 2 device=2 binding=--interleave=4,5
tu-c0r2n21 - 3 device=3 binding=--interleave=6,7
tu-c0r2n21 - 1 device=1 binding=--interleave=2,3
tu-c0r2n06 - 0 device=0 binding=--interleave=0,1
tu-c0r2n06 - 1 device=1 binding=--interleave=2,3
tu-c0r2n12 - 1 device=1 binding=--interleave=2,3
tu-c0r2n06 - 2 device=2 binding=--interleave=4,5
tu-c0r1n12 - 2 device=2 binding=--interleave=4,5
tu-c0r1n12 - 3 device=3 binding=--interleave=6,7
tu-c0r2n12 - 0 device=0 binding=--interleave=0,1
tu-c0r1n12 - 1 device=1 binding=--interleave=2,3
tu-c0r1n12 - 0 device=0 binding=--interleave=0,1
tu-c0r2n12 - 3 device=3 binding=--interleave=6,7
tu-c0r1n18 - 2 device=2 binding=--interleave=4,5
tu-c0r1n18 - 0 device=0 binding=--interleave=0,1
tu-c0r1n18 - 3 device=3 binding=--interleave=6,7
tu-c0r1n18 - 1 device=1 binding=--interleave=2,3
tu-c0r1n06 - 2 device=2 binding=--interleave=4,5
tu-c0r2n09 - 3 device=3 binding=--interleave=6,7
tu-c0r1n06 - 0 device=0 binding=--interleave=0,1
tu-c0r2n15 - 0 device=0 binding=--interleave=0,1
tu-c0r2n09 - 2 device=2 binding=--interleave=4,5
tu-c0r1n06 - 3 device=3 binding=--interleave=6,7
tu-c0r2n15 - 1 device=1 binding=--interleave=2,3
tu-c0r2n15 - 2 device=2 binding=--interleave=4,5
tu-c0r1n06 - 1 device=1 binding=--interleave=2,3
tu-c0r2n15 - 3 device=3 binding=--interleave=6,7
tu-c0r2n09 - 1 device=1 binding=--interleave=2,3
tu-c0r2n09 - 0 device=0 binding=--interleave=0,1
tu-c0r1n09 - 0 device=0 binding=--interleave=0,1
tu-c0r1n09 - 1 device=1 binding=--interleave=2,3
tu-c0r1n09 - 2 device=2 binding=--interleave=4,5
tu-c0r1n09 - 3 device=3 binding=--interleave=6,7
tu-c0r1n21 - 2 device=2 binding=--interleave=4,5
tu-c0r1n21 - 0 device=0 binding=--interleave=0,1
tu-c0r1n21 - 1 device=1 binding=--interleave=2,3
tu-c0r1n15 - 2 device=2 binding=--interleave=4,5
tu-c0r1n21 - 3 device=3 binding=--interleave=6,7
tu-c0r1n15 - 1 device=1 binding=--interleave=2,3
tu-c0r1n15 - 0 device=0 binding=--interleave=0,1
tu-c0r2n03 - 2 device=2 binding=--interleave=4,5
tu-c0r1n15 - 3 device=3 binding=--interleave=6,7
tu-c0r2n03 - 0 device=0 binding=--interleave=0,1
tu-c0r2n00 - 3 device=3 binding=--interleave=6,7
tu-c0r2n03 - 3 device=3 binding=--interleave=6,7
tu-c0r2n00 - 1 device=1 binding=--interleave=2,3
tu-c0r2n00 - 2 device=2 binding=--interleave=4,5
tu-c0r2n18 - 1 device=1 binding=--interleave=2,3
tu-c0r2n00 - 0 device=0 binding=--interleave=0,1
tu-c0r2n18 - 2 device=2 binding=--interleave=4,5
tu-c0r2n18 - 3 device=3 binding=--interleave=6,7
tu-c0r2n18 - 0 device=0 binding=--interleave=0,1
tu-c0r2n03 - 1 device=1 binding=--interleave=2,3
tu-c0r1n03 - 0 device=0 binding=--interleave=0,1
tu-c0r1n03 - 1 device=1 binding=--interleave=2,3
tu-c0r1n03 - 2 device=2 binding=--interleave=4,5
tu-c0r1n03 - 3 device=3 binding=--interleave=6,7
OPENMPI detected
AcceleratorCudaInit: using default device
AcceleratorCudaInit: assume user either uses a) IBM jsrun, or
AcceleratorCudaInit: b) invokes through a wrapping script to set CUDA_VISIBLE_DEVICES, UCX_NET_DEVICES, and numa binding
AcceleratorCudaInit: Configure options --enable-summit, --enable-select-gpu=no
AcceleratorCudaInit: ================================================
OPENMPI detected
AcceleratorCudaInit: using default device
AcceleratorCudaInit: assume user either uses a) IBM jsrun, or
AcceleratorCudaInit: b) invokes through a wrapping script to set CUDA_VISIBLE_DEVICES, UCX_NET_DEVICES, and numa binding
AcceleratorCudaInit: Configure options --enable-summit, --enable-select-gpu=no
AcceleratorCudaInit: ================================================
OPENMPI detected
AcceleratorCudaInit[0]: ========================
AcceleratorCudaInit[0]: Device Number : 0
AcceleratorCudaInit[0]: ========================
AcceleratorCudaInit[0]: Device identifier: NVIDIA A100-SXM4-40GB
AcceleratorCudaInit[0]: totalGlobalMem: 42505273344
AcceleratorCudaInit[0]: managedMemory: 1
AcceleratorCudaInit[0]: isMultiGpuBoard: 0
AcceleratorCudaInit[0]: warpSize: 32
AcceleratorCudaInit[0]: pciBusID: 3
AcceleratorCudaInit[0]: pciDeviceID: 0
AcceleratorCudaInit[0]: maxGridSize (2147483647,65535,65535)
AcceleratorCudaInit: using default device
AcceleratorCudaInit: assume user either uses a) IBM jsrun, or
AcceleratorCudaInit: b) invokes through a wrapping script to set CUDA_VISIBLE_DEVICES, UCX_NET_DEVICES, and numa binding
AcceleratorCudaInit: Configure options --enable-summit, --enable-select-gpu=no
AcceleratorCudaInit: ================================================
OPENMPI detected
AcceleratorCudaInit[0]: ========================
AcceleratorCudaInit[0]: Device Number : 0
AcceleratorCudaInit[0]: ========================
AcceleratorCudaInit[0]: Device identifier: NVIDIA A100-SXM4-40GB
AcceleratorCudaInit[0]: totalGlobalMem: 42505273344
AcceleratorCudaInit[0]: managedMemory: 1
AcceleratorCudaInit[0]: isMultiGpuBoard: 0
AcceleratorCudaInit[0]: warpSize: 32
AcceleratorCudaInit[0]: pciBusID: 3
AcceleratorCudaInit[0]: pciDeviceID: 0
AcceleratorCudaInit[0]: maxGridSize (2147483647,65535,65535)
AcceleratorCudaInit: using default device
AcceleratorCudaInit: assume user either uses a) IBM jsrun, or
AcceleratorCudaInit: b) invokes through a wrapping script to set CUDA_VISIBLE_DEVICES, UCX_NET_DEVICES, and numa binding
AcceleratorCudaInit: Configure options --enable-summit, --enable-select-gpu=no
AcceleratorCudaInit: ================================================
OPENMPI detected
AcceleratorCudaInit: using default device
AcceleratorCudaInit: assume user either uses a) IBM jsrun, or
AcceleratorCudaInit: b) invokes through a wrapping script to set CUDA_VISIBLE_DEVICES, UCX_NET_DEVICES, and numa binding
AcceleratorCudaInit: Configure options --enable-summit, --enable-select-gpu=no
AcceleratorCudaInit: ================================================
OPENMPI detected
AcceleratorCudaInit: using default device
AcceleratorCudaInit: assume user either uses a) IBM jsrun, or
AcceleratorCudaInit: b) invokes through a wrapping script to set CUDA_VISIBLE_DEVICES, UCX_NET_DEVICES, and numa binding
AcceleratorCudaInit: Configure options --enable-summit, --enable-select-gpu=no
AcceleratorCudaInit: ================================================
OPENMPI detected
AcceleratorCudaInit: using default device
AcceleratorCudaInit: assume user either uses a) IBM jsrun, or
AcceleratorCudaInit: b) invokes through a wrapping script to set CUDA_VISIBLE_DEVICES, UCX_NET_DEVICES, and numa binding
AcceleratorCudaInit: Configure options --enable-summit, --enable-select-gpu=no
AcceleratorCudaInit: ================================================
OPENMPI detected
AcceleratorCudaInit: using default device
AcceleratorCudaInit: assume user either uses a) IBM jsrun, or
AcceleratorCudaInit: b) invokes through a wrapping script to set CUDA_VISIBLE_DEVICES, UCX_NET_DEVICES, and numa binding
AcceleratorCudaInit: Configure options --enable-summit, --enable-select-gpu=no
AcceleratorCudaInit: ================================================
SharedMemoryMpi: World communicator of size 64
SharedMemoryMpi: Node communicator of size 4
0SharedMemoryMpi: SharedMemoryMPI.cc acceleratorAllocDevice 2147483648bytes at 0x7f05c0000000 for comms buffers
Setting up IPC
__|__|__|__|__|__|__|__|__|__|__|__|__|__|__
__|__|__|__|__|__|__|__|__|__|__|__|__|__|__
__|_ | | | | | | | | | | | | _|__
__|_ _|__
__|_ GGGG RRRR III DDDD _|__
__|_ G R R I D D _|__
__|_ G R R I D D _|__
__|_ G GG RRRR I D D _|__
__|_ G G R R I D D _|__
__|_ GGGG R R III DDDD _|__
__|_ _|__
__|__|__|__|__|__|__|__|__|__|__|__|__|__|__
__|__|__|__|__|__|__|__|__|__|__|__|__|__|__
| | | | | | | | | | | | | |
Copyright (C) 2015 Peter Boyle, Azusa Yamaguchi, Guido Cossu, Antonin Portelli and other authors
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
Current Grid git commit hash=9d2238148c56e3fbadfa95dcabf2b83d4bde14cd: (HEAD -> develop) uncommited changes
Grid : Message : ================================================
Grid : Message : MPI is initialised and logging filters activated
Grid : Message : ================================================
Grid : Message : Requested 2147483648 byte stencil comms buffers
Grid : Message : MemoryManager Cache 34004218675 bytes
Grid : Message : MemoryManager::Init() setting up
Grid : Message : MemoryManager::Init() cache pool for recent allocations: SMALL 32 LARGE 8
Grid : Message : MemoryManager::Init() Non unified: Caching accelerator data in dedicated memory
Grid : Message : MemoryManager::Init() Using cudaMalloc
Grid : Message : 1.814936 s : Grid Layout
Grid : Message : 1.814947 s : Global lattice size : 64 64 64 256
Grid : Message : 1.814952 s : OpenMP threads : 4
Grid : Message : 1.814955 s : MPI tasks : 2 2 2 8
Grid : Message : 1.859229 s : Making s innermost grids
Grid : Message : 1.907983 s : Initialising 4d RNG
Grid : Message : 1.999619 s : Intialising parallel RNG with unique string 'The 4D RNG'
Grid : Message : 1.999657 s : Seed SHA256: 49db4542db694e3b1a74bf2592a8c1b83bfebbe18401693c2609a4c3af1
Grid : Message : 3.786102 s : Initialising 5d RNG
Grid : Message : 5.361999 s : Intialising parallel RNG with unique string 'The 5D RNG'
Grid : Message : 5.362036 s : Seed SHA256: b6316f2fac44ce14111f93e0296389330b077bfd0a7b359f781c58589f8a
Grid : Message : 38.698345 s : Initialised RNGs
Grid : Message : 42.821728 s : Drawing gauge field
Grid : Message : 43.916364 s : Random gauge initialised
Grid : Message : 46.410003 s : Setting up Cshift based reference
Grid : Message : 54.242661 s : *****************************************************************
Grid : Message : 54.242686 s : * Kernel options --dslash-generic, --dslash-unroll, --dslash-asm
Grid : Message : 54.242688 s : *****************************************************************
Grid : Message : 54.242689 s : *****************************************************************
Grid : Message : 54.242690 s : * Benchmarking DomainWallFermionR::Dhop
Grid : Message : 54.242691 s : * Vectorising space-time by 8
Grid : Message : 54.242692 s : * VComplexF size is 64 B
Grid : Message : 54.242694 s : * SINGLE precision
Grid : Message : 54.242697 s : * Using Overlapped Comms/Compute
Grid : Message : 54.242698 s : * Using GENERIC Nc WilsonKernels
Grid : Message : 54.242699 s : *****************************************************************
Grid : Message : 56.314112 s : Called warmup
Grid : Message : 84.246354 s : Called Dw 3000 times in 2.79318e+07 us
Grid : Message : 84.246405 s : mflop/s = 1.52229e+08
Grid : Message : 84.246408 s : mflop/s per rank = 2.37857e+06
Grid : Message : 84.246412 s : mflop/s per node = 9.51428e+06
Grid : Message : 84.246414 s : RF GiB/s (base 2) = 309325
Grid : Message : 84.246417 s : mem GiB/s (base 2) = 193328
Grid : Message : 84.250016 s : norm diff 1.03478e-13
Grid : Message : 84.285132 s : #### Dhop calls report
Grid : Message : 84.285137 s : WilsonFermion5D Number of DhopEO Calls : 6002
Grid : Message : 84.285140 s : WilsonFermion5D TotalTime /Calls : 4703.27 us
Grid : Message : 84.285142 s : WilsonFermion5D CommTime /Calls : 3131.05 us
Grid : Message : 84.285144 s : WilsonFermion5D FaceTime /Calls : 492.972 us
Grid : Message : 84.285146 s : WilsonFermion5D ComputeTime1/Calls : 56.9085 us
Grid : Message : 84.285148 s : WilsonFermion5D ComputeTime2/Calls : 1099.95 us
Grid : Message : 84.285160 s : Average mflops/s per call : 1.43412e+10
Grid : Message : 84.285165 s : Average mflops/s per call per rank : 2.24082e+08
Grid : Message : 84.285170 s : Average mflops/s per call per node : 8.96328e+08
Grid : Message : 84.285173 s : Average mflops/s per call (full) : 1.53416e+08
Grid : Message : 84.285176 s : Average mflops/s per call per rank (full): 2.39712e+06
Grid : Message : 84.285194 s : Average mflops/s per call per node (full): 9.58847e+06
Grid : Message : 84.285197 s : WilsonFermion5D Stencil
Grid : Message : 84.285271 s : Stencil calls 3001
Grid : Message : 84.285275 s : Stencil halogtime 0
Grid : Message : 84.285277 s : Stencil gathertime 55.2059
Grid : Message : 84.285281 s : Stencil gathermtime 20.0923
Grid : Message : 84.285283 s : Stencil mergetime 18.9057
Grid : Message : 84.285286 s : Stencil decompresstime 0.0619793
Grid : Message : 84.285289 s : Stencil comms_bytes 4.02653e+08
Grid : Message : 84.285292 s : Stencil commtime 6323.57
Grid : Message : 84.285295 s : Stencil 63.675 GB/s per rank
Grid : Message : 84.285298 s : Stencil 254.7 GB/s per node
Grid : Message : 84.285301 s : WilsonFermion5D StencilEven
Grid : Message : 84.285316 s : WilsonFermion5D StencilOdd
Grid : Message : 84.285333 s : WilsonFermion5D Stencil Reporti()
Grid : Message : 84.285336 s : WilsonFermion5D StencilEven Reporti()
Grid : Message : 84.285337 s : WilsonFermion5D StencilOdd Reporti()
Grid : Message : 106.985790 s : Compare to naive wilson implementation Dag to verify correctness
Grid : Message : 106.985814 s : Called DwDag
Grid : Message : 106.985815 s : norm dag result 12.0421
Grid : Message : 107.188790 s : norm dag ref 12.0421
Grid : Message : 107.349010 s : norm dag diff 7.63254e-14
Grid : Message : 107.762980 s : Calling Deo and Doe and //assert Deo+Doe == Dunprec
Grid : Message : 107.458374 s : src_e0.499998
Grid : Message : 107.754073 s : src_o0.500002
Grid : Message : 107.855191 s : *********************************************************
Grid : Message : 107.855194 s : * Benchmarking DomainWallFermionF::DhopEO
Grid : Message : 107.855195 s : * Vectorising space-time by 8
Grid : Message : 107.855197 s : * SINGLE precision
Grid : Message : 107.855198 s : * Using Overlapped Comms/Compute
Grid : Message : 107.855199 s : * Using GENERIC Nc WilsonKernels
Grid : Message : 107.855200 s : *********************************************************
Grid : Message : 121.549348 s : Deo mflop/s = 1.56492e+08
Grid : Message : 121.549382 s : Deo mflop/s per rank 2.44518e+06
Grid : Message : 121.549384 s : Deo mflop/s per node 9.78074e+06
Grid : Message : 121.549387 s : #### Dhop calls report
Grid : Message : 121.549388 s : WilsonFermion5D Number of DhopEO Calls : 3001
Grid : Message : 121.549390 s : WilsonFermion5D TotalTime /Calls : 4563.01 us
Grid : Message : 121.549393 s : WilsonFermion5D CommTime /Calls : 2967.77 us
Grid : Message : 121.549395 s : WilsonFermion5D FaceTime /Calls : 601.095 us
Grid : Message : 121.549397 s : WilsonFermion5D ComputeTime1/Calls : 59.9877 us
Grid : Message : 121.549399 s : WilsonFermion5D ComputeTime2/Calls : 1038.46 us
Grid : Message : 121.549423 s : Average mflops/s per call : 1.2726e+10
Grid : Message : 121.549428 s : Average mflops/s per call per rank : 1.98843e+08
Grid : Message : 121.549430 s : Average mflops/s per call per node : 7.95373e+08
Grid : Message : 121.549432 s : Average mflops/s per call (full) : 1.58131e+08
Grid : Message : 121.549436 s : Average mflops/s per call per rank (full): 2.4708e+06
Grid : Message : 121.549440 s : Average mflops/s per call per node (full): 9.88321e+06
Grid : Message : 121.549442 s : WilsonFermion5D Stencil
Grid : Message : 121.549453 s : WilsonFermion5D StencilEven
Grid : Message : 121.549472 s : WilsonFermion5D StencilOdd
Grid : Message : 121.549484 s : Stencil calls 3001
Grid : Message : 121.549490 s : Stencil halogtime 0
Grid : Message : 121.549492 s : Stencil gathertime 55.2206
Grid : Message : 121.549496 s : Stencil gathermtime 19.4562
Grid : Message : 121.549500 s : Stencil mergetime 18.3469
Grid : Message : 121.549502 s : Stencil decompresstime 0.0646451
Grid : Message : 121.549506 s : Stencil comms_bytes 2.01327e+08
Grid : Message : 121.549510 s : Stencil commtime 2979.17
Grid : Message : 121.549512 s : Stencil 67.5782 GB/s per rank
Grid : Message : 121.549514 s : Stencil 270.313 GB/s per node
Grid : Message : 121.549517 s : WilsonFermion5D Stencil Reporti()
Grid : Message : 121.549519 s : WilsonFermion5D StencilEven Reporti()
Grid : Message : 121.549522 s : WilsonFermion5D StencilOdd Reporti()
Grid : Message : 121.625928 s : r_e6.02108
Grid : Message : 121.634489 s : r_o6.02101
Grid : Message : 121.640496 s : res12.0421
Grid : Message : 122.275455 s : norm diff 0
Grid : Message : 123.135840 s : norm diff even 0
Grid : Message : 123.389190 s : norm diff odd 0

276
systems/Tursa/dwf.perf → systems/Tursa/dwf.4node.perf
View File

@ -1,19 +1,25 @@
tu-c0r0n00 - 0 device=0 binding=--interleave=0,1
tu-c0r0n09 - 0 device=0 binding=--interleave=0,1
tu-c0r0n00 - 1 device=1 binding=--interleave=2,3
tu-c0r0n09 - 1 device=1 binding=--interleave=2,3
tu-c0r0n09 - 2 device=2 binding=--interleave=4,5
tu-c0r0n03 - 1 device=1 binding=--interleave=2,3
tu-c0r0n03 - 0 device=0 binding=--interleave=0,1
tu-c0r0n03 - 3 device=3 binding=--interleave=6,7
tu-c0r0n03 - 2 device=2 binding=--interleave=4,5
tu-c0r0n00 - 2 device=2 binding=--interleave=4,5
tu-c0r0n06 - 0 device=0 binding=--interleave=0,1
tu-c0r0n06 - 1 device=1 binding=--interleave=2,3
tu-c0r0n00 - 3 device=3 binding=--interleave=6,7
tu-c0r0n00 - 2 device=2 binding=--interleave=4,5
tu-c0r0n09 - 3 device=3 binding=--interleave=6,7
tu-c0r0n09 - 0 device=0 binding=--interleave=0,1
tu-c0r0n09 - 2 device=2 binding=--interleave=4,5
tu-c0r0n03 - 1 device=1 binding=--interleave=2,3
tu-c0r0n06 - 2 device=2 binding=--interleave=4,5
tu-c0r0n09 - 3 device=3 binding=--interleave=6,7
tu-c0r0n00 - 3 device=3 binding=--interleave=6,7
tu-c0r0n03 - 0 device=0 binding=--interleave=0,1
tu-c0r0n03 - 2 device=2 binding=--interleave=4,5
tu-c0r0n06 - 3 device=3 binding=--interleave=6,7
tu-c0r0n03 - 3 device=3 binding=--interleave=6,7
OPENMPI detected
AcceleratorCudaInit: using default device
AcceleratorCudaInit: assume user either uses a) IBM jsrun, or
AcceleratorCudaInit: b) invokes through a wrapping script to set CUDA_VISIBLE_DEVICES, UCX_NET_DEVICES, and numa binding
AcceleratorCudaInit: Configure options --enable-summit, --enable-select-gpu=no
AcceleratorCudaInit: ================================================
OPENMPI detected
AcceleratorCudaInit[0]: ========================
AcceleratorCudaInit[0]: Device Number : 0
@ -32,18 +38,6 @@ AcceleratorCudaInit: b) invokes through a wrapping script to set CUDA_VISIBLE_DE
AcceleratorCudaInit: Configure options --enable-summit, --enable-select-gpu=no
AcceleratorCudaInit: ================================================
OPENMPI detected
AcceleratorCudaInit: using default device
AcceleratorCudaInit: assume user either uses a) IBM jsrun, or
AcceleratorCudaInit: b) invokes through a wrapping script to set CUDA_VISIBLE_DEVICES, UCX_NET_DEVICES, and numa binding
AcceleratorCudaInit: Configure options --enable-summit, --enable-select-gpu=no
AcceleratorCudaInit: ================================================
OPENMPI detected
AcceleratorCudaInit: using default device
AcceleratorCudaInit: assume user either uses a) IBM jsrun, or
AcceleratorCudaInit: b) invokes through a wrapping script to set CUDA_VISIBLE_DEVICES, UCX_NET_DEVICES, and numa binding
AcceleratorCudaInit: Configure options --enable-summit, --enable-select-gpu=no
AcceleratorCudaInit: ================================================
OPENMPI detected
AcceleratorCudaInit[0]: ========================
AcceleratorCudaInit[0]: Device Number : 0
AcceleratorCudaInit[0]: ========================
@ -73,12 +67,18 @@ AcceleratorCudaInit: b) invokes through a wrapping script to set CUDA_VISIBLE_DE
AcceleratorCudaInit: Configure options --enable-summit, --enable-select-gpu=no
AcceleratorCudaInit: ================================================
OPENMPI detected
AcceleratorCudaInit: using default device
AcceleratorCudaInit: assume user either uses a) IBM jsrun, or
AcceleratorCudaInit: b) invokes through a wrapping script to set CUDA_VISIBLE_DEVICES, UCX_NET_DEVICES, and numa binding
AcceleratorCudaInit: Configure options --enable-summit, --enable-select-gpu=no
AcceleratorCudaInit: ================================================
OPENMPI detected
AcceleratorCudaInit: using default device
AcceleratorCudaInit: assume user either uses a) IBM jsrun, or
AcceleratorCudaInit: b) invokes through a wrapping script to set CUDA_VISIBLE_DEVICES, UCX_NET_DEVICES, and numa binding
AcceleratorCudaInit: Configure options --enable-summit, --enable-select-gpu=no
AcceleratorCudaInit: ================================================
OPENMPI detected
AcceleratorCudaInit: using default device
AcceleratorCudaInit: assume user either uses a) IBM jsrun, or
AcceleratorCudaInit: b) invokes through a wrapping script to set CUDA_VISIBLE_DEVICES, UCX_NET_DEVICES, and numa binding
@ -86,7 +86,7 @@ AcceleratorCudaInit: Configure options --enable-summit, --enable-select-gpu=no
AcceleratorCudaInit: ================================================
SharedMemoryMpi: World communicator of size 16
SharedMemoryMpi: Node communicator of size 4
0SharedMemoryMpi: SharedMemoryMPI.cc acceleratorAllocDevice 2147483648bytes at 0x7f84e0000000 for comms buffers
0SharedMemoryMpi: SharedMemoryMPI.cc acceleratorAllocDevice 2147483648bytes at 0x7fcd80000000 for comms buffers
Setting up IPC
__|__|__|__|__|__|__|__|__|__|__|__|__|__|__
@ -127,119 +127,119 @@ Grid : Message : MemoryManager::Init() setting up
Grid : Message : MemoryManager::Init() cache pool for recent allocations: SMALL 32 LARGE 8
Grid : Message : MemoryManager::Init() Non unified: Caching accelerator data in dedicated memory
Grid : Message : MemoryManager::Init() Using cudaMalloc
Grid : Message : 1.234535 s : Grid Layout
Grid : Message : 1.234543 s : Global lattice size : 64 64 64 64
Grid : Message : 1.234547 s : OpenMP threads : 4
Grid : Message : 1.234548 s : MPI tasks : 2 2 2 2
Grid : Message : 1.272006 s : Making s innermost grids
Grid : Message : 1.316898 s : Initialising 4d RNG
Grid : Message : 1.423692 s : Intialising parallel RNG with unique string 'The 4D RNG'
Grid : Message : 1.423718 s : Seed SHA256: 49db4542db694e3b1a74bf2592a8c1b83bfebbe18401693c2609a4c3af1
Grid : Message : 1.851400 s : Initialising 5d RNG
Grid : Message : 3.495243 s : Intialising parallel RNG with unique string 'The 5D RNG'
Grid : Message : 3.495282 s : Seed SHA256: b6316f2fac44ce14111f93e0296389330b077bfd0a7b359f781c58589f8a
Grid : Message : 11.929748 s : Initialised RNGs
Grid : Message : 15.135728 s : Drawing gauge field
Grid : Message : 16.109867 s : Random gauge initialised
Grid : Message : 17.570374 s : Setting up Cshift based reference
Grid : Message : 26.641932 s : *****************************************************************
Grid : Message : 26.641961 s : * Kernel options --dslash-generic, --dslash-unroll, --dslash-asm
Grid : Message : 26.641963 s : *****************************************************************
Grid : Message : 26.641965 s : *****************************************************************
Grid : Message : 26.641967 s : * Benchmarking DomainWallFermionR::Dhop
Grid : Message : 26.641968 s : * Vectorising space-time by 8
Grid : Message : 26.641970 s : * VComplexF size is 64 B
Grid : Message : 26.641974 s : * SINGLE precision
Grid : Message : 26.641976 s : * Using Overlapped Comms/Compute
Grid : Message : 26.641977 s : * Using GENERIC Nc WilsonKernels
Grid : Message : 26.641978 s : *****************************************************************
Grid : Message : 28.676542 s : Called warmup
Grid : Message : 56.574545 s : Called Dw 3000 times in 2.78974e+07 us
Grid : Message : 56.574570 s : mflop/s = 3.8104e+07
Grid : Message : 56.574573 s : mflop/s per rank = 2.3815e+06
Grid : Message : 56.574575 s : mflop/s per node = 9.526e+06
Grid : Message : 56.574577 s : RF GiB/s (base 2) = 77426.5
Grid : Message : 56.574579 s : mem GiB/s (base 2) = 48391.5
Grid : Message : 56.578174 s : norm diff 1.03481e-13
Grid : Message : 56.613049 s : #### Dhop calls report
Grid : Message : 56.613053 s : WilsonFermion5D Number of DhopEO Calls : 6002
Grid : Message : 56.613057 s : WilsonFermion5D TotalTime /Calls : 4692.42 us
Grid : Message : 56.613059 s : WilsonFermion5D CommTime /Calls : 3204.73 us
Grid : Message : 56.613061 s : WilsonFermion5D FaceTime /Calls : 489.82 us
Grid : Message : 56.613063 s : WilsonFermion5D ComputeTime1/Calls : 41.1853 us
Grid : Message : 56.613065 s : WilsonFermion5D ComputeTime2/Calls : 1014.2 us
Grid : Message : 56.613073 s : Average mflops/s per call : 3.56434e+09
Grid : Message : 56.613079 s : Average mflops/s per call per rank : 2.22771e+08
Grid : Message : 56.613083 s : Average mflops/s per call per node : 8.91085e+08
Grid : Message : 56.613087 s : Average mflops/s per call (full) : 3.84425e+07
Grid : Message : 56.613091 s : Average mflops/s per call per rank (full): 2.40266e+06
Grid : Message : 56.613104 s : Average mflops/s per call per node (full): 9.61063e+06
Grid : Message : 56.613106 s : WilsonFermion5D Stencil
Grid : Message : 56.613164 s : Stencil calls 3001
Grid : Message : 56.613168 s : Stencil halogtime 0
Grid : Message : 56.613170 s : Stencil gathertime 45.0563
Grid : Message : 56.613172 s : Stencil gathermtime 19.5342
Grid : Message : 56.613174 s : Stencil mergetime 18.4625
Grid : Message : 56.613176 s : Stencil decompresstime 0.0616461
Grid : Message : 56.613178 s : Stencil comms_bytes 4.02653e+08
Grid : Message : 56.613180 s : Stencil commtime 6345.15
Grid : Message : 56.613182 s : Stencil 63.4584 GB/s per rank
Grid : Message : 56.613184 s : Stencil 253.834 GB/s per node
Grid : Message : 56.613186 s : WilsonFermion5D StencilEven
Grid : Message : 56.613193 s : WilsonFermion5D StencilOdd
Grid : Message : 56.613205 s : WilsonFermion5D Stencil Reporti()
Grid : Message : 56.613211 s : WilsonFermion5D StencilEven Reporti()
Grid : Message : 56.613214 s : WilsonFermion5D StencilOdd Reporti()
Grid : Message : 79.610634 s : Compare to naive wilson implementation Dag to verify correctness
Grid : Message : 79.610656 s : Called DwDag
Grid : Message : 79.610657 s : norm dag result 12.0421
Grid : Message : 79.623325 s : norm dag ref 12.0421
Grid : Message : 79.639282 s : norm dag diff 7.63236e-14
Grid : Message : 79.679130 s : Calling Deo and Doe and //assert Deo+Doe == Dunprec
Grid : Message : 80.544440 s : src_e0.499997
Grid : Message : 80.384797 s : src_o0.500003
Grid : Message : 80.485472 s : *********************************************************
Grid : Message : 80.485477 s : * Benchmarking DomainWallFermionF::DhopEO
Grid : Message : 80.485478 s : * Vectorising space-time by 8
Grid : Message : 80.485480 s : * SINGLE precision
Grid : Message : 80.485481 s : * Using Overlapped Comms/Compute
Grid : Message : 80.485482 s : * Using GENERIC Nc WilsonKernels
Grid : Message : 80.485483 s : *********************************************************
Grid : Message : 94.142389 s : Deo mflop/s = 3.91983e+07
Grid : Message : 94.142423 s : Deo mflop/s per rank 2.4499e+06
Grid : Message : 94.142425 s : Deo mflop/s per node 9.79959e+06
Grid : Message : 94.142427 s : #### Dhop calls report
Grid : Message : 94.142428 s : WilsonFermion5D Number of DhopEO Calls : 3001
Grid : Message : 94.142430 s : WilsonFermion5D TotalTime /Calls : 4550.65 us
Grid : Message : 94.142432 s : WilsonFermion5D CommTime /Calls : 2986.58 us
Grid : Message : 94.142434 s : WilsonFermion5D FaceTime /Calls : 607.616 us
Grid : Message : 94.142436 s : WilsonFermion5D ComputeTime1/Calls : 58.1666 us
Grid : Message : 94.142438 s : WilsonFermion5D ComputeTime2/Calls : 998.146 us
Grid : Message : 94.142460 s : Average mflops/s per call : 3.01763e+09
Grid : Message : 94.142464 s : Average mflops/s per call per rank : 1.88602e+08
Grid : Message : 94.142466 s : Average mflops/s per call per node : 7.54407e+08
Grid : Message : 94.142468 s : Average mflops/s per call (full) : 3.96402e+07
Grid : Message : 94.142473 s : Average mflops/s per call per rank (full): 2.47751e+06
Grid : Message : 94.142478 s : Average mflops/s per call per node (full): 9.91005e+06
Grid : Message : 94.142481 s : WilsonFermion5D Stencil
Grid : Message : 94.142492 s : WilsonFermion5D StencilEven
Grid : Message : 94.142504 s : WilsonFermion5D StencilOdd
Grid : Message : 94.142515 s : Stencil calls 3001
Grid : Message : 94.142519 s : Stencil halogtime 0
Grid : Message : 94.142522 s : Stencil gathertime 54.0447
Grid : Message : 94.142525 s : Stencil gathermtime 19.4485
Grid : Message : 94.142530 s : Stencil mergetime 18.3146
Grid : Message : 94.142532 s : Stencil decompresstime 0.0593136
Grid : Message : 94.142535 s : Stencil comms_bytes 2.01327e+08
Grid : Message : 94.142539 s : Stencil commtime 2989.1
Grid : Message : 94.142542 s : Stencil 67.3536 GB/s per rank
Grid : Message : 94.142546 s : Stencil 269.414 GB/s per node
Grid : Message : 94.142548 s : WilsonFermion5D Stencil Reporti()
Grid : Message : 94.142551 s : WilsonFermion5D StencilEven Reporti()
Grid : Message : 94.142553 s : WilsonFermion5D StencilOdd Reporti()
Grid : Message : 94.216038 s : r_e6.02111
Grid : Message : 94.223150 s : r_o6.02102
Grid : Message : 94.229148 s : res12.0421
Grid : Message : 94.830824 s : norm diff 0
Grid : Message : 95.633245 s : norm diff even 0
Grid : Message : 95.999452 s : norm diff odd 0
Grid : Message : 1.198523 s : Grid Layout
Grid : Message : 1.198530 s : Global lattice size : 64 64 64 64
Grid : Message : 1.198534 s : OpenMP threads : 4
Grid : Message : 1.198535 s : MPI tasks : 2 2 2 2
Grid : Message : 1.397615 s : Making s innermost grids
Grid : Message : 1.441828 s : Initialising 4d RNG
Grid : Message : 1.547973 s : Intialising parallel RNG with unique string 'The 4D RNG'
Grid : Message : 1.547998 s : Seed SHA256: 49db4542db694e3b1a74bf2592a8c1b83bfebbe18401693c2609a4c3af1
Grid : Message : 1.954777 s : Initialising 5d RNG
Grid : Message : 3.633825 s : Intialising parallel RNG with unique string 'The 5D RNG'
Grid : Message : 3.633869 s : Seed SHA256: b6316f2fac44ce14111f93e0296389330b077bfd0a7b359f781c58589f8a
Grid : Message : 12.162710 s : Initialised RNGs
Grid : Message : 15.882520 s : Drawing gauge field
Grid : Message : 15.816362 s : Random gauge initialised
Grid : Message : 17.279671 s : Setting up Cshift based reference
Grid : Message : 26.331426 s : *****************************************************************
Grid : Message : 26.331452 s : * Kernel options --dslash-generic, --dslash-unroll, --dslash-asm
Grid : Message : 26.331454 s : *****************************************************************
Grid : Message : 26.331456 s : *****************************************************************
Grid : Message : 26.331458 s : * Benchmarking DomainWallFermionR::Dhop
Grid : Message : 26.331459 s : * Vectorising space-time by 8
Grid : Message : 26.331463 s : * VComplexF size is 64 B
Grid : Message : 26.331465 s : * SINGLE precision
Grid : Message : 26.331467 s : * Using Overlapped Comms/Compute
Grid : Message : 26.331468 s : * Using GENERIC Nc WilsonKernels
Grid : Message : 26.331469 s : *****************************************************************
Grid : Message : 28.413717 s : Called warmup
Grid : Message : 56.418423 s : Called Dw 3000 times in 2.80047e+07 us
Grid : Message : 56.418476 s : mflop/s = 3.79581e+07
Grid : Message : 56.418479 s : mflop/s per rank = 2.37238e+06
Grid : Message : 56.418481 s : mflop/s per node = 9.48953e+06
Grid : Message : 56.418483 s : RF GiB/s (base 2) = 77130
Grid : Message : 56.418485 s : mem GiB/s (base 2) = 48206.3
Grid : Message : 56.422076 s : norm diff 1.03481e-13
Grid : Message : 56.456894 s : #### Dhop calls report
Grid : Message : 56.456899 s : WilsonFermion5D Number of DhopEO Calls : 6002
Grid : Message : 56.456903 s : WilsonFermion5D TotalTime /Calls : 4710.93 us
Grid : Message : 56.456905 s : WilsonFermion5D CommTime /Calls : 3196.15 us
Grid : Message : 56.456908 s : WilsonFermion5D FaceTime /Calls : 494.392 us
Grid : Message : 56.456910 s : WilsonFermion5D ComputeTime1/Calls : 44.4107 us
Grid : Message : 56.456912 s : WilsonFermion5D ComputeTime2/Calls : 1037.75 us
Grid : Message : 56.456921 s : Average mflops/s per call : 3.55691e+09
Grid : Message : 56.456925 s : Average mflops/s per call per rank : 2.22307e+08
Grid : Message : 56.456928 s : Average mflops/s per call per node : 8.89228e+08
Grid : Message : 56.456930 s : Average mflops/s per call (full) : 3.82915e+07
Grid : Message : 56.456933 s : Average mflops/s per call per rank (full): 2.39322e+06
Grid : Message : 56.456952 s : Average mflops/s per call per node (full): 9.57287e+06
Grid : Message : 56.456954 s : WilsonFermion5D Stencil
Grid : Message : 56.457016 s : Stencil calls 3001
Grid : Message : 56.457022 s : Stencil halogtime 0
Grid : Message : 56.457024 s : Stencil gathertime 55.9154
Grid : Message : 56.457026 s : Stencil gathermtime 20.1073
Grid : Message : 56.457028 s : Stencil mergetime 18.5585
Grid : Message : 56.457030 s : Stencil decompresstime 0.0639787
Grid : Message : 56.457032 s : Stencil comms_bytes 4.02653e+08
Grid : Message : 56.457034 s : Stencil commtime 6379.93
Grid : Message : 56.457036 s : Stencil 63.1124 GB/s per rank
Grid : Message : 56.457038 s : Stencil 252.45 GB/s per node
Grid : Message : 56.457040 s : WilsonFermion5D StencilEven
Grid : Message : 56.457048 s : WilsonFermion5D StencilOdd
Grid : Message : 56.457062 s : WilsonFermion5D Stencil Reporti()
Grid : Message : 56.457065 s : WilsonFermion5D StencilEven Reporti()
Grid : Message : 56.457066 s : WilsonFermion5D StencilOdd Reporti()
Grid : Message : 79.259261 s : Compare to naive wilson implementation Dag to verify correctness
Grid : Message : 79.259287 s : Called DwDag
Grid : Message : 79.259288 s : norm dag result 12.0421
Grid : Message : 79.271740 s : norm dag ref 12.0421
Grid : Message : 79.287759 s : norm dag diff 7.63236e-14
Grid : Message : 79.328100 s : Calling Deo and Doe and //assert Deo+Doe == Dunprec
Grid : Message : 79.955951 s : src_e0.499997
Grid : Message : 80.633620 s : src_o0.500003
Grid : Message : 80.164163 s : *********************************************************
Grid : Message : 80.164168 s : * Benchmarking DomainWallFermionF::DhopEO
Grid : Message : 80.164170 s : * Vectorising space-time by 8
Grid : Message : 80.164172 s : * SINGLE precision
Grid : Message : 80.164174 s : * Using Overlapped Comms/Compute
Grid : Message : 80.164177 s : * Using GENERIC Nc WilsonKernels
Grid : Message : 80.164178 s : *********************************************************
Grid : Message : 93.797635 s : Deo mflop/s = 3.93231e+07
Grid : Message : 93.797670 s : Deo mflop/s per rank 2.45769e+06
Grid : Message : 93.797672 s : Deo mflop/s per node 9.83077e+06
Grid : Message : 93.797674 s : #### Dhop calls report
Grid : Message : 93.797675 s : WilsonFermion5D Number of DhopEO Calls : 3001
Grid : Message : 93.797677 s : WilsonFermion5D TotalTime /Calls : 4542.83 us
Grid : Message : 93.797679 s : WilsonFermion5D CommTime /Calls : 2978.97 us
Grid : Message : 93.797681 s : WilsonFermion5D FaceTime /Calls : 602.287 us
Grid : Message : 93.797683 s : WilsonFermion5D ComputeTime1/Calls : 67.1416 us
Grid : Message : 93.797685 s : WilsonFermion5D ComputeTime2/Calls : 1004.07 us
Grid : Message : 93.797713 s : Average mflops/s per call : 3.30731e+09
Grid : Message : 93.797717 s : Average mflops/s per call per rank : 2.06707e+08
Grid : Message : 93.797719 s : Average mflops/s per call per node : 8.26827e+08
Grid : Message : 93.797721 s : Average mflops/s per call (full) : 3.97084e+07
Grid : Message : 93.797727 s : Average mflops/s per call per rank (full): 2.48178e+06
Grid : Message : 93.797732 s : Average mflops/s per call per node (full): 9.92711e+06
Grid : Message : 93.797735 s : WilsonFermion5D Stencil
Grid : Message : 93.797746 s : WilsonFermion5D StencilEven
Grid : Message : 93.797758 s : WilsonFermion5D StencilOdd
Grid : Message : 93.797769 s : Stencil calls 3001
Grid : Message : 93.797773 s : Stencil halogtime 0
Grid : Message : 93.797776 s : Stencil gathertime 56.7458
Grid : Message : 93.797780 s : Stencil gathermtime 22.6504
Grid : Message : 93.797782 s : Stencil mergetime 21.1913
Grid : Message : 93.797786 s : Stencil decompresstime 0.0556481
Grid : Message : 93.797788 s : Stencil comms_bytes 2.01327e+08
Grid : Message : 93.797791 s : Stencil commtime 2989.33
Grid : Message : 93.797795 s : Stencil 67.3484 GB/s per rank
Grid : Message : 93.797798 s : Stencil 269.394 GB/s per node
Grid : Message : 93.797801 s : WilsonFermion5D Stencil Reporti()
Grid : Message : 93.797803 s : WilsonFermion5D StencilEven Reporti()
Grid : Message : 93.797805 s : WilsonFermion5D StencilOdd Reporti()
Grid : Message : 93.873429 s : r_e6.02111
Grid : Message : 93.879931 s : r_o6.02102
Grid : Message : 93.885912 s : res12.0421
Grid : Message : 94.876555 s : norm diff 0
Grid : Message : 95.485643 s : norm diff even 0
Grid : Message : 95.581236 s : norm diff odd 0

33
systems/Tursa/dwf16.slurm Normal file
View File

@ -0,0 +1,33 @@
#!/bin/bash
#SBATCH -J dslash
#SBATCH -A tc002
#SBATCH -t 2:20:00
#SBATCH --exclusive
#SBATCH --nodes=16
#SBATCH --ntasks=64
#SBATCH --ntasks-per-node=4
#SBATCH --cpus-per-task=8
#SBATCH --time=12:00:00
#SBATCH --partition=gpu
#SBATCH --gres=gpu:4
#SBATCH --output=%x.%j.out
#SBATCH --error=%x.%j.err
export OMP_NUM_THREADS=4
export OMPI_MCA_btl=^uct,openib
export UCX_TLS=gdr_copy,rc,rc_x,sm,cuda_copy,cuda_ipc
export UCX_RNDV_SCHEME=put_zcopy
export UCX_RNDV_THRESH=16384
export UCX_IB_GPU_DIRECT_RDMA=yes
export UCX_MEMTYPE_CACHE=n
OPT="--comms-overlap --comms-concurrent"
mpirun -np $SLURM_NTASKS -x LD_LIBRARY_PATH --bind-to none ./mpiwrapper.sh \
./benchmarks/Benchmark_dwf_fp32 \
$OPT \
--mpi 2.2.2.8 \
--accelerator-threads 8 \
--grid 64.64.64.256 \
--shm 2048 > dwf.16node.perf

2
systems/Tursa/dwf.slurm → systems/Tursa/dwf4.slurm
View File

@ -31,7 +31,7 @@ mpirun -np $SLURM_NTASKS -x LD_LIBRARY_PATH --bind-to none \
--mpi 2.2.2.2 \
--accelerator-threads 8 \
--grid 64.64.64.64 \
--shm 2048 > dwf.perf
--shm 2048 > dwf.4node.perf