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

Summit up to date

This commit is contained in:
Peter Boyle 2022-09-27 10:58:43 -04:00
parent 48165c1dc1
commit fad2f969d9
5 changed files with 112 additions and 497 deletions

View File

@ -1,179 +0,0 @@
OPENMPI detected
AcceleratorCudaInit[0]: ========================
AcceleratorCudaInit[0]: Device Number : 0
AcceleratorCudaInit[0]: ========================
AcceleratorCudaInit[0]: Device identifier: Tesla V100-SXM2-16GB
AcceleratorCudaInit[0]: totalGlobalMem: 16911433728
AcceleratorCudaInit[0]: managedMemory: 1
AcceleratorCudaInit[0]: isMultiGpuBoard: 0
AcceleratorCudaInit[0]: warpSize: 32
AcceleratorCudaInit[0]: pciBusID: 4
AcceleratorCudaInit[0]: pciDeviceID: 0
AcceleratorCudaInit[0]: maxGridSize (2147483647,65535,65535)
AcceleratorCudaInit: rank 0 setting device to node rank 0
AcceleratorCudaInit: Configure options --enable-setdevice=yes
local rank 0 device 0 bus id: 0004:04:00.0
AcceleratorCudaInit: ================================================
SharedMemoryMpi: World communicator of size 24
SharedMemoryMpi: Node communicator of size 6
0SharedMemoryMpi: SharedMemoryMPI.cc acceleratorAllocDevice 1073741824bytes at 0x200060000000 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=7cb1ff7395a5833ded6526c43891bd07a0436290: (HEAD -> develop, origin/develop, origin/HEAD) clean
Grid : Message : ================================================
Grid : Message : MPI is initialised and logging filters activated
Grid : Message : ================================================
Grid : Message : Requested 1073741824 byte stencil comms buffers
AcceleratorCudaInit: rank 1 setting device to node rank 1
AcceleratorCudaInit: Configure options --enable-setdevice=yes
local rank 1 device 1 bus id: 0004:05:00.0
AcceleratorCudaInit: rank 2 setting device to node rank 2
AcceleratorCudaInit: Configure options --enable-setdevice=yes
local rank 2 device 2 bus id: 0004:06:00.0
AcceleratorCudaInit: rank 5 setting device to node rank 5
AcceleratorCudaInit: Configure options --enable-setdevice=yes
local rank 5 device 5 bus id: 0035:05:00.0
AcceleratorCudaInit: rank 4 setting device to node rank 4
AcceleratorCudaInit: Configure options --enable-setdevice=yes
local rank 4 device 4 bus id: 0035:04:00.0
AcceleratorCudaInit: rank 3 setting device to node rank 3
AcceleratorCudaInit: Configure options --enable-setdevice=yes
local rank 3 device 3 bus id: 0035:03:00.0
Grid : Message : MemoryManager Cache 13529146982 bytes
Grid : Message : MemoryManager::Init() setting up
Grid : Message : MemoryManager::Init() cache pool for recent allocations: SMALL 8 LARGE 2
Grid : Message : MemoryManager::Init() Non unified: Caching accelerator data in dedicated memory
Grid : Message : MemoryManager::Init() Using cudaMalloc
Grid : Message : 2.137929 s : Grid is setup to use 6 threads
Grid : Message : 2.137941 s : Number of iterations to average: 250
Grid : Message : 2.137950 s : ====================================================================================================
Grid : Message : 2.137958 s : = Benchmarking sequential halo exchange from host memory
Grid : Message : 2.137966 s : ====================================================================================================
Grid : Message : 2.137974 s : L Ls bytes MB/s uni MB/s bidi
AcceleratorCudaInit: rank 22 setting device to node rank 4
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 10 setting device to node rank 4
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 15 setting device to node rank 3
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 21 setting device to node rank 3
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 20 setting device to node rank 2
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 7 setting device to node rank 1
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 9 setting device to node rank 3
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 11 setting device to node rank 5
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 8 setting device to node rank 2
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 6 setting device to node rank 0
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 19 setting device to node rank 1
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 23 setting device to node rank 5
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 18 setting device to node rank 0
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 12 setting device to node rank 0
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 16 setting device to node rank 4
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 13 setting device to node rank 1
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 14 setting device to node rank 2
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 17 setting device to node rank 5
AcceleratorCudaInit: Configure options --enable-setdevice=yes
Grid : Message : 2.604949 s : 8 8 393216 89973.9 179947.8
Grid : Message : 2.668249 s : 8 8 393216 18650.3 37300.5
Grid : Message : 2.732288 s : 8 8 393216 18428.5 36857.1
Grid : Message : 2.753565 s : 8 8 393216 55497.2 110994.4
Grid : Message : 2.808960 s : 12 8 1327104 100181.5 200363.0
Grid : Message : 3.226900 s : 12 8 1327104 20600.5 41201.0
Grid : Message : 3.167459 s : 12 8 1327104 24104.6 48209.2
Grid : Message : 3.227660 s : 12 8 1327104 66156.7 132313.5
Grid : Message : 3.413570 s : 16 8 3145728 56174.4 112348.8
Grid : Message : 3.802697 s : 16 8 3145728 24255.9 48511.7
Grid : Message : 4.190498 s : 16 8 3145728 24336.7 48673.4
Grid : Message : 4.385171 s : 16 8 3145728 48484.1 96968.2
Grid : Message : 4.805284 s : 20 8 6144000 46380.5 92761.1
Grid : Message : 5.562975 s : 20 8 6144000 24328.5 48656.9
Grid : Message : 6.322562 s : 20 8 6144000 24266.7 48533.4
Grid : Message : 6.773598 s : 20 8 6144000 40868.5 81736.9
Grid : Message : 7.600999 s : 24 8 10616832 40198.3 80396.6
Grid : Message : 8.912917 s : 24 8 10616832 24279.5 48559.1
Grid : Message : 10.220961 s : 24 8 10616832 24350.2 48700.4
Grid : Message : 11.728250 s : 24 8 10616832 37390.9 74781.8
Grid : Message : 12.497258 s : 28 8 16859136 36792.2 73584.5
Grid : Message : 14.585387 s : 28 8 16859136 24222.2 48444.3
Grid : Message : 16.664783 s : 28 8 16859136 24323.4 48646.8
Grid : Message : 17.955238 s : 28 8 16859136 39194.7 78389.4
Grid : Message : 20.136479 s : 32 8 25165824 35718.3 71436.5
Grid : Message : 23.241958 s : 32 8 25165824 24311.4 48622.9
Grid : Message : 26.344810 s : 32 8 25165824 24331.9 48663.7
Grid : Message : 28.384420 s : 32 8 25165824 37016.3 74032.7
Grid : Message : 28.388879 s : ====================================================================================================
Grid : Message : 28.388894 s : = Benchmarking sequential halo exchange from GPU memory
Grid : Message : 28.388909 s : ====================================================================================================
Grid : Message : 28.388924 s : L Ls bytes MB/s uni MB/s bidi
Grid : Message : 28.553993 s : 8 8 393216 8272.4 16544.7
Grid : Message : 28.679592 s : 8 8 393216 9395.4 18790.8
Grid : Message : 28.811112 s : 8 8 393216 8971.0 17942.0
Grid : Message : 28.843770 s : 8 8 393216 36145.6 72291.2
Grid : Message : 28.981754 s : 12 8 1327104 49591.6 99183.2
Grid : Message : 29.299764 s : 12 8 1327104 12520.8 25041.7
Grid : Message : 29.620288 s : 12 8 1327104 12422.2 24844.4
Grid : Message : 29.657645 s : 12 8 1327104 106637.5 213275.1
Grid : Message : 29.952933 s : 16 8 3145728 43939.2 87878.5
Grid : Message : 30.585411 s : 16 8 3145728 14922.1 29844.2
Grid : Message : 31.219781 s : 16 8 3145728 14877.2 29754.4
Grid : Message : 31.285017 s : 16 8 3145728 144724.3 289448.7
Grid : Message : 31.706443 s : 20 8 6144000 54676.2 109352.4
Grid : Message : 32.739205 s : 20 8 6144000 17848.0 35696.1
Grid : Message : 33.771852 s : 20 8 6144000 17849.9 35699.7
Grid : Message : 33.871981 s : 20 8 6144000 184141.4 368282.8
Grid : Message : 34.536808 s : 24 8 10616832 55784.3 111568.6
Grid : Message : 36.275648 s : 24 8 10616832 18317.6 36635.3
Grid : Message : 37.997181 s : 24 8 10616832 18501.7 37003.4
Grid : Message : 38.140442 s : 24 8 10616832 222383.9 444767.9
Grid : Message : 39.177222 s : 28 8 16859136 56609.7 113219.4
Grid : Message : 41.874755 s : 28 8 16859136 18749.9 37499.8
Grid : Message : 44.529381 s : 28 8 16859136 19052.9 38105.8
Grid : Message : 44.742192 s : 28 8 16859136 237717.1 475434.2
Grid : Message : 46.184000 s : 32 8 25165824 57091.2 114182.4
Grid : Message : 50.734740 s : 32 8 25165824 19411.0 38821.9
Grid : Message : 53.931228 s : 32 8 25165824 19570.6 39141.2
Grid : Message : 54.238467 s : 32 8 25165824 245765.6 491531.2
Grid : Message : 54.268664 s : ====================================================================================================
Grid : Message : 54.268680 s : = All done; Bye Bye
Grid : Message : 54.268691 s : ====================================================================================================

View File

@ -3,7 +3,7 @@
--enable-gen-simd-width=32 \ --enable-gen-simd-width=32 \
--enable-unified=no \ --enable-unified=no \
--enable-shm=no \ --enable-shm=no \
--disable-gparity \ --enable-gparity \
--disable-setdevice \ --disable-setdevice \
--disable-fermion-reps \ --disable-fermion-reps \
--enable-accelerator=cuda \ --enable-accelerator=cuda \

View File

@ -10,19 +10,16 @@ AcceleratorCudaInit[0]: warpSize: 32
AcceleratorCudaInit[0]: pciBusID: 4 AcceleratorCudaInit[0]: pciBusID: 4
AcceleratorCudaInit[0]: pciDeviceID: 0 AcceleratorCudaInit[0]: pciDeviceID: 0
AcceleratorCudaInit[0]: maxGridSize (2147483647,65535,65535) AcceleratorCudaInit[0]: maxGridSize (2147483647,65535,65535)
AcceleratorCudaInit: rank 0 setting device to node rank 0 AcceleratorCudaInit: using default device
AcceleratorCudaInit: Configure options --enable-setdevice=yes AcceleratorCudaInit: assume user either uses
AcceleratorCudaInit: 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-setdevice=no
local rank 0 device 0 bus id: 0004:04:00.0 local rank 0 device 0 bus id: 0004:04:00.0
AcceleratorCudaInit: ================================================ AcceleratorCudaInit: ================================================
SharedMemoryMpi: World communicator of size 24 SharedMemoryMpi: World communicator of size 24
SharedMemoryMpi: Node communicator of size 6 SharedMemoryMpi: Node communicator of size 1
0SharedMemoryMpi: SharedMemoryMPI.cc acceleratorAllocDevice 2147483648bytes at 0x200080000000 for comms buffers 0SharedMemoryMpi: SharedMemoryMPI.cc acceleratorAllocDevice 1073741824bytes at 0x200080000000 - 2000bfffffff for comms buffers
AcceleratorCudaInit: rank 3 setting device to node rank 3
AcceleratorCudaInit: Configure options --enable-setdevice=yes
local rank 3 device 3 bus id: 0035:03:00.0
AcceleratorCudaInit: rank 5 setting device to node rank 5
AcceleratorCudaInit: Configure options --enable-setdevice=yes
local rank 5 device 5 bus id: 0035:05:00.0
Setting up IPC Setting up IPC
__|__|__|__|__|__|__|__|__|__|__|__|__|__|__ __|__|__|__|__|__|__|__|__|__|__|__|__|__|__
@ -36,6 +33,11 @@ __|_ G GG RRRR I D D _|__
__|_ G G R R I D D _|__ __|_ G G R R I D D _|__
__|_ GGGG R R III DDDD _|__ __|_ GGGG R R III DDDD _|__
__|_ _|__ __|_ _|__
local rank 5 device 0 bus id: 0035:05:00.0
local rank 1 device 0 bus id: 0004:05:00.0
local rank 2 device 0 bus id: 0004:06:00.0
local rank 3 device 0 bus id: 0035:03:00.0
local rank 4 device 0 bus id: 0035:04:00.0
__|__|__|__|__|__|__|__|__|__|__|__|__|__|__ __|__|__|__|__|__|__|__|__|__|__|__|__|__|__
__|__|__|__|__|__|__|__|__|__|__|__|__|__|__ __|__|__|__|__|__|__|__|__|__|__|__|__|__|__
| | | | | | | | | | | | | | | | | | | | | | | | | | | |
@ -45,15 +47,6 @@ Copyright (C) 2015 Peter Boyle, Azusa Yamaguchi, Guido Cossu, Antonin Portelli a
This program is free software; you can redistribute it and/or modify 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 it under the terms of the GNU General Public License as published by
AcceleratorCudaInit: rank 4 setting device to node rank 4
AcceleratorCudaInit: Configure options --enable-setdevice=yes
local rank 4 device 4 bus id: 0035:04:00.0
AcceleratorCudaInit: rank 1 setting device to node rank 1
AcceleratorCudaInit: Configure options --enable-setdevice=yes
local rank 1 device 1 bus id: 0004:05:00.0
AcceleratorCudaInit: rank 2 setting device to node rank 2
AcceleratorCudaInit: Configure options --enable-setdevice=yes
local rank 2 device 2 bus id: 0004:06:00.0
the Free Software Foundation; either version 2 of the License, or the Free Software Foundation; either version 2 of the License, or
(at your option) any later version. (at your option) any later version.
@ -61,146 +54,63 @@ This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details. GNU General Public License for more details.
Current Grid git commit hash=7cb1ff7395a5833ded6526c43891bd07a0436290: (HEAD -> develop, origin/develop, origin/HEAD) clean Current Grid git commit hash=1713de35c0dc339564661dd7df8a72583f889e91: (HEAD -> feature/dirichlet) uncommited changes
Grid : Message : ================================================ Grid : Message : ================================================
Grid : Message : MPI is initialised and logging filters activated Grid : Message : MPI is initialised and logging filters activated
Grid : Message : ================================================ Grid : Message : ================================================
Grid : Message : Requested 2147483648 byte stencil comms buffers Grid : Message : Requested 1073741824 byte stencil comms buffers
Grid : Message : MemoryManager Cache 8388608000 bytes Grid : Message : MemoryManager Cache 4194304000 bytes
Grid : Message : MemoryManager::Init() setting up Grid : Message : MemoryManager::Init() setting up
Grid : Message : MemoryManager::Init() cache pool for recent allocations: SMALL 8 LARGE 2 Grid : Message : MemoryManager::Init() cache pool for recent allocations: SMALL 8 LARGE 2
Grid : Message : MemoryManager::Init() Non unified: Caching accelerator data in dedicated memory Grid : Message : MemoryManager::Init() Non unified: Caching accelerator data in dedicated memory
Grid : Message : MemoryManager::Init() Using cudaMalloc Grid : Message : MemoryManager::Init() Using cudaMalloc
Grid : Message : 1.731905 s : Grid Layout
Grid : Message : 1.731915 s : Global lattice size : 48 48 48 72
Grid : Message : 1.731928 s : OpenMP threads : 6
Grid : Message : 1.731938 s : MPI tasks : 2 2 2 3
AcceleratorCudaInit: rank 9 setting device to node rank 3
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 23 setting device to node rank 5
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 0.179000 s : ++++++++++++++++++++++++++++++++++++++++++++++++
AcceleratorCudaInit: rank 22 setting device to node rank 4 Grid : Message : 0.196000 s : Testing with full communication
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 0.211000 s : ++++++++++++++++++++++++++++++++++++++++++++++++
AcceleratorCudaInit: rank 21 setting device to node rank 3 Grid : Message : 0.225000 s : Grid Layout
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 0.233000 s : Global lattice size : 48 48 48 72
AcceleratorCudaInit: rank 18 setting device to node rank 0 Grid : Message : 0.246000 s : OpenMP threads : 6
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 0.255000 s : MPI tasks : 2 2 2 3
AcceleratorCudaInit: rank 6 setting device to node rank 0 Grid : Message : 0.182200 s : Initialising 4d RNG
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 0.233863 s : Intialising parallel RNG with unique string 'The 4D RNG'
AcceleratorCudaInit: rank 7 setting device to node rank 1 Grid : Message : 0.233886 s : Seed SHA256: 49db4542db694e3b1a74bf2592a8c1b83bfebbe18401693c2609a4c3af1
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 0.245805 s : Initialising 5d RNG
AcceleratorCudaInit: rank 10 setting device to node rank 4 Grid : Message : 1.710720 s : Intialising parallel RNG with unique string 'The 5D RNG'
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 1.710950 s : Seed SHA256: b6316f2fac44ce14111f93e0296389330b077bfd0a7b359f781c58589f8a
AcceleratorCudaInit: rank 8 setting device to node rank 2 Grid : Message : 2.220272 s : Drawing gauge field
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 2.418119 s : Random gauge initialised
AcceleratorCudaInit: rank 11 setting device to node rank 5 Grid : Message : 2.418142 s : Applying BCs for Dirichlet Block5 [0 0 0 0 0]
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 2.418156 s : Applying BCs for Dirichlet Block4 [0 0 0 0]
AcceleratorCudaInit: rank 20 setting device to node rank 2 Grid : Message : 2.489588 s : Setting up Cshift based reference
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 13.921239 s : *****************************************************************
AcceleratorCudaInit: rank 19 setting device to node rank 1 Grid : Message : 13.921261 s : * Kernel options --dslash-generic, --dslash-unroll, --dslash-asm
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 13.921270 s : *****************************************************************
AcceleratorCudaInit: rank 13 setting device to node rank 1 Grid : Message : 13.921279 s : *****************************************************************
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 13.921288 s : * Benchmarking DomainWallFermionR::Dhop
AcceleratorCudaInit: rank 12 setting device to node rank 0 Grid : Message : 13.921296 s : * Vectorising space-time by 4
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 13.921305 s : * VComplexF size is 32 B
AcceleratorCudaInit: rank 14 setting device to node rank 2 Grid : Message : 13.921314 s : * SINGLE precision
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 13.921321 s : * Using Overlapped Comms/Compute
AcceleratorCudaInit: rank 16 setting device to node rank 4 Grid : Message : 13.921328 s : * Using GENERIC Nc WilsonKernels
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 13.921335 s : *****************************************************************
AcceleratorCudaInit: rank 15 setting device to node rank 3 Grid : Message : 14.821339 s : Called warmup
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 23.975467 s : Called Dw 300 times in 9.15155e+06 us
AcceleratorCudaInit: rank 17 setting device to node rank 5 Grid : Message : 23.975528 s : mflop/s = 5.51286e+06
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 23.975543 s : mflop/s per rank = 229702
Grid : Message : 2.683494 s : Making s innermost grids Grid : Message : 23.975557 s : mflop/s per node = 229702
Grid : Message : 2.780034 s : Initialising 4d RNG Grid : Message : 23.989684 s : norm diff 5.09279e-313 Line 291
Grid : Message : 2.833099 s : Intialising parallel RNG with unique string 'The 4D RNG' Grid : Message : 39.450493 s : ----------------------------------------------------------------
Grid : Message : 2.833121 s : Seed SHA256: 49db4542db694e3b1a74bf2592a8c1b83bfebbe18401693c2609a4c3af1 Grid : Message : 39.450517 s : Compare to naive wilson implementation Dag to verify correctness
Grid : Message : 2.916841 s : Initialising 5d RNG Grid : Message : 39.450526 s : ----------------------------------------------------------------
Grid : Message : 3.762880 s : Intialising parallel RNG with unique string 'The 5D RNG' Grid : Message : 39.450534 s : Called DwDag
Grid : Message : 3.762902 s : Seed SHA256: b6316f2fac44ce14111f93e0296389330b077bfd0a7b359f781c58589f8a Grid : Message : 39.450542 s : norm dag result nan
Grid : Message : 5.264345 s : Initialised RNGs Grid : Message : 39.451564 s : norm dag ref nan
Grid : Message : 6.489904 s : Drawing gauge field Grid : Message : 39.455714 s : norm dag diff nan Line 354
Grid : Message : 6.729262 s : Random gauge initialised
Grid : Message : 7.781273 s : Setting up Cshift based reference
Grid : Message : 8.725313 s : *****************************************************************
Grid : Message : 8.725332 s : * Kernel options --dslash-generic, --dslash-unroll, --dslash-asm
Grid : Message : 8.725342 s : *****************************************************************
Grid : Message : 8.725352 s : *****************************************************************
Grid : Message : 8.725362 s : * Benchmarking DomainWallFermionR::Dhop
Grid : Message : 8.725372 s : * Vectorising space-time by 4
Grid : Message : 8.725383 s : * VComplexF size is 32 B
Grid : Message : 8.725395 s : * SINGLE precision
Grid : Message : 8.725405 s : * Using Overlapped Comms/Compute
Grid : Message : 8.725415 s : * Using GENERIC Nc WilsonKernels
Grid : Message : 8.725425 s : *****************************************************************
Grid : Message : 9.465229 s : Called warmup
Grid : Message : 58.646066 s : Called Dw 3000 times in 4.91764e+07 us
Grid : Message : 58.646121 s : mflop/s = 1.02592e+07
Grid : Message : 58.646134 s : mflop/s per rank = 427468
Grid : Message : 58.646145 s : mflop/s per node = 2.56481e+06
Grid : Message : 58.646156 s : RF GiB/s (base 2) = 20846.5
Grid : Message : 58.646166 s : mem GiB/s (base 2) = 13029.1
Grid : Message : 58.648008 s : norm diff 1.04778e-13
Grid : Message : 58.734885 s : #### Dhop calls report
Grid : Message : 58.734897 s : WilsonFermion5D Number of DhopEO Calls : 6002
Grid : Message : 58.734909 s : WilsonFermion5D TotalTime /Calls : 8217.71 us
Grid : Message : 58.734922 s : WilsonFermion5D CommTime /Calls : 7109.5 us
Grid : Message : 58.734933 s : WilsonFermion5D FaceTime /Calls : 446.623 us
Grid : Message : 58.734943 s : WilsonFermion5D ComputeTime1/Calls : 18.0558 us
Grid : Message : 58.734953 s : WilsonFermion5D ComputeTime2/Calls : 731.097 us
Grid : Message : 58.734979 s : Average mflops/s per call : 4.8157e+09
Grid : Message : 58.734989 s : Average mflops/s per call per rank : 2.00654e+08
Grid : Message : 58.734999 s : Average mflops/s per call per node : 1.20393e+09
Grid : Message : 58.735008 s : Average mflops/s per call (full) : 1.04183e+07
Grid : Message : 58.735017 s : Average mflops/s per call per rank (full): 434094
Grid : Message : 58.735026 s : Average mflops/s per call per node (full): 2.60456e+06
Grid : Message : 58.735035 s : WilsonFermion5D Stencil
Grid : Message : 58.735043 s : WilsonFermion5D StencilEven
Grid : Message : 58.735051 s : WilsonFermion5D StencilOdd
Grid : Message : 58.735059 s : WilsonFermion5D Stencil Reporti()
Grid : Message : 58.735067 s : WilsonFermion5D StencilEven Reporti()
Grid : Message : 58.735075 s : WilsonFermion5D StencilOdd Reporti()
Grid : Message : 64.934380 s : Compare to naive wilson implementation Dag to verify correctness
Grid : Message : 64.934740 s : Called DwDag
Grid : Message : 64.934870 s : norm dag result 12.0422
Grid : Message : 64.120756 s : norm dag ref 12.0422
Grid : Message : 64.149389 s : norm dag diff 7.6644e-14
Grid : Message : 64.317786 s : Calling Deo and Doe and //assert Deo+Doe == Dunprec
Grid : Message : 64.465331 s : src_e0.499995
Grid : Message : 64.524653 s : src_o0.500005
Grid : Message : 64.558706 s : *********************************************************
Grid : Message : 64.558717 s : * Benchmarking DomainWallFermionF::DhopEO
Grid : Message : 64.558727 s : * Vectorising space-time by 4
Grid : Message : 64.558737 s : * SINGLE precision
Grid : Message : 64.558745 s : * Using Overlapped Comms/Compute
Grid : Message : 64.558753 s : * Using GENERIC Nc WilsonKernels
Grid : Message : 64.558761 s : *********************************************************
Grid : Message : 92.702145 s : Deo mflop/s = 8.97692e+06
Grid : Message : 92.702185 s : Deo mflop/s per rank 374038
Grid : Message : 92.702198 s : Deo mflop/s per node 2.24423e+06
Grid : Message : 92.702209 s : #### Dhop calls report
Grid : Message : 92.702223 s : WilsonFermion5D Number of DhopEO Calls : 3001
Grid : Message : 92.702240 s : WilsonFermion5D TotalTime /Calls : 9377.88 us
Grid : Message : 92.702257 s : WilsonFermion5D CommTime /Calls : 8221.84 us
Grid : Message : 92.702277 s : WilsonFermion5D FaceTime /Calls : 543.548 us
Grid : Message : 92.702301 s : WilsonFermion5D ComputeTime1/Calls : 20.936 us
Grid : Message : 92.702322 s : WilsonFermion5D ComputeTime2/Calls : 732.33 us
Grid : Message : 92.702376 s : Average mflops/s per call : 4.13001e+09
Grid : Message : 92.702387 s : Average mflops/s per call per rank : 1.72084e+08
Grid : Message : 92.702397 s : Average mflops/s per call per node : 1.0325e+09
Grid : Message : 92.702407 s : Average mflops/s per call (full) : 9.12937e+06
Grid : Message : 92.702416 s : Average mflops/s per call per rank (full): 380391
Grid : Message : 92.702426 s : Average mflops/s per call per node (full): 2.28234e+06
Grid : Message : 92.702435 s : WilsonFermion5D Stencil
Grid : Message : 92.702443 s : WilsonFermion5D StencilEven
Grid : Message : 92.702451 s : WilsonFermion5D StencilOdd
Grid : Message : 92.702459 s : WilsonFermion5D Stencil Reporti()
Grid : Message : 92.702467 s : WilsonFermion5D StencilEven Reporti()
Grid : Message : 92.702475 s : WilsonFermion5D StencilOdd Reporti()
Grid : Message : 92.772983 s : r_e6.02121
Grid : Message : 92.786384 s : r_o6.02102
Grid : Message : 92.799622 s : res12.0422
Grid : Message : 93.860500 s : norm diff 0
Grid : Message : 93.162026 s : norm diff even 0
Grid : Message : 93.197529 s : norm diff odd 0

View File

@ -10,14 +10,21 @@ AcceleratorCudaInit[0]: warpSize: 32
AcceleratorCudaInit[0]: pciBusID: 4 AcceleratorCudaInit[0]: pciBusID: 4
AcceleratorCudaInit[0]: pciDeviceID: 0 AcceleratorCudaInit[0]: pciDeviceID: 0
AcceleratorCudaInit[0]: maxGridSize (2147483647,65535,65535) AcceleratorCudaInit[0]: maxGridSize (2147483647,65535,65535)
AcceleratorCudaInit: rank 0 setting device to node rank 0 AcceleratorCudaInit: using default device
AcceleratorCudaInit: Configure options --enable-setdevice=yes AcceleratorCudaInit: assume user either uses
AcceleratorCudaInit: 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-setdevice=no
local rank 0 device 0 bus id: 0004:04:00.0 local rank 0 device 0 bus id: 0004:04:00.0
AcceleratorCudaInit: ================================================ AcceleratorCudaInit: ================================================
SharedMemoryMpi: World communicator of size 24 SharedMemoryMpi: World communicator of size 24
SharedMemoryMpi: Node communicator of size 6 SharedMemoryMpi: Node communicator of size 1
0SharedMemoryMpi: SharedMemoryMPI.cc acceleratorAllocDevice 2147483648bytes at 0x200080000000 for comms buffers local rank 3 device 0 bus id: 0004:04:00.0
local rank 2 device 0 bus id: 0004:04:00.0
local rank 1 device 0 bus id: 0004:04:00.0
0SharedMemoryMpi: SharedMemoryMPI.cc acceleratorAllocDevice 1073741824bytes at 0x200080000000 - 2000bfffffff for comms buffers
Setting up IPC Setting up IPC
local rank 5 device 0 bus id: 0004:04:00.0
__|__|__|__|__|__|__|__|__|__|__|__|__|__|__ __|__|__|__|__|__|__|__|__|__|__|__|__|__|__
__|__|__|__|__|__|__|__|__|__|__|__|__|__|__ __|__|__|__|__|__|__|__|__|__|__|__|__|__|__
@ -39,168 +46,46 @@ Copyright (C) 2015 Peter Boyle, Azusa Yamaguchi, Guido Cossu, Antonin Portelli a
This program is free software; you can redistribute it and/or modify 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 it under the terms of the GNU General Public License as published by
local rank 4 device 0 bus id: 0004:04:00.0
the Free Software Foundation; either version 2 of the License, or the Free Software Foundation; either version 2 of the License, or
(at your option) any later version. (at your option) any later version.
This program is distributed in the hope that it will be useful, This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
AcceleratorCudaInit: rank 2 setting device to node rank 2
AcceleratorCudaInit: Configure options --enable-setdevice=yes
local rank 2 device 2 bus id: 0004:06:00.0
AcceleratorCudaInit: rank 1 setting device to node rank 1
AcceleratorCudaInit: Configure options --enable-setdevice=yes
local rank 1 device 1 bus id: 0004:05:00.0
AcceleratorCudaInit: rank 4 setting device to node rank 4
AcceleratorCudaInit: Configure options --enable-setdevice=yes
local rank 4 device 4 bus id: 0035:04:00.0
AcceleratorCudaInit: rank 3 setting device to node rank 3
AcceleratorCudaInit: Configure options --enable-setdevice=yes
local rank 3 device 3 bus id: 0035:03:00.0
AcceleratorCudaInit: rank 5 setting device to node rank 5
AcceleratorCudaInit: Configure options --enable-setdevice=yes
local rank 5 device 5 bus id: 0035:05:00.0
GNU General Public License for more details. GNU General Public License for more details.
Current Grid git commit hash=7cb1ff7395a5833ded6526c43891bd07a0436290: (HEAD -> develop, origin/develop, origin/HEAD) clean Current Grid git commit hash=1713de35c0dc339564661dd7df8a72583f889e91: (HEAD -> feature/dirichlet) uncommited changes
Grid : Message : ================================================ Grid : Message : ================================================
Grid : Message : MPI is initialised and logging filters activated Grid : Message : MPI is initialised and logging filters activated
Grid : Message : ================================================ Grid : Message : ================================================
Grid : Message : Requested 2147483648 byte stencil comms buffers Grid : Message : Requested 1073741824 byte stencil comms buffers
Grid : Message : MemoryManager Cache 8388608000 bytes
Grid : Message : MemoryManager::Init() setting up Grid : Message : MemoryManager::Init() setting up
Grid : Message : MemoryManager::Init() cache pool for recent allocations: SMALL 8 LARGE 2 Grid : Message : MemoryManager::Init() cache pool for recent allocations: SMALL 8 LARGE 2
Grid : Message : MemoryManager::Init() Non unified: Caching accelerator data in dedicated memory Grid : Message : MemoryManager::Init() Unified memory space
Grid : Message : MemoryManager::Init() Using cudaMalloc Grid : Message : MemoryManager::Init() Using cudaMallocManaged
Grid : Message : 1.544984 s : Grid Layout
Grid : Message : 1.544992 s : Global lattice size : 64 64 64 96
Grid : Message : 1.545003 s : OpenMP threads : 6
Grid : Message : 1.545011 s : MPI tasks : 2 2 2 3
AcceleratorCudaInit: rank 8 setting device to node rank 2
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 6 setting device to node rank 0
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 0.139000 s : ++++++++++++++++++++++++++++++++++++++++++++++++
AcceleratorCudaInit: rank 11 setting device to node rank 5 Grid : Message : 0.151000 s : Testing with full communication
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 0.158000 s : ++++++++++++++++++++++++++++++++++++++++++++++++
AcceleratorCudaInit: rank 16 setting device to node rank 4 Grid : Message : 0.165000 s : Grid Layout
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 0.171000 s : Global lattice size : 64 64 64 96
AcceleratorCudaInit: rank 17 setting device to node rank 5 Grid : Message : 0.181000 s : OpenMP threads : 6
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 0.189000 s : MPI tasks : 2 2 2 3
AcceleratorCudaInit: rank 13 setting device to node rank 1 Grid : Message : 0.177717 s : Initialising 4d RNG
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 0.342461 s : Intialising parallel RNG with unique string 'The 4D RNG'
AcceleratorCudaInit: rank 12 setting device to node rank 0 Grid : Message : 0.342483 s : Seed SHA256: 49db4542db694e3b1a74bf2592a8c1b83bfebbe18401693c2609a4c3af1
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 0.370454 s : Initialising 5d RNG
AcceleratorCudaInit: rank 21 setting device to node rank 3 Grid : Message : 3.174160 s : Intialising parallel RNG with unique string 'The 5D RNG'
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 3.174420 s : Seed SHA256: b6316f2fac44ce14111f93e0296389330b077bfd0a7b359f781c58589f8a
AcceleratorCudaInit: rank 23 setting device to node rank 5 Grid : Message : 22.119339 s : Drawing gauge field
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 38.113060 s : Random gauge initialised
AcceleratorCudaInit: rank 22 setting device to node rank 4 Grid : Message : 38.113320 s : Applying BCs for Dirichlet Block5 [0 0 0 0 0]
AcceleratorCudaInit: Configure options --enable-setdevice=yes Grid : Message : 38.113470 s : Applying BCs for Dirichlet Block4 [0 0 0 0]
AcceleratorCudaInit: rank 19 setting device to node rank 1 Grid : Message : 43.906786 s : Setting up Cshift based reference
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 18 setting device to node rank 0
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 7 setting device to node rank 1
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 10 setting device to node rank 4
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 9 setting device to node rank 3
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 14 setting device to node rank 2
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 15 setting device to node rank 3
AcceleratorCudaInit: Configure options --enable-setdevice=yes
AcceleratorCudaInit: rank 20 setting device to node rank 2
AcceleratorCudaInit: Configure options --enable-setdevice=yes
Grid : Message : 2.994920 s : Making s innermost grids
Grid : Message : 2.232502 s : Initialising 4d RNG
Grid : Message : 2.397047 s : Intialising parallel RNG with unique string 'The 4D RNG'
Grid : Message : 2.397069 s : Seed SHA256: 49db4542db694e3b1a74bf2592a8c1b83bfebbe18401693c2609a4c3af1
Grid : Message : 2.653140 s : Initialising 5d RNG
Grid : Message : 5.285347 s : Intialising parallel RNG with unique string 'The 5D RNG'
Grid : Message : 5.285369 s : Seed SHA256: b6316f2fac44ce14111f93e0296389330b077bfd0a7b359f781c58589f8a
Grid : Message : 9.994738 s : Initialised RNGs
Grid : Message : 13.153426 s : Drawing gauge field
Grid : Message : 13.825697 s : Random gauge initialised
Grid : Message : 18.537657 s : Setting up Cshift based reference
Grid : Message : 22.296755 s : *****************************************************************
Grid : Message : 22.296781 s : * Kernel options --dslash-generic, --dslash-unroll, --dslash-asm
Grid : Message : 22.296791 s : *****************************************************************
Grid : Message : 22.296800 s : *****************************************************************
Grid : Message : 22.296809 s : * Benchmarking DomainWallFermionR::Dhop
Grid : Message : 22.296818 s : * Vectorising space-time by 4
Grid : Message : 22.296828 s : * VComplexF size is 32 B
Grid : Message : 22.296838 s : * SINGLE precision
Grid : Message : 22.296847 s : * Using Overlapped Comms/Compute
Grid : Message : 22.296855 s : * Using GENERIC Nc WilsonKernels
Grid : Message : 22.296863 s : *****************************************************************
Grid : Message : 24.746452 s : Called warmup
Grid : Message : 137.525756 s : Called Dw 3000 times in 1.12779e+08 us
Grid : Message : 137.525818 s : mflop/s = 1.41383e+07
Grid : Message : 137.525831 s : mflop/s per rank = 589097
Grid : Message : 137.525843 s : mflop/s per node = 3.53458e+06
Grid : Message : 137.525854 s : RF GiB/s (base 2) = 28728.7
Grid : Message : 137.525864 s : mem GiB/s (base 2) = 17955.5
Grid : Message : 137.693645 s : norm diff 1.04885e-13
Grid : Message : 137.965585 s : #### Dhop calls report
Grid : Message : 137.965598 s : WilsonFermion5D Number of DhopEO Calls : 6002
Grid : Message : 137.965612 s : WilsonFermion5D TotalTime /Calls : 18899.7 us
Grid : Message : 137.965624 s : WilsonFermion5D CommTime /Calls : 16041.4 us
Grid : Message : 137.965634 s : WilsonFermion5D FaceTime /Calls : 859.705 us
Grid : Message : 137.965644 s : WilsonFermion5D ComputeTime1/Calls : 70.5881 us
Grid : Message : 137.965654 s : WilsonFermion5D ComputeTime2/Calls : 2094.8 us
Grid : Message : 137.965682 s : Average mflops/s per call : 3.87638e+09
Grid : Message : 137.965692 s : Average mflops/s per call per rank : 1.61516e+08
Grid : Message : 137.965702 s : Average mflops/s per call per node : 9.69095e+08
Grid : Message : 137.965712 s : Average mflops/s per call (full) : 1.43168e+07
Grid : Message : 137.965721 s : Average mflops/s per call per rank (full): 596533
Grid : Message : 137.965730 s : Average mflops/s per call per node (full): 3.5792e+06
Grid : Message : 137.965740 s : WilsonFermion5D Stencil
Grid : Message : 137.965748 s : WilsonFermion5D StencilEven
Grid : Message : 137.965756 s : WilsonFermion5D StencilOdd
Grid : Message : 137.965764 s : WilsonFermion5D Stencil Reporti()
Grid : Message : 137.965772 s : WilsonFermion5D StencilEven Reporti()
Grid : Message : 137.965780 s : WilsonFermion5D StencilOdd Reporti()
Grid : Message : 156.554605 s : Compare to naive wilson implementation Dag to verify correctness
Grid : Message : 156.554632 s : Called DwDag
Grid : Message : 156.554642 s : norm dag result 12.0421
Grid : Message : 156.639265 s : norm dag ref 12.0421
Grid : Message : 156.888281 s : norm dag diff 7.62057e-14
Grid : Message : 157.609797 s : Calling Deo and Doe and //assert Deo+Doe == Dunprec
Grid : Message : 158.208630 s : src_e0.499996
Grid : Message : 158.162447 s : src_o0.500004
Grid : Message : 158.267780 s : *********************************************************
Grid : Message : 158.267791 s : * Benchmarking DomainWallFermionF::DhopEO
Grid : Message : 158.267801 s : * Vectorising space-time by 4
Grid : Message : 158.267811 s : * SINGLE precision
Grid : Message : 158.267820 s : * Using Overlapped Comms/Compute
Grid : Message : 158.267828 s : * Using GENERIC Nc WilsonKernels
Grid : Message : 158.267836 s : *********************************************************
Grid : Message : 216.487829 s : Deo mflop/s = 1.37283e+07
Grid : Message : 216.487869 s : Deo mflop/s per rank 572011
Grid : Message : 216.487881 s : Deo mflop/s per node 3.43206e+06
Grid : Message : 216.487893 s : #### Dhop calls report
Grid : Message : 216.487903 s : WilsonFermion5D Number of DhopEO Calls : 3001
Grid : Message : 216.487913 s : WilsonFermion5D TotalTime /Calls : 19399.6 us
Grid : Message : 216.487923 s : WilsonFermion5D CommTime /Calls : 16475.4 us
Grid : Message : 216.487933 s : WilsonFermion5D FaceTime /Calls : 972.393 us
Grid : Message : 216.487943 s : WilsonFermion5D ComputeTime1/Calls : 49.8474 us
Grid : Message : 216.487953 s : WilsonFermion5D ComputeTime2/Calls : 2089.93 us
Grid : Message : 216.488001 s : Average mflops/s per call : 5.39682e+09
Grid : Message : 216.488011 s : Average mflops/s per call per rank : 2.24867e+08
Grid : Message : 216.488020 s : Average mflops/s per call per node : 1.3492e+09
Grid : Message : 216.488030 s : Average mflops/s per call (full) : 1.39479e+07
Grid : Message : 216.488039 s : Average mflops/s per call per rank (full): 581162
Grid : Message : 216.488048 s : Average mflops/s per call per node (full): 3.48697e+06
Grid : Message : 216.488057 s : WilsonFermion5D Stencil
Grid : Message : 216.488065 s : WilsonFermion5D StencilEven
Grid : Message : 216.488073 s : WilsonFermion5D StencilOdd
Grid : Message : 216.488081 s : WilsonFermion5D Stencil Reporti()
Grid : Message : 216.488089 s : WilsonFermion5D StencilEven Reporti()
Grid : Message : 216.488097 s : WilsonFermion5D StencilOdd Reporti()
Grid : Message : 217.384495 s : r_e6.02113
Grid : Message : 217.426121 s : r_o6.02096
Grid : Message : 217.472636 s : res12.0421
Grid : Message : 218.200068 s : norm diff 0
Grid : Message : 218.645673 s : norm diff even 0
Grid : Message : 218.816561 s : norm diff odd 0

View File

@ -7,16 +7,15 @@
export OMP_NUM_THREADS=6 export OMP_NUM_THREADS=6
export PAMI_IBV_ADAPTER_AFFINITY=1 export PAMI_IBV_ADAPTER_AFFINITY=1
export PAMI_ENABLE_STRIPING=1 export PAMI_ENABLE_STRIPING=1
export PAMI_DISABLE_IPC=1
export OPT="--comms-concurrent --comms-overlap " export OPT="--comms-concurrent --comms-overlap "
#export GRID_ALLOC_NCACHE_LARGE=1
export APP="./benchmarks/Benchmark_comms_host_device --mpi 2.2.2.3 "
jsrun --nrs 4 -a6 -g6 -c42 -dpacked -b packed:7 --latency_priority gpu-cpu --smpiargs=-gpu $APP > comms.4node
APP="./benchmarks/Benchmark_dwf_fp32 --grid 48.48.48.72 --mpi 2.2.2.3 --shm 2048 --shm-force-mpi 1 --device-mem 8000 --shm-force-mpi 1 $OPT "
jsrun --nrs 4 -a6 -g6 -c42 -dpacked -b packed:7 --latency_priority gpu-cpu --smpiargs=-gpu $APP > dwf.24.4node
APP="./benchmarks/Benchmark_dwf_fp32 --grid 64.64.64.96 --mpi 2.2.2.3 --shm 2048 --shm-force-mpi 1 --device-mem 8000 --shm-force-mpi 1 $OPT " APP="./wrap.sh ./benchmarks/Benchmark_dwf_fp32 --grid 48.48.48.72 --mpi 2.2.2.3 --shm 1024 --device-mem 4000 --shm-force-mpi 1 $OPT "
jsrun --nrs 4 -a6 -g6 -c42 -dpacked -b packed:7 --latency_priority gpu-cpu --smpiargs=-gpu $APP > dwf.32.4node jsrun --nrs 24 -a1 -g1 -c6 -dpacked -b packed:6 --latency_priority gpu-cpu --smpiargs="-gpu" $APP > dwf.24.4node
APP="./wrap.sh ./benchmarks/Benchmark_comms_host_device --grid 48.48.48.72 --mpi 2.2.2.3 --shm 1024 --device-mem 4000 --shm-force-mpi 1 $OPT "
jsrun --smpiargs="-gpu" --nrs 4 -a6 -g6 -c42 -dpacked -b packed:6 $APP > comms.24.4node