From fad2f969d9ec25ccaf028ee4aca65874a3f778b2 Mon Sep 17 00:00:00 2001 From: Peter Boyle Date: Tue, 27 Sep 2022 10:58:43 -0400 Subject: [PATCH] Summit up to date --- systems/Summit/comms.4node | 179 ---------------------------- systems/Summit/config-command | 2 +- systems/Summit/dwf.24.4node | 218 ++++++++++------------------------ systems/Summit/dwf.32.4node | 197 +++++++----------------------- systems/Summit/dwf4.lsf | 13 +- 5 files changed, 112 insertions(+), 497 deletions(-) delete mode 100644 systems/Summit/comms.4node diff --git a/systems/Summit/comms.4node b/systems/Summit/comms.4node deleted file mode 100644 index b0df0801..00000000 --- a/systems/Summit/comms.4node +++ /dev/null @@ -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 : ==================================================================================================== diff --git a/systems/Summit/config-command b/systems/Summit/config-command index 4caf652e..2a856be0 100644 --- a/systems/Summit/config-command +++ b/systems/Summit/config-command @@ -3,7 +3,7 @@ --enable-gen-simd-width=32 \ --enable-unified=no \ --enable-shm=no \ - --disable-gparity \ + --enable-gparity \ --disable-setdevice \ --disable-fermion-reps \ --enable-accelerator=cuda \ diff --git a/systems/Summit/dwf.24.4node b/systems/Summit/dwf.24.4node index 212e471c..d0ca9697 100644 --- a/systems/Summit/dwf.24.4node +++ b/systems/Summit/dwf.24.4node @@ -10,19 +10,16 @@ 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 +AcceleratorCudaInit: using default device +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 AcceleratorCudaInit: ================================================ SharedMemoryMpi: World communicator of size 24 -SharedMemoryMpi: Node communicator of size 6 -0SharedMemoryMpi: SharedMemoryMPI.cc acceleratorAllocDevice 2147483648bytes at 0x200080000000 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 +SharedMemoryMpi: Node communicator of size 1 +0SharedMemoryMpi: SharedMemoryMPI.cc acceleratorAllocDevice 1073741824bytes at 0x200080000000 - 2000bfffffff for comms buffers Setting up IPC __|__|__|__|__|__|__|__|__|__|__|__|__|__|__ @@ -36,6 +33,11 @@ __|_ G GG RRRR I D D _|__ __|_ G G R R I D D _|__ __|_ 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 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 (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 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 +Current Grid git commit hash=1713de35c0dc339564661dd7df8a72583f889e91: (HEAD -> feature/dirichlet) 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 8388608000 bytes +Grid : Message : Requested 1073741824 byte stencil comms buffers +Grid : Message : MemoryManager Cache 4194304000 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 : 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 -AcceleratorCudaInit: rank 22 setting device to node rank 4 -AcceleratorCudaInit: Configure options --enable-setdevice=yes -AcceleratorCudaInit: rank 21 setting device to node rank 3 -AcceleratorCudaInit: Configure options --enable-setdevice=yes -AcceleratorCudaInit: rank 18 setting device to node rank 0 -AcceleratorCudaInit: Configure options --enable-setdevice=yes -AcceleratorCudaInit: rank 6 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 8 setting device to node rank 2 -AcceleratorCudaInit: Configure options --enable-setdevice=yes -AcceleratorCudaInit: rank 11 setting device to node rank 5 -AcceleratorCudaInit: Configure options --enable-setdevice=yes -AcceleratorCudaInit: rank 20 setting device to node rank 2 -AcceleratorCudaInit: Configure options --enable-setdevice=yes -AcceleratorCudaInit: rank 19 setting device to node rank 1 -AcceleratorCudaInit: Configure options --enable-setdevice=yes -AcceleratorCudaInit: rank 13 setting device to node rank 1 -AcceleratorCudaInit: Configure options --enable-setdevice=yes -AcceleratorCudaInit: rank 12 setting device to node rank 0 -AcceleratorCudaInit: Configure options --enable-setdevice=yes -AcceleratorCudaInit: rank 14 setting device to node rank 2 -AcceleratorCudaInit: Configure options --enable-setdevice=yes -AcceleratorCudaInit: rank 16 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 17 setting device to node rank 5 -AcceleratorCudaInit: Configure options --enable-setdevice=yes -Grid : Message : 2.683494 s : Making s innermost grids -Grid : Message : 2.780034 s : Initialising 4d RNG -Grid : Message : 2.833099 s : Intialising parallel RNG with unique string 'The 4D RNG' -Grid : Message : 2.833121 s : Seed SHA256: 49db4542db694e3b1a74bf2592a8c1b83bfebbe18401693c2609a4c3af1 -Grid : Message : 2.916841 s : Initialising 5d RNG -Grid : Message : 3.762880 s : Intialising parallel RNG with unique string 'The 5D RNG' -Grid : Message : 3.762902 s : Seed SHA256: b6316f2fac44ce14111f93e0296389330b077bfd0a7b359f781c58589f8a -Grid : Message : 5.264345 s : Initialised RNGs -Grid : Message : 6.489904 s : Drawing gauge field -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 + + + + + + + +Grid : Message : 0.179000 s : ++++++++++++++++++++++++++++++++++++++++++++++++ +Grid : Message : 0.196000 s : Testing with full communication +Grid : Message : 0.211000 s : ++++++++++++++++++++++++++++++++++++++++++++++++ +Grid : Message : 0.225000 s : Grid Layout +Grid : Message : 0.233000 s : Global lattice size : 48 48 48 72 +Grid : Message : 0.246000 s : OpenMP threads : 6 +Grid : Message : 0.255000 s : MPI tasks : 2 2 2 3 +Grid : Message : 0.182200 s : Initialising 4d RNG +Grid : Message : 0.233863 s : Intialising parallel RNG with unique string 'The 4D RNG' +Grid : Message : 0.233886 s : Seed SHA256: 49db4542db694e3b1a74bf2592a8c1b83bfebbe18401693c2609a4c3af1 +Grid : Message : 0.245805 s : Initialising 5d RNG +Grid : Message : 1.710720 s : Intialising parallel RNG with unique string 'The 5D RNG' +Grid : Message : 1.710950 s : Seed SHA256: b6316f2fac44ce14111f93e0296389330b077bfd0a7b359f781c58589f8a +Grid : Message : 2.220272 s : Drawing gauge field +Grid : Message : 2.418119 s : Random gauge initialised +Grid : Message : 2.418142 s : Applying BCs for Dirichlet Block5 [0 0 0 0 0] +Grid : Message : 2.418156 s : Applying BCs for Dirichlet Block4 [0 0 0 0] +Grid : Message : 2.489588 s : Setting up Cshift based reference +Grid : Message : 13.921239 s : ***************************************************************** +Grid : Message : 13.921261 s : * Kernel options --dslash-generic, --dslash-unroll, --dslash-asm +Grid : Message : 13.921270 s : ***************************************************************** +Grid : Message : 13.921279 s : ***************************************************************** +Grid : Message : 13.921288 s : * Benchmarking DomainWallFermionR::Dhop +Grid : Message : 13.921296 s : * Vectorising space-time by 4 +Grid : Message : 13.921305 s : * VComplexF size is 32 B +Grid : Message : 13.921314 s : * SINGLE precision +Grid : Message : 13.921321 s : * Using Overlapped Comms/Compute +Grid : Message : 13.921328 s : * Using GENERIC Nc WilsonKernels +Grid : Message : 13.921335 s : ***************************************************************** +Grid : Message : 14.821339 s : Called warmup +Grid : Message : 23.975467 s : Called Dw 300 times in 9.15155e+06 us +Grid : Message : 23.975528 s : mflop/s = 5.51286e+06 +Grid : Message : 23.975543 s : mflop/s per rank = 229702 +Grid : Message : 23.975557 s : mflop/s per node = 229702 +Grid : Message : 23.989684 s : norm diff 5.09279e-313 Line 291 +Grid : Message : 39.450493 s : ---------------------------------------------------------------- +Grid : Message : 39.450517 s : Compare to naive wilson implementation Dag to verify correctness +Grid : Message : 39.450526 s : ---------------------------------------------------------------- +Grid : Message : 39.450534 s : Called DwDag +Grid : Message : 39.450542 s : norm dag result nan +Grid : Message : 39.451564 s : norm dag ref nan +Grid : Message : 39.455714 s : norm dag diff nan Line 354 diff --git a/systems/Summit/dwf.32.4node b/systems/Summit/dwf.32.4node index eed54f2d..fe21bad8 100644 --- a/systems/Summit/dwf.32.4node +++ b/systems/Summit/dwf.32.4node @@ -10,14 +10,21 @@ 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 +AcceleratorCudaInit: using default device +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 AcceleratorCudaInit: ================================================ SharedMemoryMpi: World communicator of size 24 -SharedMemoryMpi: Node communicator of size 6 -0SharedMemoryMpi: SharedMemoryMPI.cc acceleratorAllocDevice 2147483648bytes at 0x200080000000 for comms buffers +SharedMemoryMpi: Node communicator of size 1 +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 +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 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 (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 -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. -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 : MPI is initialised and logging filters activated Grid : Message : ================================================ -Grid : Message : Requested 2147483648 byte stencil comms buffers -Grid : Message : MemoryManager Cache 8388608000 bytes +Grid : Message : Requested 1073741824 byte stencil comms buffers 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 : 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 -AcceleratorCudaInit: rank 11 setting device to node rank 5 -AcceleratorCudaInit: Configure options --enable-setdevice=yes -AcceleratorCudaInit: rank 16 setting device to node rank 4 -AcceleratorCudaInit: Configure options --enable-setdevice=yes -AcceleratorCudaInit: rank 17 setting device to node rank 5 -AcceleratorCudaInit: Configure options --enable-setdevice=yes -AcceleratorCudaInit: rank 13 setting device to node rank 1 -AcceleratorCudaInit: Configure options --enable-setdevice=yes -AcceleratorCudaInit: rank 12 setting device to node rank 0 -AcceleratorCudaInit: Configure options --enable-setdevice=yes -AcceleratorCudaInit: rank 21 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 -AcceleratorCudaInit: rank 22 setting device to node rank 4 -AcceleratorCudaInit: Configure options --enable-setdevice=yes -AcceleratorCudaInit: rank 19 setting device to node rank 1 -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 +Grid : Message : MemoryManager::Init() Unified memory space +Grid : Message : MemoryManager::Init() Using cudaMallocManaged + + + + + + + +Grid : Message : 0.139000 s : ++++++++++++++++++++++++++++++++++++++++++++++++ +Grid : Message : 0.151000 s : Testing with full communication +Grid : Message : 0.158000 s : ++++++++++++++++++++++++++++++++++++++++++++++++ +Grid : Message : 0.165000 s : Grid Layout +Grid : Message : 0.171000 s : Global lattice size : 64 64 64 96 +Grid : Message : 0.181000 s : OpenMP threads : 6 +Grid : Message : 0.189000 s : MPI tasks : 2 2 2 3 +Grid : Message : 0.177717 s : Initialising 4d RNG +Grid : Message : 0.342461 s : Intialising parallel RNG with unique string 'The 4D RNG' +Grid : Message : 0.342483 s : Seed SHA256: 49db4542db694e3b1a74bf2592a8c1b83bfebbe18401693c2609a4c3af1 +Grid : Message : 0.370454 s : Initialising 5d RNG +Grid : Message : 3.174160 s : Intialising parallel RNG with unique string 'The 5D RNG' +Grid : Message : 3.174420 s : Seed SHA256: b6316f2fac44ce14111f93e0296389330b077bfd0a7b359f781c58589f8a +Grid : Message : 22.119339 s : Drawing gauge field +Grid : Message : 38.113060 s : Random gauge initialised +Grid : Message : 38.113320 s : Applying BCs for Dirichlet Block5 [0 0 0 0 0] +Grid : Message : 38.113470 s : Applying BCs for Dirichlet Block4 [0 0 0 0] +Grid : Message : 43.906786 s : Setting up Cshift based reference diff --git a/systems/Summit/dwf4.lsf b/systems/Summit/dwf4.lsf index 7d940338..51560f71 100644 --- a/systems/Summit/dwf4.lsf +++ b/systems/Summit/dwf4.lsf @@ -7,16 +7,15 @@ export OMP_NUM_THREADS=6 export PAMI_IBV_ADAPTER_AFFINITY=1 export PAMI_ENABLE_STRIPING=1 +export PAMI_DISABLE_IPC=1 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 " -jsrun --nrs 4 -a6 -g6 -c42 -dpacked -b packed:7 --latency_priority gpu-cpu --smpiargs=-gpu $APP > dwf.32.4node +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 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