Added osu_accumulate and osu_get_accumulate benchmark

mpiexecjl -n 2 julia --project -e 'using MPIBenchmarks; benchmark(OSUAccumulateLatency())'

Running benchmark OSU Accumulate Latency with type Float32 on 2 MPI ranks with window synchronization: lock


size (bytes),iterations,time (seconds)
0,262144,4.2557439883239567e-7
4,262144,7.505532266804948e-7
8,262144,7.353046385105699e-7
16,262144,8.963224900071509e-7
32,262144,8.514280125382356e-7
64,262144,8.501310730935074e-7
128,262144,8.633387551526539e-7
256,262144,1.3205008144723251e-6
512,262144,1.4137822290649638e-6
1024,262144,1.1022139005945064e-6
2048,131072,1.3272365322336555e-6
4096,65536,1.8548853404354304e-6
8192,32768,2.80239328276366e-6
16384,16384,4.678906407207251e-6
32768,8192,7.670663762837648e-6
65536,4096,1.6773934476077557e-5
131072,2048,2.6918831281363964e-5
262144,1024,4.532025195658207e-5
524288,512,8.638901636004448e-5
1048576,256,0.00023242272436618805

mpiexecjl -n 2 julia --project -e 'using MPIBenchmarks; benchmark(OSUAccumulateLatency(; synchronization_option = "fence"))'

Running benchmark OSU Accumulate Latency with type Float32 on 2 MPI ranks with window synchronization: fence

size (bytes),iterations,time (seconds)
0,262144,1.7035054042935371e-6
4,262144,2.0018487703055143e-6
8,262144,1.9923036234104075e-6
16,262144,2.1017522158217616e-6
32,262144,2.106946340063587e-6
64,262144,2.2691619960824028e-6
128,262144,2.1824926079716533e-6
256,262144,2.1935620679869317e-6
512,262144,2.307415343238972e-6
1024,262144,2.4210612536990084e-6
2048,131072,2.8457288863137364e-6
4096,65536,3.2039606594480574e-6
8192,32768,5.109548510517925e-6
16384,16384,5.495123332366347e-6
32768,8192,8.472357876598835e-6
65536,4096,1.758075086399913e-5
131072,2048,3.30568291246891e-5
262144,1024,0.00011241133324801922
524288,512,0.00016682641580700874
1048576,256,0.0004568202421069145

mpiexecjl -n 2 julia --project -e 'using MPIBenchmarks; benchmark(OSUGetAccumulateLatency())'

Running benchmark OSU Get Accumulate Latency with type Float32 on 2 MPI ranks with window synchronization: lock


size (bytes),iterations,time (seconds)
0,262144,4.0541362977819517e-7
4,262144,7.935223038657568e-7
8,262144,8.085598892648704e-7
16,262144,8.599718057666905e-7
32,262144,9.329878594144247e-7
64,262144,8.676952347741462e-7
128,262144,9.018449418363161e-7
256,262144,9.466557457926683e-7
512,262144,1.0367020877311006e-6
1024,262144,1.1986585377599113e-6
2048,131072,1.6608446458121762e-6
4096,65536,2.4620239855721593e-6
8192,32768,4.1051607695408165e-6
16384,16384,7.180002285167575e-6
32768,8192,1.164109562523663e-5
65536,4096,2.3475266061723232e-5
131072,2048,4.280440043658018e-5
262144,1024,7.905182428658009e-5
524288,512,0.00015766825526952744
1048576,256,0.0003815107047557831

mpiexecjl -n 2 julia --project -e 'using MPIBenchmarks; #benchmark(OSUGetAccumulateLatency(; synchronization_option = "fence"))'

Running benchmark OSU Get Accumulate Latency with type Float32 on 2 MPI ranks with window synchronization: fence

size (bytes),iterations,time (seconds)
0,262144,1.7923339328262955e-6
4,262144,2.316199243068695e-6
8,262144,2.433920599287376e-6
16,262144,2.4982991817523725e-6
32,262144,2.3783932192600332e-6
64,262144,2.3855409381212667e-6
128,262144,2.3913289624033496e-6
256,262144,2.4216651581809856e-6
512,262144,2.8349531930871308e-6
1024,262144,2.8990361897740513e-6
2048,131072,3.442642992013134e-6
4096,65536,5.094829248264432e-6
8192,32768,6.671085429843515e-6
16384,16384,7.578099030070007e-6
32768,8192,1.308973878622055e-5
65536,4096,2.8241949621587992e-5
131072,2048,5.411496385931969e-5
262144,1024,0.00014675199054181576
524288,512,0.0003619040362536907
1048576,256,0.0008867727592587471

JuliaParallel / MPIBenchmarks.jl

Added osu_accumulate and osu_get_accumulate benchmark #8