As discussed with Carsten Bauer on the Julia Slack, I try to follow the tutorial https://github.com/RRZE-HPC/likwid/wiki/Tutorial%3A-Empirical-Roofline-Model in Julia. However, operational intensities vary significantly between several runs.
I followed the Julia setup at https://juliaperf.github.io/LIKWID.jl/stable/marker/#Example. A slightly modified version below serves as MWE.
| # This file is machine-generated - editing it directly is not advised | |
| julia_version = "1.7.0-rc3" | |
| manifest_format = "2.0" | |
| [[deps.AbstractFFTs]] | |
| deps = ["LinearAlgebra"] | |
| git-tree-sha1 = "485ee0867925449198280d4af84bdb46a2a404d0" | |
| uuid = "621f4979-c628-5d54-868e-fcf4e3e8185c" | |
| version = "1.0.1" |
| ISM-V2 | |
| 289 544 256 1 | |
| -10.0000000000000 -10.0000000000000 | |
| -8.75000000000000 -10.0000000000000 | |
| -7.50000000000000 -10.0000000000000 | |
| -6.25000000000000 -10.0000000000000 | |
| -5.00000000000000 -10.0000000000000 | |
| -3.75000000000000 -10.0000000000000 | |
| -2.50000000000000 -10.0000000000000 | |
| -1.25000000000000 -10.0000000000000 |
| using Plots | |
| #= | |
| results obtained from running | |
| for polydeg in 1:7 | |
| @show polydeg | |
| tabulate_benchmarks(benchmark_euler, levels=0:8, polydeg=polydeg) | |
| end | |
| with https://github.com/jlchan/ESDG.jl/blob/refactor/Trixi_comparison/benchmark_JC/benchmark_euler2D.jl |
| #!/usr/bin/env julia | |
| # Benchmark some common RHS evaluations on a uniform 2D mesh in Trixi.jl | |
| using Pkg | |
| Pkg.activate(temp=true) | |
| Pkg.add("BenchmarkTools") | |
| Pkg.add("Trixi") | |
| using LinearAlgebra | |
| BLAS.set_num_threads(Threads.nthreads()) |
| // compile and run with | |
| // gcc -I../include -O2 -Wall sgemv.c -o sgemv -lOpenCL -L . -lclblast; ./sgemv | |
| // in the build directory of CLBlast | |
| #include <stdlib.h> | |
| #include <stdio.h> | |
| #include <string.h> | |
| #define CL_USE_DEPRECATED_OPENCL_1_2_APIS // to disable deprecation warnings | |
| #include <clblast_c.h> |
| [ 2%] Built target test_correctness_common | |
| [ 46%] Built target clblast | |
| [ 47%] Built target clblast_test_xgemmbatched | |
| [ 48%] Built target clblast_test_xspr | |
| [ 48%] Built target clblast_test_xsymm | |
| [ 50%] Built target clblast_test_xspmv | |
| [ 51%] Built target clblast_test_xscal | |
| [ 52%] Built target clblast_test_xtrmv | |
| [ 53%] Built target clblast_test_xcopy | |
| [ 55%] Built target clblast_test_xgeru |