Closed wilfonba closed 1 month ago
The benchmarking "passes" even though this is the result:
Run . ./mfc.sh load -c p -m g
mfc: Loading modules (& env variables) for GT Phoenix on GPUs:
mfc: $ module load python/3.9.12-rkxvr6
mfc: $ module load cmake/3.23.1-327dbl
mfc: $ module load cuda/11.7.0-7sdye3
mfc: $ module load nvhpc/22.11
mfc: $ export MFC_CUDA_CC=70,80
mfc: $ export CC=nvc
mfc: $ export CXX=nvc++
mfc: $ export FC=nvfortran
mfc: OK > All modules and environment variables have been loaded.
mfc: OK > (venv) Entered the Python 3.9.12 virtual environment (>= 3.8).
.=++*: -+*+=. | sbryngelson3@login-phoenix-slurm-4.pace.gatech.edu [Linux]
:+ -*- == =* . | ----------------------------------------------------------
:*+ == ++ .+- | --jobs 1
:*##-.....:*+ .#%+++=--+=:::. | --mpi
-=-++-======#=--**+++==+*++=::-:. | --no-gpu
.:++=----------====+*= ==..:%..... | --no-debug
.:-=++++===--==+=-+= +. := | --targets pre_process, simulation, and post_process
+#=::::::::=%=. -+: =+ *: | ----------------------------------------------------------
.*=-=*=.. :=+*+: -...-- | $ ./mfc.sh (build, run, test, clean, count, packer) --help
Comparing Benchmarks:
1.[5](https://github.com/MFlowCode/MFC/actions/runs/9081893351/job/24956905746#step:5:6)x indicates pr/bench-cpu.yaml is 1.5-times as fast as master/bench-cpu.yaml (so pr/bench-cpu.yaml is faster than master/bench-cpu.yaml).
0.5x indicates pr/bench-cpu.yaml is 0.5-times as fast as master/bench-cpu.yaml (so pr/bench-cpu.yaml is slower than master/bench-cpu.yaml).
Case Pre Process Simulation Post Process
──────────────────────────────────────────────────────────────────
viscous_weno5_sgb_mono 1.00x N/A N/A
5eq_rk3_weno3_hllc 1.00x N/A N/A
ibm 2.00x N/A N/A
hypo_hll 0.50x N/A N/A
I opened an issue for this https://github.com/MFlowCode/MFC/issues/415
Description
Fixes a performance regression on Frontier between the optimized code at the head of @abbotts fork of MFC and the current master branch.
Type of change
Scope
How Has This Been Tested?
Ran the test suite on Frontier
Checklist
If your code changes any code source files (anything in
src/simulation
)To make sure the code is performing as expected on GPU devices, I have:
./mfc.sh run XXXX --gpu -t simulation --omniperf
, and have attached the output file and plain text results to this PR.