@@ -19,16 +19,23 @@ jobs:
1919 - name : nvidia-h100
2020 runner : cern-nextgen-h100
2121 cmake_args : -DENABLE_CUDA=1 -DENABLE_HIP=0 -DCUDA_COMPUTETARGET=90
22+ profiler_runs : 21
23+ standalone_runs : 42
2224 - name : nvidia-l40s
2325 runner : cern-nextgen-l40s
2426 cmake_args : -DENABLE_CUDA=1 -DENABLE_HIP=0 -DCUDA_COMPUTETARGET=89
27+ profiler_runs : 42
28+ standalone_runs : 42
2529 - name : amd-mi300x
2630 runner : cern-nextgen-mi300x
2731 cmake_args : -DENABLE_CUDA=0 -DENABLE_HIP=1 -DHIP_AMDGPUTARGET=gfx942
32+ profiler_runs : 42
33+ standalone_runs : 42
2834 - name : amd-w7900
2935 runner : cern-nextgen-w7900
3036 cmake_args : -DENABLE_CUDA=0 -DENABLE_HIP=1 -DHIP_AMDGPUTARGET=gfx1100
31-
37+ profiler_runs : 42
38+ standalone_runs : 42
3239 env :
3340 WORK_DIR : /cvmfs/alice.cern.ch
3441 ALIBUILD_ARCH_PREFIX : el9-x86_64/Packages
8794 source /etc/profile.d/modules.sh
8895 module load ninja/fortran-v1.11.1.g9-15 Vc/1.4.5-10 boost/v1.83.0-alice2-57 fmt/11.1.2-14 CMake/v3.31.6-10 ms_gsl/4.2.1-3 Clang/v20.1.7-9 TBB/v2022.3.0-3 ROOT/v6-36-04-alice9-15 ONNXRuntime/v1.22.0-71 GLFW/3.3.2-25
8996 cd ${STANDALONE_DIR}
90- ${TIMING_CA} --debug 1 --runs 42 --runsInit 2 --PROCdebugMarkdown 1 --PROCresetTimers 1 --PROCdebugCSV /root/${BENCHMARK_CSV}
97+ ${TIMING_CA} --debug 1 --runs ${{ matrix.standalone_runs }} --runsInit 2 --PROCdebugMarkdown 1 --PROCresetTimers 1 --PROCdebugCSV /root/${BENCHMARK_CSV}
9198 python3 ${GITHUB_WORKSPACE}/.github/scripts/merge_runs.py --discard 2 --input /root/${BENCHMARK_CSV} --output /root/${BENCHMARK_CSV}
9299
93100 - name : Profiler - Nsight Compute
@@ -97,10 +104,10 @@ jobs:
97104 source /etc/profile.d/modules.sh
98105 module load ninja/fortran-v1.11.1.g9-15 Vc/1.4.5-10 boost/v1.83.0-alice2-57 fmt/11.1.2-14 CMake/v3.31.6-10 ms_gsl/4.2.1-3 Clang/v20.1.7-9 TBB/v2022.3.0-3 ROOT/v6-36-04-alice9-15 ONNXRuntime/v1.22.0-71 GLFW/3.3.2-25
99106 cd ${STANDALONE_DIR}
100- ncu --set none --metrics gpu__time_duration.avg --export ${{ matrix.name }} --clock-control none --force-overwrite ${TIMING_CA} --runs 21 --debug 1 --PROCdebugMarkdown 1 # Generates ${{ matrix.name }}.ncu-rep
107+ ncu --set none --metrics gpu__time_duration.avg --export ${{ matrix.name }} --clock-control none --force-overwrite ${TIMING_CA} --runs ${{ matrix.profiler_runs }} --debug 1 --PROCdebugMarkdown 1 # Generates ${{ matrix.name }}.ncu-rep
101108 ncu --import ${STANDALONE_DIR}/${{ matrix.name }}.ncu-rep --print-units base --csv > /root/${PROFILER_CSV}
102109 rm -rf ${STANDALONE_DIR}/events/50kHz ${STANDALONE_DIR}/build
103- python3 ${GITHUB_WORKSPACE}/.github/scripts/profiler_ncu.py --runs 21 --input /root/${PROFILER_CSV} --output /root/${PROFILER_CSV}
110+ python3 ${GITHUB_WORKSPACE}/.github/scripts/profiler_ncu.py --runs ${{ matrix.profiler_runs }} --input /root/${PROFILER_CSV} --output /root/${PROFILER_CSV}
104111
105112 - name : Profiler - Nsight Systems
106113 if : ${{ matrix.name == 'nvidia-l40s' }}
@@ -110,20 +117,20 @@ jobs:
110117 source /etc/profile.d/modules.sh
111118 module load ninja/fortran-v1.11.1.g9-15 Vc/1.4.5-10 boost/v1.83.0-alice2-57 fmt/11.1.2-14 CMake/v3.31.6-10 ms_gsl/4.2.1-3 Clang/v20.1.7-9 TBB/v2022.3.0-3 ROOT/v6-36-04-alice9-15 ONNXRuntime/v1.22.0-71 GLFW/3.3.2-25
112119 cd ${STANDALONE_DIR}
113- nsys profile -o ${{ matrix.name }} ${TIMING_CA} --runs 42 --debug 1 --PROCdebugMarkdown 1 # Generates ${{ matrix.name }}.nsys-rep
120+ nsys profile -o ${{ matrix.name }} ${TIMING_CA} --runs ${{ matrix.profiler_runs }} --debug 1 --PROCdebugMarkdown 1 # Generates ${{ matrix.name }}.nsys-rep
114121 nsys stats --report cuda_gpu_kern_sum --timeunit usec --force-export=true --format csv ${{ matrix.name }}.nsys-rep > /root/${PROFILER_CSV}
115122 rm -rf ${STANDALONE_DIR}/events/50kHz ${STANDALONE_DIR}/build
116- python3 ${GITHUB_WORKSPACE}/.github/scripts/profiler_nsys.py --runs 42 --input /root/${PROFILER_CSV} --output /root/${PROFILER_CSV}
123+ python3 ${GITHUB_WORKSPACE}/.github/scripts/profiler_nsys.py --runs ${{ matrix.profiler_runs }} --input /root/${PROFILER_CSV} --output /root/${PROFILER_CSV}
117124
118125 - name : Profiler - rocprofv2
119126 if : ${{ matrix.name == 'amd-mi300x' || matrix.name == 'amd-w7900' }}
120127 run : |
121128 source /etc/profile.d/modules.sh
122129 module load ninja/fortran-v1.11.1.g9-15 Vc/1.4.5-10 boost/v1.83.0-alice2-57 fmt/11.1.2-14 CMake/v3.31.6-10 ms_gsl/4.2.1-3 Clang/v20.1.7-9 TBB/v2022.3.0-3 ROOT/v6-36-04-alice9-15 ONNXRuntime/v1.22.0-71 GLFW/3.3.2-25
123130 cd ${STANDALONE_DIR}
124- rocprofv2 --output-directory /root --output-file-name ${{ matrix.name }} ${TIMING_CA} --runs 42 --debug 1 --PROCdebugMarkdown 1 # Generates results_${{ matrix.name }}.csv == ${PROFILER_CSV}
131+ rocprofv2 --output-directory /root --output-file-name ${{ matrix.name }} ${TIMING_CA} --runs ${{ matrix.standalone_runs }} --debug 1 --PROCdebugMarkdown 1 # Generates results_${{ matrix.name }}.csv == ${PROFILER_CSV}
125132 rm -rf ${STANDALONE_DIR}/events/50kHz ${STANDALONE_DIR}/build
126- python3 ${GITHUB_WORKSPACE}/.github/scripts/profiler_rocprofv2.py --runs 42 --input /root/${PROFILER_CSV} --output /root/${PROFILER_CSV}
133+ python3 ${GITHUB_WORKSPACE}/.github/scripts/profiler_rocprofv2.py --runs ${{ matrix.profiler_runs }} --input /root/${PROFILER_CSV} --output /root/${PROFILER_CSV}
127134
128135 - name : Upload Artifact
129136 uses : actions/upload-artifact@v6
@@ -138,7 +145,7 @@ jobs:
138145 mkdir -p ${STANDALONE_DIR}/baseline
139146 curl -fL --retry 3 -o ${STANDALONE_DIR}/baseline/${PROFILER_CSV} https://cernbox.cern.ch/remote.php/dav/public-files/SfYXgQOHFga2w75/baseline/${PROFILER_CSV}
140147 curl -fL --retry 3 -o ${STANDALONE_DIR}/baseline/${BENCHMARK_CSV} https://cernbox.cern.ch/remote.php/dav/public-files/SfYXgQOHFga2w75/baseline/${BENCHMARK_CSV}
141- python3 ${GITHUB_WORKSPACE}/.github/scripts/csv_to_md.py --baseline ${STANDALONE_DIR}/baseline/${PROFILER_CSV} --current /root/${PROFILER_CSV} >> ${GITHUB_STEP_SUMMARY}
148+ python3 ${GITHUB_WORKSPACE}/.github/scripts/csv_to_md.py --runs ${{ matrix.profiler_runs }} -- baseline ${STANDALONE_DIR}/baseline/${PROFILER_CSV} --current /root/${PROFILER_CSV} >> ${GITHUB_STEP_SUMMARY}
142149 echo -e "\n\n" >> ${GITHUB_STEP_SUMMARY}
143- python3 ${GITHUB_WORKSPACE}/.github/scripts/csv_to_md.py --baseline ${STANDALONE_DIR}/baseline/${BENCHMARK_CSV} --current /root/${BENCHMARK_CSV} >> ${GITHUB_STEP_SUMMARY}
150+ python3 ${GITHUB_WORKSPACE}/.github/scripts/csv_to_md.py --runs ${{ matrix.standalone_runs }} -- baseline ${STANDALONE_DIR}/baseline/${BENCHMARK_CSV} --current /root/${BENCHMARK_CSV} >> ${GITHUB_STEP_SUMMARY}
144151 rm -rf ${STANDALONE_DIR}/baseline
0 commit comments