HPCToolkit Profiling Information
Application: dunetpc v06_57_00
- LArSoft/protoDune: proton_6p0GeV_mono (proton_6GeV)
- LArSoft/protoDune: beam_cosmics_p1GeV (cosmic_1GeV)
- LArSoft/Dune-FD: prodgenie_nue_dune10kt_1x2x6 (nue_dune10kt)
Performance Profiling Reports
- CPU: Inclusive and Exclusive Time (sampling with 200Hz frequency)
- CPI: Cycle per Instruction (PAPI_TOT_CYC/PAPI_TOT_INS)
- PAPI_TOT_CYC: Total cycles
- PAPI_TOT_INS: Instructions completed
- FMO: FLOPS/Memory Operations (Computational Intensity = PAPI_FP_OPS/(PAPI_LD_INS+PAPI_SR_INS)
- PAPI_FP_OPS: Floating point operations
- PAPI_LD_INS: Load instructions
- PAPI_SR_INS: Store instructions
CPU
Hardware Counters (CPI, FMO)
Processor: Intel(R) Xeon(R) CPU E5-2620 0 @ 2.00GHz, cache size : 15360 KB, MemTotal: 32845168 kB
Database (download for a full GUI analysis)
Example steps (for detail instructions, refer to HPCToolkit Users Manual [pdf]
1) click a link in the talbe and download the database file
2) tar xzf lar-xyz.dat.tgz
3) hpcviewer lar-xyz.dat
4) select View - Calling (topdown call paths), Callers (bottom up call paths), or Flat
5) navigate metrics with control buttons