-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Hello Erik, all,
it seems that PAPI 5.3 has options to compile for the mics. One only has to pass --with-mic to configure. Are you aware of incompatibilities of PAPI 5.3 or should it be fine to update the tarball in ExternalLibraries? (CC'd to users@einsteintoolkit.org)
I can compile PAPI 5.3 on stampede and run the papi.par test in ExternalLibraries/PAPI and get Cache misses (but no timing, but lots of event-does-not exist warnings):
INFO (PAPI): PAPI_register_clock INFO (PAPI): Running local single-threaded DGEMM benchmark with N=1000... INFO (PAPI): Floating point operations: 2 Gflop over 1 threads INFO (PAPI): Pure performance: 0.651531 Gflop/sec INFO (PAPI): Overall performance: 0.643122 Gflop/sec INFO (PAPI): Running local multi-threaded DGEMM benchmark with N=1000... INFO (PAPI): Floating point operations: 488 Gflop over 244 threads INFO (PAPI): Pure performance: 36.7471 Gflop/sec INFO (PAPI): Overall performance: 27.8211 Gflop/sec INFO (PAPI): NOTE: These performance numbers do not have to be large, but they need to agree with what PAPI reports INFO (PAPI): PAPI Statistics: CPU time 5042.929817665 sec CPU cycles 5520.440487605 Gcyc PAPI_TOT_CYC 0.324841654 cyc^(-1) PAPI_TOT_INS 0.027041240 cyc^(-1) PAPI_L1_ICA 0.033998239 cyc^(-1) PAPI_L1_ICM 0.000011858 cyc^(-1) PAPI_L1_DCA 0.011631986 cyc^(-1) PAPI_L1_DCM 0.000893145 cyc^(-1) PAPI_LD_INS 562.116089268 cyc^(-1) PAPI_SR_INS 0.001194933 cyc^(-1) INFO (PAPI): PAPI Statistics: CPU time 4.165410795 sec CPU cycles 4.559431145 Gcyc PAPI_TOT_CYC 393.906331451 cyc^(-1) PAPI_TOT_INS 32.842575465 cyc^(-1) PAPI_L1_ICA 41.205762993 cyc^(-1) PAPI_L1_ICM 0.014362943 cyc^(-1) PAPI_L1_DCA 14.119073876 cyc^(-1) PAPI_L1_DCM 1.093455247 cyc^(-1) PAPI_LD_INS 680595.542337835 cyc^(-1) PAPI_SR_INS 1.448845164 cyc^(-1)
Yours, Roland
- -- My email is as private as my paper mail. I therefore support encrypting and signing email messages. Get my PGP key from http://keys.gnupg.net.