Improved QE clocks
URL: https://docs.google.com/document/d/1JSefW-9qIJfqN7CreQhA2sCM4SWJOjb5Ij1TCxeLqMg/edit?usp=sharing
-
Remove ‘_gpu’ variants and adjust reporting -
Remove all CUDA Fortran dependencies (if any) related to clock invocation -
Prettify clocks_handler.f90 -
Identify a maximum depth for standard clocks and replace lower-level ones with instrumentation routines for profiling purposes only -
(Optional) Deprecate __TRACE from clocks (still exists in mp, it should be deprecated there too) -
(Optional) Deprecate PRINT_AVG_CPU_TIME_PER_THREAD -
Convert mytime as proper module (proposed name: timing) containing all relevant subroutines (public and private) -
(Optional) extend start_clock API to count GPU executions (not time, only occurrences) - require a new module structure
Edited by Laura Bellentani