Intel® Advisor Help
Estimate invocation taxes assuming the invocation tax is paid only for the first kernel launch.
Analysis Properties > Offload Modeling > Performance Modeling > Single Kernel Launch Tax
--assume-hide-taxes --no-assume-hide-taxes |
Off (no-assume-hide-taxes)
You can control how to model kernel invocation taxes for your application. When a high call count value is detected for a potentially profitable code region, Intel® Advisor assumes that the kernel invocation tax is paid as many times as the kernel is launched. This results in high invocation tax and cost of offloading, which means that this code region cannot benefit from offloading.
For simple applications where there is no need to wait for a kernel instance to finish, use the assume-hide-taxes to hide this cost every time except the very first one and minimize the invocation tax.
Enable the assume-hide-taxes option to analyze oneAPI applications (DPC++ or OpenMP* target) running on CPU.
Run Survey Analysis.
Run Trip Counts and FLOP analyses of the Characterization stage.
Model your application performance on a target device assuming the invocation tax is paid only when the kernel is launched for the first time.
advisor --collect=survey --static-instruction-mix --project-dir=./advi_results -- ./myApplication
advisor --collect=tripcounts --flop --enable-data-transfer-analysis --project-dir=./advi_results -- ./myApplicationadvisor --collect=projection --assume-hide-taxes --project-dir=./advi_results