Any profiling API for tool developers?

[Jie__J_]

Hi,

One can use API __Offload_report ( int )  or the environmental variable OFFLOAD_REPORT to get the timing information ( and the amount of data transferred ) for all the offloaded regions. And the profiling information is output to the stdout.

But for profiling tool developers, the following features may be desired:

Is it possible to provide a profiling API that can be used to timing a specific offloaded region , instread of all offloaded regions?

And is there any way to record the measured profiling information from the profiling API  into memory buffer, instead of writing to stdout?

[Loc_N_Intel]

The current implementation of __Offload_report API (or the equivalent environment variable OFFLOAD_REPORT) just reports all offloaded regions in stdout. I already forwarded your feature request.