I'm trying to measure performance of Brook+ samples shipped with the SDK, but it seems that setting BRT_RUNTIME to "cal" or "cpu" does not affect the timing information at all.
Any idea what might be wrong? We're running on a HD3870, latest drivers.
The code executes correctly and the timings obtained using Setup(0), Start(0), Stop(0) and PrintTime(0) are simply the same regardless of BRT_RUNTIME.
StreamRead and StreamWrite were excluded from measurements, only kernel call is between Start(0) and Stop(0).
Thank you.
PS. You seem to have a bug in the SDK's timer.cpp file, where PrintTime always outputs timer 0th time as opposed to timer ith time.