Hello Steve, > VK_KHR_calibrated_timestamps says it should query 'quasi simultaneously > from two time domains' - but this is purely providing an interface to > read a single counter at a time. > > It _may_ be useful to report the GPU's view of time and at the same time > (or as near as possible) the architectural counters. That can be used to > deal with drift between the GPU's counters and arch counters[1]. > > In general we try to avoid providing an interface to something which is > unrelated to the GPU, especially when user space already has a mechanism. > > Steve > > [1] See Mihail's response to the panthor patches for details of > differences that might occur. I initially didn't saw the register to get the GPU timestamp and wrongly assumed I would have to query it from the generic arch timer. I will make SYSTEM_TIMESTAMP returns the value of the timestamp register on v2 and keep SYSTEM_TIMESTAMP_FREQUENCY the same way as it is at the moment. Thanks for the review, Mary
