commit 07f9e6973f84de68a581a5733473198b56cc08c7
author Felipe Archondo <felipearchondo@google.com> Fri Jun 14 14:42:54 2019 +0000
committer CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Fri Jun 14 14:42:54 2019 +0000
tree 5a64f548e9b197a8e6a879f7d4435d6cd2c79c70
parent 4f49f6bbb2de199294349f47db44fe20c3438563

[tracing][escher] Reland "use vkTimestamps correctly in TimestampProfiler"

This is a reland of dc4a163b160d6448aedd1a3ac55ed70c50e0fe95

Instead of kCleanupDelay.get() we use kCleanupDelay.to_msecs(), since we
are working in milliseonds, and .get() returns the value in nanoseconds
which led to the bug.

PT-206 #done
SCN-1477 #done

Original change's description:
> [tracing][escher] use vkTimestamps correctly in TimestampProfiler
>
> We no longer assume that the timebase given to us in the VkTimestamp
> is the same as the CPU's. While the previous version worked on available
> hardware, it led to undefined behavior on AEMU which made the GPU events
> on the traces unusable.
>
> Instead, we now use the fact that CleanupEscher() runs every millisecond
> to offset the durations we get back from the GPU by 0.5ms, guaranteeing
> us to be precise to +/- 0.5ms.
>
> Ideally we would use a Vulkan timings extension to get even more precise
> data, but this should be fine until then to get GPU work included in AEMU
> flows. See SCN-1460 for details.
>
> SCN-1460 #done
>
> Test: unit tests, tracing on real hardware and AEMU
>
> Change-Id: I3766ceebca686c5a2f20746afee78f4f7c5a29db

Change-Id: Ibcefa49022ea0944d50dfc70e557019d97af0224