Enforce barrier when sampling timestamps

[PENGUINLIONG]

Some context: #2235.

[CLAassistant]

All committers have signed the CLA.

[SRSaunders]

Will there be a performance hit by doing this? If so, then perhaps it needs to be an optional parm.

[PENGUINLIONG]

@SRSaunders Is there anything I could refer to?

[billhollings]

Interesting catch! Good work. And thanks for submitting this fix!

[billhollings]

Will there be a performance hit by doing this? If so, then perhaps it needs to be an optional parm.

This is a good point, and probably the reason we originally set barrier to NO.

Apple's doc does indicate that it may affect GPU performance, as one would expect from any barrier.

However, the Vulkan spec explicitly states:

When vkCmdWriteTimestamp is submitted to a queue, it defines an execution dependency on commands that were submitted before it, and writes a timestamp to a query pool.

which would indicate that a barrier is required and expected here, and that timestamp scope accuracy is more important than absolute performance.