ANGLE bots were getting lots of corrupted GMs - we set fPreferVRAMUseOverFlushes
to false. In that case, multiple deferred proxies were instantiating to the same
scratch resource. Any proxy that we're going to fill with an ASAP upload needs
to have no pending IO - we hoist all those loads to the front of the flush, so
normal IO tracking doesn't really help.
Bug: skia:
Change-Id: Id36fd8700e522db412a3c992b93c778e2ebb1188
Reviewed-on: https://skia-review.googlesource.com/53940
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
099fa0fb98