5fbd18370f
Previously, we were reporting numbers that were far too low, because we were getting way ahead of the GPU, and then spending all of our time in finish (which isn't timed). That led to us picking very high loop counts, so our wall clock time to run nanobench was very high, and our reported times were very low. This fixes all of that, and removes all the spam about not having fence support. Change-Id: Ib9dfc043da82bf8ee6645b8627cfade66eb9864e Reviewed-on: https://skia-review.googlesource.com/58001 Reviewed-by: Brian Salomon <bsalomon@google.com> Commit-Queue: Brian Osman <brianosman@google.com> |
||
---|---|---|
.. | ||
gl | ||
mock | ||
mtl | ||
vk | ||
FenceSync.h | ||
GpuTimer.h | ||
GrContextFactory.cpp | ||
GrContextFactory.h | ||
GrTest.cpp | ||
GrTest.h | ||
TestContext.cpp | ||
TestContext.h |