072e09b26d
The implementation is inconsistent across CPU and GPU. CPU currently always implements kStrict even when kFast is passed. For mipmapping this means computing a set of levels from the subset, which is an expensive. GPU limits the texture sampling coordinates to the subset in the base level. However, higher level mipmap pixels that map back to a footprint outside the subset are still sampled. So GPU pays a higher shader complexity cost without really providing the benefit of kStrict. We are adding anisotropic filtering, which will not work well with kStrict, and thus will also be disabled by kStrict. Must land after https://chromium-review.googlesource.com/c/chromium/src/+/3571661 Bug: skia:10481 Bug: skia:13078 Bug: skia:13036 Canary-Chromium-CL: 3571661 Change-Id: I011b8698a3f9fafa9b819486873c8ff54df6299e Reviewed-on: https://skia-review.googlesource.com/c/skia/+/527284 Reviewed-by: Michael Ludwig <michaelludwig@google.com> Commit-Queue: Brian Salomon <bsalomon@google.com> |
||
---|---|---|
.. | ||
android | ||
c | ||
codec | ||
config | ||
core | ||
docs | ||
effects | ||
encode | ||
gpu | ||
pathops | ||
ports | ||
private | ||
sksl | ||
svg | ||
third_party | ||
utils | ||
OWNERS |