52a6a40535
The performance regression only happened on iOS devices. Benchmarking on Desktops or Android devices doesn't seem to show any significant regressions. Tested this on iPhone XS with nanobench and saw a 10ms->30ms regression: regression: 61/63 MB 1 10.5ms 10.7ms 10.8ms 12.2ms 5% █▂▁▁▂▁▂▂▂▁ 8888 GM_savelayer_with_backdrop 101/101 MB 1 36ms 36.5ms 36.4ms 36.8ms 1% ▁▅▃▆▆▄▆▃▃█ gles GM_savelayer_with_backdrop before regression: 61/63 MB 1 10.3ms 10.4ms 10.6ms 12.4ms 6% █▂▁▂▁▁▁▁▁▁ 8888 GM_savelayer_with_backdrop 96/96 MB 1 10.6ms 10.7ms 10.7ms 10.8ms 1% ▄▁▄▂▆█▅▃▂▅ gles GM_savelayer_with_backdrop It's interesting to see that 8888 is always fast. Is the regressing CL changing blur from software to GPU? Bug: https://github.com/flutter/flutter/issues/36064 Change-Id: Iea2305e41a31bd5b4267f45bfd835231d49d639a Reviewed-on: https://skia-review.googlesource.com/c/skia/+/234413 Commit-Queue: Yuqian Li <liyuqian@google.com> Reviewed-by: Michael Ludwig <michaelludwig@google.com> |
||
|---|---|---|
| animations | ||
| bench | ||
| bin | ||
| build_overrides | ||
| dm | ||
| docker | ||
| docs/examples | ||
| example | ||
| experimental | ||
| fuzz | ||
| gm | ||
| gn | ||
| include | ||
| infra | ||
| modules | ||
| platform_tools | ||
| resources | ||
| samplecode | ||
| site | ||
| specs | ||
| src | ||
| tests | ||
| third_party | ||
| tools | ||
| .clang-format | ||
| .clang-tidy | ||
| .gitignore | ||
| .gn | ||
| AUTHORS | ||
| BUILD.gn | ||
| codereview.settings | ||
| CONTRIBUTING | ||
| CQ_COMMITTERS | ||
| DEPS | ||
| go.mod | ||
| go.sum | ||
| LICENSE | ||
| OWNERS | ||
| PRESUBMIT.py | ||
| public.bzl | ||
| README | ||
| README.chromium | ||
| RELEASE_NOTES.txt | ||
| whitespace.txt | ||
Skia is a complete 2D graphic library for drawing Text, Geometries, and Images. See full details, and build instructions, at https://skia.org.