Oleg Oshmyan
Oleg Oshmyan
Thanks for putting this up here. But I believe we concluded the reason was different when we previously discussed this elsewhere: [Old Microsoft docs](https://learn.microsoft.com/en-us/previous-versions/ms969909%28v=msdn.10%29?redirectedfrom=MSDN) say a penalty of 9000 (matching...
Apparently, Uniscribe doesn’t apply TrueType `kern` tables—only OpenType `GPOS` tables. Or _at least in some cases_ doesn’t apply them. See https://github.com/libass/libass/issues/706#issuecomment-1712814726 and the font in that issue’s initial post. (It’s...
> GDI applies kerning to punctuation regardless of the surrounding script, while HarfBuzz seems to act as if the punctuation has inherited script and only applies kerning defined for that...
Rebased on master and addressed two nits. (More substantial changes still pending.) Meanwhile, we’ve run performance tests on several other CPUs. Full times of `profile` built from [this source code](https://github.com/astiob/libass/commits/5ee32ab19ceb2fa32ac272bef53659e2b2ced1bc)...
> Actually it's not a boolean parameter: there're small enough blurs to be between `true` and `false`. I'm thinking of replacing boolean with interpolation parameter (0–255) What would be the...
It has just occurred to me that there’s another opportunity to explore here. Most of the pixels have either `g[x] == 255`, where the entire pixel is covered by the...
I’ve rerun some tests in a slightly more stable environment. I ran each test 30 times on the Ryzen. Most of the results still annoyingly failed normality tests, so take...
On the M1: ``` vectorized float: 9.27s user 3.06s system 99% cpu 12.365 total conditional int: 11.35s user 2.59s system 99% cpu 13.957 total conditional float: 11.51s user 2.67s system...
> I suppose you could hand-write vectorized assembly that also performs this check against the whole vector: if it’s all zeros or all 255s, no division is necessary. That might...
> It would be "effective blurriness". I'm going to do some math to find a good approximation for it depending on blur radius, but we can use only 0 and...