Commit Graph
14392 Commits
Author SHA1 Message Date
Behdad Esfahbod ba9a61597f [hvgl] Fix MINI build
HVF disabling to be configured better. But fix build for now.
2025-03-02 17:03:51 -07:00
Behdad Esfahbod 758ee5cff9 [COLR] Regain speed if all (normalized) coords are 0 2025-03-02 17:03:51 -07:00
Behdad Esfahbod da1fba8863 [cff] Regain speed re skipping applying variations 2025-03-02 17:03:51 -07:00
Behdad Esfahbod f883d370af [glyf] Regain speed if variations are all zero
We always have coords array populated now... Check for non-zero.
2025-03-02 17:03:49 -07:00
Behdad Esfahbod d4bf8a7734 [glyf] Regain speed if variations are all zero
We always have coords array populated now... Check for non-zero.
2025-03-02 17:03:15 -07:00
Behdad Esfahbod cf7f5ad0ed [font] Remove avar->is_biased() short-circuit
That was not considering avar2 fonts. Let the has_nonzero_coords
magic take care of optimizing for default instance. A couple extra
allocations during hb_font_create(), but no big deal.
2025-03-02 17:00:17 -07:00
Behdad Esfahbod 5f93435d01 [ot-font/hvgl] Guard include 2025-03-02 17:00:17 -07:00
Behdad Esfahbod aa3d37e8df [avar] Rewrite the "extended" logic more robustly 2025-03-02 17:00:17 -07:00
Behdad Esfahbod a0dcca6719 [avar] Non-change 2025-03-02 17:00:17 -07:00
Behdad Esfahbod e049666552 [avar] Fix bug introduced ... 2025-03-02 17:00:17 -07:00
Behdad Esfahbod 19b4c6e559 [subset] "Fix" subset tests after recent fvar/avar precision changes 2025-03-02 17:00:17 -07:00
Behdad Esfahbod d7b15b0e21 [font/fvar/avar] Increase axis coordinate normalization precision
Spec requires 16.16; we were doing 2.14. Fix up.
Now we do math in float, but do the 16.16 and final 2.14 rounding
as required by the spec.
2025-03-02 17:00:17 -07:00
Behdad Esfahbod bfcbb439cb [avar] Remove some unused code
Unfortunately subsetter doesn't support avar2 yet. So, more
cruft still left in there.
2025-03-02 17:00:17 -07:00
Behdad Esfahbod 29991730c6 [fvar] Remove unused code 2025-03-02 17:00:17 -07:00
Behdad Esfahbod 9ccb714217 [font] Return NAN for design coords, if normalized coords were set 2025-03-02 17:00:17 -07:00
Behdad Esfahbod 117b79fd92 [font] Change docs to match recent behavior change
We always allocate coords vectors full-sized.
2025-03-02 17:00:17 -07:00
Behdad Esfahbod feb899bb93 [ft] Fix rounding of normalized coords
According to spec...
https://learn.microsoft.com/en-us/typography/opentype/spec/otvaroverview#coordinate-scales-and-normalization
2025-03-02 17:00:17 -07:00
Behdad Esfahbod 3a489fd7c0 [avar] Another hack, to match CoreText
Until further notice, this mess remains...
2025-03-02 17:00:17 -07:00
Behdad Esfahbod 9b6da29971 [font/var] Try to regain performance of all-normalized-coords=0 2025-03-02 17:00:14 -07:00
Behdad Esfahbod 1f0aae5280 [font/var] Try to allow "biased" avar's
That map 0 to non-0!
2025-03-02 16:59:24 -07:00
Behdad Esfahbod 04be808886 [font/var] Always allocate coords vector the length of fvar axes
In prep for black magic.
2025-03-02 16:59:23 -07:00
Behdad Esfahbod 74b0455e34 [avar] Bend to some weirdo font's wght mapping 2025-03-02 16:59:23 -07:00
Behdad Esfahbod f3a1ec291c [open-type] Minor float math 2025-03-02 16:59:23 -07:00
Behdad Esfahbod f3500cb3b4 [hvgl] I *think* I got transforms right this time! 2025-03-02 16:59:23 -07:00
Behdad Esfahbod 7bf8ce5834 [hvgl] Fixup
Oops. Still buggy.
2025-03-02 16:59:23 -07:00
Behdad Esfahbod c6ec6eb194 [geometry] Whitespace 2025-03-02 16:59:23 -07:00
Behdad Esfahbod f286d33d67 [hvgl] Back to composing transforms the way the paper says
...
2025-03-02 16:59:23 -07:00
Behdad Esfahbod 2bcbfd35ea [hvgl] Indent 2025-03-02 16:59:23 -07:00
Behdad Esfahbod fc0966d90a [hvgl] Fix the complex-number transform code path
Tested by U+9DAD at wght=900 of PingFangUI.
2025-03-02 16:59:23 -07:00
Behdad Esfahbod d4511648db [draw] Simplify quadratic-to-cubic conversion
Hopefully introduces less error.
2025-03-02 16:59:23 -07:00
Behdad Esfahbod 9685d39315 [hvgl] Another clamp
Huh. Errors down to just 1!
2025-03-02 16:59:23 -07:00
Behdad Esfahbod 40b9930c97 [hvgl] Minor move condition
Previous code was safe but was taking address of an out-of-range
array member if end was 0. Was never dereferenced though.
2025-03-02 16:59:23 -07:00
Behdad Esfahbod a71ad88629 [hvgl] Rename a variable 2025-03-02 16:59:23 -07:00
Behdad Esfahbod 32a6940df6 [hvgl] Minor const 2025-03-02 16:59:23 -07:00
Behdad Esfahbod 3c84aa52f2 [hvgl] Move transforms around
This doesn't seem to quite match the spec. But produces fewer wrong
shapes.  Down to 834 out of 34485 characters.
2025-03-02 16:59:23 -07:00
Behdad Esfahbod 3e1e2ab393 [hvgl] Another transform fix 2025-03-02 16:59:23 -07:00
Behdad Esfahbod 7e84cf811f [hvgl] Simplify transform 2025-03-02 16:59:23 -07:00
Behdad Esfahbod d5174c41ae [hvgl] Fix transform order 2025-03-02 16:59:23 -07:00
Behdad Esfahbod 76dbae15e7 [hvgl] Minor 2025-03-02 16:59:23 -07:00
Behdad Esfahbod 453aea05b0 [hvgl] Use double instead of float for coords & transforms
As "spec" requires.
2025-03-02 16:59:19 -07:00
Behdad Esfahbod 00bc3ecf25 [geometry] Use int literals instead of float ones
In anticipation of templatizing on float type.
2025-03-02 16:58:37 -07:00
Behdad Esfahbod 9f85e97401 [hvgl] Remove a resize(0) that was not doing any good
In hb_vector_t<double> we use hb_realloc() for resizing, which
doesn't care what the vector current length is.
2025-03-02 16:57:55 -07:00
Behdad Esfahbod dbef4e324d [hvgl] Name a value 2025-03-02 16:57:55 -07:00
Behdad Esfahbod da3fca4685 [hvgl] Reduce another malloc 2025-03-02 16:57:55 -07:00
Behdad Esfahbod 44780db50d [hvgl] Remove unused parameter 2025-03-02 16:57:55 -07:00
Behdad Esfahbod 9579cee8b0 [hvgl] Add table version check 2025-03-02 16:57:55 -07:00
Behdad Esfahbod d4b29817c8 [hvgl] Remove use of a transient variable 2025-03-02 16:57:55 -07:00
Behdad Esfahbod 9b36117203 [hvgl] Micro-optimize vector allocation 2025-03-02 16:57:55 -07:00
Behdad Esfahbod a2fc952342 [hvgl] Use a common "scratch" vector
To avoid reallocation.
2025-03-02 16:57:55 -07:00
Behdad Esfahbod de0e1bddff [hvgl] Minor simplify sanitize 2025-03-02 16:57:55 -07:00