hb_set_t optimizations #23
Conversation
hb_set_t previously used a statically allocated bit vector big enough to store the whole range of codepoints. now, pages are created for each small bitvector dynamically as needed.
|
Humm. Is github confused about this branch? |
|
Very confused. Ignore it until I finish this task. This branch was just for the review questions I asked last Thursday. |
|
Ok. Just checked the patch. I think I have more comments for you. Wait for that before spending much more time on it. Thanks! |
| @@ -324,7 +324,7 @@ struct hb_set_t | |||
| static const unsigned int SHIFT = 5; | |||
| static const unsigned int BITS = (1 << SHIFT); | |||
| static const unsigned int MASK = BITS - 1; | |||
| static const unsigned int ELTS = (MAX_G + 1 + (BITS - 1)) / BITS; | |||
| static const unsigned int ELTS = (MAX_G + 1) / BITS; | |||
There was a problem hiding this comment.
FWIW, old cold was implementing ceil(), which is correct, since it does not make assumption on MAX_G + 1 being divisible by BITS.
|
Heh. I have had commented on that before... I'll try to clean this up and merge finally. :) |
|
Would be nice to see this go in. It was my biggest contribution to harfbuzz :) Let me know if I can help you merge this. |
|
Hi Behdad. Any news about this PR? Anything I can help you with? Thanks 😄 |
|
@timofonic let's see what @behdad says 😄 |
Add binding for kerning callbacks. Make feature fields public.
First stage of the optimizations changes.
Switched the data structure used for the set to a dynamic array of bit vectors like fontconfig's FcCharSet. Basic functions supported (creating, setting, checking, etc). A good moment to get it reviewed before switching the rest of the functions (algebra and iter).
Let me know when you have time to discuss this. I have a few ideas for further improvements in stage 2.
There are 2 smal bugs I am aware of, don't worry about them :)