|
9df07c3c
|
2023-06-08T11:46:06
|
|
[gsubgpos] Prevent a leak in closure
If the push wasn't successful we were constructing an
hb_set_t on the Crap data... At least that's my reading
of the code.
|
|
|
82741304
|
2023-06-04T09:41:41
|
|
[subset] start_embed never returns nullptr
Remove checks.
|
|
|
31949636
|
2023-05-04T16:29:48
|
|
Revert "[layout] Cache subtable coverages in hb_map_t"
This reverts commit 7a715d74e06720c17d28ba7b4c3da0b583f8d1d3.
|
|
|
7a715d74
|
2023-05-04T16:28:11
|
|
[layout] Cache subtable coverages in hb_map_t
Proof-of-concept. Going to revert. Memory consumption is more
than I like. It does speed up Roboto shaping another 15% though.
Perhaps if we could add logic to choose which subtables to
cache, this might be a useful approach.
|
|
|
975980d3
|
2023-05-04T15:56:06
|
|
[gsubgpos] Add apply_cached() as separate method
In prep for some other work.
|
|
|
3f2401e2
|
2023-05-03T16:56:52
|
|
[layout] Don't init iters successively multiple times
|
|
|
959f1634
|
2023-05-03T16:27:41
|
|
[gsubgpos] Minor reduce variable scope
|
|
|
fb795dc3
|
2023-05-03T15:17:20
|
|
[GPOS] Optimize iterator reset
Speeds up Gulzar shaping 5%.
|
|
|
8a8fc37c
|
2023-05-03T14:57:00
|
|
[GDEF] Use a cache for glyph classes
Shows 5% speedup for Roboto shaping.
|
|
|
49ac5e11
|
2023-05-02T12:12:26
|
|
[match_input] Micro-optimize
|
|
|
63afb4f2
|
2023-04-26T10:29:25
|
|
[syllabic] Better fix for previous issue
With previous fix the GPOS application was still reading the syllable()
member, which was already freed. This fix is more correct.
|
|
125450d2
|
2023-03-13T15:43:29
|
|
[subset] collect name_ids for FeratureParams
|
|
|
8708b9e0
|
2023-02-06T14:51:25
|
|
[GPOS] Avoid O(n^2) behavior in mark-attachment
Better implementation; avoids arbitrary limit on look-back.
|
|
|
661050b4
|
2023-02-06T12:38:17
|
|
Revert "[layout] Limit how far we skip when looking back"
This reverts commit 85be877925ddbf34f74a1229f3ca1716bb6170dc.
|
|
|
b29fbd16
|
2023-02-06T13:08:52
|
|
[gsubgpos] Refactor skippy_iter.match()
|
|
|
85be8779
|
2023-02-01T20:00:43
|
|
[layout] Limit how far we skip when looking back
See comments.
|
|
|
d18fd3f7
|
2023-01-31T16:28:49
|
|
[layout] Comment
|
|
|
7a4bd97e
|
2023-01-31T14:59:39
|
|
[layout] Build lookup accelerators lazily on-demand
Reduces memory consumption for large multi-script fonts
drastically.
|
|
|
83353f13
|
2023-01-31T14:32:14
|
|
[layout] Reduce memory use slightly
By using raw pointer instead of vector for subtable accelerator.
To be used for more memory saving by making subtable accelerators
lazy-loaded by shape-plans for large fonts.
|
|
|
49d75ef3
|
2023-01-27T15:37:11
|
|
[gsubgpos] Fix bug in cached ChainContextFormat2 application
|
|
|
d2279a20
|
2023-01-27T12:32:55
|
|
[gsubgpos] Avoid a copy into the vector
|
|
|
8d29be39
|
2023-01-26T11:41:58
|
|
[gsubgpos] Minor drop an unnecessary hb_iter
|
|
|
be8a87c4
|
2023-01-21T15:00:41
|
|
Move TRACE_DISPATCH after may_recurse()
Such that we don't get memory access issues if DEBUG_SANITIZE is
on and may_recurse() returns false.
|
|
ea1ee0d4
|
2023-01-14T00:23:53
|
|
[instance] drop and don't collect lookups from feature variations when fully instancing.
This previously incorrectly collected lookups that could be reached via feature variations that are dropped and not activated by the current instance position.
|
|
|
ed023f66
|
2023-01-12T17:04:24
|
|
Enable -Wformat-signedness
And fix the codebase.
|
|
|
1a00ab69
|
2023-01-11T10:56:16
|
|
[subset-plan] Simplify gsub_lookups and gpos_lookups allocation
|
|
|
2764a616
|
2023-01-05T15:14:54
|
|
Revert "[gsubgpos] Use swap instead of move"
This reverts commit 8a17cc4ecf21f6754e2d90562d0ced7496870f74.
|
|
|
167b7c60
|
2023-01-05T14:48:20
|
|
Revert "[gsubgpos] Reduce hb_set_t allocations"
This reverts commit 0b7f6d6cf0e2deba637783ab3880fdfb90ca8ac3.
Not much benefit as the main allocations come from other places.
|
|
|
3947cedd
|
2023-01-05T14:47:47
|
|
Revert "[gsubgpos] Cache pos_glyphs allocation in closure"
This reverts commit 3961cc46bf438947b19063cb7e735247358f1d4f.
This was wrong...
|
|
|
8a17cc4e
|
2023-01-05T14:15:38
|
|
[gsubgpos] Use swap instead of move
Move is wrong when we want to reuse the object.
|
|
|
4401dd24
|
2023-01-05T14:13:57
|
|
[gsubgpos] Minor use ->clear() directly
|
|
|
0b7f6d6c
|
2023-01-05T13:58:57
|
|
[gsubgpos] Reduce hb_set_t allocations
|
|
|
3961cc46
|
2023-01-05T13:58:57
|
|
[gsubgpos] Cache pos_glyphs allocation in closure
Saves some 3% in Gulzar-Regular subsetting.
|
|
|
a90f149e
|
2023-01-05T13:52:11
|
|
[gsubgpos] Minor drop an allocation
|
|
|
2eacc37e
|
2022-12-31T12:27:13
|
|
[vector] Add internal API for exact-size allocation
Use it from a couple of places.
|
|
|
36bef5dc
|
2022-12-30T13:17:23
|
|
[gsubgpos] Prealloc subtables vector
|
|
|
630b874a
|
2022-12-03T12:13:15
|
|
[gsubgpos] Add a cache to intersected_class_glyphs
30% gain on subset_codepoints/NotoNastaliqUrdu-Regular.ttf/nohinting/1400.
|
|
|
c044f4af
|
2022-12-03T11:58:04
|
|
[gsubgpos] Remove wrong const
|
|
|
2680be1f
|
2022-12-03T11:53:14
|
|
[gsubgpos] Don't set unnecessary funcs
|
|
|
ac8b232a
|
2022-11-30T16:42:15
|
|
[gsub] Cache intersects_class results for closure
Benchmark Time CPU Time Old Time New CPU Old CPU New
----------------------------------------------------------------------------------------------------------------------------------------------------------------------
BM_subset/subset_codepoints/NotoNastaliqUrdu-Regular.ttf/nohinting/10 +0.0246 +0.0240 0 0 0 0
BM_subset/subset_codepoints/NotoNastaliqUrdu-Regular.ttf/nohinting/64 -0.5541 -0.5544 4 2 4 2
BM_subset/subset_codepoints/NotoNastaliqUrdu-Regular.ttf/nohinting/512 -0.1120 -0.1123 43 38 43 38
BM_subset/subset_codepoints/NotoNastaliqUrdu-Regular.ttf/nohinting/1400 -0.1154 -0.1159 43 38 43 38
|
|
|
c6a4b601
|
2022-11-30T14:09:59
|
|
[gsubgpos] Add an unlikely
|
|
|
a4a40786
|
2022-11-24T11:56:58
|
|
Merge pull request #3893 from googlefonts/preprocess_test
[subset] Fix testing of preprocess
|
|
|
0c70bc7f
|
2022-11-24T11:48:48
|
|
[skippy-iter] Fix two logic errors
First, a signed underflow.
Second, a wrong condition.
Both were introduced in 42681bdb55a75520d4ac194302fe936d1ce3cb34
|
|
|
1fa64c0c
|
2022-11-23T21:38:51
|
|
[gsubgpos] Conditionalize skippy on unsafe-to-concat
|
|
|
42681bdb
|
2022-11-23T21:36:43
|
|
[gsubgpos] No logic-change minor rewrite
|
|
|
c15efdec
|
2022-11-23T21:14:28
|
|
[gsubgpos] Comment
|
|
|
7ec1c41a
|
2022-11-23T21:12:41
|
|
[gsubgpos] Skippy-iter: Prefer correctness to performance
Prefer unsafe-to-concat correctness, over performance.
|
|
|
ff3cac0c
|
2022-11-23T23:50:49
|
|
[subset] Fix unecessary trailing 0 bytes left by ContextFormat2 pruning.
Uneeded rules where beind removed from the count by the bytes for them was being left in the font.
|
|
|
01f961ac
|
2022-11-18T15:47:17
|
|
[gsubgpos] Minor call hb_iter() instead of ->iter() directly
|
|
|
bba57655
|
2022-11-18T14:52:17
|
|
[gsubgpos] Remove a few unnecessary namespace invocations
|
|
|
b4d0d160
|
2022-11-18T14:47:33
|
|
[gsubgpos] Rewrite a couple apply() functions as daggers
|
|
|
76c8214e
|
2022-11-18T12:30:46
|
|
[gsubgpos] Move member around
|
|
|
1610008e
|
2022-11-16T20:02:36
|
|
[gsubgpos] Minor remove call to hb_iter
|
|
|
19ec01d2
|
2022-11-16T19:14:42
|
|
[gsubgpos] Sprinkle const around
|
|
|
8b2a2111
|
2022-11-16T16:57:44
|
|
[layout] Keep digest updated in the context
Don't recompute digest after every (applied) GSUB lookup.
|
|
|
15b6c325
|
2022-11-16T15:59:13
|
|
[layout] Use a buffer digest for GPOS to skip whole lookups
|
|
|
a053b84c
|
2022-11-16T14:39:25
|
|
[gsubgpos] Optimize set-digest initialization
Previously we were once collecting set-digest for each subtable,
and another time for each lookup.
Now we compute the one for each lookup simply from the ones for
its subtables.
|
|
|
f4813e3b
|
2022-09-09T10:34:09
|
|
[instance] update collect_lookups/prune_features()/closure_features() with variations
Some features will be substituted with variations, so we do not collect
lookups from the original feature tables.
|
|
|
64e2f2fc
|
2022-09-09T09:36:19
|
|
[instance] store active featurevariation record/condition idxes in the plan
|
|
|
a3e6dbbb
|
2022-09-02T13:53:10
|
|
[subset] skip the script if it's not included in the final subset
|
|
|
8d611a7f
|
2022-07-22T22:49:40
|
|
[repacker] remove temporary visibility overrides in gsubgpos.
|
|
|
b1d38a6d
|
2022-07-19T23:33:16
|
|
[repack] WIP implement extension promotion mechanism.
|
|
|
3f7a74ff
|
2022-07-19T21:50:13
|
|
[repacker] WIP extension promotion implementation.
|
|
|
94de3280
|
2022-07-24T18:51:55
|
|
[GSUB/GPOS] Adjust buffer message
|
|
|
bc80e4b9
|
2022-07-24T18:48:38
|
|
[GSUB/GPOS] More buffer message for contextuals
Fixes https://github.com/harfbuzz/harfbuzz/issues/3753
|
|
|
5a9c7930
|
2022-07-22T21:33:15
|
|
Add HB_NO_BEYOND_64K
|
|
|
00dfbbce
|
2022-07-21T11:39:32
|
|
[Coverage] Rename and templatize intersected_coverage_glyphs
|
|
|
6826b2c3
|
2022-07-20T13:25:34
|
|
[gsubgpos/closure] Minor condition use bool operator
|
|
|
1665cf6b
|
2022-07-19T13:56:23
|
|
[>64k:layout] Split GDEFVersion1
https://github.com/be-fonts/boring-expansion-spec/issues/36
|
|
|
99f017f4
|
2022-07-11T13:13:19
|
|
[>64k:layout:GSUBGPOS] Implement format 2
Implements https://github.com/be-fonts/boring-expansion-spec/issues/58
Subset does NOT lower format.
|
|
|
9ef9fc01
|
2022-07-11T13:29:23
|
|
[>64k:layout] Templatize GSUBGPOSFormat1
|
|
|
5fd0a3f0
|
2022-07-11T13:10:04
|
|
[>64k:layout] Templatize GSUBGPOSFormat1
|
|
|
6d0e3e67
|
2022-07-11T12:28:53
|
|
[>64k:layout] Split GSUBGPOSVersion1 into own struct
|
|
|
04c5cd40
|
2022-07-11T11:39:41
|
|
[subset/layout] Move find_duplicate_features to subset from layout
|
|
|
1bf8fa2f
|
2022-07-08T14:22:21
|
|
[>64k:layout:(Chain)Context] Implement format 4
Implements rest of https://github.com/be-fonts/boring-expansion-spec/issues/34
Subset does NOT lower format.
|
|
|
0f13eb1f
|
2022-07-08T13:43:33
|
|
[>64k:layout] Templatize (Chain)ContextFormat1
|
|
|
d1f58e59
|
2022-07-08T13:25:07
|
|
[>64k:layout:(Chain)Context] Implement format 5
Implements part of https://github.com/be-fonts/boring-expansion-spec/issues/34
Subset does NOT lower format.
|
|
|
a90c5af9
|
2022-07-08T13:11:47
|
|
[>64k:layout] Templatize (Chain)ContextFormat2
|
|
|
9d0e9faa
|
2022-07-06T17:49:07
|
|
[>64k:layout] Templatable bunch of GSUBGPOS internal functions
|
|
|
429b387a
|
2022-07-06T17:37:11
|
|
[>64k:layout] Support HBUINT24 in skippy_iter
|
|
|
1ef67a6d
|
2022-07-06T17:28:02
|
|
[gsubgpos] Remove HBUINT16 from matcher
|
|
|
6d051f40
|
2022-07-08T14:08:51
|
|
[layout] Simplify StructAfter<> usage
|
|
|
5192294f
|
2022-07-08T14:00:24
|
|
.
|
|
|
8b349e11
|
2022-07-06T17:22:34
|
|
[gsubgpos] Remove HBUINT16 from match functions signatures
|
|
|
f114b18c
|
2022-07-06T17:31:46
|
|
[gsubgpos] Break skippy_iter set_match_func into two
|
|
|
4119f73c
|
2022-06-08T06:34:48
|
|
[subset/layout] Rename dispatch_closure_lookups_recurse_func to dispatch_recurse_func<>
|
|
|
21346af0
|
2022-06-07T14:41:39
|
|
[layout-cache] Adjust cost-function for recent change
|
|
|
bfee6839
|
2022-06-07T11:41:05
|
|
[layout-cache] Cache lookahead, not input, classdef in ChainContextFormat2
From the commit:
+ /* For ChainContextFormat2 we cache the LookaheadClassDef instead of InputClassDef.
+ * The reason is that most heavy fonts want to identify a glyph in context and apply
+ * a lookup to it. In this scenario, the length of the input sequence is one, whereas
+ * the lookahead / backtrack are typically longer. The one glyph in input sequence is
+ * looked-up below and no input glyph is looked up in individual rules, whereas the
+ * lookahead and backtrack glyphs are tried. Since we match lookahead before backtrack,
+ * we should cache lookahead. This decisions showed a 20% improvement in shaping of
+ * the Gulzar font.
https://github.com/harfbuzz/harfbuzz/pull/3636
|
|
|
39820af7
|
2022-06-07T10:18:38
|
|
[layout] Add HB_NO_OT_LAYOUT_LOOKUP_CACHE to disable caching lookups
Enabled when optimize-size profile is enabled.
|
|
|
c8fb048f
|
2022-06-07T09:20:27
|
|
[gsubgpos] Document caching
|
|
|
5963cf44
|
2022-06-07T09:12:45
|
|
[gsubgpos] Merge cache_enter and cache_leave entry points
Saves a pointer per subtable
|
|
|
d4c09e9a
|
2022-06-07T09:03:30
|
|
[gsubgpos] Remove apply_cached() entry point
Just use a bool to apply()
|
|
|
b96622d1
|
2022-06-05T02:45:41
|
|
[layout] Use a cache for main input ClassDef of (Chain)ContextLookupFormat2
This commit adds a per-lookup caching infrastructure to GSUB/GPOS, and
uses it to cache input ClassDef.get_class value for (Chain)ContextLookupFormat2.
For fonts heavy on use of heave class-based2 context matching, this shows
a good speedup. For NotoNastaliqUrdu for example, I observe 17% speedup.
Unfortunately not many other lookups can use a cache like this :(.
https://github.com/harfbuzz/harfbuzz/pull/3636
|
|
|
e6409d39
|
2022-06-05T06:57:37
|
|
Revert "[layout] Use a cache for main input ClassDef of (Chain)ContextLookups"
This reverts commit 57d1c08739d0acd94b96da2f9d5dd6c0ff3b3722.
Err. This was an accident.
|
|
|
57d1c087
|
2022-06-05T02:45:41
|
|
[layout] Use a cache for main input ClassDef of (Chain)ContextLookups
|
|
|
f9b643f6
|
2022-06-04T07:29:40
|
|
[layout] s/hb_get_subtables_context_t/hb_accelerate_subtables_context_t/g
|
|
|
a42c624f
|
2022-06-03T01:22:34
|
|
Convert one final use of hashmap to unique_ptr
|
|
|
f13a7954
|
2022-06-03T01:17:20
|
|
[subset] Convert another use of hashmap to unique_ptr
|
|
|
997d9cc4
|
2022-06-02T18:04:12
|
|
[map] Make unique_ptr hashable
|
|
|
bca710e8
|
2022-06-02T12:06:25
|
|
[gsubgpos] Use map has() instead of get() when appropriate
|