Commit Graph
16342 Commits
Author SHA1 Message Date
Behdad Esfahbod 4d4792ce3d [subset/hdmx] Speed up more for retaingids 2023-06-03 20:34:51 -06:00
Behdad Esfahbod a3f7ed709b [subset/hdmx] Don't clear memory unnecessarily 2023-06-03 20:25:14 -06:00
Behdad Esfahbod e263c3ccbb [subset/hdmx] Speed up 2023-06-03 20:20:57 -06:00
Behdad Esfahbod 0e16be9b2c [subset-plan] Reuse glyph_for_gid result 2023-06-03 19:45:44 -06:00
Behdad Esfahbod 590fb3529a [subset] Reduce memory pressure 2023-06-03 19:21:22 -06:00
Behdad Esfahbod c03c300016 [subset] Tweak memory allocation
Reduces chances of rework, specially for glyf table if padding
is needed...
2023-06-03 19:05:10 -06:00
Behdad Esfahbod 38fa032330 Revert "[subset/loca] Build in the serializer memory"
This reverts commit 546b518186.
2023-06-03 18:51:09 -06:00
Behdad Esfahbod 546b518186 [subset/loca] Build in the serializer memory
Because it's hot. Doesn't seem to speed up though. Slight
slowdown. Going to revert.
2023-06-03 18:50:35 -06:00
Behdad Esfahbod 3cb2b49e61 [subset/loca] Shuffle code around 2023-06-03 18:45:20 -06:00
Behdad Esfahbod f014112d16 Revert "[subset] Allow adding table directly from subset()"
This reverts commit 25a6514887.
2023-06-03 18:27:24 -06:00
Behdad Esfahbod 25a6514887 [subset] Allow adding table directly from subset()
And do it for hmtx. Going to revert since it slows things down.
Apparently the serializer buffer is in caches and hot, so faster
to write to and copy even.
2023-06-03 18:26:36 -06:00
Behdad Esfahbod b0d0babb2e [subset/loca] Rewrite loop faster 2023-06-03 17:14:55 -06:00
Behdad Esfahbod 8dab444102 [subset/hmtx] Micro-optimize last commit
hb_zip has its own overhead...
2023-06-03 17:06:30 -06:00
Behdad Esfahbod e981dc24c8 [subset/hmtx] Rewrite loop faster for retaingid 2023-06-03 17:04:00 -06:00
Behdad Esfahbod 4a9da02619 [subset/hmtx] Micro-optimize 2023-06-03 16:46:29 -06:00
Behdad Esfahbod 6a3fcc64f3 [map] Don't call hash() if map is empty 2023-06-03 16:39:55 -06:00
Behdad Esfahbod 735d249639 [map] Remove an unlikely
Empty maps are common in some subsetting operations.
2023-06-03 16:35:03 -06:00
Behdad Esfahbod ca27925d55 Use hb_codepoint_pair_t in more places 2023-06-03 16:18:15 -06:00
Behdad Esfahbod 841e86fd98 [subset/hmtx] Further speedup! 2023-06-03 16:09:45 -06:00
Behdad Esfahbod de729ec105 [meta] Return reference from reference-wrapper 2023-06-03 15:48:54 -06:00
Behdad Esfahbod c7493efeaf [subset/loca] Micro-optimize 2023-06-03 15:39:27 -06:00
Behdad Esfahbod 13a4c7b3d3 [subset/hmtx] Comment 2023-06-03 15:12:53 -06:00
Behdad Esfahbod 4da2996fe2 [doc] Fix 2023-06-03 15:08:20 -06:00
Behdad Esfahbod c3e7f45166 Add HB_CODEPOINT_INVALID
Fixes https://github.com/harfbuzz/harfbuzz/issues/4262
2023-06-03 14:56:47 -06:00
Behdad Esfahbod efefec13cc [subset/glyf] Remove unneeded codepath 2023-06-02 20:33:52 -06:00
Behdad Esfahbod ac1f5ca0d9 [subset/loca] Micro-optimize 2023-06-02 17:07:44 -06:00
Behdad Esfahbod 6eae932566 [subset/loca] Reduce a vector allocation 2023-06-02 17:05:27 -06:00
Behdad Esfahbod cd249d2364 [subset/hmtx] Micro-optimize 2023-06-02 16:15:05 -06:00
Behdad Esfahbod ec8965e432 [subset/glyf] Don't clear loca allocation 2023-06-02 16:10:25 -06:00
Behdad Esfahbod 916629d182 [subset/hmtx] Don't clear allocation 2023-06-02 16:08:49 -06:00
Behdad Esfahbod 24b069cd53 [subset] Pre-alloc a few maps 2023-06-02 15:55:53 -06:00
Behdad Esfahbod 7319d0d712 [subset-plan] Add a couple of map pre-allocations
Approximate...
2023-06-02 15:41:47 -06:00
Behdad Esfahbod 76904b0423 [subset-plan] Simplify / speedup planning glyphset 2023-06-02 15:30:12 -06:00
Behdad Esfahbod 965ce7bba4 [subset-plan] Speed up planning new_to_old_gid_list 2023-06-02 15:24:00 -06:00
Behdad Esfahbod b67e464b1c [subset/retaingid] Regain perf lost
In 0f12fd5a66
2023-06-02 15:11:06 -06:00
Behdad Esfahbod 0f12fd5a66 [subset/glyf] Fix for non-monotonic glyphmap
Slows things down again, but is correct. Still a good win
combined with the previous changes.
2023-06-02 13:58:55 -06:00
Behdad Esfahbod 66ce902c3a [glyf] Reduce allocation again 2023-06-02 13:53:21 -06:00
Behdad Esfahbod 0e0110e7f2 [glyf] Avoid a SubsetGlyph copy
Unfortunately hb_second() always does a copy...
2023-06-02 13:34:58 -06:00
Behdad Esfahbod 33ea8b5e5d [subset/glyf] Don't allocate empty SubsetGlyf's
Shows 33% speedup on Mplus/retaingid/10 benchmark:

Comparing before to after
Benchmark                                                                   Time             CPU      Time Old      Time New       CPU Old       CPU New
--------------------------------------------------------------------------------------------------------------------------------------------------------
BM_subset/subset_glyphs/Mplus1p-Regular.ttf/retaingids/10                -0.3296         -0.3294             0             0             0             0
2023-06-02 13:28:07 -06:00
Garret Rieger ad872e2313 [subset] Optimize hmtx/vmtx serialization.
Make serializer allocation up front to avoid bounds checking overhead for each metric.

Benchmarks:
BM_subset/subset_glyphs/Mplus1p-Regular.ttf/retaingids/10_median                   -0.1005         -0.1005             0             0             0             0
BM_subset/subset_glyphs/Mplus1p-Regular.ttf/retaingids/64_median                   -0.0693         -0.0692             0             0             0             0
BM_subset/subset_glyphs/Mplus1p-Regular.ttf/retaingids/512_median                  -0.0294         -0.0293             1             1             1             1
BM_subset/subset_glyphs/Mplus1p-Regular.ttf/retaingids/4096_median                 -0.0033         -0.0032             3             3             3             3
BM_subset/subset_glyphs/Mplus1p-Regular.ttf/retaingids/10000_median                +0.0170         +0.0171             7             7             7             7
2023-06-02 12:41:44 -06:00
Behdad Esfahbod c6368e014d [map] Return const reference from operator()
Like we do in operator[].
2023-06-01 20:51:17 -06:00
Behdad Esfahbod b04ca1c188 Revert "[subset/glyf] Exact allocation"
This reverts commit a830f085c8.

Weirdly enough in a benchmark I'm running I see this show
some slowdown. Makes zero sense.
2023-06-01 20:43:07 -06:00
Behdad Esfahbod a830f085c8 [subset/glyf] Exact allocation 2023-06-01 20:41:46 -06:00
Behdad Esfahbod ff0b85cf57 [subset/glyf] Empty .notdef only if old-gid was also 0
Otherwise it wasn't a .notdef, even if new-gid is 0.
2023-06-01 20:30:56 -06:00
Behdad Esfahbod 65a7d8c6f0 [glyf] Empty glyphs need no padding 2023-06-01 20:20:00 -06:00
Behdad Esfahbod 73c18d1d42 [glyf] Micro-optimize padded_offsets 2023-06-01 20:11:22 -06:00
Behdad Esfahbod 88d8a6d69f [BEInt] Use packed-int trick for writing as well 2023-06-01 19:49:44 -06:00
Behdad Esfahbod 4d853b8ba6 [subset/regaingid/glyf] Add a fastpath for empty glyphs
Mplus benchmark:
Comparing before to after
Benchmark                                                                      Time             CPU      Time Old      Time New       CPU Old       CPU New
-----------------------------------------------------------------------------------------------------------------------------------------------------------
BM_subset/subset_glyphs/Mplus1p-Regular.ttf/retaingids/10                   -0.1764         -0.1771             0             0             0             0
BM_subset/subset_glyphs/Mplus1p-Regular.ttf/retaingids/64                   -0.1394         -0.1394             0             0             0             0
BM_subset/subset_glyphs/Mplus1p-Regular.ttf/retaingids/512                  -0.0644         -0.0645             0             0             0             0
BM_subset/subset_glyphs/Mplus1p-Regular.ttf/retaingids/4096                 +0.0132         +0.0131             1             1             1             1
BM_subset/subset_glyphs/Mplus1p-Regular.ttf/retaingids/10000                -0.0029         -0.0034             3             3             3             3
BM_subset/subset_glyphs/Mplus1p-Regular.ttf/10                              +0.0364         +0.0362             0             0             0             0
BM_subset/subset_glyphs/Mplus1p-Regular.ttf/64                              +0.0346         +0.0343             0             0             0             0
BM_subset/subset_glyphs/Mplus1p-Regular.ttf/512                             +0.0271         +0.0268             0             0             0             0
BM_subset/subset_glyphs/Mplus1p-Regular.ttf/4096                            +0.0193         +0.0193             1             1             1             1
BM_subset/subset_glyphs/Mplus1p-Regular.ttf/10000                           -0.0240         -0.0243             2             2             2             2
OVERALL_GEOMEAN                                                             -0.0305         -0.0307             0             0             0             0
2023-06-01 19:24:36 -06:00
Behdad Esfahbod 6ec7629fba [subset/glyf] Add a const 2023-06-01 18:18:30 -06:00
Behdad Esfahbod 9f54ab922f [subset-plan] Keep a reverse-gid vector
To speed up iteration.
2023-06-01 18:07:59 -06:00