From 3ff9ebc868355a40f8d560e2692c4887417efd78 Mon Sep 17 00:00:00 2001
From: Behdad Esfahbod <behdad@behdad.org>
Date: Mon, 13 May 2024 14:46:55 -0600
Subject: [PATCH] [aat] For short words, use buffer digest to skip morx
 subtables

Shows 3 / 4% speedup in en-words test case with Helvetica Neue.
---
 src/hb-aat-layout-common.hh     |  9 +++++++++
 src/hb-aat-layout-morx-table.hh | 31 +++++++++++++++++++++++++++++++
 2 files changed, 40 insertions(+)

diff --git a/src/hb-aat-layout-common.hh b/src/hb-aat-layout-common.hh
index dbee48717..6a4952f0b 100644
--- a/src/hb-aat-layout-common.hh
+++ b/src/hb-aat-layout-common.hh
@@ -60,6 +60,7 @@ struct hb_aat_apply_context_t :
   const ankr *ankr_table;
   const OT::GDEF *gdef_table;
   const hb_sorted_vector_t<hb_aat_map_t::range_flags_t> *range_flags = nullptr;
+  hb_set_digest_t buffer_digest = hb_set_digest_t::full ();
   hb_set_digest_t machine_glyph_set = hb_set_digest_t::full ();
   hb_set_digest_t left_set = hb_set_digest_t::full ();
   hb_set_digest_t right_set = hb_set_digest_t::full ();
@@ -927,6 +928,14 @@ struct StateTableDriver
 	      machine (machine_),
 	      num_glyphs (face_->get_num_glyphs ()) {}
 
+  template <typename context_t>
+  bool is_idempotent_on_all_out_of_bounds (context_t *c, hb_aat_apply_context_t *ac)
+  {
+    const auto entry = machine.get_entry (StateTableT::STATE_START_OF_TEXT, CLASS_OUT_OF_BOUNDS);
+    return !c->is_actionable (ac->buffer, this, entry) &&
+	    machine.new_state (entry.newState) == StateTableT::STATE_START_OF_TEXT;
+  }
+
   template <typename context_t, typename set_t = hb_set_digest_t>
   void drive (context_t *c, hb_aat_apply_context_t *ac)
   {
diff --git a/src/hb-aat-layout-morx-table.hh b/src/hb-aat-layout-morx-table.hh
index 4a94e6a8f..f3d42f4e6 100644
--- a/src/hb-aat-layout-morx-table.hh
+++ b/src/hb-aat-layout-morx-table.hh
@@ -170,6 +170,11 @@ struct RearrangementSubtable
     driver_context_t dc (this);
 
     StateTableDriver<Types, EntryData> driver (machine, c->face);
+
+    if (driver.is_idempotent_on_all_out_of_bounds (&dc, c) &&
+	!c->buffer_digest.may_have (c->machine_glyph_set))
+      return_trace (false);
+
     driver.drive (&dc, c);
 
     return_trace (dc.ret);
@@ -267,6 +272,7 @@ struct ContextualSubtable
       {
 	buffer->unsafe_to_break (mark, hb_min (buffer->idx + 1, buffer->len));
 	buffer->info[mark].codepoint = *replacement;
+	c->buffer_digest.add (*replacement);
 	if (has_glyph_classes)
 	  _hb_glyph_info_set_glyph_props (&buffer->info[mark],
 					  gdef.get_glyph_props (*replacement));
@@ -296,6 +302,7 @@ struct ContextualSubtable
       if (replacement)
       {
 	buffer->info[idx].codepoint = *replacement;
+	c->buffer_digest.add (*replacement);
 	if (has_glyph_classes)
 	  _hb_glyph_info_set_glyph_props (&buffer->info[idx],
 					  gdef.get_glyph_props (*replacement));
@@ -328,6 +335,11 @@ struct ContextualSubtable
     driver_context_t dc (this, c);
 
     StateTableDriver<Types, EntryData> driver (machine, c->face);
+
+    if (driver.is_idempotent_on_all_out_of_bounds (&dc, c) &&
+	!c->buffer_digest.may_have (c->machine_glyph_set))
+      return_trace (false);
+
     driver.drive (&dc, c);
 
     return_trace (dc.ret);
@@ -586,6 +598,11 @@ struct LigatureSubtable
     driver_context_t dc (this, c);
 
     StateTableDriver<Types, EntryData> driver (machine, c->face);
+
+    if (driver.is_idempotent_on_all_out_of_bounds (&dc, c) &&
+	!c->buffer_digest.may_have (c->machine_glyph_set))
+      return_trace (false);
+
     driver.drive (&dc, c);
 
     return_trace (dc.ret);
@@ -654,6 +671,7 @@ struct NoncontextualSubtable
       if (replacement)
       {
 	info[i].codepoint = *replacement;
+	c->buffer_digest.add (*replacement);
 	if (has_glyph_classes)
 	  _hb_glyph_info_set_glyph_props (&info[i],
 					  gdef.get_glyph_props (*replacement));
@@ -788,6 +806,9 @@ struct InsertionSubtable
 	  if (unlikely (!buffer->copy_glyph ())) return;
 	/* TODO We ignore KashidaLike setting. */
 	if (unlikely (!buffer->replace_glyphs (0, count, glyphs))) return;
+	for (unsigned int i = 0; i < count; i++)
+	  c->buffer_digest.add (glyphs[i]);
+	ret = true;
 	if (buffer->idx < buffer->len && !before)
 	  buffer->skip_glyph ();
 
@@ -853,6 +874,11 @@ struct InsertionSubtable
     driver_context_t dc (this, c);
 
     StateTableDriver<Types, EntryData> driver (machine, c->face);
+
+    if (driver.is_idempotent_on_all_out_of_bounds (&dc, c) &&
+	!c->buffer_digest.may_have (c->machine_glyph_set))
+      return_trace (false);
+
     driver.drive (&dc, c);
 
     return_trace (dc.ret);
@@ -1348,6 +1374,11 @@ struct mortmorx
 
     c->buffer->unsafe_to_concat ();
 
+    if (c->buffer->len < 16)
+      c->buffer_digest = c->buffer->digest ();
+    else
+      c->buffer_digest = hb_set_digest_t::full ();
+
     c->set_lookup_index (0);
     const Chain<Types> *chain = &firstChain;
     unsigned int count = chainCount;