[subset] speed up feature collection when tags are specified.

Precompute a feature index filter to avoid needing to iterate the feature tag list for each encountered feature index. For this particular fuzzer case speeds up feature collection from 50s to 2s.

src/hb-ot-layout.cc

diff --git a/src/hb-ot-layout.cc b/src/hb-ot-layout.cc
index 97d2971..bc80586 100644
--- a/src/hb-ot-layout.cc
+++ b/src/hb-ot-layout.cc
@@ -993,10 +993,46 @@
 {
   hb_collect_features_context_t (hb_face_t *face,
 				 hb_tag_t   table_tag,
-				 hb_set_t  *feature_indexes_)
+				 hb_set_t  *feature_indices_,
+                                 const hb_tag_t *features)
+
     : g (get_gsubgpos_table (face, table_tag)),
-      feature_indexes (feature_indexes_),
-      script_count (0),langsys_count (0), feature_index_count (0) {}
+      feature_indices (feature_indices_),
+      has_feature_filter (false),
+      script_count (0),langsys_count (0), feature_index_count (0)
+  {
+    compute_feature_filter (features);
+  }
+
+  void compute_feature_filter (const hb_tag_t *features)
+  {
+    if (features == nullptr)
+    {
+      has_feature_filter = false;
+      return;
+    }
+
+    has_feature_filter = true;
+    for (; *features; features++)
+    {
+      hb_tag_t tag = *features;
+      unsigned index;
+      g.find_feature_index (tag, &index);
+      if (index == OT::Index::NOT_FOUND_INDEX) continue;
+
+      feature_indices_filter.add(index);
+      for (int i = (int) index - 1; i >= 0; i--)
+      {
+        if (g.get_feature_tag (i) != tag) break;
+        feature_indices_filter.add(i);
+      }
+      for (unsigned i = index + 1; i < g.get_feature_count (); i++)
+      {
+        if (g.get_feature_tag (i) != tag) break;
+        feature_indices_filter.add(i);
+      }
+    }
+  }
 
   bool visited (const OT::Script &s)
   {
@@ -1045,7 +1081,9 @@
 
   public:
   const OT::GSUBGPOS &g;
-  hb_set_t           *feature_indexes;
+  hb_set_t *feature_indices;
+  hb_set_t  feature_indices_filter;
+  bool has_feature_filter;
 
   private:
   hb_set_t visited_script;
@@ -1057,37 +1095,31 @@
 
 static void
 langsys_collect_features (hb_collect_features_context_t *c,
-			  const OT::LangSys  &l,
-			  const hb_tag_t     *features)
+			  const OT::LangSys  &l)
 {
   if (c->visited (l)) return;
 
-  if (!features)
+  if (!c->has_feature_filter)
   {
     /* All features. */
     if (l.has_required_feature () && !c->visited_feature_indices (1))
-      c->feature_indexes->add (l.get_required_feature_index ());
+      c->feature_indices->add (l.get_required_feature_index ());
 
+    // TODO(garretrieger): filter out indices >= feature count?
     if (!c->visited_feature_indices (l.featureIndex.len))
-      l.add_feature_indexes_to (c->feature_indexes);
+      l.add_feature_indexes_to (c->feature_indices);
   }
   else
   {
-    /* Ugh. Any faster way? */
-    for (; *features; features++)
+    if (c->feature_indices_filter.is_empty()) return;
+    unsigned int num_features = l.get_feature_count ();
+    for (unsigned int i = 0; i < num_features; i++)
     {
-      hb_tag_t feature_tag = *features;
-      unsigned int num_features = l.get_feature_count ();
-      for (unsigned int i = 0; i < num_features; i++)
-      {
-	unsigned int feature_index = l.get_feature_index (i);
+      unsigned int feature_index = l.get_feature_index (i);
+      if (!c->feature_indices_filter.has (feature_index)) continue;
 
-	if (feature_tag == c->g.get_feature_tag (feature_index))
-	{
-	  c->feature_indexes->add (feature_index);
-	  break;
-	}
-      }
+      c->feature_indices->add (feature_index);
+      c->feature_indices_filter.del (feature_index);
     }
   }
 }
@@ -1095,8 +1127,7 @@
 static void
 script_collect_features (hb_collect_features_context_t *c,
 			 const OT::Script   &s,
-			 const hb_tag_t *languages,
-			 const hb_tag_t *features)
+			 const hb_tag_t *languages)
 {
   if (c->visited (s)) return;
 
@@ -1105,14 +1136,13 @@
     /* All languages. */
     if (s.has_default_lang_sys ())
       langsys_collect_features (c,
-				s.get_default_lang_sys (),
-				features);
+				s.get_default_lang_sys ());
+
 
     unsigned int count = s.get_lang_sys_count ();
     for (unsigned int language_index = 0; language_index < count; language_index++)
       langsys_collect_features (c,
-				s.get_lang_sys (language_index),
-				features);
+				s.get_lang_sys (language_index));
   }
   else
   {
@@ -1121,8 +1151,8 @@
       unsigned int language_index;
       if (s.find_lang_sys_index (*languages, &language_index))
 	langsys_collect_features (c,
-				  s.get_lang_sys (language_index),
-				  features);
+				  s.get_lang_sys (language_index));
+
     }
   }
 }
@@ -1153,7 +1183,7 @@
 			       const hb_tag_t *features,
 			       hb_set_t       *feature_indexes /* OUT */)
 {
-  hb_collect_features_context_t c (face, table_tag, feature_indexes);
+  hb_collect_features_context_t c (face, table_tag, feature_indexes, features);
   if (!scripts)
   {
     /* All scripts. */
@@ -1161,8 +1191,7 @@
     for (unsigned int script_index = 0; script_index < count; script_index++)
       script_collect_features (&c,
 			       c.g.get_script (script_index),
-			       languages,
-			       features);
+			       languages);
   }
   else
   {
@@ -1172,8 +1201,7 @@
       if (c.g.find_script_index (*scripts, &script_index))
 	script_collect_features (&c,
 				 c.g.get_script (script_index),
-				 languages,
-				 features);
+				 languages);
     }
   }
 }

test/fuzzing/fonts/clusterfuzz-testcase-minimized-hb-subset-fuzzer-5241922561114112

diff --git a/test/fuzzing/fonts/clusterfuzz-testcase-minimized-hb-subset-fuzzer-5241922561114112 b/test/fuzzing/fonts/clusterfuzz-testcase-minimized-hb-subset-fuzzer-5241922561114112
new file mode 100644
index 0000000..122a968
--- /dev/null
+++ b/test/fuzzing/fonts/clusterfuzz-testcase-minimized-hb-subset-fuzzer-5241922561114112
Binary files differ