Functions: parallelize materializing arrays after field evaluation

This improves performance e.g. when creating an integer attribute
based on an index field. For 4 million vertices, I measured a speedup
from 3.5 ms to 1.2 ms.

1 files changed, 11 insertions, 6 deletions

diff --git a/source/blender/functions/intern/field.cc b/source/blender/functions/intern/field.cc
index 986d6ddc19e..944674c23a9 100644
--- a/source/blender/functions/intern/field.cc
+++ b/source/blender/functions/intern/field.cc
@@ -468,16 +468,21 @@ Vector<GVArray> evaluate_fields(ResourceScope &scope,
       /* Still have to copy over the data in the destination provided by the caller. */
       if (dst_varray.is_span()) {
         /* Materialize into a span. */
-        computed_varray.materialize_to_uninitialized(mask, dst_varray.get_internal_span().data());
+        threading::parallel_for(mask.index_range(), 2048, [&](const IndexRange range) {
+          computed_varray.materialize_to_uninitialized(mask.slice(range),
+                                                       dst_varray.get_internal_span().data());
+        });
       }
       else {
         /* Slower materialize into a different structure. */
         const CPPType &type = computed_varray.type();
-        BUFFER_FOR_CPP_TYPE_VALUE(type, buffer);
-        for (const int i : mask) {
-          computed_varray.get_to_uninitialized(i, buffer);
-          dst_varray.set_by_relocate(i, buffer);
-        }
+        threading::parallel_for(mask.index_range(), 2048, [&](const IndexRange range) {
+          BUFFER_FOR_CPP_TYPE_VALUE(type, buffer);
+          for (const int i : mask.slice(range)) {
+            computed_varray.get_to_uninitialized(i, buffer);
+            dst_varray.set_by_relocate(i, buffer);
+          }
+        });
       }
       r_varrays[out_index] = dst_varray;
     }