BLI: Avoid invoking tbb for small parallel_reduce calls

Apply a change similar to e1309030603980c6b2f33486 for
`parallel_reduce`, just like `parallel_for`. I measured a performance
improvement in viewport FPS of at least 10% with 1 million small
instances (one bottleneck was computing many small bounding boxes).

1 files changed, 10 insertions, 8 deletions

diff --git a/source/blender/blenlib/BLI_task.hh b/source/blender/blenlib/BLI_task.hh
index d87a86ce696..904dea66f7a 100644
--- a/source/blender/blenlib/BLI_task.hh
+++ b/source/blender/blenlib/BLI_task.hh
@@ -77,17 +77,19 @@ Value parallel_reduce(IndexRange range,
                       const Reduction &reduction)
 {
 #ifdef WITH_TBB
-  return tbb::parallel_reduce(
-      tbb::blocked_range<int64_t>(range.first(), range.one_after_last(), grain_size),
-      identity,
-      [&](const tbb::blocked_range<int64_t> &subrange, const Value &ident) {
-        return function(IndexRange(subrange.begin(), subrange.size()), ident);
-      },
-      reduction);
+  if (range.size() >= grain_size) {
+    return tbb::parallel_reduce(
+        tbb::blocked_range<int64_t>(range.first(), range.one_after_last(), grain_size),
+        identity,
+        [&](const tbb::blocked_range<int64_t> &subrange, const Value &ident) {
+          return function(IndexRange(subrange.begin(), subrange.size()), ident);
+        },
+        reduction);
+  }
 #else
   UNUSED_VARS(grain_size, reduction);
-  return function(range, identity);
 #endif
+  return function(range, identity);
 }
 
 /**