Tune I8

gevtushenko · Jun 6, 2023 · db1eb3d · db1eb3d
1 parent 0f57445
commit db1eb3d
Showing 1 changed file with 9 additions and 0 deletions.
diff --git a/cub/device/dispatch/dispatch_scan.cuh b/cub/device/dispatch/dispatch_scan.cuh
@@ -215,6 +215,15 @@ struct sm90_tuning
   using delay_constructor = detail::default_delay_constructor_t<AccumT>;
 };
 
+template <class AccumT>
+struct sm90_tuning<AccumT, 1>
+{
+  static constexpr int threads = 192;
+  static constexpr int items   = 22;
+
+  using delay_constructor = detail::fixed_delay_constructor_t<288, 1200>;
+};
+
 template <class AccumT>
 struct sm90_tuning<AccumT, 2>
 {