Sync from v0.13

csrc/cub_helpers.h

@@ -0,0 +1,18 @@
+#pragma once
+
+#ifndef USE_ROCM
+  #include <cub/cub.cuh>
+  #if CUB_VERSION >= 200800
+    #include <cuda/std/functional>
+using CubAddOp = cuda::std::plus<>;
+using CubMaxOp = cuda::maximum<>;
+  #else   // if CUB_VERSION < 200800
+using CubAddOp = cub::Sum;
+using CubMaxOp = cub::Max;
+  #endif  // CUB_VERSION
+#else
+  #include <hipcub/hipcub.hpp>
+namespace cub = hipcub;
+using CubAddOp = hipcub::Sum;
+using CubMaxOp = hipcub::Max;
+#endif  // USE_ROCM