wp

moritz-2001 · Oct 5, 2024 · ec14b4b · ec14b4b
1 parent 4e7dbc7
commit ec14b4b
Showing 1 changed file with 3 additions and 4 deletions.
diff --git a/group/group_inclusive_scan.cpp b/group/group_inclusive_scan.cpp
@@ -46,12 +46,11 @@ class MicroBenchGroupInclusiveScan {
       cgh.parallel_for<MicroBenchGroupInclusiveScanKernel<DataT, Iterations>>(
           s::nd_range<1>{num_groups * args.local_size, args.local_size}, [=](cl::sycl::nd_item<1> item) {
             auto g = item.get_group();
-            size_t gid = item.get_global_linear_id();
             DataT d = initialize_type<DataT>(0);
 
-            for(int i = Iterations; i >= 0; --i) {
-              d = s::inclusive_scan_over_group(g, a_[item.get_local_linear_id() + i % args.local_size], cl::sycl::plus<DataT>{});
-              out[gid] = d;
+            for(int i = 0; i < Iterations; ++i) {
+              d = s::inclusive_scan_over_group(g, a_[item.get_local_linear_id()], cl::sycl::plus<DataT>{});
+              out[item.get_local_linear_id()] = d;
             }
           });
     }));