Fix GPU version computation

ginkgo-project · May 19, 2020 · 0e4fc49 · 0e4fc49
1 parent 6e0d7ca
commit 0e4fc49
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 4 deletions.
diff --git a/cuda/matrix/csr_kernels.cu b/cuda/matrix/csr_kernels.cu
@@ -168,8 +168,8 @@ GKO_ENABLE_IMPLEMENTATION_SELECTION(select_merge_path_spmv, merge_path_spmv);
 template <typename ValueType, typename IndexType>
 int compute_items_per_thread(std::shared_ptr<const CudaExecutor> exec)
 {
-    const int version = exec->get_major_version()
-                        << 4 + exec->get_minor_version();
+    const int version =
+        (exec->get_major_version() << 4) + exec->get_minor_version();
     // The num_item is decided to make the occupancy 100%
     // TODO: Extend this list when new GPU is released
     //       Tune this parameter

diff --git a/hip/matrix/csr_kernels.hip.cpp b/hip/matrix/csr_kernels.hip.cpp
@@ -181,8 +181,8 @@ int compute_items_per_thread(std::shared_ptr<const HipExecutor> exec)
 #if GINKGO_HIP_PLATFORM_NVCC
 
 
-    const int version = exec->get_major_version()
-                        << 4 + exec->get_minor_version();
+    const int version =
+        (exec->get_major_version() << 4) + exec->get_minor_version();
     // The num_item is decided to make the occupancy 100%
     // TODO: Extend this list when new GPU is released
     //       Tune this parameter