diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle_streamk_v3.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle_streamk_v3.hpp
old mode 100644
new mode 100755
index 3171208830..eb0e76a76f
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle_streamk_v3.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle_streamk_v3.hpp
@@ -777,7 +777,7 @@ struct DeviceGemm_Xdl_CShuffle_Streamk_V3 : public DeviceGemm_Streamk_V2<ALayout
             {BlockGemmPipelineVersion::v5, "v5"}};
 
         // clang-format off
-        str << "DeviceGemmXdlUniversal"
+        str << "DeviceGemmXdlStreamKUniversal"
             << "<"
             << getGemmSpecializationString(GemmSpec) << ", "
             << std::string(ALayout::name)[0]
@@ -799,7 +799,9 @@ struct DeviceGemm_Xdl_CShuffle_Streamk_V3 : public DeviceGemm_Streamk_V2<ALayout
             << "BlkGemmPipelineVersion: "
             << BlkGemmPipelineVersionToString[BlkGemmPipelineVer] << ", "
             << "BlkGemmPipelinePrefetchStages: "
-            << GridwiseGemm::BlockwiseGemmPipe::PrefetchStages;
+            << GridwiseGemm::BlockwiseGemmPipe::PrefetchStages << ", "
+            << "GridDim: "
+            << grid_dims.x << "x" << grid_dims.y << "x" << grid_dims.z;
         // clang-format on
 
         return str.str();