3 files changed, 5 insertions, 7 deletions
diff --git a/llvm/tools/llvm-gpu-loader/amdhsa.cpp b/llvm/tools/llvm-gpu-loader/amdhsa.cpp
index be1b6b7..5715058 100644
--- a/llvm/tools/llvm-gpu-loader/amdhsa.cpp
+++ b/llvm/tools/llvm-gpu-loader/amdhsa.cpp
@@ -192,7 +192,7 @@ hsa_status_t launch_kernel(hsa_agent_t dev_agent, hsa_executable_t executable,
   // Initialize all the arguments (explicit and implicit) to zero, then set the
   // explicit arguments to the values created above.
   std::memset(args, 0, args_size);
-  std::memcpy(args, &kernel_args, sizeof(args_t));
+  std::memcpy(args, &kernel_args, std::is_empty_v<args_t> ? 0 : sizeof(args_t));
 
   // Initialize the necessary implicit arguments to the proper values.
   int dims = 1 + (params.num_blocks_y * params.num_threads_y != 1) +
@@ -563,7 +563,7 @@ int load_amdhsa(int argc, const char **argv, const char **envp, void *image,
   // Save the return value and perform basic clean-up.
   int ret = *static_cast<int *>(host_ret);
 
-  end_args_t fini_args = {ret};
+  end_args_t fini_args = {};
   if (hsa_status_t err = launch_kernel(
           dev_agent, executable, kernargs_pool, coarsegrained_pool, queue,
           server, single_threaded_params, "_end.kd", fini_args,
diff --git a/llvm/tools/llvm-gpu-loader/llvm-gpu-loader.h b/llvm/tools/llvm-gpu-loader/llvm-gpu-loader.h
index ed34d0b..08861c2 100644
--- a/llvm/tools/llvm-gpu-loader/llvm-gpu-loader.h
+++ b/llvm/tools/llvm-gpu-loader/llvm-gpu-loader.h
@@ -41,9 +41,7 @@ struct start_args_t {
 };
 
 /// The arguments to the '_end' kernel.
-struct end_args_t {
-  int argc;
-};
+struct end_args_t {};
 
 /// Generic interface to load the \p image and launch execution of the _start
 /// kernel on the target device. Copies \p argc and \p argv to the device.
diff --git a/llvm/tools/llvm-gpu-loader/nvptx.cpp b/llvm/tools/llvm-gpu-loader/nvptx.cpp
index 781a045..82b4552 100644
--- a/llvm/tools/llvm-gpu-loader/nvptx.cpp
+++ b/llvm/tools/llvm-gpu-loader/nvptx.cpp
@@ -177,7 +177,7 @@ CUresult launch_kernel(CUmodule binary, CUstream stream, rpc::Server &server,
     handle_error(err);
 
   // Set up the arguments to the '_start' kernel on the GPU.
-  uint64_t args_size = sizeof(args_t);
+  uint64_t args_size = std::is_empty_v<args_t> ? 0 : sizeof(args_t);
   void *args_config[] = {CU_LAUNCH_PARAM_BUFFER_POINTER, &kernel_args,
                          CU_LAUNCH_PARAM_BUFFER_SIZE, &args_size,
                          CU_LAUNCH_PARAM_END};
@@ -342,7 +342,7 @@ int load_nvptx(int argc, const char **argv, const char **envp, void *image,
   if (CUresult err = cuStreamSynchronize(stream))
     handle_error(err);
 
-  end_args_t fini_args = {host_ret};
+  end_args_t fini_args = {};
   if (CUresult err =
           launch_kernel(binary, stream, server, single_threaded_params, "_end",
                         fini_args, print_resource_usage))