Sync from upstream llama.cpp repository

ggml/src/ggml-cuda/set-rows.cuh

@@ -0,0 +1,7 @@
+#pragma once
+
+#include "common.cuh"
+
+#define CUDA_SET_ROWS_BLOCK_SIZE 256
+
+void ggml_cuda_op_set_rows(ggml_backend_cuda_context & ctx, ggml_tensor * dst);