ggml: bypass code incompatible with CUDA < 11.1 (#2020)

`cudaHostRegisterReadOnly` parameter was only introduced in CUDA 11.1

See this issue for more details:
https://github.com/ggerganov/whisper.cpp/issues/2007

diff --git a/ggml-cuda.cu b/ggml-cuda.cu
index be8e33a56c40f45dac8807b419925379c72f3c0f..5607386cee9b5be72b233961c1f2ba0f3d1c3b5c 100644 (file)
--- a/ggml-cuda.cu
+++ b/ggml-cuda.cu
@@ -2797,6 +2797,7 @@ GGML_CALL bool ggml_backend_cuda_register_host_buffer(void * buffer, size_t size
         return false;
     }
 
+#if CUDART_VERSION >= 11100
     cudaError_t err = cudaHostRegister(buffer, size, cudaHostRegisterPortable | cudaHostRegisterReadOnly);
     if (err != cudaSuccess) {
         // clear the error
@@ -2807,6 +2808,9 @@ GGML_CALL bool ggml_backend_cuda_register_host_buffer(void * buffer, size_t size
         return false;
     }
     return true;
+#else
+    return false;
+#endif
 }
 
 GGML_CALL void ggml_backend_cuda_unregister_host_buffer(void * buffer) {