Merge pull request #16376 from JulienMaille:cuda-dnn-for-older-gpus

5 years ago · 282fcb90dc
parent d85e67d3ec a696348ec5
commit 282fcb90dc
1 changed files with 2 additions and 0 deletions
--- a/modules/dnn/src/cuda/shortcut.cu
+++ b/modules/dnn/src/cuda/shortcut.cu
@ -103,7 +103,9 @@ void input_shortcut(const csl::Stream& stream, csl::TensorSpan<T> output, csl::T
    }
 }

+#if !defined(__CUDA_ARCH__) || (__CUDA_ARCH__ >= 530)
 template void input_shortcut(const Stream&, TensorSpan<__half>, TensorView<__half>, TensorView<__half>);
+#endif
 template void input_shortcut(const Stream&, TensorSpan<float>, TensorView<float>, TensorView<float>);

 }}}} /* namespace cv::dnn::cuda4dnn::kernels */