choose proper thread number per block according to CUDA architecture.

diff --git a/include/caffe/common.hpp b/include/caffe/common.hpp
index e7c5abe..7fd7ea6 100644 (file)
--- a/include/caffe/common.hpp
+++ b/include/caffe/common.hpp
@@ -49,7 +49,12 @@ using boost::shared_ptr;
 
 
 // We will use 1024 threads per block, which requires cuda sm_2x or above.
-const int CAFFE_CUDA_NUM_THREADS = 1024;
+#if __CUDA_ARCH__ >= 200
+    const int CAFFE_CUDA_NUM_THREADS = 1024;
+#else
+    const int CAFFE_CUDA_NUM_THREADS = 512;
+#endif
+
 
 
 inline int CAFFE_GET_BLOCKS(const int N) {