From 8cef9f1584cccfb27d986b0c09f7face6766d1fc Mon Sep 17 00:00:00 2001 From: Liang-Chi Hsieh Date: Wed, 29 Jan 2014 17:37:57 +0800 Subject: [PATCH 1/2] fix program path. --- data/get_mnist.sh | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/data/get_mnist.sh b/data/get_mnist.sh index 2a9ad083fb4..c807a7b9439 100755 --- a/data/get_mnist.sh +++ b/data/get_mnist.sh @@ -17,7 +17,7 @@ gunzip t10k-labels-idx1-ubyte.gz echo "Creating leveldb..." -../examples/convert_mnist_data.bin train-images-idx3-ubyte train-labels-idx1-ubyte mnist-train-leveldb -../examples/convert_mnist_data.bin t10k-images-idx3-ubyte t10k-labels-idx1-ubyte mnist-test-leveldb +../build/examples/convert_mnist_data.bin train-images-idx3-ubyte train-labels-idx1-ubyte mnist-train-leveldb +../build/examples/convert_mnist_data.bin t10k-images-idx3-ubyte t10k-labels-idx1-ubyte mnist-test-leveldb echo "Done." From 7d723b6b3037e8aa370bdbb0a28facd09ee77494 Mon Sep 17 00:00:00 2001 From: Liang-Chi Hsieh Date: Wed, 29 Jan 2014 18:39:16 +0800 Subject: [PATCH 2/2] choose proper thread number per block according to CUDA architecture. --- include/caffe/common.hpp | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/include/caffe/common.hpp b/include/caffe/common.hpp index e7c5abe7435..7fd7ea6329c 100644 --- 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) {