NativeLstm2.so:未定义符号:sgemm_

Question

尝试运行22_train.sh 时

OpCodeCompiler call: /home/end2end/cuda-8.0/bin/nvcc -shared -O2 -std=c++11 -I /home/anaconda2/lib/python2.7/site-packages/tensorflow/include -I /anaconda2/lib/python2.7/site-packages/tensorflow/include/external/nsync/public -I /home/end2endcuda-8.0/include -L /home/end2endcuda-8.0/lib64 -x cu -DGOOGLE_CUDA=1 -Xcompiler -fPIC -arch compute_30 -D_GLIBCXX_USE_CXX11_ABI=0 -g /tmp/returnn_tf_cache/ops/NativeLstm2/357cf3b2da/NativeLstm2.cc -o /tmp/returnn_tf_cache/ops/NativeLstm2/357cf3b2da/NativeLstm2.so
Exception creating layer root/'lstm0_fw' of class RecLayer with opts:
{'direction': 1,
'loss': None,
'n_out': 1024,
'name': 'lstm0_fw',
'network': <TFNetwork 'root' train=<tf.Tensor 'globals/train_flag:0' shape=() dtype=bool>>,
'output': Data(name='lstm0_fw_output', shape=(None, 1024), dtype='float32', batch_dim_axis=1),
'sources': [<EvalLayer 'source' out_type=Data(shape=(None, 40), dtype='float32')>],
'unit': 'nativelstm2'}
Unhandled exception <class 'tensorflow.python.framework.errors_impl.NotFoundError'> in thread <_MainThread(MainThread, started 140224295745280)>, proc 18758.
Thread current, main, <_MainThread(MainThread, started 140224295745280)>:
(Excluded thread.)
That were all threads.
NotFoundError: /tmp/returnn_tf_cache/ops/NativeLstm2/357cf3b2da/NativeLstm2.so: undefined symbol: sgemm_

配置了Cuda 8.0和CUDNN路径。

Answer 1

将注释放入答案：

这可能是由于旧版本造成的，现在应该已经修复了。RETURNN查找sgemm_符号的方式发生了一些变化。