提交 3f5ed4bc authored 作者: notoraptor's avatar notoraptor

Use new debug options to print runtime chosen algo for dnn_gw.

上级 1ffa497a
...@@ -202,6 +202,15 @@ APPLY_SPECIFIC(conv_gw)(PyGpuArrayObject *input, PyGpuArrayObject *output, ...@@ -202,6 +202,15 @@ APPLY_SPECIFIC(conv_gw)(PyGpuArrayObject *input, PyGpuArrayObject *output,
prev_top_dims[i] = PyGpuArray_DIM(output, i); prev_top_dims[i] = PyGpuArray_DIM(output, i);
} }
} }
#ifdef DEBUG
char algorithm_name[128];
if (0 != theano_enum_to_string_cudnnConvolutionBwdFilterAlgo_t(algo, algorithm_name)) {
return 1;
};
// NB: This is printed only when algorithm is chosen at runtime.
fprintf(stderr, "(using %s) ", algorithm_name);
#endif
} }
// The FFT implementation does not support strides, 1x1 filters or inputs // The FFT implementation does not support strides, 1x1 filters or inputs
......
Markdown 格式
0%
您添加了 0 到此讨论。请谨慎行事。
请先完成此评论的编辑!
注册 或者 后发表评论