提交 f6dd8251 authored 作者: notoraptor's avatar notoraptor

Update documentation and config flags about supported cuDNN algorithms.

上级 c5cd87fa
...@@ -668,12 +668,12 @@ import theano and print the config variable, as in: ...@@ -668,12 +668,12 @@ import theano and print the config variable, as in:
String value: String value:
``'small'``, ``'none'``, ``'large'``, ``'fft'``, ``'fft_tiling'``, ``'small'``, ``'none'``, ``'large'``, ``'fft'``, ``'fft_tiling'``,
``'winograd'``, ``'guess_once'``, ``'guess_on_shape_change'``, ``'winograd'``, ``'winograd_non_fused'``, ``'guess_once'``, ``'guess_on_shape_change'``,
``'time_once'``, ``'time_on_shape_change'``. ``'time_once'``, ``'time_on_shape_change'``.
Default: ``'small'`` Default: ``'small'``
3d convolution only support ``'none'``, ``'winograd'``, ``'guess_once'``, 3d convolution only support ``'none'``, ``'small'``, ``'fft_tiling'``, ``'guess_once'``,
``'guess_on_shape_change'``, ``'time_once'``, ``'time_on_shape_change'``. ``'guess_on_shape_change'``, ``'time_once'``, ``'time_on_shape_change'``.
.. attribute:: config.dnn.conv.algo_bwd .. attribute:: config.dnn.conv.algo_bwd
...@@ -684,24 +684,24 @@ import theano and print the config variable, as in: ...@@ -684,24 +684,24 @@ import theano and print the config variable, as in:
.. attribute:: config.dnn.conv.algo_bwd_filter .. attribute:: config.dnn.conv.algo_bwd_filter
String value: String value:
``'none'``, ``'deterministic'``, ``'fft'``, ``'small'``, ``'guess_once'``, ``'none'``, ``'deterministic'``, ``'fft'``, ``'small'``, ``'winograd_non_fused'``, ``'fft_tiling'``, ``'guess_once'``,
``'guess_on_shape_change'``, ``'time_once'``, ``'time_on_shape_change'``. ``'guess_on_shape_change'``, ``'time_once'``, ``'time_on_shape_change'``.
Default: ``'none'`` Default: ``'none'``
3d convolution only supports ``'none'``, ``'guess_once'``, 3d convolution only supports ``'none'``, ``'small'``, ``'guess_once'``,
``'guess_on_shape_change'``, ``'time_once'``, ``'time_on_shape_change'``. ``'guess_on_shape_change'``, ``'time_once'``, ``'time_on_shape_change'``.
.. attribute:: config.dnn.conv.algo_bwd_data .. attribute:: config.dnn.conv.algo_bwd_data
String value: String value:
``'none'``, ``'deterministic'``, ``'fft'``, ``'fft_tiling'``, ``'winograd'``, ``'none'``, ``'deterministic'``, ``'fft'``, ``'fft_tiling'``, ``'winograd'``,
``'guess_once'``, ``'guess_on_shape_change'``, ``'time_once'``, ``'winograd_non_fused'``, ``'guess_once'``, ``'guess_on_shape_change'``, ``'time_once'``,
``'time_on_shape_change'``. ``'time_on_shape_change'``.
Default: ``'none'`` Default: ``'none'``
3d convolution only support ``'none'``, ``'winograd'``, 3d convolution only support ``'none'``, ``'deterministic'``, ``'fft_tiling'``
``'guess_once'``, ``'guess_on_shape_change'``, ``'time_once'``, ``'guess_once'``, ``'guess_on_shape_change'``, ``'time_once'``,
``'time_on_shape_change'``. ``'time_on_shape_change'``.
......
...@@ -273,11 +273,11 @@ SUPPORTED_DNN_CONV_ALGO_RUNTIME = ('guess_once', 'guess_on_shape_change', 'time_ ...@@ -273,11 +273,11 @@ SUPPORTED_DNN_CONV_ALGO_RUNTIME = ('guess_once', 'guess_on_shape_change', 'time_
# Those are the supported algorithm by Theano, # Those are the supported algorithm by Theano,
# The tests will reference those lists. # The tests will reference those lists.
SUPPORTED_DNN_CONV_ALGO_FWD = ('small', 'none', 'large', 'fft', 'fft_tiling', 'winograd') + SUPPORTED_DNN_CONV_ALGO_RUNTIME SUPPORTED_DNN_CONV_ALGO_FWD = ('small', 'none', 'large', 'fft', 'fft_tiling', 'winograd', 'winograd_non_fused') + SUPPORTED_DNN_CONV_ALGO_RUNTIME
SUPPORTED_DNN_CONV_ALGO_BWD_DATA = ('none', 'deterministic', 'fft', 'fft_tiling', 'winograd') + SUPPORTED_DNN_CONV_ALGO_RUNTIME SUPPORTED_DNN_CONV_ALGO_BWD_DATA = ('none', 'deterministic', 'fft', 'fft_tiling', 'winograd', 'winograd_non_fused') + SUPPORTED_DNN_CONV_ALGO_RUNTIME
SUPPORTED_DNN_CONV_ALGO_BWD_FILTER = ('none', 'deterministic', 'fft', 'small') + SUPPORTED_DNN_CONV_ALGO_RUNTIME SUPPORTED_DNN_CONV_ALGO_BWD_FILTER = ('none', 'deterministic', 'fft', 'small', 'winograd_non_fused', 'fft_tiling') + SUPPORTED_DNN_CONV_ALGO_RUNTIME
SUPPORTED_DNN_CONV_PRECISION = ('as_input_f32', 'as_input', 'float16', 'float32', 'float64') SUPPORTED_DNN_CONV_PRECISION = ('as_input_f32', 'as_input', 'float16', 'float32', 'float64')
......
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