Fix typos and forbide float16 precision for gradinput deterministic.

theano/gpuarray/cudnn_defs.py

@@ -9,9 +9,9 @@ for a given cuDNN version.
 
 Currently supported cuDNN APIs:
 
- - v5.1
- - v6.0
- - v7.0
+ - v5.1*
+ - v6.0*
+ - v7.0*
 
 """
 
@@ -125,7 +125,7 @@ class CuDNNV51(object):
     def get_supported_dtype_configs(self, check_runtime=None):
         """
         Return the tuple of data type configurations supported by this version of cuDNN.
-        This is currently convenient for both cuDNN V5.1 and V6, as Theano does not
+        This is currently convenient for all supported cuDNN versions, as Theano does not
         yet support new data types (like INT8, INT8x4, etc.).
 
         ``check_runtime`` may be a function that tests if a data type configuration is supported.::
@@ -199,7 +199,9 @@ class CuDNNV51(object):
             return not is_true_half_config(dtype, precision)
         if algo == algorithms.CUDNN_CONVOLUTION_BWD_DATA_ALGO_1:
             # CUDNN_CONVOLUTION_BWD_DATA_ALGO_1: all data type configs supported.
-            return True
+            # NB: Let's avoid float16 precision, as some strange errors may be encountered
+            # with that precision ( see https://github.com/Theano/Theano/pull/5932/ )
+            return not is_true_half_config(dtype, precision)
         if algo == algorithms.CUDNN_CONVOLUTION_BWD_DATA_ALGO_FFT:
             return ndim == 2 and (is_pseudo_half_config(dtype, precision) or is_float_config(dtype, precision))
         if algo == algorithms.CUDNN_CONVOLUTION_BWD_DATA_ALGO_FFT_TILING: