Allow to concatenate conv case generators.

theano/gpuarray/cudnn_defs.py

@@ -165,11 +165,6 @@ class CuDNNV51(object):
         if algo == algorithms.CUDNN_CONVOLUTION_FWD_ALGO_FFT_TILING:
             if ndim == 2:
                 return is_pseudo_half_config(dtype, precision) or is_float_config(dtype, precision)
-                # NB: For cuDNN V6:
-                # " Data Type Config Support: PSEUDO_HALF_CONFIG, FLOAT_CONFIG
-                # (DOUBLE_CONFIG is also supported when the task can be handled by 1D FFT,
-                # ie, one of the filter dimension, width or height is 1)"
-                # Could be checked only when being in C code.
             if ndim == 3:
                 return not is_true_half_config(dtype, precision)
         if algo == algorithms.CUDNN_CONVOLUTION_FWD_ALGO_WINOGRAD:
@@ -210,9 +205,6 @@ class CuDNNV51(object):
         if algo == algorithms.CUDNN_CONVOLUTION_BWD_DATA_ALGO_FFT_TILING:
             if ndim == 2:
                 return is_pseudo_half_config(dtype, precision) or is_float_config(dtype, precision)
-                # NB: For cuDNN V6: "(DOUBLE_CONFIG is also supported when the task can be handled by 1D FFT,
-                # ie, one of the filter dimension, width or height is 1)"
-                # Could be checked only when being in C code.
             if ndim == 3:
                 return not is_true_half_config(dtype, precision)
         if algo == algorithms.CUDNN_CONVOLUTION_BWD_DATA_ALGO_WINOGRAD:
@@ -265,6 +257,22 @@ class CuDNNV6(CuDNNV51):
                                       ('CUDNN_REDUCE_TENSOR_NORM2', 'norm2'),
                                       ctype='cudnnReduceTensorOp_t')
 
+    def fwd_algo_supports_dtype_config(self, algo, dtype, precision, ndim):
+        is_supported = super(CuDNNV6, self).fwd_algo_supports_dtype_config(algo, dtype, precision, ndim)
+        if not is_supported:
+            algorithms = self.cudnnConvolutionFwdAlgo_t
+            algo = algorithms.fromalias(algo)
+            if algo == algorithms.CUDNN_CONVOLUTION_FWD_ALGO_FFT_TILING:
+                # NB: For cuDNN V6:
+                # "Data Type Config Support: PSEUDO_HALF_CONFIG, FLOAT_CONFIG
+                # (DOUBLE_CONFIG is also supported when the task can be handled by 1D FFT,
+                # ie, one of the filter dimension, width or height is 1)"
+                # Could be checked only in C code. By default, let's allow DOUBLE_CONFIG.
+                return ndim == 2 and (is_pseudo_half_config(dtype, precision) or
+                                      is_float_config(dtype, precision) or
+                                      is_double_config(dtype, precision))
+        return is_supported
+
     def bwd_filter_algo_supports_dtype_config(self, algo, dtype, precision, ndim):
         is_supported = super(CuDNNV6, self).bwd_filter_algo_supports_dtype_config(algo, dtype, precision, ndim)
         if not is_supported:
@@ -276,6 +284,22 @@ class CuDNNV6(CuDNNV51):
                                       is_double_config(dtype, precision))
         return is_supported
 
+    def bwd_data_algo_supports_dtype_config(self, algo, dtype, precision, ndim):
+        is_supported = super(CuDNNV6, self).bwd_data_algo_supports_dtype_config(algo, dtype, precision, ndim)
+        if not is_supported:
+            algorithms = self.cudnnConvolutionBwdDataAlgo_t
+            algo = algorithms.fromalias(algo)
+            if algo == algorithms.CUDNN_CONVOLUTION_BWD_DATA_ALGO_FFT_TILING:
+                # NB: For cuDNN V6:
+                # "Data Type Config Support: PSEUDO_HALF_CONFIG, FLOAT_CONFIG
+                # (DOUBLE_CONFIG is also supported when the task can be handled by 1D FFT,
+                # ie, one of the filter dimension, width or height is 1)"
+                # Could be checked only in C code. By default, let's allow DOUBLE_CONFIG.
+                return ndim == 2 and (is_pseudo_half_config(dtype, precision) or
+                                      is_float_config(dtype, precision) or
+                                      is_double_config(dtype, precision))
+        return is_supported
+
 
 class CuDNNV7(CuDNNV6):
     version = 7

theano/gpuarray/tests/check_dnn_conv.py

@@ -16,7 +16,7 @@
 from __future__ import absolute_import, print_function, division
 
 import sys
-from itertools import product
+from itertools import product, chain
 
 import nose
 import numpy as np
@@ -316,6 +316,19 @@ class ConvCaseGenerator:
                                all_border_modes, all_conv_modes, all_alphas, all_betas))
 
 
+class ConvCaseGeneratorChain:
+    """
+    Help class to concatenate many conv case generators.
+    """
+
+    def __init__(self, *conv_case_generators):
+        assert all(isinstance(g, ConvCaseGenerator) for g in conv_case_generators)
+        self.generators = conv_case_generators
+
+    def get_cases(self, filter=None):
+        return chain(*[generator.get_cases(filter) for generator in self.generators])
+
+
 class CuDNNV51ConvCaseGenerator(object):
     """
     Helper class to generate specific test cases for every algorithm supported by cuDNN V5.1.
@@ -430,14 +443,18 @@ class CuDNNV6ConvCaseGenerator(CuDNNV51ConvCaseGenerator):
 
     def _fwd_fft_tiling(self, ndim):
         if ndim == 2:
-            filters_sizes = [(32, 5), (256, 1), (10, 10), (5, 1)]
             subsamples = [(1, 1)]
-            borders = [(1, 1), (2, 1)]
-            return ConvCaseGenerator(ndim=ndim,
-                                     filters_sizes=filters_sizes,
-                                     subsamples=subsamples,
-                                     borders=borders,
-                                     dilations=self._dilations(ndim))
+            # wDesc's filter height must be greater than convDesc's zero-padding height
+            # wDesc's filter width must be greater than convDesc's zero-padding width
+            filters_sizes = [(32, 5), (10, 10)]
+            borders = [(1, 1), (6, 4)]
+            generator1 = ConvCaseGenerator(ndim=ndim, dilations=self._dilations(ndim), subsamples=subsamples,
+                                           filters_sizes=filters_sizes, borders=borders)
+            filters_sizes = [(256, 1), (5, 1)]
+            borders = [(1, 0), (2, 0)]
+            generator2 = ConvCaseGenerator(ndim=ndim, dilations=self._dilations(ndim), subsamples=subsamples,
+                                           filters_sizes=filters_sizes, borders=borders)
+            return ConvCaseGeneratorChain(generator1, generator2)
         if ndim == 3:
             return super(CuDNNV6ConvCaseGenerator, self)._fwd_fft_tiling(ndim)
 
@@ -445,10 +462,10 @@ class CuDNNV6ConvCaseGenerator(CuDNNV51ConvCaseGenerator):
         return self._fwd_none(ndim)
 
     def _gw_fft_tiling(self, ndim):
-        inputs_sizes = [(256, 1), (20, 1)]
+        inputs_sizes = [(247, 1), (20, 1)]
         filters_sizes = [(3, 1), (10, 1)]
         subsamples = [(1,) * ndim]
-        borders = [(1, 1), (2, 1)]
+        borders = [(1, 0), (2, 0)]
         return ConvCaseGenerator(ndim=ndim,
                                  inputs_sizes=inputs_sizes,
                                  filters_sizes=filters_sizes,
@@ -467,6 +484,8 @@ class CuDNNV6ConvCaseGenerator(CuDNNV51ConvCaseGenerator):
     def gw(self, algo, ndim):
         if algo == self.NONE:
             return self._gw_none(ndim)
+        if algo == self.FFT_TILING:
+            return self._gw_fft_tiling(ndim)
         return super(CuDNNV6ConvCaseGenerator, self).gw(algo, ndim)
 
     def gi(self, algo, ndim):

theano/gpuarray/tests/run_dnn_conv.py

@@ -126,8 +126,8 @@ if args.algo not in SUPPORTED_DNN_CONV_ALGO_RUNTIME:
     if test == BWD_DATA:
         check_config = cudnn.bwd_data_algo_supports_dtype_config(args.algo, args.dtype, args.precision, ndim)
     if not check_config:
-        raise ValueError('%s computation does not support configuration (%s, %s) for algo %s.' % (
-            test, args.dtype, args.precision, args.algo))
+        print('Warning: %s computation does not normally support configuration (%s, %s) for algo %s.' % (
+            test, args.dtype, args.precision, args.algo), file=sys.stderr)
 algo = args.algo
 dtype = args.dtype
 precision = args.precision