padding support for dnn_conv

theano/sandbox/cuda/dnn.py

@@ -141,13 +141,15 @@ class GpuDnnConvDesc(GpuOp):
     def c_compiler(self):
         return NVCC_compiler
 
-    def __init__(self, border_mode, subsample=(1, 1), conv_mode='conv'):
-        assert border_mode in ('valid', 'full')
+    def __init__(self, border_mode, subsample=(1, 1), conv_mode='conv',
+                 padding=(0, 0)):
+        assert border_mode in ('valid', 'full', 'padding')
         self.border_mode = border_mode
         assert len(subsample) == 2
         self.subsample = subsample
         assert conv_mode in ('conv', 'cross')
         self.conv_mode = conv_mode
+        self.padding = padding
 
     def make_node(self, img_shape, kern_shape):
         if img_shape.type.ndim != 1 or img_shape.type.dtype != 'int64':
@@ -162,11 +164,19 @@ class GpuDnnConvDesc(GpuOp):
         img_shape, kern_shape = inputs
         desc, = outputs
 
-        if self.border_mode == "valid":
-            bmode = 1
+        pad_h_spec, pad_w_spec = map(int, self.padding)
+        assert pad_h_spec >= 0 and pad_w_spec >= 0
+
+        if self.border_mode == 'padding':
+            bmode = 2
         else:
-            assert self.border_mode == "full"
-            bmode = 0
+            assert pad_h_spec == 0 and pad_w_spec == 0, \
+                'padding not zero, but border_mode != "padding"'
+            if self.border_mode == "valid":
+                bmode = 1
+            else:
+                assert self.border_mode == "full"
+                bmode = 0
 
         if self.conv_mode == 'conv':
             conv_flag = 'CUDNN_CONVOLUTION'
@@ -185,7 +195,10 @@ class GpuDnnConvDesc(GpuOp):
     %(fail)s
   }
 
-  if (%(bmode)d == 1) {
+  if (%(bmode)d == 2) {
+    pad_h%(name)s = %(pad_h_spec)d;
+    pad_w%(name)s = %(pad_w_spec)d;
+  } else if (%(bmode)d == 1) {
     pad_h%(name)s = 0;
     pad_w%(name)s = 0;
   } else if (%(bmode)d == 0) {
@@ -218,10 +231,11 @@ class GpuDnnConvDesc(GpuOp):
 }
 """ % dict(name=name, img_shape=img_shape, kern_shape=kern_shape, desc=desc,
            bmode=bmode, conv_flag=conv_flag, fail=sub['fail'],
-           subsx=self.subsample[0], subsy=self.subsample[1])
+           subsx=self.subsample[0], subsy=self.subsample[1],
+           pad_h_spec=pad_h_spec, pad_w_spec=pad_w_spec)
 
     def c_code_cache_version(self):
-        return (1,)
+        return (2,)
 
 
 class GpuDnnConvBase(DnnBase):
@@ -450,7 +464,7 @@ class GpuDnnConvGradI(GpuDnnConvBase):
 
 
 def dnn_conv(img, kerns, border_mode='valid', subsample=(1, 1),
-             conv_mode='conv'):
+             conv_mode='conv', padding=(0, 0)):
     """
     GPU convolution using cuDNN from NVIDIA.
 
@@ -459,7 +473,7 @@ def dnn_conv(img, kerns, border_mode='valid', subsample=(1, 1),
 
     :param img: images to do the convolution over
     :param kerns: convolution filters
-    :param border_mode: one of 'valid', 'full' (default: 'valid')
+    :param border_mode: one of 'valid', 'full' or 'padding'(default: 'valid')
     :param subsample: perform subsampling of the output (default: (1, 1))
     :param conv_mode: perform convolution (kernels flipped) or cross-correlation.  One of 'conv', 'cross'. (default: 'conv')
 
@@ -470,7 +484,7 @@ def dnn_conv(img, kerns, border_mode='valid', subsample=(1, 1),
     img = gpu_contiguous(img)
     kerns = gpu_contiguous(kerns)
     desc = GpuDnnConvDesc(border_mode=border_mode, subsample=subsample,
-                          conv_mode=conv_mode)(img.shape, kerns.shape)
+                          conv_mode=conv_mode, padding=padding)(img.shape, kerns.shape)
     return GpuDnnConv()(img, kerns, desc)