Update docs to explain CPU incompatibilities for Convolution layouts

src/operator/nn/convolution-inl.h

@@ -103,7 +103,8 @@ struct ConvolutionParam : public dmlc::Parameter<ConvolutionParam> {
     .add_enum("NDHWC", mshadow::kNDHWC)
     .set_default(dmlc::optional<int>())
     .describe("Set layout for input, output and weight. Empty for\n    "
-              "default layout: NCW for 1d, NCHW for 2d and NCDHW for 3d.");
+              "default layout: NCW for 1d, NCHW for 2d and NCDHW for 3d."
+              "NHWC and NDHWC are only supported on GPU.");
   }
   // Adjusts kernel size for effects of dilation in the dimension `dim`.
   index_t DilatedKernelSize(int dim) const {

src/operator/nn/convolution.cc

@@ -426,7 +426,7 @@ then we have::
 If ``no_bias`` is set to be true, then the ``bias`` term is ignored.
 
 The default data ``layout`` is *NCHW*, namely *(batch_size, channel, height,
-width)*. We can choose other layouts such as *NHWC*.
+width)*. We can choose other layouts such as *NWC*.
 
 If ``num_group`` is larger than 1, denoted by *g*, then split the input ``data``
 evenly into *g* parts along the channel axis, and also evenly split ``weight``

src/operator/nn/deconvolution-inl.h

@@ -113,7 +113,8 @@ struct DeconvolutionParam : public dmlc::Parameter<DeconvolutionParam> {
       .add_enum("NDHWC", mshadow::kNDHWC)
       .set_default(dmlc::optional<int>())
       .describe("Set layout for input, output and weight. Empty for "
-                "default layout, NCW for 1d, NCHW for 2d and NCDHW for 3d.");
+                "default layout, NCW for 1d, NCHW for 2d and NCDHW for 3d."
+                "NHWC and NDHWC are only supported on GPU.");
   }
 
   template<size_t ndim>