Merge branch 'add_get_work_space' into bwd_weight_bf16_splitk

example/20_convnd_bwd_weight_xdl/convnd_bwd_weight_xdl.cpp

@@ -299,7 +299,7 @@ int main(int argc, char* argv[])
     // alloc work space
     size_t bwd_weight_workspace_size = conv->GetWorkSpaceSize(argument.get());
     float ave_time                   = 0.f;
-    if(bwd_weight_workspace_size > 0)
+    if(std::is_same<InDataType, ck::bhalf_t>::value && split_k > 1)
     {
         DeviceMem wei_work_space_device_buf(bwd_weight_workspace_size);
         wei_work_space_device_buf.SetZero();

include/ck/tensor_operation/gpu/device/device_convnd_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp

@@ -1269,14 +1269,6 @@ struct DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_
                                 arg.filter_spatial_lengths_[1] * arg.filter_spatial_lengths_[2] *
                                 sizeof(float);
             }
-            else
-            {
-                WorkSpaceSize = arg.Conv_K_ * 0;
-            }
-        }
-        else
-        {
-            WorkSpaceSize = arg.Conv_K_ * 0;
         }
         return WorkSpaceSize;
     }