simplify the gpu gather implementation

src/targets/gpu/device/gather.cpp

@@ -23,14 +23,6 @@ argument gather(hipStream_t stream,
             const auto* indices_ptr = device_cast(indices.data());
             auto* out_ptr           = device_cast(output.data());
             const auto* in_ptr      = device_cast(input.data());
-            if(output_shape.scalar())
-            {
-                gs_launch(stream, 1)(
-                    [=](auto i) { out_ptr[i] = in_ptr[static_cast<int>(indices_ptr[0])]; });
-            }
-            else
-            {
-                // if indices are a scalar, output has one dim smaller than input
             auto& input_shape = args[0].get_shape();
             auto lens         = input_shape.lens();
             lens[axis_index]  = args[1].get_shape().elements();
@@ -44,7 +36,6 @@ argument gather(hipStream_t stream,
                     out_ptr[ii]        = in_ptr[desc_input.linear(in_idx)];
                 });
             });
-            }
         });
     });