Add missing output_alias to miopen_fusion op (#1076)

This causes incorrect memory coloring, which was causing the accuracy failures in the vision model when enabling the pointwise fusions. Resnet50, inceptionv3 and inceptionv4 do verify now in the driver.

src/targets/gpu/fuse_ops.cpp

@@ -587,6 +587,11 @@ struct miopen_fusion
         return pack(f(self.ops, "ops"));
     }
 
+    std::ptrdiff_t output_alias(const std::vector<shape>& shapes) const
+    {
+        return shapes.size() - 1;
+    }
+
     value compile(context& ctx, const shape&, std::vector<shape> inputs)
     {
         // Compensate for allocation