Why is the optimum latency incorrect for symmetric hybrid-based multipliers when targeting Virtex-II, Spartan-3, or Spartan-3E?
Hybrid multipliers on all device families might not reach peak performance when the "Optimum pipeline stages" value displayed on the final page of the GUI is used. This value can be one less than the true value for a fully-pipelined multiplier, for some configurations.
This is fixed in Multiplier Generator v10.0.
You can work around this issue, but you should consider your application before making changes to the core pipeline stages.
If your application is performance driven, you should increase the number of pipeline stages by 1 to achieve maximum performance for multipliers larger than 20x20.
If your application is area driven, a lower degree of pipelining might save resources in some cases.
The recommendation is to first check to see if the core will meet the designed timing in your larger design before adding more pipeline stages.