For statically scheduled superscalars that don't perform fusion, which is probably the common choice for statically scheduled designs, this change will generally be a deoptimization. For dynamically scheduled designs that don't perform fusion, it's probably more or less neutral. Not sure how these facts should drive the decision; just pointing them out.
On Wed, Jul 9, 2025 at 6:28 AM Jeff Law <jeffreya...@gmail.com> wrote: > > One thing I forgot to bring up in the patchwork meeting yesterday. > > Philip or Craig asked if we should add the most basic fusions to the > generic tuning models for the two toolchains. > > I'm generally in favor of making that kind of change. I don't think > anyone believes it'd be a major performance driver, but it does slightly > reduce the search space when we do need to chase things down. > > lui/auipc+addi would fall into that set. It's unclear if any others would. > > Thoughts? > > Jeff >
