On Mon May 29 16:18:57 2023 +0000, Nikolay Sivov wrote:
Depends on what that pass is going to do. The obvious way I can thing of is to have separate jump types, if e.g. we'll want to check for dead texkill/discards later, and for that you'll have to know the meaning of its argument.
Yes, but the mere fact that it *can* happen is a problem for me, because it requires that every developer keeps in mind that `struct hlsl_ir_jump` changes meaning depending on the SM, which I think it's a bad idea. Especially, I don't want this to become a precedent for other IR instructions to be assigned different meanings depending on the SM. I don't know whether this is already happening for floating point rules, but if it is it's already bad enough, and if it isn't then let's not start it now.