You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Jun 10, 2024. It is now read-only.
The GLSL back end is currently ad hoc in terms of phi removal, and looks like it has to keep getting more ad hoc as more complex topologies are handled. (To make multiple names in LLVM IR map back again to a single name in GLSL, somehow). It is logical that executing a reg-to-mem pass would just do the whole job in one go, and end the need to ad hoc handle new topologies as they show up.
So, this needs an experiment, and in particular, checking out what happens to the names LunarGOO comes up with in the final GLSL, as it is desirable to have the same names when two similar shaders to the same thing with the same names (helps compression).
The text was updated successfully, but these errors were encountered:
The GLSL back end is currently ad hoc in terms of phi removal, and looks like it has to keep getting more ad hoc as more complex topologies are handled. (To make multiple names in LLVM IR map back again to a single name in GLSL, somehow). It is logical that executing a reg-to-mem pass would just do the whole job in one go, and end the need to ad hoc handle new topologies as they show up.
So, this needs an experiment, and in particular, checking out what happens to the names LunarGOO comes up with in the final GLSL, as it is desirable to have the same names when two similar shaders to the same thing with the same names (helps compression).
The text was updated successfully, but these errors were encountered: