On Fri Jan 13 00:08:40 2023 +0000, Zebediah Figura wrote:
This still wasn't looking quite like I imagined, so I ended up playing with it myself for a while. I ended up with the following branch: https://gitlab.winehq.org/zfigura/vkd3d/-/tree/copyprop Thoughts?
A couple non-obvious notes regarding the changes I made to your patches—
* I added a comment explaining why we have these copy-propagation passes. Especially because even now, every time I look at the pass I have to look up and remember why we can't just vectorize *all* stores.
* I got rid of the type check from copy_propagation_compute_constant_replacement(). We already filter out complex types in the caller, and there's not much point preventing object types from showing up here since they can't have constant values anyway.