On Wed, Dec 22, 2021 at 12:08 AM Zebediah Figura zfigura@codeweavers.com wrote:
This patch is something of an argument for storing the resource and sampler variables as hlsl_src rather than hlsl_deref. I'm still inclined to avoid this, because that would require the variable to be unchanged between the load and resource load, and that feels like a tricky and dangerous invariant to preserve.
I don't think it's a big deal either way. Sticking to hlsl_deref for now seems fine to me, I don't think we're expecting a lot of juggling of those objects aside from what we already have (which is sort of an artifact of our general IR setup). At any rate, it shouldn't be too big of a mess if we change our mind at some point, as I'd expect the changes to be decently self-contained.