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
I'd love to be able to pack opaque types like depth into structs, but whenever I do, this happens:
error: error:0:0 - [VUID-StandaloneSpirv-None-04667] In Vulkan, OpTypeStruct must not contain an opaque type.
%material__screen_space_shadows__directional__SSSDirectionalLight____ = OpTypeStruct %renderer__camera__Camera %v3float %v2uint %v2uint %uint %uint %float %_ptr_UniformConstant_192
|
= note: spirv-val failed
I know you're already inlining all methods that reference opaque types, as one cannot simply pass an opaque type to a function in glsl. Could we also get that for structs?
Example use-case: Custom Samplers (or buffer loaders). Right now you are forced to pass the image to sample from to every single sample call, as you can't store the image in the custom sampler's struct. Often when constructing your sampler you want to pass in the image anyway, to load some metadata about it (eg. size) that will be constant throughout all sampling operations. And let's just hope noone accidentally passes in a different image for sampling than the one used to construct the sampler...
The text was updated successfully, but these errors were encountered:
I'd love to be able to pack opaque types like depth into structs, but whenever I do, this happens:
I know you're already inlining all methods that reference opaque types, as one cannot simply pass an opaque type to a function in glsl. Could we also get that for structs?
Example use-case: Custom Samplers (or buffer loaders). Right now you are forced to pass the image to sample from to every single sample call, as you can't store the image in the custom sampler's struct. Often when constructing your sampler you want to pass in the image anyway, to load some metadata about it (eg. size) that will be constant throughout all sampling operations. And let's just hope noone accidentally passes in a different image for sampling than the one used to construct the sampler...
The text was updated successfully, but these errors were encountered: