On Thu Sep 28 15:16:55 2023 +0000, Giovanni Mascellani wrote:
This means that the DXBC signature is replaced with the DXIL signature as far as the parser is concerned. That means that the VSIR code (and, say, the SPIR-V code downstream) is generated from a potentially different signature than what is returned by `vkd3d_shader_scan()`. I guess that could be confusing for the client.
The same applies to TPF due to signature normalisation. The DXBC signature expresses the same info in different form. `struct vkd3d_shader_signature_element` doesn't support register counts, so if a reason arises for clients to access SM 6 signature info, we would need to extend the ABI.