alan-baker
alan-baker
The spec is currently contradictory on this point. `struct` where `N` is greater than zero, but the grammar has: >struct_body_decl : BRACE_LEFT struct_member* BRACE_RIGHT
https://gpuweb.github.io/gpuweb/wgsl/#matrix-builtin-functions already lists it. Suggest we close this issue.
The spec also sometimes says that the attribute decorates something. We should be consistent with how they are "attached" to things too.
When I last looked at libclc conversions, there were some obvious bugs in some of the rounding modes, but I got pushback on trying to fix them. Let's make sure...
I expect this is no longer desirable given where we ended up with diagnostics. It would also be awkward to annotate a variable as uniform only for compute shaders to...
> Compute shaders have to respect the uniformity analysis though? Yes when we did the diagnostics and uniformity opt out, you cannot opt out of the analysis from control barriers...
Those revisions all look ok to me. I'm glad to hear there aren't more issues.
This work is now complete. Any remaining problems should be filed as new bugs.
I've filed #1061.
Can you provide an example of a translation from the spirv-llvm-translator you wish to replace? Would the replacement be directly using a SPIR-V instruction (or extended instruction) or an emulation?...