lenary wrote:

Conventionally for a feature like this the implementation order is Assembly, 
then CodeGen, then ABI, because you can have CodeGen without ABI (for TU-local 
operations).

The reason we avoided Q was because of lack of implementations, but we realised 
it was worth implementing assembly to ensure we had it (you've found that 
discussion).

If you are working on a hardware implementation of Q you would like codegen 
for, and have the capacity to commit to helping maintain and review the CodeGen 
and ABI implementation, I don't think we'd have a problem having it implemented 
upstream. If there's no commitment to hardware, then it's less likely to 
justify itself. 

If you don't have a hardware roadmap for Q, then my personal preference would 
be that you focus on increasing GlobalISel support for RISC-V for things 
already supported by SDag, rather than implementing codegen for Q, but this is 
merely my preference, and I am not in charge of you :) 

https://github.com/llvm/llvm-project/pull/195166
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to