[llvm-dev] how to widen vector in target defined intrinsics

余瑶 via llvm-dev llvm-dev at lists.llvm.org
Fri Oct 29 02:35:47 PDT 2021

Hello Developers,
    I add an Intrinsic in my target called int_npp_veltadd using llvm-11.0.0,  
    Its declaration is <64 x i8> @llvm.nnp.veltadd(<64 x i8> %a, <64 x i8> %b, <64 x i8> %c, <64 x i8> %d).
    But when the input is <28 x i8>, my backend get an error in LegalizeVectorTypes.cpp line 2733, it says "Do not know how to widen the result of this operator!".  Because of the vector <28 x i8> is extended,  the function TLI.getOperationAction return EXPAND. So the target's ReplaceNodeResults cannot be called. so I cannot widen the vector to 64 x i8 in my target.
  how can I widen vector in this situation?
  thanks a lot!~
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20211029/105fa662/attachment.html>

More information about the llvm-dev mailing list