kparzysz-quic commented on PR #104: URL: https://github.com/apache/tvm-rfcs/pull/104#issuecomment-1751038124
Sorry for the delay... What I'm aiming at is to be able to lower the TIR to a generic CPU, that is to an architecture that does not support SVE. The TIR will need to have some default lowering in CodeGenLLVM/CodeGenCPU, so being able to do that is important. For that, we should be able to assume that `vscale` is 1. The `vscale` would simply be an indicator to the codegen (in TVM) that the code **may** be lowered to SVE. What I wrote earlier about `vfactor` was that the value of it depended on the data type for which it was calculated. If you're sticking with `vscale`, then it may seem like we don't need it, but the issue is with using "x * vscale" as an idiom: if you have several occurrences of "4 * vscale" in an expression, it may end up being rearranged to something like "(4*vi + 4) * vscale", or "ceildiv(128, 4 * vscale)" may end up being "ceildiv(32, vscale)". So, instead of "x * vscale", I suggest "vscale(x)". -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
