[PATCH] D89697: * [x86] Implement smarter instruction lowering for FP_TO_UINT from vXf32 to vXi32 for SSE2 and AVX2 by using the exact semantic of the CVTTPS2SI instruction.

[Tom Hender via Phabricator via llvm-commits]

TomHender added a comment.

Sorry for the delayed response from me. I was not actively checking this anymore since there was no further input for a while.

It think back in October I was still looking for a resolution of the apparent cost model inconsistencies. @RKSimon suggested using the LLVM-MCA numbers but it seemed to me that I was adding to the mess, due to apparent inconsistencies in comparison to the other numbers. Thus I was hoping for either a clarification of why this is indeed correct or alternatively an acknowledgement of the issue and that the other costs are still to be updated to be LLVM-MCA-like in the future so this change 'fits in' in the longterm.

In any case you are welcome to take this over as I currently don't expect to have time for LLVM this month.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D89697/new/

https://reviews.llvm.org/D89697