[PATCH] D28508: [NVPTX] Implement NVPTXTargetLowering::getSqrtEstimate.

[Justin Lebar via Phabricator via llvm-commits]

jlebar added a comment.

All right, I think this should work.  I verified by hand that the approximation routines do something sane --namely, that when the error is zero, we get the intended exact result.  I don't have the numerical chops to say whether it's actually a refinement or not...  :)


https://reviews.llvm.org/D28508