[PATCH] D126158: [MLIR][GPU] Replace fdiv on fp16 with promoted (fp32) multiplication with reciprocal plus one (conditional) Newton iteration.

[Stephan Herhut via Phabricator via llvm-commits]

herhut accepted this revision.
herhut added a comment.
This revision is now accepted and ready to land.

Separate pass works for me.



================
Comment at: mlir/include/mlir/Dialect/LLVMIR/Transforms/Passes.td:19
 
+def NVVMOptimize : Pass<"nvvm-optimize"> {
+  let summary = "Optimize NVVM IR";
----------------
Maybe `llvm-optimize-for-nvvm`? Or even `llvm-optimize-for-nvvm-target`? 

This does not really optimize `nvvm` but rewrites `llvm` ir. 


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D126158/new/

https://reviews.llvm.org/D126158