[Mlir-commits] [mlir] [NvGpu Dialect] add rcp approxe op (PR #100965)

[Guray Ozen]

================
@@ -802,4 +803,16 @@ def NVGPU_WarpgroupMmaInitAccumulatorOp : NVGPU_Op<"warpgroup.mma.init.accumulat
   let hasVerifier = 1;
 }
 
+def NVGPU_RcpApproxOp : NVGPU_Op<"rcp_approx", [
+  Pure, SameOperandsAndResultType
+]> {
+  let summary = "A wrapper of nvvm rcp.approx.ftz.f";
----------------
grypp wrote:

Nit: It's beyond a wrapper. The lowering transforms multi dimensional vector, and generates nvvm OP for each of them. Maybe let's say something like
F32 rcp calculation for vector types using `nvvm.rcp.*` OP

https://github.com/llvm/llvm-project/pull/100965