[Mlir-commits] [mlir] [MLIR][NVVM] Add new narrow FP convert Ops (PR #184291)

[Durgadoss R]

================
@@ -2164,6 +2242,99 @@ def NVVM_ConvertF6x2ToF16x2Op :
 def NVVM_ConvertF4x2ToF16x2Op :
   NVVM_ConvertToFP16x2Op_Base<"F4", I8, "F16">;
 
+def NVVM_ConvertF32x2ToS2F6x2Op : NVVM_Op<"convert.f32x2.to.s2f6x2"> {
+  let summary = "Convert a pair of f32 inputs to S2F6x2";
+  let description = [{
+    This Op converts each of the given f32 inputs to the
+    S2F6x2 type. The result `dst` can be an I16 or vector<2xi8>.
+    The `relu` attribute, when set, lowers to the '.relu' variant
+    of the cvt instruction. The optional scaling-factor for the
+    conversion is provided through the operand `scaleFactor`.
+    Only `ue8m0` is supported as the type of the scale-factor currently.
----------------
durga4github wrote:

Let us append "...as an i16 type" to this line since ue8m0 is only 8-bits wide

https://github.com/llvm/llvm-project/pull/184291