[Mlir-commits] [mlir] [MLIR][NVGPU] Add convert.fpext and convert.fptrunc Ops (PR #199700)

[Guray Ozen]

================
@@ -671,4 +671,78 @@ def NVGPU_RcpOp : NVGPU_Op<"rcp", [Pure,
   let hasVerifier = 1;
 }
 
+//===----------------------------------------------------------------------===//
+// NVGPU Conversion Ops
+//===----------------------------------------------------------------------===//
+
+def AnyI32Like : TypeOrValueSemanticsContainer<I32, "scalar i32 or vector of i32">;
+
+// nvgpu.convert.fptrunc only supports the satfinite and none saturation modes.
+def NVGPU_SaturationModeSatfiniteOrNone :
+  ConfinedAttr<SaturationModeAttr, [EnumAttrIsOneOf<SaturationModeAttr,
+                [SaturationModeNone, SaturationModeFinite]>]>;
+
+def NVGPU_ConvertFPTruncOp : NVGPU_Op<"convert.fptrunc", [Pure]> {
----------------
grypp wrote:

```suggestion
def NVGPU_TruncfOp : NVGPU_Op<"truncf", [Pure]> {
```

https://github.com/llvm/llvm-project/pull/199700