[MLIR][Math] Improve docs for round op (NFC)

[chelini]

As per discussion in: https://reviews.llvm.org/D127286/new/

[antiagainst]

Thanks for making this clearer!

mlir/include/mlir/Dialect/Math/IR/MathOps.td

671

I don't think we need to explicitly mention "this lowers to .." given that's irrelevant to the definition of this op in particular; we only need to define its semantics here. (And I do intend to support this op's lowering in SPIR-V for example.)

The rest LGTM!

[chelini]

Address comment.

mlir/include/mlir/Dialect/Math/IR/MathOps.td

let description = [{
 
     ```
     operation ::= ssa-id `=` `math.round` ssa-use `:` type
     ```
 
     The `round` operation returns the operand rounded to the nearest integer
     value in floating-point format. It takes one operand of floating point type
     (i.e., scalar, tensor or vector) and produces one result of the same type.
+    Currently, the operation lowers to an llvm intrinsics (llvm.round.*) or to Libm
+    calls. In both cases, the operation rounds the argument to the nearest integer
+    value in floating-point format, rounding halfway cases away from zero,
+    regardless of the current rounding direction.
 
     Example:
 
     ```mlir
     // Scalar round operation.
     %a = math.round %b : f64
     ```
   }];
 }
 
 #endif // MATH_OPS