[Mlir-commits] [mlir] [mlir] Add FP software implementation lowering pass: `arith-to-apfloat` (PR #166618)

[Renato Golin]

================
@@ -0,0 +1,81 @@
+//===- APFloatWrappers.cpp - Software Implementation of FP Arithmetics --- ===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file exposes the APFloat infrastructure to MLIR programs as a runtime
+// library. APFloat is a software implementation of floating point arithmetics.
+//
+// On the MLIR side, floating-point values must be bitcasted to 64-bit integers
+// before calling a runtime function. If a floating-point type has less than
+// 64 bits, it must be zero-extended to 64 bits after bitcasting it to an
+// integer.
+//
+// Runtime functions receive the floating-point operands of the arithmeic
+// operation in the form of 64-bit integers, along with the APFloat semantics
+// in the form of a 32-bit integer, which will be interpreted as an
+// APFloatBase::Semantics enum value.
+//
+#include "llvm/ADT/APFloat.h"
+
+#if (defined(_WIN32) || defined(__CYGWIN__))
+#define MLIR_APFLOAT_WRAPPERS_EXPORTED __declspec(dllexport)
----------------
rengolin wrote:

Don't we have these defined elsewhere? Or is it just to keep self contained in this runtime source?

https://github.com/llvm/llvm-project/pull/166618