[flang-commits] [flang] [flang][FIR] add a new fir.bitcast operation (PR #187793)

[Razvan Lupusoru via flang-commits]

================
@@ -1481,6 +1481,54 @@ static mlir::ParseResult parseCmpOp(mlir::OpAsmParser &parser,
   return mlir::success();
 }
 
+//===----------------------------------------------------------------------===//
+// BitcastOp
+//===----------------------------------------------------------------------===//
+
+static bool isBitcastCompatibleType(mlir::Type ty) {
+  return mlir::isa<mlir::IntegerType, mlir::FloatType, fir::LogicalType>(ty) ||
+         (mlir::isa<fir::CharacterType>(ty) &&
+          mlir::cast<fir::CharacterType>(ty).getLen() ==
+              fir::CharacterType::singleton());
+}
+
+static std::optional<unsigned> getBitcastBitSize(mlir::Type ty) {
+  if (auto intTy = mlir::dyn_cast<mlir::IntegerType>(ty))
+    return intTy.getWidth();
+  if (auto floatTy = mlir::dyn_cast<mlir::FloatType>(ty))
+    return floatTy.getWidth();
+  // Bit size of fir.logical and fir.char depends on the kind map which is not
----------------
razvanlupusoru wrote:

I think it would be great if we had the size in lieu of the kind directly from lowering - which should know the size representation anyway, right?

https://github.com/llvm/llvm-project/pull/187793