[clang] [CIR] Add cir.builtin_int_cast operation to replace uses of builtin.unrealized_conversion_cast (PR #201592)

[Erich Keane via cfe-commits]

================
@@ -1430,6 +1430,40 @@ mlir::LogicalResult CIRToLLVMCastOpLowering::matchAndRewrite(
   return mlir::success();
 }
 
+mlir::LogicalResult CIRToLLVMBuiltinIntCastOpLowering::matchAndRewrite(
+    cir::BuiltinIntCastOp op, OpAdaptor adaptor,
+    mlir::ConversionPatternRewriter &rewriter) const {
+  // Both the CIR integer and the builtin integer/index lower to LLVM integer
+  // types, so this cast becomes an integer resize. Signedness is taken from
+  // the CIR integer side (the builtin/index side is treated as signless).
+  bool isUnsigned = true;
+  if (auto cirSrc = mlir::dyn_cast<cir::IntType>(op.getSrc().getType()))
+    isUnsigned = cirSrc.isUnsigned();
+  else if (auto cirDst = mlir::dyn_cast<cir::IntType>(op.getType()))
+    isUnsigned = cirDst.isUnsigned();
+
+  mlir::Value llvmSrc = adaptor.getSrc();
+  mlir::Type llvmDstTy = getTypeConverter()->convertType(op.getType());
+  auto srcIntTy = mlir::cast<mlir::IntegerType>(llvmSrc.getType());
+  auto dstIntTy = mlir::cast<mlir::IntegerType>(llvmDstTy);
+  unsigned srcWidth = srcIntTy.getWidth();
+  unsigned dstWidth = dstIntTy.getWidth();
+
+  // Fixed-width builtin integers must match the CIR integer wdith.
----------------
erichkeane wrote:

```suggestion
  // Fixed-width builtin integers must match the CIR integer width.
```

https://github.com/llvm/llvm-project/pull/201592