[clang] [CIR] Upstream missing support for floating point unary operator (PR #193215)

[Erich Keane via cfe-commits]

================
@@ -825,14 +818,23 @@ class ScalarExprEmitter : public StmtVisitor<ScalarExprEmitter, mlir::Value> {
     return builder.createOrFold<cir::MinusOp>(loc, operand, nsw);
   }
 
-  mlir::Value emitIncOrDec(const UnaryOperator *e, mlir::Value input,
-                           bool nsw = false) {
+  mlir::Value emitIntIncOrDec(const UnaryOperator *e, mlir::Value input,
+                              bool nsw = false) {
     mlir::Location loc = cgf.getLoc(e->getSourceRange().getBegin());
     return e->isIncrementOp()
                ? builder.createOrFold<cir::IncOp>(loc, input, nsw)
                : builder.createOrFold<cir::DecOp>(loc, input, nsw);
   }
 
+  mlir::Value emitFloatIncOrDec(const UnaryOperator *e, mlir::Value input) {
+    mlir::Location loc = cgf.getLoc(e->getSourceRange().getBegin());
+    auto fpType = mlir::cast<cir::FPTypeInterface>(input.getType());
+    mlir::Value amount = builder.getConstFP(
+        loc, input.getType(), llvm::APFloat(fpType.getFloatSemantics(), 1));
+    return e->isIncrementOp() ? builder.createFAdd(loc, input, amount)
----------------
erichkeane wrote:

Huh, we do!  I think my mind is changed, inc/dec add little/no value (And perhaps are harmful!).  I DO think we probably could/should/will have a pattern to recognize inc/dec for the above purposes, but that is irrelevant here/etc.  

So I'll resolve this one.

https://github.com/llvm/llvm-project/pull/193215