[clang] [CIR] Upstream lowering of conditional operators to TernaryOp (PR #138156)

[Andy Kaylor via cfe-commits]

================
@@ -875,6 +877,174 @@ class ScalarExprEmitter : public StmtVisitor<ScalarExprEmitter, mlir::Value> {
     // NOTE: We don't need to EnsureInsertPoint() like LLVM codegen.
     return Visit(e->getRHS());
   }
+
+  mlir::Value VisitBinLAnd(const clang::BinaryOperator *e) {
+    if (e->getType()->isVectorType()) {
+      assert(!cir::MissingFeatures::vectorType());
+      return {};
+    }
+
+    bool instrumentRegions = cgf.cgm.getCodeGenOpts().hasProfileClangInstr();
+    mlir::Type resTy = cgf.convertType(e->getType());
+    mlir::Location loc = cgf.getLoc(e->getExprLoc());
+
+    // If we have 0 && RHS, see if we can elide RHS, if so, just return 0.
+    // If we have 1 && X, just emit X without inserting the control flow.
----------------
andykaylor wrote:

This results in inconsistent behavior. We generate a pair of `cir.ternary` ops for `x && true` but we simplify `true && x`. 

https://godbolt.org/z/1rfYT8E46

It seems to me that it would be better to leave both alone and let the optimizer do the optimization.

https://github.com/llvm/llvm-project/pull/138156