[clang] [CIR] Add support for __builtin_alloca (PR #157116)

[Morris Hafner via cfe-commits]

================
@@ -149,6 +149,57 @@ RValue CIRGenFunction::emitBuiltinExpr(const GlobalDecl &gd, unsigned builtinID,
     emitVAEnd(emitVAListRef(e->getArg(0)).getPointer());
     return {};
 
+  case Builtin::BIalloca:
+  case Builtin::BI_alloca:
+  case Builtin::BI__builtin_alloca_uninitialized:
+  case Builtin::BI__builtin_alloca: {
+    // Get alloca size input
+    mlir::Value size = emitScalarExpr(e->getArg(0));
+
+    // The alignment of the alloca should correspond to __BIGGEST_ALIGNMENT__.
+    const TargetInfo &TI = getContext().getTargetInfo();
+    const CharUnits SuitableAlignmentInBytes =
+        getContext().toCharUnitsFromBits(TI.getSuitableAlign());
+
+    // Emit the alloca op with type `u8 *` to match the semantics of
+    // `llvm.alloca`. We later bitcast the type to `void *` to match the
+    // semantics of C/C++
+    // FIXME(cir): It may make sense to allow AllocaOp of type `u8` to return a
+    // pointer of type `void *`. This will require a change to the allocaOp
+    // verifier.
+    auto allocaAddr = builder.createAlloca(
+        getLoc(e->getSourceRange()), builder.getUInt8PtrTy(),
+        builder.getUInt8Ty(), "bi_alloca", SuitableAlignmentInBytes, size);
----------------
mmha wrote:

The name string attribute isn't used for anything (or at least I couldn't find a use) so this shouldn't be an issue.

Also note that we have some precedence with array ctors and `__array_idx`: https://godbolt.org/z/dT3G4neYT

https://github.com/llvm/llvm-project/pull/157116