================
@@ -92,6 +159,16 @@ mlir::LogicalResult
CIRToLLVMGlobalOpLowering::matchAndRewrite(
mlir::dyn_cast<cir::IntAttr>(init.value())) {
// Initializer is a constant array: convert it to a compatible llvm init.
init = rewriter.getIntegerAttr(llvmType, intAttr.getValue());
+ } else if (isa<cir::ConstPtrAttr>(init.value())) {
+ // TODO(cir): once LLVM's dialect has proper equivalent attributes this
+ // should be updated. For now, we use a custom op to initialize globals
+ // to the appropriate value.
+ setupRegionInitializedLLVMGlobalOp();
+ CIRAttrToValue attrVisitor;
+ mlir::Value value = attrVisitor.lowerCirAttrAsValue(
+ op, init.value(), rewriter, typeConverter, dataLayout);
+ rewriter.create<mlir::LLVM::ReturnOp>(loc, value);
+ return mlir::success();
----------------
erichkeane wrote:
Hmm... this already changing to make these not 'equal' in some way is causing
me consternation. I realize this is the reason for an inability to skip the
visitor, which I see the reasoning for here, but this function is going to
QUICKLY get unreadable without some sort of better organization.
We should noodle on a way to better organize this into something that isn't
going to get unwieldy. And by 'we' I believe that is a
group-homework-assignment :)
https://github.com/llvm/llvm-project/pull/125619
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
