erichkeane wrote:
> > Shouldn't there be a pointer-vs-intty difference in the LLVM-IR here?
>
> No, because in this case tryEmitAbstract() is only using the size of the type
> that's passed in. After a few visits, it eventually calls this function:
>
> ```
> llvm::Constant *ProduceIntToIntCast(const Expr *E, QualType DestType) {
> QualType FromType = E->getType();
> // See also HandleIntToIntCast in ExprConstant.cpp
> if (FromType->isIntegerType())
> if (llvm::Constant *C = Visit(E, FromType))
> if (auto *CI = dyn_cast<llvm::ConstantInt>(C)) {
> unsigned SrcWidth = CGM.getContext().getIntWidth(FromType);
> unsigned DstWidth = CGM.getContext().getIntWidth(DestType);
> if (DstWidth == SrcWidth)
> return CI;
> llvm::APInt A = FromType->isSignedIntegerType()
> ? CI->getValue().sextOrTrunc(DstWidth)
> : CI->getValue().zextOrTrunc(DstWidth);
> return llvm::ConstantInt::get(CGM.getLLVMContext(), A);
> }
> return nullptr;
> }
> ```
>
> The DestType parameter is only used to resize the constant if necessary
> (which in this case it won't be). The equivalent function in CIR was
> asserting that DestType was an integer type, which is what led me to this
> change.
Oooh, hrmph. I find myself thinking that `ProduceIntToIntCast` shouldn't be
necessary in most cases, since the `size_t is pointer type` is pretty uniform.
Though I think I saw above you found some platforms we support doing this, so
an assert wouldn't be a good replacement.
I can see now why this isn't really testable, this is really quite unfortunate
here.
https://github.com/llvm/llvm-project/pull/123433
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
