[PATCH][ARM] Label offset crashes llvm integrated assembler

[Jiangning Liu]

A question, can you explain why when SOExpr->EvaluateAsAbsolute(Value)
returns false, it's safe to return Match_Success?

How do we guarantee EvaluateAsAbsolute can always solve the expression and
return an immediate value in encoding stage via MCStreamer? Looking at your
patch, it seems there might be some assumptions here, or you want to simply
use "assertion" in encoding stage to detect any invalid expression in
future? If it is yes, it doesn't sound a good solution to me from design
level of point.

Thanks,
-Jiangning



2014-03-13 14:23 GMT+08:00 Saleem Abdulrasool <compnerd at compnerd.org>:

> On Wed, Mar 12, 2014 at 11:12 PM, Jiangning Liu <liujiangning1 at gmail.com>wrote:
>
>> Hi,
>>
>> Attached patch is to raise an error message when finding an expression at
>> compile time can't really be solved as a constant for immediate operand.
>> Without this patch, llvm compiler would crash.
>>
>
> The attached patch seems wrong.  With the attached
> change, validateTargetOperandClass always returns Match_InvalidOperand for
> any immediate in this addressing mode.  If the value is a constant value,
> then it should return Match_Success.
>
>
>> --
>> Thanks,
>> -Jiangning
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>
>>
>
>
> --
> Saleem Abdulrasool
> compnerd (at) compnerd (dot) org
>



-- 
Thanks,
-Jiangning
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140313/3241d55b/attachment.html>