[llvm-commits] [patch][pr11029] fix for internal crash due to ExpandUnalignedLoad/Store not handling indexed loads correctly

[James Molloy]

Hi,

Could someone please code review this patch? It fixes PR11029, an internal
fault.

Once reviewed I can commit on Kristof's behalf, but I'm not experienced
enough with SelectionDAG to make a good review.

Cheers,

James

-----Original Message-----
From: llvm-commits-bounces at cs.uiuc.edu
[mailto:llvm-commits-bounces at cs.uiuc.edu] On Behalf Of Kristof Beyls
Sent: 02 November 2011 11:30
To: llvm-commits at cs.uiuc.edu
Subject: Re: [llvm-commits] [patch][pr11029] fix for internal crash due to
ExpandUnalignedLoad/Store not handling indexed loads correctly

ping

> -----Original Message-----
> From: llvm-commits-bounces at cs.uiuc.edu [mailto:llvm-commits-
> bounces at cs.uiuc.edu] On Behalf Of Kristof Beyls
> Sent: 26 October 2011 16:40
> To: llvm-commits at cs.uiuc.edu
> Subject: [llvm-commits] [patch][pr11029] fix for internal crash due to
> ExpandUnalignedLoad/Store not handling indexed loads correctly
> 
> The attached patch fixes PR11029 (
> http://llvm.org/bugs/show_bug.cgi?id=11029 ).
> 
> The root cause of the problem seems to be that ExpandUnalignedLoad and
> ExpandUnalignedStore doesn't handle indexed loads or stores correctly.
> 
> There seem to be 2 possible ways to fix this:
> 1. Implement support for indexed loads/stores in the above functions.
> 2. Don't generate indexed load/stores in cases where these will need to
> be
> expanded.
> 
> The attached patch implements the second approach. The reasons for
> going
> with the second approach are:
> * The generation of indexed loads/stores seems to be an optimization.
> The
> second approach chooses not to do the optimization when it's not
> supported.
> * There will not be a regression in code generation, since in the cases
> that
> the patch prohibits optimization, the compiler would crash or generate
> incorrect code with a high probability.
> * It's unclear whether generating indexed loads/stores for unaligned
> loads/stores that need expanding would actually result in better
> quality of
> the generated code.
> 
> Could someone review the patch and commit it if fine?
> 
> Thanks,
> 
> Kristof




_______________________________________________
llvm-commits mailing list
llvm-commits at cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits