llvm dev - Feb 2015 - [LLVMdev] AliasInfo for MachineMemOperands

Hi,

when looking at the available Alias Information after lowering,
it seems that the expansion of 'llvm.memcpy.p0i8.p0i8.i32' (and
variants)
onto
load/store machine instructions is resulting in load/stores for which no
AAInfo is available in the MachineMemOperand.

Because of this we need to assume that they may alias with anything... This
is painful if these copies come from copying smaller structs around..

Would the correct solution here be to propagate the tbaa ref on the
intrinsic to the load/stores ? Or are there still more subtle dangers (like
having a Value in the MachineMemOperand that does not relate to anything ?)


Greetings,

Jeroen Dobbelaere
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20150204/000eb344/attachment.html>

----- Original Message -----
> From: "Jeroen Dobbelaere" <jeroen.dobbelaere at gmail.com>
> To: LLVMdev at cs.uiuc.edu
> Sent: Wednesday, February 4, 2015 9:47:31 AM
> Subject: [LLVMdev] AliasInfo for MachineMemOperands
> 
> Hi,
> 
> when looking at the available Alias Information after lowering,
> it seems that the expansion of 'llvm.memcpy.p0i8.p0i8.i32' (and
> variants) onto
> load/store machine instructions is resulting in load/stores for which
> no AAInfo is available in the MachineMemOperand.
> 
> Because of this we need to assume that they may alias with
> anything... This is painful if these copies come from copying
> smaller structs around..
> 
> Would the correct solution here be to propagate the tbaa ref on the
> intrinsic to the load/stores ? Or are there still more subtle
> dangers (like having a Value in the MachineMemOperand that does not
> relate to anything ?)
I don't see any reason why that should be problematic. You can use the base
operand as the Value*, we just need to make sure that we set the size and offset
correctly for each of the instructions in the expansion.

 -Hal
> 
> Greetings,
> 
> 
> 
> Jeroen Dobbelaere
> 
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
> 
-- 
Hal Finkel
Assistant Computational Scientist
Leadership Computing Facility
Argonne National Laboratory