llvm dev - Nov 2015 - One element vectors versus scalars

I am curious as to why LLVM has explicit support for 1 element vectors, for
example 'v1i32'.  Are there some contexts where handling a value as a 1
element vector can yield better code quality than the corresponding scalar
'i32'?

 

Thanks,

 

            MartinO

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20151110/8ede523c/attachment.html>

----- Original Message -----
> From: "Martin J. O'Riordan via llvm-dev" <llvm-dev at
lists.llvm.org>
> To: "LLVM Developers" <llvm-dev at lists.llvm.org>
> Sent: Tuesday, November 10, 2015 6:00:27 AM
> Subject: [llvm-dev] One element vectors versus scalars
> 
> I am curious as to why LLVM has explicit support for 1 element
> vectors, for example ‘ v1i32 ’. Are there some contexts where
> handling a value as a 1 element vector can yield better code quality
> than the corresponding scalar ‘ i32 ’?
> 
Hi Martin,

No, but I see two reasons we support 1-element vectors:

 1. Having 1-element vectors makes a lot of the SDAG legalization code simpler,
allowing us to separate the code that splits vector types from the code that
transforms 1-element vectors into scalars.

 2. On some platforms, there are ABI differences between the two types. (e.g.
i128 and <1 x i128> are passed in different registers across function-call
boundaries).

 -Hal
> 
> 
> Thanks,
> 
> 
> 
> MartinO
> 
> 
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
> 
-- 
Hal Finkel
Assistant Computational Scientist
Leadership Computing Facility
Argonne National Laboratory

Thanks Hal.  In our case the ABI issue doesn't arise, but I'll have to
think about the legalization to see if maybe we could write our type definitions
better.

	MartinO

-----Original Message-----
From: Hal Finkel [mailto:hfinkel at anl.gov] 
Sent: 10 November 2015 12:34

----- Original Message -----
> From: "Martin J. O'Riordan via llvm-dev" <llvm-dev at
lists.llvm.org>
> 
> I am curious as to why LLVM has explicit support for 1 element 
> vectors, for example ‘ v1i32 ’. Are there some contexts where handling 
> a value as a 1 element vector can yield better code quality than the 
> corresponding scalar ‘ i32 ’?
Hi Martin,

No, but I see two reasons we support 1-element vectors:

 1. Having 1-element vectors makes a lot of the SDAG legalization code simpler,
allowing us to separate the code that splits vector types from the code that
transforms 1-element vectors into scalars.

 2. On some platforms, there are ABI differences between the two types. (e.g.
i128 and <1 x i128> are passed in different registers across function-call
boundaries).

 -Hal