[llvm-commits] (no subject)

[Ana Pazos]

Hi Anton,

 

I synced to the latest LLVM code and noted that the revised patch I provided
to generate fused instructions conditioned by FPContractions flag has not
been merged yet.

 

So there are two patches to be merged:  fusedMACpatch.diff (code change),
fusedMACtestpatch.diff (test cases).  See emails attached with the patches.

 

I think I do not have permission to commit changes.  Can you do it? Or we
can ask my colleague Sebastian to do it. Are you ok with it?

 

>Ok, but this is user-level interface. Are you interested in adding

>clang support for vfpv4 ?

Not at the moment. Note that if a user invokes clang setting -mcpu flag to a
CPU that includes FeatureNEONVFP4 and has the target option
NoExcessFPPrecision  turned off, the fused instructions will be generated.
No additional change in clang is needed.

The idea of supporting the neon-vfpv4 flag in clang is to be able to turn
neon-vfpv4 on/off. But that does not work well when you also set -mcpu  flag
(it seems the flags the cpu implies prevail over fpu flags). So I think
there are more changes required in clang to support -mfpu and -mcpu flags
correctly for the new the neon-vfpv4 flag to be useful.

 

Thank you,

Ana.

 

 

-----Original Message-----
From: anton at korobeynikov.info [mailto:anton at korobeynikov.info] On Behalf Of
Anton Korobeynikov
Sent: Monday, February 20, 2012 11:21 PM
To: Ana Pazos
Cc: llvm-commits at cs.uiuc.edu
Subject: Re: [llvm-commits] LLVM patch to support ARM fused multiply
add/subtract instructions

 

Hi Ana,

 

Sorry, I somehow missed this e-mail earlier.

 

> - I have updated fusedMAC.ll to include Neon fused multiply add/sub test
cases. The patch is attached.

Ok. Please commit.

 

> So I think it just relies on -mfpu=neon-vfpv4 to imply Neon2.

Ok, but this is user-level interface. Are you interested in adding

clang support for vfpv4 ?

 

-- 

With best regards, Anton Korobeynikov

Faculty of Mathematics and Mechanics, Saint Petersburg State University

 

 

 

 

--

Qualcomm Innovation Center, Inc is a member of Code Aurora Forum

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20120227/4e17da3b/attachment.html>
-------------- next part --------------
An embedded message was scrubbed...
From: "Ana Pazos" <apazos at codeaurora.org>
Subject: Re: [llvm-commits] LLVM patch to support ARM fused	multiply	add/subtract instructions
Date: Wed, 25 Jan 2012 15:55:32 -0800
Size: 9414
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20120227/4e17da3b/attachment.mht>
-------------- next part --------------
An embedded message was scrubbed...
From: "Ana Pazos" <apazos at codeaurora.org>
Subject: Re: [llvm-commits] LLVM patch to support ARM fused	multiply	add/subtract instructions
Date: Tue, 24 Jan 2012 16:18:45 -0800
Size: 22351
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20120227/4e17da3b/attachment-0001.mht>