[llvm-commits] LLVM patch to support ARM fused multiply add/subtract instructions

[Ana Pazos]

Hi folks!

 

Thanks for the feedback on the fused multiply add/sub patch.

 

I will work with Anton on an updated patch.

 

Anton, 

Reply to your question below: FMA mode.

The reason that the fused operation (FMA) is more accurate than the chained (VMLA) is the chained operation performs a round after the  multiply and before the add. This is only an error of +/-2-24 (for single precision) but it can have a huge impact on the results returned.

 

Ana.

 

-----Original Message-----
From: Anton Korobeynikov [mailto:anton at korobeynikov.info] 
Sent: Monday, January 23, 2012 9:07 PM
To: Ana Pazos
Cc: llvm-commits at cs.uiuc.edu; rajav at codeaurora.org
Subject: Re: [llvm-commits] LLVM patch to support ARM fused multiply add/subtract instructions

 

Hi Ana,

 

> Regarding accuracy, what I know is that Qualcomm provides IEEE-754 2008 specified result. This means the multiply is performed without any loss of accuracy (i.e., no rounding) and then the add/subtract operation happens. The final result is rounded according to the configured rounding mode in the VFP unit.

In which case multiply is performed w/o loss of accuracy? FMA mode? Of

usual VMLA?

 

-- 

With best regards, Anton Korobeynikov

Faculty of Mathematics and Mechanics, Saint Petersburg State University

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20120124/da637e43/attachment.html>