Displaying 8 results from an estimated 8 matches for "connoisseurs".
Did you mean:
connoisseur
2013 Feb 11
2
[LLVMdev] Is there any llvm neon intrinsic that maps to vmla.f32 instruction ?
In theory, the backend should choose the best instructions for the selected target processor. VMLA is not always the best choice. Lang Hames did some measurements a while back to come up with the current behavior, but I don't remember exactly what he found. CC'ing Lang.
On Feb 11, 2013, at 8:12 AM, Renato Golin <renato.golin at linaro.org> wrote:
> On 11 February 2013 15:51,
2013 Feb 11
0
[LLVMdev] Is there any llvm neon intrinsic that maps to vmla.f32 instruction ?
On 11 February 2013 15:51, Sebastien DELDON-GNB <sebastien.deldon at st.com>wrote:
> Indeed problem is with generation of vmla.f64. Affected benchmark is MILC
> from SPEC 2006 suite and disabling vmlx forwarding gives a 10% speed-up on
> complete benchmark execution ! So it is worth a try.
>
Hi Sebastien,
Ineed, worth having a look. Including Bob Wilson (who introduced the
2013 Feb 11
3
[LLVMdev] Is there any llvm neon intrinsic that maps to vmla.f32 instruction ?
Hi Renato,
Indeed problem is with generation of vmla.f64. Affected benchmark is MILC from SPEC 2006 suite and disabling vmlx forwarding gives a 10% speed-up on complete benchmark execution ! So it is worth a try. Now going back to vmla generation through LLMV intrinsic usage. I've looked at .td file and it seems to me that when there is a "pattern" to generate instruction, no
2013 Feb 11
0
[LLVMdev] Is there any llvm neon intrinsic that maps to vmla.f32 instruction ?
Hi Bob, Seb, Renalto,
My VMLA performance work was on Swift, rather than Cortex-A9.
Sebastian - is vmlx-forwarding really the only variable you changed between
your tests?
As far as I can see the VMLx forwarding attribute only exists to restrict
the application of one DAG combine optimization: PerformVMULCombine in
ARMISelLowering.cpp, which turns (A + B) * C into (A * C) + (B * C). This
2011 Jul 24
0
repeated execution of svm(e1071) gives different results, if probability = TRUE is set
Hello, Connoisseurs!
Please explain to novices, why svm model gives different results in the
loop with the same data? As a result, I can not find the best gamma and
cost parameters. Also tune.svm yields results that can not be repeated.
How can I avoid this?
My sessionInfo:
R version 2.11.1 (2010-05-31)
x86_64-pc-l...
2005 Mar 29
0
setAs between a new S4 class and "POSIXlt"
Dear R core team
Please apologize for posting the same question twice on R-help and
R-devel. Since I was not sure which list is appropriate I tried R-help (Tue
Mar 22), but got no answer.
Now I do not know if the formulation of my question was unclear or the
question is not so easy to answer or to easy (what I do not hope).
My problem: I create a new S4 class, containing one slot, data (of
2010 Dec 17
3
Moving from Fedora -- Advice??
I'm running Fedora14 on all machines, including my wife's -- and
I'm the nearest (distant) thing there is to tech support.
She's far more likely to outlive me than I her; so I want to
install something requiring a lot less maintenance on her machine, so
that she'll have it and be used to it, years ahead of need.
I'm thinking CentOS 6, whenever it's ready, is
2013 Feb 12
2
[LLVMdev] Is there any llvm neon intrinsic that maps to vmla.f32 instruction ?
Hi all,
Sorry for my naïve question but what is Swift ?
Yes vmlx-forwarding is the only variable I changed in my tests.
I did the experiment on another popular FP benchmark and observe a 14% speed-up only by disabling vmlx-forwarding.
Best Regards
Seb
My VMLA performance work was on Swift, rather than Cortex-A9.
Sebastian - is vmlx-forwarding really the only variable you changed between your