[GenABEL-dev] [Genabel-commits] r1600 - pkg/OmicABEL/src

[Yury Aulchenko]

wow! from n^2 to n*p is not a trivial speedup then - assuming n is the sample size :)

Y

On Mar 20, 2014, at 00:02, Diego Fabregat <fabregat at aices.rwth-aachen.de> wrote:

> From the log, it looks like by p I mean the width of XL (i.e., intercept + covariates).
> 
> On 03/19/2014 11:57 PM, Yurii Aulchenko wrote:
>> Deiego, what is the "p" (n^2 -> np) in the Log message?
>> 
>> 
>> On Fri, Feb 7, 2014 at 6:32 PM, <noreply at r-forge.r-project.org> wrote:
>> Author: dfabregat
>> Date: 2014-02-07 18:32:24 +0100 (Fri, 07 Feb 2014)
>> New Revision: 1600
>> 
>> Modified:
>>    pkg/OmicABEL/src/REML.c
>> Log:
>> Performance improvement for REML estimation.
>> Reusing precomputed data to replace an expensive
>> n^2 gemv for a cheaper n*p gemv.
>> 
>> 
>> Modified: pkg/OmicABEL/src/REML.c
>> ===================================================================
>> --- pkg/OmicABEL/src/REML.c     2014-02-06 21:29:08 UTC (rev 1599)
>> +++ pkg/OmicABEL/src/REML.c     2014-02-07 17:32:24 UTC (rev 1600)
>> @@ -194,10 +194,15 @@
>>         // loglik = a + b
>>         //  a -> log(det(M))
>>         //  b -> YmXB' inv(M) YmXB
>> -    dgemv_(TRANS,
>> +    /*dgemv_(TRANS,
>>              &n, &n,
>>              &ONE, Z, &n, YmXB, &iONE,
>> -            &ZERO, ZtY_upd, &iONE);
>> +            &ZERO, ZtY_upd, &iONE);*/
>> +    memcpy( ZtY_upd, ZtY, n * sizeof(double) );
>> +    dgemv_( NO_TRANS,
>> +            &n, &wXL,
>> +            &MINUS_ONE, ZtX, &n, beta, &iONE,
>> +            &ONE, ZtY_upd, &iONE );
>>         // YmXB' * inv( M ) * YmXB
>>         *loglik = 0.0;
>>         for (i = 0; i < n; i++ )
>> 
>> _______________________________________________
>> Genabel-commits mailing list
>> Genabel-commits at lists.r-forge.r-project.org
>> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-commits
>> 
>> 
>> 
>> -- 
>> -----------------------------------------------------
>> Yurii S. Aulchenko
>> 
>> [ LinkedIn ] [ Twitter ] [ Blog ]
>> 
>> 
>> _______________________________________________
>> genabel-devel mailing list
>> genabel-devel at lists.r-forge.r-project.org
>> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-devel
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/genabel-devel/attachments/20140320/33762985/attachment.html>