Wow, thanks guys. I am back to original speed with matrix operations as documented here. As noted previously, this includes the 4 fold basic slowing of matrix operations on a single processor that occurred at the time of the change ~2012(?). So I have approx 80/5 = 16 fold improvement!!! I can do experiments and do a new experiment tomorrow based on the calculation!!!
Tests applied today II_5_2016: *Test without vecLib option > source("http://www.bio.umass.edu/biology/kunkel/pub/R/CuriousResult.R") matrix multiplication 80.516 1.338 81.22 tcrossprod 79.999 1.366 80.734 transposition and reuse 14.303 2.323 16.52 elementwise after reshape 11.088 1.776 12.774 columnwise sapply 22.781 9.631 32.137 for loop over columns 23.976 7.811 31.507 >q() *apply unix adjustment # for vecLib use > source("http://www.bio.umass.edu/biology/kunkel/pub/R/CuriousResult.R") matrix multiplication 19.595 1.188 5.389 tcrossprod 19 1.175 4.702 transposition and reuse 14.113 2.416 16.432 elementwise after reshape 11.102 1.764 12.798 columnwise sapply 21.747 9.125 30.683 for loop over columns 23.243 6.458 29.515 I have renewed faith! Joe > On Feb 5, 2016, at 10:20 AM, Bryan Hanson <han...@depauw.edu> wrote: > > On El Cap you can still pull off the symlink approach, as described here, > without building R fresh: > > http://blog.quadrivio.com/2015/06/improved-r-performance-with-openblas.html > > The next time you launch R you will have the library you specified. > > Bryan > >> On Feb 5, 2016, at 10:03 AM, peter dalgaard <pda...@gmail.com> wrote: >> >> Why me..? >> >> Probably Simon and maybe Brian has the full recollection of the story, but >> as I understood it, once upon a time you could switch out the BLAS on CRAN R >> just by editing a symlink in the R installation. For some reason, that >> cannot work anymore (Apple considered symlinked libraries a security risk?). >> I think the current situation is that you can still do it, but you have to >> physically overwrite the default BLAS(? Simon will know.). >> >> At any rate, you can certainly still do a local compile with the Accelerate >> framework, once you get all the tools in place: >> >>> M <- matrix(rnorm(1e8),10000) >>> system.time(M %*% t(M)) >> user system elapsed >> 219.566 0.806 21.752 >>> M <- crossprod(M) >>> system.time(solve(M)) >> user system elapsed >> 310.874 1.477 29.998 >> >> (To tell the truth, I actually don't have all the tools in place on that >> machine, so this was from a build of 3.2.1 patched) >> >> -pd >> >> On 05 Feb 2016, at 14:52 , Joseph Kunkel <j...@bio.umass.edu> wrote: >> >>> To me there are big gorillas in the room and I need to know why I can not >>> use them all. >>> >>> • Testing for physical and logical cpus on Joe's MacBook Pro. >>> Josephs-MacBook-Pro:~ josephgkunkel$ /usr/sbin/sysctl -n hw.physicalcpu >>> 4 >>> Josephs-MacBook-Pro:~ josephgkunkel$ /usr/sbin/sysctl -n hw.logicalcpu >>> 8 >>> >>> Prior to about 2012 my multicore Macs would use all (physical) cores >>> automagically in R. %*% was multicore automatically. >>> >>> A 24 hour heavy matrix calculation would take a little over 6 hours on a 4 >>> core Mac. >>> >>> Then some problem with the BLAS library changed everything and colleague >>> stats people and I got really mad that we could not do our calculations as >>> fast in R. >>> >>> Many work-around libraries were devised which did not seem to be that >>> useful for freewielding matrix operations. >>> >>> Then Revolution R seemed to solve the problem and patented(?) it. … but >>> not for Macs. >>> >>> Recently they provided a free Mac version but using their R ‘open' version >>> messes up my computer for updating the libraries I am addicted to using. >>> >>> My question after this appologeticlay long narrative is: >>> >>> Why has no satisfactory and transparent method for multicore use been >>> available to CRAN R? >>> >>> Secondarily, how could our R open software system be hijacked by >>> Revolution and now Microsoft? >>> >>> I would be pleased to know that there are colleagues out there who are >>> similarly hoping for an R core solution within CRAN. >>> >>> I can do primitive matrix things faster with Julia, which is encouraging, >>> but the libraries and flexability for me are not there yet. >>> >>> Joe >>> >>> -·. .· ·. .><((((º>·. .· ·. .><((((º>·. .· ·. .><((((º> .··.· >=- >>> =º}}}}}>< >>> Joseph G. Kunkel, Research Professor >>> 112A Marine Science Center >>> University of New England >>> Biddeford ME 04005 >>> http://www.bio.umass.edu/biology/kunkel/ >>> >>> >>> >> >> -- >> Peter Dalgaard, Professor, >> Center for Statistics, Copenhagen Business School >> Solbjerg Plads 3, 2000 Frederiksberg, Denmark >> Phone: (+45)38153501 >> Office: A 4.23 >> Email: pd....@cbs.dk Priv: pda...@gmail.com >> >> _______________________________________________ >> R-SIG-Mac mailing list >> R-SIG-Mac@r-project.org >> https://stat.ethz.ch/mailman/listinfo/r-sig-mac > _______________________________________________ R-SIG-Mac mailing list R-SIG-Mac@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-mac