1 Reply Latest reply on Jul 4, 2014 9:11 AM by lejeczek

    make R to use ACML for blas and lapack, compilation + runtime(broken?)


      I really believe I had it, it all compiled and this benchmark - http://r.research.att.com/benchmarks/ - gave very very positive results, but now something has changed either in my compilation or runtime or both environments, although I believe have checked everything thoroughly it renders R binary that is somewhat greatly impaired FFT over 2,400,000 random values____________________ (sec):  0.438333333333333 Eigenvalues of a 640x640 random matrix______________ (sec):  134.277333333333  <--- here Determinant of a 2500x2500 random matrix____________ (sec):  2.67433333333332 and slower than "regular" R, by a lot!! before I try to debug/trace it, and I am not! a developer/programmer thus it's going to be a mission, maybe an expert could suggest where to look and what to double check, I'd be grateful. I'm trying Scientific Linux'es 3.0.2 and 3.1.0 Rs and they compile ok, I see all: ./lib64/R/lib/libR.so ./lib64/R/lib/libRblas.so ./lib64/R/lib/libRlapack.so depend on libacml_mp.so but the total time when I had it OK was under 10sec now is: Total time for all 15 tests_________________________ (sec):  152.896 I'd be be happy to include all the logs and output, simply ask for what you might need regards