20 Juli 2011 15:33
I've been doing some numerical testing of Sho, and am puzzled by how slow the matrix manipulations are. Matrix multiplication, for example, is
many times slower than Numpy/Matlab that both allegedly use the MKL multi-processor libraries. You can see from the task manager that Sho is not using all available cores.
20 Juli 2011 15:59
As a specific example:
Sho: m=rand(2000,2000), tic(),m*m,toc()
result takes 10.36s on my icore7 dual core
this takes 1.47 s
21 Juli 2011 6:09
OK, I found that upon initialization of Sho, the initial statement indicated that for some reason "fast math" was not enabled. After completely
re-installing Sho, the statement went away, and now the matrix multiply speed for large arrays are comparable (if not slightly better!) than Matlab7.6.
14 Oktober 2011 14:12
Could you compare it to a newer version of MATLAB? How about other functions?