Recently I upgraded the OS of my (geriatric 
) laptop from Windows 7 (running, of necessity, Scilab 6.1.1) to Windows 10, and, naturally I installed Scilab 2026. To ensure things run smooth, I ran a few apps I’d written last year, basically comparisons of solution times of large linear systems of equations using stock GMRES & BiCGStab and my own implementation of the Irons-Aitken accelerator on a vectorized Jacobi iterative procedure. I was in for a big shock! The very same app ran in just 1/10 of the time it took under Scilab 6.1.1. I could hardly believe my eyes and I repeated the test many times. To be sure the comparison was fair, I instalIed Scilab 6.1.1 in Windows 10 and repeated the testing. 6.1.1 was still 10 times slower, so the improvement cannot be attributed to the change of Windows version. I even ran a largish problem (involving extensive use of virtual memory) to measure the timings using a handheld timer, which confirmed the 10:1 speedup of Scilab 2026. Certainly the code of GMRES and BiCGStab has not changed, neither has my own accelerator code, so, logically, the improvement can only be attributed to the inner workings of Scilab 2026. Kudos to the Scilab programmers, you’ve done an amazing job!
1 Like
Hello,
Glad to hear that. Maybe Speedup sparse double product (!718) · Merge requests · scilab / scilab · GitLab helped.
S.
Indeed. Terrific idea! Thank you.