Hello All, Thanks Douglas... I have not idea, really, how it is done internally. I do not really mind this 'bug', but maybe something could be done to mke it better??.
Greg, the sky vector is MF4 (2305x1); the DC matrices are all (2048x2305). What I was doing before was: *Results = DC1*sky + DC2*sky + DC3*sky + DC3*sky* rmtxop DC1 sky_vec > lux1 rmtxop DC2 sky_vec > lux2 rmtxop DC3 sky_vec > lux3 rmtxop DC4 sky_vec > lux4 rmtxop lux1 + lux2 + lux3 + lux4 > total_lux *Results = (DC1+DC2+DC3+DC4)*sky* rmtxop DC1 + DC2 + DC3 + DC4 > TotalDC rmtxop TotalDC sky_vec > total_lux When replacing, on the first case, rmtxop by dctimestep I got slightly faster results. Best, Germán 2017-05-10 13:09 GMT-03:00 Gregory J. Ward <[email protected]>: > Can you reduce this to the actual commands, noting the matrix dimensions > for us? > > Thanks! > -Greg > > *From: *Germán Molina Larrain <[email protected]> > > *Date: *May 10, 2017 5:52:47 AM PDT > > > Hello all, > > I am working on a research project that requires a lot of matrix matrix > operations (i.e. add, multiply, substract). Since I am working with the > ouptuts of RFLUXMTX, I am using RMTXOP. > > Since I have to do this lots of times, I designed my scripts so they are > efficient. Accordingly, avoiding matrix multiplication was a priority... so > I rearranged the equation: > > Results = DC1*sky + DC2*sky + DC3*sky + DC3*sky > > into > > Results = (DC1+DC2+DC3+DC4)*sky > > I thought the latter one would be much faster... but it takes double the > time! Is there any reason for that? I am somehow surprised since, I > understand, matrix multiplication is a O(n^3) algorithm while adding is a > O(n^2) one. > > The DC matrices are the contribution from different windows to the > workplane (using the 3-phase method... VTD) and the sky is a vector, not a > matrix. > > Best, > > Germán > > > _______________________________________________ > Radiance-dev mailing list > [email protected] > https://www.radiance-online.org/mailman/listinfo/radiance-dev > >
_______________________________________________ Radiance-dev mailing list [email protected] https://www.radiance-online.org/mailman/listinfo/radiance-dev
