Photon sharing and dual grid give inconsistent output when used with more than one pattern (demo_photon_sharing.m)

[andreafarina]

Dear Qinqian,
I've tried to change the cfg.method from 'elem' to 'grid' to test the dual grid in demo_photon_sharing.m. As expected the flux.data output is a 5D array having the following dimensions: [Num_pattern, Nx, Ny, Nz, NTime]. After summing up the last (time) I tried to plot the fluence in the z dimension for each pattern. I obtain inconsistent results, which does not happen if only one pattern is simulated in the source. Probably there is some bug in reshaping the output data.
Thanks a lot!
Andrea

[fangq]

thanks @andreafarina for reporting this - can you provide a reproducer code to show the problem?

[andreafarina]

For sure! It is a slight variation of the example demo_photon_sharing.m adapted to grid instead of elem.

cfg.nphoton=3e6;
[cfg.node face cfg.elem]=meshabox([0 0 0],[60 60 20],2, 2);
cfg.elemprop=ones(size(cfg.elem,1),1);
cfg.srcpos=[10 10 -2];
cfg.srcdir=[0 0 1];
cfg.srctype='pattern';
cfg.srcparam1=[40.0 0.0 0.0 40];
cfg.srcparam2=[0.0 40.0 0.0 40];
cfg.prop=[0 0 1 1;0.01 10 0.9 1.37];
cfg.tstart=0;
cfg.tend=5e-9;
cfg.tstep=1e-10;
cfg.gpuid=-1;
cfg.method='grid';
cfg.debuglevel='TP';

%% 5 patterns represented by a 3D matrix

pat=zeros(40,40,5);
pat(:,:,1) = ones(40);  % full field illumination pattern
pat(1:20,:,2) = 1;      % pattern with left half bright
pat(:,1:20,3) = 1;      % pattern with top half bright
pat(11:30,11:30,4) = 1; % pattern with bright square in the middle
pat(16:25,:,5) = 1;  
pat(:,16:25,5) = 1;     % pattern with a bright cross

cfg.srcpattern = pat;

%% run the simulation

flux=mmclab(cfg);

data = flux.data;
disp(['size(flux.data) = ',num2str(size(flux.data))]);
% 5D array: [Npatt,Nx,Ny,Nz,Ntime]
cwdata = squeeze(sum(data,5)); %sum over time


%% look z = 3
z = 3;
for i = 1:size(cwdata,1) % changed cw to cwdata to avoid an error
    figure(1),
    subplot(2,3,i),
    imagesc(squeeze(cwdata(i,:,:,z))),
    title(['pattern ',num2str(i)]),
    axis image
end

[fangq]

@andreafarina, I was able to reproduce the issue and fixed it with the following commit

2a7e9dd

the screenshot of the output is attached below.

the photon-sharing feature is now supported with DMMC on the CPU - still need to work on GPU port of photon-sharing (#86).

please test and let me know if you see any additional problem.

[andreafarina]

Thanks a lot!! It works now!!
Best regards!
Andrea

[fangq]

@andreafarina, FYI, the photon-sharing feature is now working on the GPU (for both OpenCL and CUDA).

see #86 and #99 - please note that the srcpattern array now uses the 1st dimension (left-most) as the pattern dimension, instead of the right-most dimension. This way, it is consistent with the photon-sharing in mcx.

[andreafarina]

@fangq thank you very much! It works great! And I can run also using CUDA with the make cudamex mex file generated. There is still the error in matlab Unrecognized function or variable 'stdout'. but it arrives at the end and the patterns are ok.
It's still not clear to me, as described in #98 (comment) how to work with the mmciii.mexa64 trinity binary generated with CMake and why the reply with cuda doesn't work...
Thank you again for the fix!
Best regards

Andrea