Replay failed

71 views
Skip to first unread message

IRENE

unread,
Oct 31, 2024, 10:02:55 PM10/31/24
to mcx-users
Dear Dr.Fang,

I tried to run the demo_example_replay.m in mmclab, but the replay is always failed. I did see the solution in 2021 (https://groups.google.com/g/mcx-users/c/90T-E6kfe7g/m/zB59eyQkBgAJ) . I tried to use cfg.gpuid =-1 and it is successful, but the simulation time is very long for another simulation model I created by myself. The MMC I downloaded should be the latest one, then I tried to reduce the cfg.nphotons from 1e6 to 1e5, the replay is successful. I am not sure what's the reason for that relay is failed when the cfg.nphotons is higher. Can you help me solve this problem? Thank you.

Best wishes,
Xingmin.

Fang, Qianqian

unread,
Nov 3, 2024, 12:23:26 PM11/3/24
to mcx-users
Hi Xingmin,

Please do not use the nightly build version of mmc, there is a know bug related to this.

The replay demo should work for mmc v2024.2 or 2023 releases.

However, this feature was recently broken after I implemented GPU-based photon-sharing feature about 1 months ago in 
the photon-sharing feature was only implemented in the CPU. need to port the code to the OpenCL/CUDA kernel in order to use this feature on a GPU.

This is a know bug reported by Andrea Farina in this ticket
The example mclab/example/demo_example_replay.m fails in testing the replay data when run on GPU. It works under CPU. Apparently the replied jacobian looks good but I would just notify you about th...

The partial path output from mmc's nightly build is also broken, as reported in 


I believe both of these bugs are regressions from my new addition. I am still working on a fix. 

Qianqian


From: mcx-...@googlegroups.com <mcx-...@googlegroups.com> on behalf of IRENE <IRENE...@OUTLOOK.COM>
Sent: Thursday, October 31, 2024 10:35 AM
To: mcx-users <mcx-...@googlegroups.com>
Subject: [mcx-users] Replay failed
 
--
You received this message because you are subscribed to the Google Groups "mcx-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mcx-users+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/mcx-users/17104637-88f2-4fd7-8706-a5aa62c94504n%40googlegroups.com.

Fang, Qianqian

unread,
Nov 4, 2024, 5:23:00 PM11/4/24
to mcx-users, Andrea Farina
Hi Xingmin and Andrea

The replay benchmark in mmc has been fixed, it is working both under OpenCL and CUDA (mmc-trinity) - tested with both mmc binary and mex file.

These fixed issues include


Please redownload the latest github CI builds from


Qianqian


From: 'Fang, Qianqian' via mcx-users <mcx-...@googlegroups.com>
Sent: Sunday, November 3, 2024 12:23 PM
To: mcx-users <mcx-...@googlegroups.com>
Subject: Re: [mcx-users] Replay failed
 

Andrea Farina

unread,
Nov 5, 2024, 2:47:59 PM11/5/24
to Fang, Qianqian, mcx-users
Thanks a lot Qianqian, this is very helpful because it allowed me to work with native cuda and to avoid the memory saturation.
Just to understand: when you compile with cmake .. -DBUILD_CUDA = on there are 2 more binaries compiled: /bin/mmciii and /mmclab/mmciii.mexa64.
Under matlab mmciii(‘gpuinfo’) gives the following error:

Invalid MEX-file '/home/andreafarina/Documents/MCXStudio/MATLAB/club/mmciii.mexa64':
/home/andreafarina/Documents/MCXStudio/MATLAB/mmclab/mmciii.mexa64: undefined symbol:
mesh_saveweight

If I compile with make cuda and make cudamex all works: a single mex file is generated, and I can switch computation with cfg.compute = ‘cuda’ or ‘opecl’.

Is there any difference between the two compiled versions?

Thanks a lot again!

Best regards

Andrea

Fang, Qianqian

unread,
Nov 5, 2024, 2:50:16 PM11/5/24
to Andrea Farina, mcx-users
Hi Andrea, cmake for mmc-trinity binaries are now fixed, see


From: Andrea Farina <andreaf...@gmail.com>
Sent: Tuesday, November 5, 2024 10:49 AM
To: Fang, Qianqian <q.f...@northeastern.edu>
Cc: mcx-users <mcx-...@googlegroups.com>
Reply all
Reply to author
Forward
0 new messages