search

UWB-Biocomputing / BrainGrid

A project to facilitate construction of high-performance neural simulations.
https://uwb-biocomputing.github.io/BrainGrid/
Apache License 2.0
32 stars 17 forks source link

Enable seeding of Mersenne Twister RNG from configuration file #73

Closed stiber closed 10 years ago

stiber commented 10 years ago

InitMTGPU() is seeded with a single number and then goes on to initialize the per-thread RNG states. So, we "just" need to use the a seed variable in simulationInfo and make sure it gets loaded from the configuration file. Actually, that may already happen; we just need to use it in the call to initMTGPU() and verify that the change works.

fumik commented 10 years ago

Do you remember why we are not using Mersenne Twister in CURAND library? If no performance degrading, we should move to the CURAND library. If you agree with this, add this work to this issue.

stiber commented 10 years ago

I remember that it was broken. I forget the details. I think that every time you called the kernel, you got the same set of n random numbers, rather than a continuation of n randoms sequences of random numbers.

Of course, it may have been fixed. I do remember being surprised at how bad it was. Either they've realized that and fixed it, or its state at that point in time was an indication of how low a priority it was.

fumik commented 10 years ago

Now InitMTGPU() uses seed value specified in configuration file.