search

ComputationalRadiationPhysics / haseongpu

HASEonGPU: High performance Amplified Spontaneous Emission on GPU
http://www.hzdr.de/crp
Other
7 stars 6 forks source link

Number of rays is limited to GPU memory #2

Open slizzered opened 10 years ago

slizzered commented 10 years ago

Most basic, yet powerful solution: use a multiplicator for the rays (for each ray in the schedule, start N rays from that prism)

Will enhance the --rays= switch, since it is no longer limited to a magic constant

slizzered commented 9 years ago

Other solution (thanks to Malte): Split the prism array, then compute the prefix sum for each sub-array, correct the indices, and have several ray-schedules. Each of these can be started in its own kernel. Will need more kernels, but can keep the kernel code itself unchanged

erikzenker commented 9 years ago

We really have to talk about this solution personally ^^ ...I really do not understand it :santa:

bussmann commented 9 years ago

Why is there a Santa? I really don't understand it. Personal chat about Santa anyone?

erikzenker commented 9 years ago

This is the hint to Santa, that he shall give me a bigger graphics card with more memory than I can think of. So, there is no need to split arrays ...Santa do you here me ???

slizzered commented 9 years ago

I probably explained it badly ^^. We can talk in person and I do some nice drawings and all will become clear :icecream: