how to reuse BoxMullerGPU() generate N normal distributed numbers

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi, I have to generate PATH_N normal distributed numbers on GPU.
sdk has a sample for this: MersenneTwister’s RandomGPU (for uniform) followed by BoxMullerGPU (for normal).
the sdk sample Monte Carlo also shows how to use this.
But above 2 functions involve MT_RNG_COUNT, whose meaning hasn’t been given. It’s defined as 4096. My question: Can I also use 4096 for my code?
Thanks!

ps. another parameter is

const int N_PER_RNG = iAlignUp(iDivUp(PATH_N, MT_RNG_COUNT), 2);
and RandomGPU ( )'s comment is
// Write MT_RNG_COUNT vertical lanes of NPerRng random numbers to *d_Random.

ps2. Actually I am to generate an American put option price matrix: Mat[WxH]. As I learn from MonteCarlo sample, I should let PATH_N = WxH, right? Thanks!
2

It is the number of parallel streams generated.
The implementation is following the DCMT approach (described in the Mersenne Twister example).
If you need to change that value, you may need to generate off-line the initial states.

Massimiliano