Direct Access to PRNG_STATE

[MathMan]

Hi all,

I would like to know if there is a way to directly access the prng-state of the mersenne-twister (current PRNG used in FPC iiuc)?

Background:

1 - I would like to retrieve the lowest 128-256 bit, after a call to Random(), consider them as a fractional and then do an upward multiplication with a 128 bit integer to generate a range restricted 128 bit integer random.

2 - currently I have to do this by first generating a 128 bit random integer (via 4 individual calls to Random( $100000000 )) and then do a downward modulo reduction to the required range.

Solution 2 has some disadvantages, as it requires substantially more time and, via the downward modulo, I am loosing the uniform distribution qualities of the Mersenne Twister (if my math serves me right).

Regards,
MathMan

[PascalDragon]

I would like to know if there is a way to directly access the prng-state of the mersenne-twister (current PRNG used in FPC iiuc)?

No, there is not. In addition to that the PRNG used is an implementation detail, so this may change depending on the version or platform.

If you need specific properties, then use your own PRNG.

[440bx]

@MathMan,

If you need specific properties, then use your own PRNG.

If you decide to follow that advice, check out ACORN if you haven't already.

HTH.

[MathMan]

I would like to know if there is a way to directly access the prng-state of the mersenne-twister (current PRNG used in FPC iiuc)?

No, there is not. In addition to that the PRNG used is an implementation detail, so this may change depending on the version or platform.

If you need specific properties, then use your own PRNG.

Understood, will use my own PRNG then.

[MathMan]

@MathMan,

If you need specific properties, then use your own PRNG.

If you decide to follow that advice, check out ACORN if you haven't already.

HTH.

Thanks for the hint - I'll take a look at it.

[MathMan]

@MathMan,

If you need specific properties, then use your own PRNG.

If you decide to follow that advice, check out ACORN if you haven't already.

HTH.

I looked at it, but it does provide too little state for my requirements. I finally opted for an xoshiro-variant, which is very fast and passes e.g. TestU01.

Kind regards,
MathMan

[440bx]

I looked at it, but it does provide too little state for my requirements. I finally opted for an xoshiro-variant, which is very fast and passes e.g. TestU01.

Kind regards,
MathMan

I'm pleased you found an algorithm that meets your needs. 

[valdir.marcos]

@MathMan,

If you need specific properties, then use your own PRNG.

If you decide to follow that advice, check out ACORN if you haven't already.
HTH.

I looked at it, but it does provide too little state for my requirements. I finally opted for an xoshiro-variant, which is very fast and passes e.g. TestU01.

Kind regards,
MathMan

Is this xoshiro-variant solution you used open source?

[MathMan]

Is this xoshiro-variant solution you used open source?

Looking at http://vigna.di.unimi.it/xorshift/ I got the impression, that the algorithm and the C-implementation are FOSS. I'll have to translate to FPC of course, but this is very little effort it seems.