PCG generators are easily “crackable”
I have already previously published a C program that discovers the internal state of a PCG generator with 64 bits of state and 32 bits of output in a few seconds (exhaustive search would require centuries).
I am now providing a C++ program that will discover in a few seconds the internal state of a PCG generator with 128 bits of state and 64 bits of output, even faster, looking at a few outputs of the generator. You will need Victor Shoup's amazing NTL library to compile it.
The programs can be found here: http://pcg.di.unimi.it/pcg.php#claims
Both programs use the same logic: guessing exhaustively a few bits, deriving a lot of other bits, and solving a simple modular equation. However, in the 64-bit case the equation can be solved by trying all possible solutions, whereas in the 128-bit case I use a standard technique based on lattice reduction: as a result, discovering initial state takes usually _less_ time than in the 64-bit case (in fact, the computation time can be brought down to well below a second if you are willing to examine more outputs).
This settles once and for all, negatively, the question "have PCG generators any cryptographic strength?".
PS: I'm purposely putting quotes around "crackable" because you cannot "crack" a non-cryptographic generator in the same way cannot crack a fig. But in view of the claims of the author of PCG generators, it seems appropriate.
His own PRNG is of course much weaker, we don't need any state to predict the next number.