I found this algorithm while I was looking for a high quality PRNG that can also handle zeros in the seed bytes. Initial test were promising. I modified the calculation to save a clc instruction over the original code.
I talked to a cryptographic expert about the algorithm and he stated that given the small resources, it is a well-designed algorithm. It comes with four internal states: X, A, B, C For an 8 bit number, the value of C (which is also returned in Accu) should be used. For 16 bit, the value of B should be used as high byte.
The implementation was optimized using self-modifying code, which yields good speed and low resources, but this implementation won't work in ROM therefore.
In a test, I plotted the output of 51200 random values, which don't seem to reveal any visible patterns:
;; X ABC Algorithm Random Number Generator for 8-Bit Devices ;; ;; Algorithm from EternityForest, slight modification by Wil ;; https://www.electro-tech-online.com/threads/ultra-fast-pseudorandom-number-generator-for-8-bit.124249/ ;; Implementation and test: Wil ;; This version stores the seed as arguments and uses self-modifying code ;; Routine requires 38 cycles (without the rts) / 28 bytes ;; Return values are in A and, if a 16 bit value is needed also in _rand8_highbyte .export _rand8:=rand8, _rand8_highbyte:=b1 rand8: inc x1 clc x1=*+1 lda #$00 ;x1 c1=*+1 eor #$c2 ;c1 a1=*+1 eor #$11 ;a1 sta a1 b1=*+1 adc #$37 ;b1 sta b1 lsr eor a1 adc c1 sta c1 rts