Hi Jon, > I just noticed a surprisingly big difference in execution speed > between 32- and 64-bit PicoLisp when doing divisions.
In general, arithmetics is much faster in pil64 than in pil32. One reason is the more efficient implementation in assembly, where you have direct access to the CPU flags like "sign" and "carry". But more important is that pil32 must represent all numbers internally as bignums (i.e. in cell structures), due to the restricted word length and tag bits. Pil64, on the other hand, represents short numbers (up to 60 bits plus sign) directly in the tagged pointer. This has the advantage that results of arithmetic operations won't cause cell allocation in the heap, avoiding increased garbage collections. > The first line sets P to (6402373705728000 18 17 16 15 14 13 12 11 10 9 8 7 6 > 5 4 3 2). The first number is still in the range of a short num. So the whole computation takes place as short numbers in pil64. ♪♫ Alex -- UNSUBSCRIBE: mailto:picolisp@software-lab.de?subject=Unsubscribe
