Bruno Haible <[EMAIL PROTECTED]> wrote: > Jim Meyering wrote: >> I've just changed xalloc's x2nrealloc to do n = 3n/2, rather than n *= 2, > > Which means that the time needed for realloc() will grow by a factor of 1.7 > on average. If it matters - haven't measured it -, I would suggest to use > - n = 2*n for n < 1000000, > - n = 3*n/2 for n >= 1000000, > so as to not penalize the frequent small allocations.
If you (or anyone) find a trend in applications where that slight increase in the number of realloc calls makes a significant difference, please let us know. But bear in mind that having to *re*allocate is relatively unusual. Minimizing the cost of the much more frequent range and overflow tests is important too. >> - In the following implementation, nonzero sizes are doubled so that >> - repeated reallocations have O(N log N) overall cost rather than >> - O(N**2) cost, but the specification for this function does not >> - guarantee that sizes are doubled. >> + In the following implementation, nonzero sizes are increased by a >> + factor of approximately 1.5 so that repeated reallocations have >> + O(N log N) overall cost rather than O(N**2) cost, but the >> + specification for this function does not guarantee that rate. > > How do you arrive at O(N log N) cost? I get > O(N) + O(N/c) + O(N/c^2) + O(N/c^3) + ... = O(N) + O(log N) = O(N) > where c is = 2 or = 1.5 respectively. I didn't, but I should have noticed. That comment was there before, and I knew that changing a constant (c) would not affect big-O. Thanks for the correction.
