Re: filling area by color atack safety - worst memory size

On Thu, 11 Apr 2024 22:09:51 -0700
Tim Rentsch <tr.17687@z991.linuxsc.com> wrote:

Michael S <already5chosen@yahoo.com> writes:
 

On Wed, 10 Apr 2024 19:47:11 -0700
Tim Rentsch <tr.17687@z991.linuxsc.com> wrote:
 

Stack-based methods tend to do well on long skinny patterns and
tend to do not as well on fatter patterns such as circles or
squares.  The fractal pattern is ideal for a stack-based method.
Conversely, patterns that are mostly solid shapes don't fare as
well under stack-based methods, at least not the ones that have
been posted in this thread, and also they tend to use more memory
in those cases. 

>
Indeed, with solid shapes it uses more memory.  But at least in my
tests on my hardware with this sort of shapes it is easily faster
than anything else.  The difference vs the best of the rest is
especially big at 4K images on AMD Zen3 based hardware, but even on
Intel Skylake which generally serves as equalizer between different
algorithms, the speed advantage of 2x2 stack is significant. 

 
This comment makes me wonder if I should post my timing results.
Maybe I will (and including an appropriate disclaimer).
 
I do timings over these sizes:
 
      25 x 19
      19 x 25
     200 x 200
    1280 x 760
     760 x 1280
    1920 x 1080
    1080 x 1920
    3840 x 2160
    2160 x 3840
    4096 x 4096
   38400 x 21600
   21600 x 38400
   32767 x 32767
   32768 x 32768
 

I didn't went that far up (ended at 4K) and I only test landscape sizes.
May be, I'd add portrait option to see anisotropic behaviors.
For bigger sizes, correctness is interesting, speed - not so much, since
they are unlikely to be edited in interactive manner.

with these patterns:
 
   fractal
   slalom
   rotated slalom
   horizontal snake and vertical snake
   cross in cross
   donut aka ellipse with hole
   entire field starting from center
 
If you have other patterns to suggest that would be great,
I can try to incorporate them (especially if there is
code to generate the pattern).
 
Patterns are allowed to include a nominal start point,
otherwise I can make an arbitrary choice and assign one.

My suit is about the same with following exceptions:
1. I didn't add donut yet
2. + 3 greeds with cell size 2, 3 and 4
3. + fractal tree
4. + entire field starting from corner
It seems, neither of us tests the cases in which linear dimensions of
the shape are much smaller than those of the field.

static void make_grid(
  unsigned char *image,
  int width, int height,
  unsigned char background_c,
  unsigned char pen_c, int cell_sz)
{
  for (int y = 0; y < height; ++y) {
    unsigned char* p = &image[y*width];
    if (y % cell_sz == 0) {
      memset(p, pen_c, width);
    } else {
      for (int x = 0; x < width; ++x)
        p[x] = x % cell_sz ? background_c : pen_c;
    }
  }
}