The following is from a Vista machine:

Your benchmark is slightly inaccurate.

$ticks does not yield constant results; so your benchmark runs for a variable number of times-- meaning you very well could be measuring 6.3 doing less or more actual work than 6.31.

I modified the benchmark to render a *constant* number of frames but calculate the time variably, dividing the constant number of frames into time to get the fps:

Code:
testvel2 {
  var %t = $ticks , %f = 0 , %n = 0 , %x = 0
  window -kpBwz +d @testvel 100 100 400 300 
  drawpic -c
  while (%f < 10000) {
    drawpic -cn @testvel 0 0 sprite.bmp
    drawpic -cnt @testvel 65280 $int(%x) 1 0 0 160 16 sprite.bmp
    drawline @testvel
    inc %x 0.1 | if (%x > 400) { %x = 0 }
    inc %f 
  }
  echo -s Drawpic FPS: $calc(10000 / (($ticks - %t) / 1000))
  window -c @testvel
}


I ran the test in three scenarios, to verify if the PNG lib might be responsible. One with (sprite.bmp,background.png), another with (sprite.bmp,sprite.bmp) and the last with (background.png,background.png), referring to the 2 images used as the sprite and the background image respectively. Here are the results:

sprite.bmp, background.png
6.31: Drawpic FPS: 1178.272652
6.3 : Drawpic FPS: 1540.832049
Diff: 23.5%

sprite.bmp, sprite.bmp
6.31: Drawpic FPS: 1366.680333
6.3 : Drawpic FPS: 1760.873393
Diff: 22.38%

background.png, background.png
6.31: Drawpic FPS: 1116.819299
6.3 : Drawpic FPS: 1541.069502
Diff: 27.5%

Conclusion:

I should first note here that the sprite.bmp and background.png files I used were different from yours. Granted, that should not affect the test in the global scheme of things but it does have effects on my specific resultset that change the way they should be intepreted.

My background.png was considerably larger than sprite.bmp. Therefore, the BMP only test was *expected* to be faster, and the PNG only test was *expected* to be slower. Instead, the PNG only test ran faster than the BMP/PNG test, meaning the PNG library actually improves performance. The BMP test is consistent with the speed improvement expected. While 6.31 is slightly slower, it's barely a factor of 1.3x.

It's important to note that when I was running tests, the first test would always be the fatest. Any further calls to the alias would be slower by nearly 40% until the window was deactivated and activated again. This happened on both 6.3 and 6.31.

I confirm that 6.31 is slower, but not by the amount you claim. The results are rather inconclusive due to the amount of fluxuation of the data I've seen, and nothing points to the PNG library being the problem. Perhaps custom window code was redone in 6.31?


- argv[0] on EFnet #mIRC
- "Life is a pointer to an integer without a cast"