igpSP 0.9-2Xb K5 - GameBoy Advance Emulator [50-80%]

Joined: 17 Sep 2006 Location: /mnt/zowki

There is a MASSIVE speedup in this release! I was so amazed I yelled and woke up the whole neighborhood. How can you say the speed has not increased? I especially liked the pixel smoothing (It doesn't actually effect the speed at all). Everything is much more clearer on my ipod nano now.

Here are my stopwatch tests for the Final Fantasy intro (This is no joke, the speed increase is massive):

Code:

No-sound build tests:

Test 1:
CPU Speed: 81MHz
Alpha Blending: ON
Frameskip: 4
Result: 1 min 12 seconds

Test 2:
CPU Speed: 78MHz
Alpha Blending: ON
Frameskip: 4
Result: 1 min 14 seconds

Test 3:
CPU Speed: 81MHz
Alpha Blending: ON
Frameskip: 2
Result: 1 min 22 seconds

Test 4:
CPU Speed: 78MHz
Alpha Blending: ON
Frameskip: 2
Result: 1 min 25 seconds

Here are my test results for the sound build:

Code:

Test 1:
CPU Speed: 81MHz
Alpha Blending: ON
Frameskip: 4
Result: 1 min 25 seconds

Test 2:
CPU Speed: 78MHz
Alpha Blending: ON
Frameskip: 4
Result: 1 min 27 seconds

Contributor Joined: 11 Apr 2006 Location: Ontario, Canada

zowki wrote:

Here are my stopwatch tests for the Final Fantasy intro

Those are around the same times that I had for both K4 and K5 (I just didn't use game_config.txt for the numbers I posted for K4 so that the differences between modes could be easier seen and for consistency with K3 tests). For your info, my stopwatch tests start when the Squeenix logo appears (predicted start for precision) until the "Press start" reaches the exact point where it stops oving (and starts blinking afterwards). I usually try two or three times then round to the nearest second.

Either way, I realized I screwed up a bit for the pixel blending/smoothing code (such that things were more "blurred" than "smooth" and edge pixels were screwy). I've fixed (or rather, improved) that with my current K6 code though I still need to test it with my 5G test iPod.

From now on, things will only get slower as I add new features and have to add additional checking loops for them. Not noticeably slower (only a few extra CPU cycles at max), but still slower.