Optimizing polygonfill bitcopy

[sparhawk]

Originally Posted by a/b

    asl.w    #2,d7        ; longwords to bytes
....

    lsr.w    #1,d3

Replace the above with this

    add.w    d7,d7        ; longwords to bytes
    add.w    d7,d7        ; Saves two cycles

....

    add.w    d3,d3        ; Saves four cycles

This works only for word operands though. Long operands are slower.

[Hedeon]

I am not an expert, but add.w d3,d3 doesn't replace lsr.w #1 d3 i think

[sparhawk]

You are right. I missed the "right" shift part. Works only for the left shift.

[Don_Adan]

Originally Posted by TCH

No, at least i did not. But maybe i was in error. Now i am not sure if this approach is really that reliable:



It shows ~3-~3% differences for the very same routines. With the external measure, the difference between runs was under 0.1%.I agree with this. But this experience is entirely independent of the discussion about the measure methods.

You tested speed on WinUAE. Better test on real A500. Im not 100% sure that WinUAE timing is OK for all 68000 commands.

[TCH]

@roondar:
Perhaps the problem lies in that i run only once the measured code, instead running it in a cycle. But to do that, i'll have to rewrite the C part, either to be encapsulated in a subroutine, or restore the variables it is trashing during work. (Intentionally, last part of the function, none of the variables are needed anymore.) However, this means an overhead, which will degrade the C part's performance, which is not fair, as it not's there without the measuring...

@sparhawk:
Thanks for the tip, i've put it into the algorithm.

@Don_Adan:
Even if we assume, that WinUAE's timings are not close enough to perfect (i'm running this in cycle-exact mode though, which is able to run demos which needs perfect timing), the difference would be appear between the results in emulator and on a real A500 and not between two results in the emulator. Unless WinUAE's timings are really "sliding" and the VM has "hiccups".

Regardless of this, i would agree with you about testing it on real hardware, but for that i would need a handler which is able to mount the PC on the Amiga. I have no ethernet on my Amigas and i still did not had the time and mood to write my parallel port "network drive" handler. Of course, i could just use my old LPT filetransfer program between them, but that would be much more inconvenient. So, it's WinUAE now. I'll do the first option eventually...

[Don_Adan]

Originally Posted by TCH

@roondar:
Perhaps the problem lies in that i run only once the measured code, instead running it in a cycle. But to do that, i'll have to rewrite the C part, either to be encapsulated in a subroutine, or restore the variables it is trashing during work. (Intentionally, last part of the function, none of the variables are needed anymore.) However, this means an overhead, which will degrade the C part's performance, which is not fair, as it not's there without the measuring...

@sparhawk:
Thanks for the tip, i've put it into the algorithm.

@Don_Adan:
Even if we assume, that WinUAE's timings are not close enough to perfect (i'm running this in cycle-exact mode though, which is able to run demos which needs perfect timing), the difference would be appear between the results in emulator and on a real A500 and not between two results in the emulator. Unless WinUAE's timings are really "sliding" and the VM has "hiccups".

Regardless of this, i would agree with you about testing it on real hardware, but for that i would need a handler which is able to mount the PC on the Amiga. I have no ethernet on my Amigas and i still did not had the time and mood to write my parallel port "network drive" handler. Of course, i could just use my old LPT filetransfer program between them, but that would be much more inconvenient. So, it's WinUAE now. I'll do the first option eventually...

Ok, anyway it will be interesting if you can test latest 5 versions on real A500 too. You need access to PC with disk drive, this is the easiest method to share files between PC and Amiga 500.

[TCH]

Why would i do floppy swaping between them, when i have a simple transfer program for parallel port? Besides i do not have a PC with floppy anymore.