Coding Competition #1

[meynaf]

Originally Posted by ross

Each cycle counts in fast routine

Don't worry, this routine can not be fast

Knowing the intended use case would also help - mere palette fades don't need fast code and 68000 is too slow for alpha blending (which would be faster in 24-bit anyway) so i don't see any use.
But perhaps it's just for the sake of the coding compo ?

[ross]

Originally Posted by meynaf

Don't worry, this routine can not be fast

Knowing the intended use case would also help - mere palette fades don't need fast code and 68000 is too slow for alpha blending (which would be faster in 24-bit anyway) so i don't see any use.
But perhaps it's just for the sake of the coding compo ?

I always hope that routines written for fun can serve somewhere (or that may point the way to most significative routines).
I remember when I was fourteen (1984..) I wrote an interpolation routine in 6502 for the C64 that just made me realize what was linear interpolation.
My editor was a sheet of paper.

So to answer You: I do not know if it will never be anything, I do it for fun and competition.
I'm so enthusiastic to returning to write code to 68000 after more than 20 years. An architecture so beautiful!
I hope I never get tired soon, burdened by my work and life.

Bye,


PS: another thing I like is that this forum is full of old farts, I feel at ease

[DanScott]

Originally Posted by ross

Hi Dan, sorry to be pedantic...
all other registers are in an undetermined state

d0-d2 can be trashed

result needs to be returned in d0

all other registers will need saving / restoring if you use them

[DanScott]

Originally Posted by meynaf

Don't worry, this routine can not be fast

Knowing the intended use case would also help - mere palette fades don't need fast code and 68000 is too slow for alpha blending (which would be faster in 24-bit anyway) so i don't see any use.
But perhaps it's just for the sake of the coding compo ?

I thought it was a nice one for the compo... but it could certainly be useful to have a fast or short routine...

[alpine9000]

Originally Posted by meynaf

Don't worry, this routine can not be fast

Knowing the intended use case would also help - mere palette fades don't need fast code and 68000 is too slow for alpha blending (which would be faster in 24-bit anyway) so i don't see any use.
But perhaps it's just for the sake of the coding compo ?

I would use the small version in my games if it is accurate.

I agree the fast version doesn't have an immediately obvious use case (but I guess there might be a good use).

For me, after I call the equivalent of this for each color, I wait for up to 100 scan lines to control the speed of the color fade.

Looking forward to seeing the results.

[Thorham]

Originally Posted by DanScott

it could certainly be useful to have a fast or short routine...

Then it would be better if d2 would range from 0 - 16 (or 0 - 17). That way you don't need divisions.

[meynaf]

Originally Posted by alpine9000

I would use the small version in my games if it is accurate.

Unless someone has made something really unexpected, it will be accurate enough for a game. At least my version is. Fade in/out is very common in games and is always more or less the same thing.

Originally Posted by alpine9000

For me, after I call the equivalent of this for each color, I wait for up to 100 scan lines to control the speed of the color fade.

You should be waiting a whole VBL even if your fade must be faster than that. Else you'll end up with display tearing due to parts of the screen having a different lightness (or you will needlessly alter your coplist).
It is also more OS friendly.

Originally Posted by Thorham

Then it would be better if d2 would range from 0 - 16 (or 0 - 17). That way you don't need divisions.

Assuredly.
There is such a routine in my last game port. It uses range 0-128 instead.

[alpine9000]

Originally Posted by meynaf

You should be waiting a whole VBL even if your fade must be faster than that. Else you'll end up with display tearing due to parts of the screen having a different lightness (or you will needlessly alter your coplist).
It is also more OS friendly.

Yes, as soon as I posted that I wondered why I wasn't waiting for the vblank, and I think it was just because my first fade routine had too many steps so I just sped it up and never went back and fixed it.

[Thorham]

Originally Posted by meynaf

68000 is too slow for alpha blending

256kb table

[PeterK]

Deleted. Sorry for being OT ! No more questions.

[Thorham]

Originally Posted by PeterK

Do you mean a 256 byte table?

No, I mean 256 kilobytes (68020, untested):

;
; in:
;
; d0.w = 0rgb
; d1.w = 0rgb
; d2.w = alpha (0 - 15)
;
; out:
;
; d0.w = 0rgb
; 
    lsl.w   #4,d0
    or.w    d2,d0
    lsl.w   #4,d1
    or.b    #15,d1
    sub.b   d2,d1
    move.l  (a0,d0.w*4),d0
    add.l   (a0,d1.w*4),d0
    lsr.l   #4,d0
    lsl.b   #4,d0
    lsl.w   #4,d0
    lsr.l   #8,d0

Not very practical, I know

[meynaf]

Originally Posted by PeterK

What?? Math precise Accuracy ??
I could only imagine integer calculations in this task. Simple Multiplies and Divisions with completely accurate integer results. No floating point required at all! Wrong?

No floating point required at all, but fixed point yes.

[ross]

Originally Posted by PeterK

I don't want to join this competition

Why? The reward is so good!

but why do you need more than 15x15=225 or if you really want 16x16=256 bytes for a table? (maybe 512 bytes, if you want to avoid another substract or neg command)

I ended up just under 1KB table, but if I had more I would have used them.

What?? Math precise Accuracy ??
I could only imagine integer calculations in this task. Simple Multiplies and Divisions with completely accurate integer results. No floating point required at all!

With enough fractional bit in fixed-point arithmetic You can have math-precise results (easy demonstrable).

Bye!
ross

[DanScott]

Please refrain from turning this thread into a rambling discussion and potentially going off topic....

[DanScott]

Entries will close tomorrow morning (UK time) at around 9am... still time for that kick-ass fast or small routine.

Results will be posted during the week, after I've had time to check through them all.

[DanScott]

Still tied up with things post "Revision" (got 2nd Place in the Amiga demo comp with an OCS demo btw!!) 6 points behind the winner (a rather nice AGA production)

Will try and get the results of THIS competition tied up and posted here this week

[pandy71]

Correctly fading shall be done in linearized RGB color space (and all computer hardware also Amiga use non linear - gamma corrected RGB color space - first RGB need to have applied inverted gamma function fade then apply gamma correction).
https://en.wikipedia.org/wiki/Gamma_correction

[ross]

Originally Posted by pandy71

Correctly fading shall be done in linearized RGB color space (and all computer hardware also Amiga use non linear - gamma corrected RGB color space - first RGB need to have applied inverted gamma function fade then apply gamma correction).

Yes, and then make 100 cycle, 50 byte, 68k code! Sorry pandy71, I'm just joking

You are right, but in this competition we are subjected to severe constraints (small code, integer math, pure 68k, 12 bit RGB, calculated table..).
And in any case I think nobody would notices the difference by using a linear interpolation compared to a real RGB space fading on Amiga OCS.

Bye
ross

[pandy71]

Originally Posted by ross

Yes, and then make 100 cycle, 50 byte, 68k code! Sorry pandy71, I'm just joking

You are right, but in this competition we are subjected to severe constraints (small code, integer math, pure 68k, 12 bit RGB, calculated table..).
And in any case I think nobody would notices the difference by using a linear interpolation compared to a real RGB space fading on Amiga OCS.

Seem you allowed for LUT at least for larger code class. Visibility depends on viewer but some people can notice this (only 16 levels may emphasize this).

[meynaf]

Originally Posted by pandy71

Seem you allowed for LUT at least for larger code class. Visibility depends on viewer but some people can notice this (only 16 levels may emphasize this).

It will be visible as lag (irregular timing) because 16 levels aren't enough.