NetSurf AGA optimizing

[Thorham]

Originally Posted by NovaCoder

As I mentioned above, I cannot build NetSurf so I cannot test any optimizations myself unfortunately

Why not ask arti to help you compile it?

Originally Posted by NovaCoder

The AGA SDL which NetSuf is built on only support 256 color AGA displays and so it will not support HAM8 without making changes to the SDL library.....please be my guest if you want to prove me wrong

The changes should be very simple if you know your way around SDL AGA's code. The idea is to simply open a HAM8 screen instead of a 256 color screen, and setup a 64 color palette instead of a 256 color palette.

Since HAM8 is an 8 bit mode, SDL AGA shouldn't know the difference, except that there are fewer palette colors available. After that, you simply use HAM8 render code to render images and what not, and blit that to the screen as normal. The only thing the browser needs to know is to only use the first 64 colors for the GUI instead of 256.

Hypothetically

Originally Posted by NovaCoder

I'd have thought rendering speed should be given priority over quality so why are we even discussing HAM displays?

Yes. The problem is that the moment you don't optimize the fixed palette rendering routine with tables, that you're going to get HAM8 rendering speeds, so that you may as well use that instead (even though that also uses a table).

[NovaCoder]

Arti once sent me his entire workspace but I still couldn't build it because he has his machine configured in a very strange way.

NetSurf is a massive project with lots of dependencies, it's really hard to get it to compile for 68k.

[Thorham]

Originally Posted by NovaCoder

Arti once sent me his entire workspace but I still couldn't build it because he has his machine configured in a very strange way.

No kidding, that sucks

Originally Posted by NovaCoder

NetSurf is a massive project with lots of dependencies, it's really hard to get it to compile for 68k.

And I thought Dopus 5.82 was hard to build (still can't)

[britelite]

Originally Posted by Thorham

Since HAM8 is an 8 bit mode, SDL AGA shouldn't know the difference, except that there are fewer palette colors available. After that, you simply use HAM8 render code to render images and what not, and blit that to the screen as normal. The only thing the browser needs to know is to only use the first 64 colors for the GUI instead of 256.

I'd say quite a lot of webpages might look quite bad if images are rendered in ham8 with all other elements mapped to the 64 color palette, rather than mapping everything to the same 256 color palette. Consider images that are used as elements on a webpage, with the ham8 they will most likely not match the rest of the page anymore.

[Thorham]

Originally Posted by britelite

I'd say quite a lot of webpages might look quite bad if images are rendered in ham8 with all other elements mapped to the 64 color palette, rather than mapping everything to the same 256 color palette. Consider images that are used as elements on a webpage, with the ham8 they will most likely not match the rest of the page anymore.

I mean that you render the GUI using the palette and the page in HAM8. Didn't write that very clearly

[britelite]

Originally Posted by Thorham

I mean that you render the GUI using the palette and the page in HAM8. Didn't write that very clearly

Wouldn't text look even worse then?

[Thorham]

Originally Posted by britelite

Wouldn't text look even worse then?

Not with good HAM8 rendering.

[britelite]

Originally Posted by Thorham

Not with good HAM8 rendering.

Care to show any examples?

EDIT: And with examples I mean different colored texts on different colored backgrounds, without using the colors from 64 color palette.

[Thorham]

Originally Posted by britelite

Care to show any examples?

Perhaps, I'll see what I can do.

Originally Posted by britelite

EDIT: And with examples I mean different colored texts on different colored backgrounds, without using the colors from 64 color palette.

That's a little hard to control, because the HAM renderer I use (part of an open source viewer called V, by meynaf) will choose between palette colors and HAM colors automatically.

[britelite]

Originally Posted by Thorham

That's a little hard to control, because the HAM renderer I use (part of an open source viewer called V, by meynaf) will choose between palette colors and HAM colors automatically.

Well, you could set the palette to all black, that way it won't use the palette in most cases. Another, more realistic, idea would be to set the palette to 64 shades of grey. I mainly want to see how well it handles different color combinations in HAM8.

[Thorham]

Originally Posted by britelite

Another, more realistic, idea would be to set the palette to 64 shades of grey. I mainly want to see how well it handles different color combinations in HAM8.

I can tell you right now that that won't work well. The algorithm relies on a proper palette to handle hard color combinations reasonably well. Using only greys reduces the rendering quality too much.

Anyway, you can try the algorithm with meynaf's picture viewer: http://meynaf.free.fr/pr/mv.lzx. Needs a 68020+ and AOS3. Supports JPG, PNG, BMP, IFF, GIF and some other formats.

[britelite]

Originally Posted by Thorham

I can tell you right now that that won't work well. The algorithm relies on a proper palette to handle hard color combinations reasonably well. Using only greys reduces the rendering quality too much.

That pretty much was my point. Considering the web isn't black text on a white background anymore, you need to be able to map both text AND background to the fixed 64 color palette, or the text will get ugly pretty quick.

[Thorham]

Originally Posted by britelite

That pretty much was my point. Considering the web isn't black text on a white background anymore, you need to be able to map both text AND background to the fixed 64 color palette, or the text will get ugly pretty quick.

I meant a proper fixed palette. A gray scale palette really isn't very good, and will certainly get ugly fast. With a proper fixed palette (which it uses by default), it probably won't do so badly.

Why don't you try the viewer with a bunch of screen grabs of web pages which you think the algorithm will screw up? Or, you could link to a few pages which you think will look crappy.

[Thorham]

To PeterK:

If you're interested, here's the 676 version of my render program: render24bit676.s. You can make the image brighter or darker by increasing or decreasing a2 before the outer loop in the code below.

I managed to optimize the render loop further (discovered displacements are free):

;
; render loop
;
	move.l	#512-1,d7	; image height
.loopy
	move.l	#640-1,d6	; image width
.loopx
	move.b	(a0)+,d0
	move.b	(a2,d0.w,256*6.w),d1

	move.b	(a0)+,d0
	add.b	(a2,d0.w,256*3.w),d1

	move.b	(a0)+,d0
	add.b	(a2,d0.w),d1

	sub.l	d2,a2

	move.b	d1,(a1)+

; dithering
	subq.l	#1,d3
	bge	.l1
	moveq	#2,d3
	add.l	d4,a2

.l1
	dbra	d6,.loopx
	sub.l	#640*6,a0
	dbra	d7,.loopy

Changing the code to handle ABGR is no problem, but BMP files are BGR

[arti]

Sorry I'm not posting anything but I don't have time atm. Maybe tommorrow.

[PeterK]

Hello arti,
how far could you get with my C coding efforts? Please, ask me to explain anything that looks confusing.

I've already found some small bugs and tried to optimize the code a bit by commenting out some lines which may not be required for Netsurf. The most important changes are here:

nsfb_palette_generate_cube_676(struct nsfb_palette_s *palette);
/* this call has to be used in the function set_palette(nsfb_t *nsfb) in sdl.c */
/* I'm not sure how this has to be handled to avoid the NetSurf dithering for my palette */

case NSFB_PALETTE_CUBE_676:
// unused	*r_error = 0;
// unused	*g_error = 0;
// unused	*b_error = 0;
// optional	if (c < 0x46000000) return 0; /* alpha < 70 gets pen 0 */
// optional	if (!(c & 0xF8F8F8)) return 1; /*this color is black */

	if (pushRGBlevel = ~pushRGBlevel) { /* push up every 2. pixel  = primitive dithering */
		if (((c & 0xFF) + 0x16) < 0x100) c += 0x16;
		if (((c & 0xFF00) + 0x1600) < 0x10000) c+= 0x1600;
		if (((c & 0xFF0000) + 0x160000) < 0x1000000) c+= 0x160000;
	}
	
	/* the following code can be replaced with Thorham's table later */
	best_col = 4;
	dr = (c & 0xFF) - 0x35;
	while (dr >= 0) {
		best_col +=0x2A;
		dr -= 0x2B;
	}
	dg = ((c >> 8) & 0xFF) - 0x29;
	while (dg >= 0) {
		best_col +=6;
		dg -= 0x25;
	}
	db = ((c >> 16) & 0xFF) - 0x35;
	while (db >= 0) {
		best_col ++;
		db -= 0x2B;
	} /* Thorham's end will come here ;) */
	
	if (best_col == 255)
		if (!(~(c | 0xFF070707))) return 2; /* this color is pure white */
	break;

@Thorham
It should be possible to merge my RGB level pushing code into the table by simply adding $16 extra entries for the color component saturation to each of the 3 subtables. So, this should result in a combined table (256 + 22) * 3 = 834 bytes.

[Thorham]

Originally Posted by PeterK

It should be possible to merge my RGB level pushing code into the table by simply adding $16 extra entries for the color component saturation to each of the 3 subtables. So, this should result in a combined table (256 + 22) * 3 = 834 bytes.

That's certainly possible. My routine already does that The tables are larger than they need to be for testing (fiddling with the dithering value).

[PeterK]

Since my C coding knowledge is very poor, I just ask if somebody could port this simple table generating ASM code to C ?

/* DataConstantBlock.Byte count, data */

 DCB.B 53,   4
 DCB.B 43,   5
 DCB.B 43,   6
 DCB.B 43,   7
 DCB.B 43,   8
 DCB.B 53,   9

 DCB.B 41,   0
 DCB.B 37,   6
 DCB.B 37,  12
 DCB.B 37,  18
 DCB.B 37,  24
 DCB.B 37,  30
 DCB.B 52,  36
 
 DCB.B 53,   0
 DCB.B 43,  42
 DCB.B 43,  84
 DCB.B 43, 126
 DCB.B 43, 168
 DCB.B 53, 210

uint8_t table_for_cube_676[834];

generate_table_for_cube_676() {
????
}

[Thorham]

Originally Posted by PeterK

Since my C coding knowledge is very poor, I just ask if somebody could port this simple table generating ASM code to C ?

So is mine, but I googled 'c array' and found this:

int billy [] = { 16, 2, 77, 40, 12071 };

[Mrs Beanbag]

instead of

	sub.l	#640*6,a0

how about

        lea -640*6(A0),A0