Hi,

I'm heaving problems with getting my triangle drawing routine to clip from the left and top edges. The right and bottom edges already work.

This is what it's doing:


After over a week I got a different result, but I don't really know which one is the closest to good clipping, but in this screenshot, I added 100 to the x coordinates before the calculating part of the routine, then substracting 100 again from the coordinates before it draws the pixels. This is what it looks like:


Here is the code of my triangle drawing routine (sorry, it isn't commented well and comments are in Dutch, and it's long too):

Code:

I haven't looked through the code (it is somewhat long, and I don't read Dutch), but I played around with the second screenshot, and here's what I found:

When the rectangle is clipped on the left side, the triangle on the left looks correct, but the triangle(s) on the right side appear to be shifted 64 pixels to the right and 2 pixels down. This could be the result of shifting the image (just the right-side triangles) right by 256 pixels (2*96+64 = 256), because the LCD memory essentially wraps around and goes down by one pixel on the right side of the screen.

It looks like you're using fixed-point numbers. Are those in 8.8 format? Is that your own implementation, and have you checked that code thoroughly for bugs? 256 is a highly suspicious value here, so I wouldn't be surprised if there's some kind of bug with handling overflows or negative values in your fixed-point code. Fixed-point operations can be a good place for bugs to hide if you're not too careful.

The fixed point math is indeed 8.8, and the math routines are signed. Should I upload the math routines too? And the algorithm was based on a gauroud triangle drawing algorithm (which is a triangle of which the color of each pixel is interpolated between the colors given for the corners) which I modifies to interpolate texture coordinates, instaed of colors.

BTW: the code I posted was the code of the secound screenshot, remove the code between the two lines of ';' chars at the beginning and the lines of code with a row of ';' characters bihind them to make it like the first screenshot.

Also, I can easely translate the comments to English and add more comments, if that's nessicary.

Can you explain your logic behind subtracting 100 the way you're doing in the code you presented? Also, are those coordinates in that section of code "world" coordinates or screen coordinates?

The only thing I can think of is that it'll be a signed comparison mistake somewhere, as it works when dealing with positive bounds but not when things stray into negatives. What does your cpHLDE look like? It's clunky, but here's what I use for signed values:

Code:

The substracting 100 was just a test to see if the problems were caused by the sign, and it turned out some, but not all, of them were. Also, this routine uses affine texture mapping, which means that it only handles screen coordinates (which makes the texture look bad when looking at it from an angle)


Code:
I tried your cpHLDE routine, and it made it draw random pixels.

Note that my cpHLDE routine destroys HL, D and A, whereas yours preserves them. You would need to find a way to preserve them; it'll be slow, but


Code:

may be worth trying to see if it is a signed comparison bug.

It works exactely the same as with my cpHLDE routine. both with and witouth the sub 100's

EDIT: I almost forgot to thank you for your fast reply.

Hrm, it probably isn't that, then. I'm not much good at running Z80 in my head, and tend to prototype more complex routines in a high-level language first, get them working, then translate to assembly - that way if I encounter a bug I have something to compare the assembly code to and can step through until I find a value that's incorrect or similar.

I think the problem might be that your GetPixel routine doesn't handle negative X coordinates. Your clipping routine assumes that the pixel pointer will be to the left, when in fact it is 32 bytes too far to the right.

I made the getpixel routine only work with positive coordinates, and it was a bit better:

It clippes some scanlines correctely, but doesn't draws the other ones. Also the top triangle isn't drawn.

I just put clipping asside for a moment, and tried to make the textures of a better resulution. It now has 16x16 textures, but is slower than the 8x8 version. But that isn't what I wanted to tell, what I wanted to tell is that when I was testing it, I accidentally let it go of the top border of the screen, and this is what happened:


The god of assembly finally starts being friendly to me

EDIT: as you can see, clipping still doesn't work for the left border

Congratulations on at least getting that figured out; any idea what the issue is If you're using 16x16 textures, does that mean some sign bit somewhere is a different location, which could be why the top edge suddenly worked? I'm shooting in the dark here; I can't think of a particularly good reason why you'd be able to properly go off the top of the screen. Have you checked that it's actually getting clipped, and you're not just corrupting the memory above the graphbuffer?

I really don't know why it suddenly works. The difference between 8x8 and 16x16 are only 5 instructions and a few different numbers as texture coordinates. I also already checked if it really clipped, and it looked like it didn't corrupt any memory when clipping of the top side.

I also did some more attempts on clipping of the left side, and I really couldn't get it to work.

Here's the most recent code of my triangle drawing routine:
Code:


And here's a link to the backup of my project, in case someone wants to see the full code: http://dl.dropbox.com/u/11215358/invasion.z80

Reading through it, I think I need quite a few more passes to make sense of that. I'd say you need a lot more comments, but since they probably wouldn't be in English, I don't know if they'd help. Do you have a reference page that you used for the math for this?

I used a slightely optimized version of the gauroud triangle drawing code on this page, of which I replaced the color values with texture coordinates. I'll also try to add English comments ASAP.

I finished commenting the code. Here is now a fully commented version of the triangle drawing routine:

Code: