OSDev.org

Intelligent people... in intelligence company S3.
Writing a video card... With arrows and glitches in middle of text mode screen.
Is it really that hard to make a video card with nor arrows nor glitches in text mode? With tons of other people which are all specialized in that stuff?

It was the only time I ever saw anything like that happen in my life honestly. I won't blame them if they never stumbled upon that glitch, it legitimately looks like a fluke.

My guess is that it was a hardware cursor and it can be displayed regardless of video mode, and the only reason it looked glitchy is because the video driver had assumed the cursor had been cleared and hence wasn't touching it anymore. I'm going to guess it's actually possible to get a working graphics cursor properly on that video hardware provided you actually provide code for it.

It was a S3 Unichrome if I recall correctly (or something like that, I don't have that machine since long ago).

That phrase means something quite different in English than you seem think it does. Unless you meant to imply that the company is run by spies, I suppose.

I think this is related to data collection (spying as you said), right?

It's not related to spying.
I thought on smart-ness.
Sarcasm of "Intelligent people". In its times it has thousands of workers and then all of them could not create working text-mode in their video card. Anyways they failed in previous millenium.

Hi,

This is supposed to be a nice white triangle:



I suspect overflows, or precision loss, or overflowing precision losses.


Cheers.

Brendan

@Brendan, this looks like an abstractionist masterpiece!

Well, it surely is a nice white triangle, but who threw it into the water?

...OK. How did you do that? What kind of algorithm is it? o_O

Hi,



The algorithm is something I call "mastery of deep research"...

I found the bugs and fixed them. Here's what it was supposed to look like:



Note: You are looking at a simple white triangle (with perfect anti-aliasing, focal blur, gamma correction, colour space independence, resolution independence and Floyd–Steinberg dithering).


Cheers,

Brendan

So wait, the bug was in the triangle itself or on the effects applied to it then? (・・？) (also I guess I should ask what kind of hardware is rendering it, since that would affect the algorithm of choice too)

Hi,



The triangle is just a triangle (3 vertices), and there are no effects applied to it (the anti-aliasing and focal blur are just a "happy side-effect" of the method I'm using to render).

The main bug was in a "convert vector to unit vector" routine, which calculated a 32-bit fixed point divisor ("divisor = sqrt(x*x + y*y + z*z);") and then scaled the vector ("x = x/divisor; y = y/divisor; y = y/divisor;"), where 32-bit fixed point didn't have enough precision and caused artefacts. The fix was to replace it with a 64-bit "inverse square root" and multiply instead (conveniently avoiding the need for "divisor larger than 32-bits" which is very expensive in 32-bit code).



The only hardware is "single 80486 (or later) CPU without any FPU" and a framebuffer from VBE.


Cheers,

Brendan

Wow, good job Brendan! I didn't knew you are good with trigonometry...

accidentally made NASM generate a bunch of trash











with VESA-sized glitches in text mode, in the locations of VGA.
overwrote the IVT and did not triple fault.

Meh... I kinda liked the first one better...

Hope you kept the code..