Direct3D: how does it work? old source code?

For the most part it's just a translation layer that simplifies and does some front end checking, before forwarding calls to the driver. If you download the Windows DDK you can look up the interfaces that the driver exports to the D3D runtime. There's a bunch of rules and specifications about how the driver and D3D will interact.

Oh, I think I'll move this to DirectX.

I guess I should be more specific..I am interested in these things:

-how directx transformation works
-where and how does it use SSE/MMX
-does it use fixed point math or just regular floats
-where it does trivial reject, and how
-does it clip in 3D, or in screen space
-the structure of its pipeline
-does it use multithreading
-is the rasterizer/transformation/clipping even done in CPP? or is it all made in assembly?
-source code!

thanks

Quote:

Original post by CPPNick -how directx transformation works -where and how does it use SSE/MMX -does it use fixed point math or just regular floats -where it does trivial reject, and how -does it clip in 3D, or in screen space -the structure of its pipeline -does it use multithreading -is the rasterizer/transformation/clipping even done in CPP? or is it all made in assembly? -source code!

For most of that stuff Direct3D doesn't do anything. It gives the info to the graphics card which does all of that however it pleases. You don't get source code because it's secret to the various hardware vendors. The short is that most is done in very parallel hardware (not threads, just actual parallel circuits)

There's probably some white papers floating about somewhere describing how some particular graphics card chipset does it's business. Search around on NVidia or ATI's websites.

-me

I think I understand, but this is why I ask is that I downloaded the open source 3D engine from Irrlicht, and the quake3 map loader file reports about 350 frames per second at 640x480 with torches, dynamic lighting all over the place....
my project is a very basic rendering engine with static gouraud shading, no collision detection, no filtering, and I can barely squeeze 30-40 fps out of it at 640x480...even "burning's software rendering engine" that is in the Irrlicht project is pretty slow..it does also handle filtering, ect...but its still slow
What is it that makes Direct3D so fast?

btw: im on a laptop with an integrated ATI 200M graphics card, which I doubt has any hardware rendering capabilities...

Quote:

Original post by CPPNick I think I understand, but this is why I ask is that I downloaded the open source 3D engine from Irrlicht, and the quake3 map loader file reports about 350 frames per second at 640x480 with torches, dynamic lighting all over the place.... my project is a very basic rendering engine with static gouraud shading, no collision detection, no filtering, and I can barely squeeze 30-40 fps out of it at 640x480...even "burning's software rendering engine" that is in the Irrlicht project is pretty slow..it does also handle filtering, ect...but its still slow What is it that makes Direct3D so fast? btw: im on a laptop with an integrated ATI 200M graphics card, which I doubt has any hardware rendering capabilities...

So you wonder why hardware rendering is faster than software? Well that's because the graphics cards are tailored for rendering, while the CPU is not.

Quote:

Original post by CPPNick What is it that makes Direct3D so fast?

It uses dedicated hardware to do the rendering, such as your GPU. The CPU is used for minimal processing.

Quote:

Original post by CPPNick btw: im on a laptop with an integrated ATI 200M graphics card, which I doubt has any hardware rendering capabilities...

It actually does...

Quote:

Original post by CPPNick I guess I should be more specific..I am interested in these things: -how directx transformation works -where and how does it use SSE/MMX -does it use fixed point math or just regular floats -where it does trivial reject, and how -does it clip in 3D, or in screen space -the structure of its pipeline -does it use multithreading -is the rasterizer/transformation/clipping even done in CPP? or is it all made in assembly? -source code! thanks

These days almost all of those things you've mentioned are done entirely on the GPU. Like it's been said several times already, D3D is mostly just an interface layer that lets your app talk to different hardware types using a consistent API.

Direct3D does come with a math/utility library called D3DX, which allows you do to some of these operations on the CPU. However no source code is available for this. You might want to check out the brand new XNA Math library available in the March 2009 SDK...it's a brand-new SSE-accelerated math library with full source code included (no libraries or DLL's, just headers you can include).

Quote:

Original post by CPPNick I think I understand, but this is why I ask is that I downloaded the open source 3D engine from Irrlicht, and the quake3 map loader file reports about 350 frames per second at 640x480 with torches, dynamic lighting all over the place.... my project is a very basic rendering engine with static gouraud shading, no collision detection, no filtering, and I can barely squeeze 30-40 fps out of it at 640x480...even "burning's software rendering engine" that is in the Irrlicht project is pretty slow..it does also handle filtering, ect...but its still slow What is it that makes Direct3D so fast? btw: im on a laptop with an integrated ATI 200M graphics card, which I doubt has any hardware rendering capabilities...

Short answer: You do/it does.
Long answer: It can beat the hell out of a CPU any day, but integrated graphics hardware gets shown the door by more or less any modern dedicated solution. Just because it's *relatively* weak doesn't mean it lacks any sort of capability. Chances are you're doing things in a serial fashion. GFX cards are fast because they do things in a *massively* parallel fashion. Like, several hundred 'cores' at once on fairly modern/capable silicon.

okay, but: I remeber years and years ago, when I had a Pentium I 200Mhz, and was able to run Dave Mirra Freestyle BMX () at 800x600...I have a 1.6ghz processor now, which can out-perform (x3) the Pentium I AND any graphics hardware I may have been lucky enough to have in whatever junky video card I had at the time...It just boggles my mind how they could have gotten that to work...they have at least some kind of filtering...I can't image how they got it that fast..

Quote:

Original post by CPPNick I can't image how they got it that fast..

Cause they're better at programming than you [smile]

If you're curious why your program is slow, run it through a profiling tool and/or use NVidia's performance tools to see where the slowdown is. Often for beginners it's "dumb" things: like: loading a mesh every time you use it instead of caching it, using immediate mode instead of VBOs or whatever, or using an O(n^2) algorithm when a constant time one exists.

Even for "good" programmers you can often get a 2-4x speed in your "good" coode by changing a few small things once the profiler points them out as big problems.

-me

my absolute first attempt at textured 3d bjects, were two triangles, with a 100x100 bitmap on them. when I zoomed in so the triangles were full screen, it got so slow the program froze... I have gained more speed than I thought possible after that first attempt..I'm wondering if its futil to go on any further, and if I should just start using Direct3D. I was hoping for a much better result without it.

Wait...a you writing a software renderer?

yes, software renderer. I was asking about source code for an older version of direct3D or something because I thought it had a built in software renderer..but it looks like direct3D doesn't actually have any internat rendering code, and only forwards the info to the appropriate hardware, right?

It has the reference rasterizer, but that's strictly for testing purposes and tends to run at 2-3 FPS at best. There's the new WARP softrast coming in, but I'm not sure when that gets out of beta.

neither WARP or the reference rasterizer are open source though =/

It's unfortunate that SwShader is no longer open...maybe source code for it is still floating around?

If you are writing a software renderer, maybe you do good by having a look at those two:

• Agner Fog's Manuals
• Intel's Manuals (not really Intel specific, except for some small parts)

hey...thanks guys! I found a few different versions of swShader here, and Fog's manual looks interesting too. I havn't had much time to look it over yet, but swShader looks pretty impressive so far...I read some posts somewhere while looking for the source code that suggested the project kinda fell through, and that it wouldn't be able to compete with the new Pixomatic engine, which is being worked on by the Zen master himself..is this still true? anywho, thanks again. hopefully swShader wont be as complicated to look through as irrlicht's engine..

Check out Tricks of the 3D Game Programming Gurus by Andre LaMothe, it covers writing a full software rasterizer in baby steps and also comes with a CD which includes source code.