Anyone currently developing on a Katana?

Just wondering if anyone was developing anything on a katana kit and if they wanted to share information about what they are working on.

I'm curious I guess to how many units are now in collections versus actively being actively used. I'm also just curious to hear if anyone has anything cool brewing.

 

I plan on getting one to use for code testing for nullDC, but I don't have one yet...

 

It's disappointing, but people actually making stuff with their devkits is pretty much unheard of. This is pretty frustrating for someone like me who wants to see games/demos coded but sees so many collected just for the sake of having the thing.

 

i use mine to code. mainly highly optimized transform and lighting + a custom 3d model format. Right now i get 80000 vertices per frame at 60fps, this is roughy 4 times the geometry of DOA2, but with no time left for game logic. I have a new model format in the works but i have no time to implement it right now. It should give a significant performace boost.

I would some time like to work on a wipeout-like game for the dc, but i do not see it happening anytime soon so i am just coding superfast modules that will help in the building of a tight and very fast 3d engine.

Right now my workbench is a mess as i am working on my final project for the university, i will try to compile the exporter and post some screenshots if i remember it.

 

I've always been under the impression that the general consensus it it's highly frowned upon to use the official development tools / hardware outside of compiling the examples and running them for personal use.

What is the stance now (especially interested in hearing if those who had considered it strictly off limits think differently now)?

KOS is great and I agree that it's been easier to get up and running with it than trying to use the official SDK but why shouldn't it be an option? Naturally, use of the licensed third-party software like encoders, etc, should be off limits, but what use to SEGA are the first-party libraries designed specifically for the Dreamcast's abandoned and now outdated (at least, in comparison to the next generation systems) hardware?

 

oh i just found a screenshot....6 models, a total of 61000 vertices with one dynamic light and one ambient occlusion texture, 78% of cpu time at 60 fps...DOA2 which is one of the most geometry heavy games pushes around 25000 vertices (characters and background) per frame at 60 fps. I just hope to find some genuinely relaxed free time to work on my new format and reach new levels of performance My current format evaluates the shader on all vertices and still is very fast. The new one will use vertex sharing and the workload will drop significantly

 

well i have really no idea what the general stance about using an official devkit is....i am really preoccupied by coding so i do not really care. KOS might be good and all that, no doubt. But the thing is, if you are really serious about pushing the hardware to the limit, then your only choice is a katana devbox as it gives you access to the sh4 pipeline simulator.

My current transform and lighting code is all in sh4 asm and takes 40 cycles per vertex (loading of vertex data, transform, light, project, send to the powervr) and the pipeline stalls for 1-2 cycles only. The code is 95% free of any kind of stall, and manages to complete 2 instructions every cycle with loads of parallelism from cache prefetches from RAM and cache flushes to the tile accelerator. Achieving this level of performance with no hardware feedback on stalls and cache misses is virtually impossible, and tha katana devbox provides this vital information.

 

awesome job

 

It's really sad because it is special hardware. It's not like a simple debug station or something like that, it's a tool that can be used to write fast code that would be very very difficult (if not impossible) to write in any other way.

 

oh i actually forgot it. I included NvTriStrip in my exporter to enable triangle stripping, and this usually cuts the amount of vertices in half with no quality loss at all. So my pipeline can push 80k vertices * 2 , almost 160k "effective" vertices per frame at 60 fps, almost 7 times the amount of DOA2

the screenshot above features un-stripped geometry. If i had enabled stripping, the cpu time would have fallen to 40%

Oh well...i really can't wait to implement the new format

 

Awesome!

@VitAmen

This is exactly what I was wanting to hear about. I wasn't looking for things people were even planning on distributing which is what most of the concern over using the Sega owned IP is about. Just wanted to know more about what cool things people were working on.

Nice job if you do continue work I would be very interested in seeing what you are doing and hearing about how it progresses.

Did you write the model format from scratch or base it on something else?

 

Very impressive!

 

Very impressive indeed, Vitamen.

The purpose of my devkit is also for programming, but it has been on hold for a while now due to various reasons.
Migrating to Windows 7 & make the devkit work on that OS was one of the reasons... I was even prepared to go back to XP if I couldn't make the devkit work on Win7

Like Vitamen, I don't plan to use the SDK, but custom-made libraries.
The tools like codescape however have no equivalent in the homebrew world, so they are really interesting.

I might port my PSX movie player PsxTulz to Dreamcast one day, as easy start-project, but finding the time & will to program after a hard day's work is more and more difficult... so it might take years

 

the model exporter was written from scratch using the maya sdk...the only external library i used is nvidia's nviTriStrip that converts triangles to triangle strips...

my only concern right now is that it does not support exporting of skeletal animation, mainly because i am not a skilled animator to produce a good data set that will motivate me to see it animating in real time on the dc...if i could find a good animator to supply me with a high res model featuring detailed skeletal animation i would be more than willing to give it a try. So for the time being, it's just pure geometry...

i am also glad that you like it...if anybody knows how to produce a selfboot dreamcast disc i could provide you with the binary and model or two so you can check it out. I have never made a dc disc and have not the time to do it now, so if you are interested let me know...

 

If you plan to play with the triangle strips a little bit more, you might want to have a look at tristripper... opensource library to do it realtime
http://users.telenet.be/tfautre/softdev/tristripper/
The link contains benchmarks on a PIII

 

Except for optimizing I actually prefer to use a regular Dreamcast. Mainly because I do all coding on Linux with my own SDK, and the development environment for the Katana is Windows only. So the Katana gets booted up if just adding a few timers isn't enough to identify the hot-spot.

 

For a while there I thought you wrote "vertex shading"(!) and was about to ask if I could have some of that goodness. ;-)

 

i would but what knida of 3d software needed i would like to make a bus simulator
something like http://omnibussimulator.de/english.htm
OMSI BUS SIMULATOR.

 

You can use any 3D software you want as long as you have a way of importing its 3D data. All of the 3D examples in the official SDK used the included 3DS Max export plugin to format the data it used.

If you want to play a bus driving simulator on the Dreamcast, there's always Tokyo Bus Guide.

 

yes i do vertex shading in sh4 asm. i may release my source code sometime in the future, maybe when i finish my new rendering pipeline. By vertex sharing i mean i plan to use indexed vertex arrays with high performance cache management, in contrast to what i am doing now which is evaluate the shader for every single vertex sent to the pvr.