I'm back home in the Bay Area. I was just playing a 1v1 game in Starcraft 2, TvP on Desert Oasis. The game loaded and I thought welp, what should I do. I decided to give my 'block the choke and hellion harass into Banshee harass into Viking/Banshee' build which I haven't really tested that much a go, and the game ended up being over an hour long with me failing to micro very well in certain situations. Lots of Dark Templars too.

Uh, I've been trying to work on SKY, playing Minecraft and SC2 and Recettear and Mario Galaxy 2, just bouncing between things. I'll be going to Redmond for another interview with Microsoft on Thursday, so I'm throwing reading old class notes into the mix. I've been spreading myself thin over a lot of different things to do lately. I'm not really sure what to make of it... hm.

Before I continue with implementing the game itself, I want to iron out all the standing issues with the game. Today I've been working on the load times. Loading the game, before todays work, was taking 51.5 seconds to complete. While all the textures would load pretty much instantly, all the models would take all sorts of time to load, namely because the models were stored in plain text which the application had to tokenize and translate into numbers. Doing all this for what was 427 KB of text was taking quite a while.

I changed things around so instead of the models being stored in plain text, they're stored in binary values. I had to fiddle around making my C++ model exporter program output them in Big Endian, and I had to fiddle around in Java with reading this values and implementing some more efficient reading and caching strategies. The end result is that the models now only take up 95.5 KB and takes only 4.2 seconds to load! Hurray!

There's a little more optimization I can do to make it load faster as well. I'm currently just using a DataInputStream object, which does not let me read an array of floats. If I could get something which reads in an array of a give datatype all at once, that would be great. It does have a method for reading in an array of bytes... I wonder if the processing from those bytes to floats will make it a slower process. The type unsafe-ness of C++ can be nice sometimes...

Some screenshots of the menus I've been putting together. I'm going to have a Yaku reference in the game. You can see the entry for Riichi expanded in the first screenshot, and the middle of the list on the second. Also if you're starting a custom game (as opposed to a Quickplay game) you can set a bunch of settings. The list of settings will be pretty huge, and you can see a few of them in the third screenshot. Yes, I'll have Open Riichi in the game.

It's coming along. The plan is to work on the game more full-time during the fall/winter period while I look for game development jobs. Perhaps I can borrow a friend's Mac for a month once I put out this game on Android and port it to iPhone. I'll get myself an iPod Touch or something in order to test it.

SKY runs really well on the G1. The framerate is very smooth, at least until I start rendering the tiles themselves. I was noticing when I was exporting the tiles from blender to a model that the amount of vertices was way higher than it needed to be. I fiddled with it for a bit, but couldn't make it budge. I figured I'd wait and see what kind of effect it would have... turns out quite a bit one! So I need to fiddle around with optimizing the models.

But the next step for the actual game section of Sky is to actually start implementing the game - picking, drawing, having a hand in front of you that you can move around and discard tiles from, etc. Then implement some sort of win listener system. I'll have it so there's a listener for each Yaku that analyzes your hand - if the listener for Pinfu, for example, finds 4 melds, all sequences, and with no other minipoints, it'll return true or something. I have no idea how I'm going to do AI though.

Although I would really like voice chat, it's a feature I'll probably cut to start out with. For now I've just been working on menus. They need to be made, and they won't take as long to put together as the rest of the game. In theory, there's good progress to be made by having my menu system up and running! So there we go.

It's been a long time in the wanting, but I've finally acquired a smartphone! This G1 was acquired in order to do debugging for Mahjong SKY, but after hacking the thing to activate without the need for a data plan, it turns out that my pre-paid sim will work in it! So as long as there is wi-fi, I have access to everything I could want out of this thing!

I was watching the video of and played a bit of Metagun, made by Notch of Minecraft fame. It's impressive, that's for sure, and actually pretty difficult. I should do something like make a game with all these tools I've written. SKY will be good! And I'll make other things too!

The title here is the name of the research project I've been working on over the summer. I've been putting in some finishing touches of sorts, making it work better before a demo of the system tomorrow. The idea of the system is that it's a plugin-based framework for visualizing network security data for use in the Allosphere here at UCSB. It's a surround view 3D visualization system with stereo and all the works. I'm working hard to make it visually awesome, as this project will be really great for my portfolio. I will need to update my professional site with it once the system is done and I have some good screenshots of it in action.

I spent a large part of today transitioning the system from using GLUT to using GLFW in order to have anti-aliasing on Windows. An overwhelmingly large chunk of that was spent making multisampled framebuffer objects work, and working out a dumb bug that got in there. Works great, if not slow on my modestly powered laptop. I definitely need to work on optimizing this framework, as it's a bit of a slow implementation, brute forcing a lot of stuff. My math library doesn't have a complete frustum class, so I cant, for instance, do frustum culling on all the thingies that are in the scene. That would be a good idea though.

This link was very helpful in getting it working in the end today, for doing multisampled FBOs. I got it working so I can render to multisampled cubemap render targets as well, since I'm using cubemaps to do my 360 degree surround mode. There's a lot of modes to this project, which I can enumerate on the professional site when I get to that. Lots of features, not enough time, that's for sure.

I can think of a bunch more things I could get working before tomorrow, but I'm not sure if I could go for them all right now... hm.