This is a routine I wrote quite some time ago but never got around to releasing it to the public. The project initially was to simply create a public-domain 65816-based assembly code for LZW compression/decompression, but it quickly expanded to trying to optimize the algorithm and implement it into something slightly useful. I don't know if I've accomplished either of these, but here's my attempt. The algorithm is sorta explained below and the program is simply a GIF saver that saves most any Super Hires image file as a GIF image. I could spend a few kilobytes of memory trying to explain what I've done, but instead I'll just give a REAL brief overview of it and let it go from there. If there is further interest, then I'd love to hear about it. Especially if there is a better way of doing this. Right now, all I have ready to release is the compression part of the project. The decompression is done and works, but it has yet to be implemented into anything useful. Currently it's just a GIF loader that only loads very specific formats. I'll try to get something going with it fairly soon. Now for the good part, the part where I attempt to explain what I've done. At first I simply followed the standard LZW decompression algorithm found in tons of documentation materials, but found it to be quite slow and inefficient. This is the best I could come up with to get around this problem. At this point, I'll assume you understand the concept behind LZW compression. If you do not, then check with other documentation to get a basis on which to build. Now with that out of the way, let's consider the standard string table. It's a table listing strings of data that have occurred within the file being compressed. Once a string has been added to the table, characters are again read in from the source file and added to a current string until the current string can not be found in the string table. This occurs when one character has been added to the current string (which IS in the table) to create a string that is not in the table. Using this fact, it's possible to represent any string in the table as a code for a string in the table (a prefix) plus a character. This alone increases the speed of searching and significantly decreases the size of the string table (since every string is now represented by two codes regardless of the string's actual length). But searching still consists of searching the entire list of strings to find the right combination of prefix and character. The routine included with this file works on the concept of only searching one prefix to see if that prefix has occured with the character in question. If not, then another branch is added to that prefix and the character is placed there. This greatly increases speed by allowing the search routine to completely ignore any string that doesn't even begin with the prefix in question. That's basically it. I ain't the best person to try to explain something, so my suggestion if you're interested in working with this is to simply follow the source code and I'm sure you'll figure it out. Let me know if this source code has helped in anyway. I'm not asking that any credit be given if this routine is used in your own code, but I won't deny it if I am mentioned. Also, the desktop shell that this program works with was written by Jonah Stich. Thanks Jonah, saved me ALOT of time.