(r7) JaysTermTwoJournal < BETA

Tags: view all tags
---+++01/05/2011
Started reading more in-depth into CUDA. Made some notes (I'll post them later).

Things to do:
   * Finish up the local aligner changes and commit
   * Create a small test driver that ties together the aligners and the has index to verify correctness.

---+++01/06/2011
Finished redoing the banded local aligner. Now I just have to add =FullAlignment= methods into the vectorized aligners. Note that only the =ScoreOnlyMax= methods of the vectorized aligners actually are vectorized. The =FullAlignment= method is just a straight copy from the corresponding non-vectorized aligners, since =FullAlignment= is used much less.

Also, I took out all the kmer index/mapper stuff, since it was still using the old aligner and I didn't want to update it. Those files are really old anyway, and really haven't been updated much for a long time...

To do:
   * Create a small test driver
   * Add =FullAlignment= methods into the vectorized aligners.

---+++01/07/2011
Finished adding in full alignment methods into the vectorized aligners. Now all the aligners are completely redone and I just have to do up the test driver.

To do:
   * Create a small test driver

---+++01/10/2011
Nearly finished the small test driver with a (very) small test set. I realized there's an issue with the hash index that doesn't actually allow me to get a sequence out for the local aligner, only a position, so I'll have to correct that before finishing. Otherwise, the very basic tests look good and it looks like everything is working properly.

To do:
   * Finish the test driver

---+++01/13/2011
Finished the test driver, and it looks like it works, even when simulating low memory conditions in the hash index. I tested using a manually generated test sequence + some test reads, and also using the E. coli bacterial genome and both work fine. I decided to add an =ExtractString= method to the hash index to allow extraction of the string from a position (much like the =getPosition= method in the FMIndex) to make it all work. I feel like that method could be optimized, but that can be for later.

To do:
   * Start a very basic GPU implementation of the aligners!

---+++01/17/2011
Started a new branch for the GPU stuff. I think the first thing I'm going to implement is an "aligner" that just finds how many matches straight-up matches there are. I got the emulators and everything working again, so I have my development environment set up; I forgot to write down what I did last time to get everything to compile and run properly, so I ended up having to re-research it all again, which was a pain. Also started reading through guides and tutorials and taking notes.

To do:
   * GPU stuff.


-- Main.jayzhang - 05 Jan 2011
Topic revision: r7 - 2011-01-18 - jayzhang