Apply PFAC to DNA application

[Cheng-Hung Lin]

We can think about how to apply PFAC library to accelerate DNA
applications.

[Lung-Sheng Chien]

We can focus a huge problem, say find EST fragment in a complete
genome human sequence.

size of EST database is 1GB
size of genome sequence is 3GB

it is impossible to complete this task in one kernel call.

In CUDA 4.0, we can try
1) one PFAC context binds to several GPUs and divide patterns and
inputs into small ones and
iterate all combinations, then merge.
The overhead of merging is huge. so if we can have a space-
efficient version of PFAC, then
we can reduce number of merging.
2) how about MPI? We need a MPI + PFAC example.

GPU computing is useful when app is HUGE.
If we only demo small app running ms, then no one cares about this
because CPU can do it very well.

[Cheng-Hung Lin]

We can refer to the paper "Tandem repeats finder: a problem to analyze
DNA sequences".
The paper is to find common subsequence without using dynamic
programming.
I think we can apply PFAC to accelerate the method proposed in that
paper.

[Lung-Sheng Chien]

common subsequence is only a part of the paper,
do you want to implement whole algorithm of that paper?
I don't think so.

If you want to extract common subsequence of [A,T,C,G]^4, then
the problem is the same as
pattern = aaaa
input = aaa......aaaa

because each thread reports a match of a pattern of length 4.

> > > applications.- 隱藏被引用文字 -
>
> - 顯示被引用文字 -