BFAST using GPUs

[nikhil.stephen]

Dear Sir,

We are Computer Engineering Students. We have read the BFAST paper

http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0007767

We are trying to implement it using GPUs.(CUDA) But we are facing some trouble with the indexing bit.
Could you explain or provide sources for more information on how BFAST creates a reference genome and how indexing is done?

[nilshomer]

Dear Sir,

We are Computer Engineering Students. We have read the BFAST paper

http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0007767

We are trying to implement it using GPUs.(CUDA) But we are facing some trouble with the indexing bit.
Could you explain or provide sources for more information on how BFAST creates a reference genome and how indexing is done?

The best place to look would be the source code found at http://bfast.sourceforge.net, especially "RGIndex.{c,h}" and "RGBinary.{c,h}".

[csoong]

this will be interesting. good luck!

[nikhil.stephen]

difficulty in understanding code

@nilshomer
we are 4th year students doing our final year project and we tried understanding the code given in "RGIndex.{c,h}" and "RGBinary.{c,h}". but its almost incomprehensible for us
Could you email us the algorithm in detail so that we can try coding it on our own? We're confident about coding since we have a fully functional GPU parallelized Smith-Waterman code, written by us

sorry for the trouble.. thank you for ur time

[nilshomer]

@nilshomer
we are 4th year students doing our final year project and we tried understanding the code given in "RGIndex.{c,h}" and "RGBinary.{c,h}". but its almost incomprehensible for us
Could you email us the algorithm in detail so that we can try coding it on our own? We're confident about coding since we have a fully functional GPU parallelized Smith-Waterman code, written by us

sorry for the trouble.. thank you for ur time

This will be beyond my ability to help,

Nils

[god_particle]

Dear Sir,

We are Computer Engineering Students. We have read the BFAST paper

http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0007767

We are trying to implement it using GPUs.(CUDA) But we are facing some trouble with the indexing bit.
Could you explain or provide sources for more information on how BFAST creates a reference genome and how indexing is done?

Hi, Interesting work. I am working in a R&D lab on various High Performance Computing applications. Would like to see if we can collaborate on this effort. Please contact me if you are interested ([email protected]).

[nilshomer]

Why not try to make the core parts vectorized (i.e the Smith Waterman)? The extra money spent on GPUs could be used to buy multi-core processors (not to mention saving rack space). The vectorized implementation would have a greater impact on users than GPUs. I think SHRiMP has vectorized code embedded. This is on my wishlist for BFAST ahead of GPU support.

My 2 cents.

[god_particle]

Why not try to make the core parts vectorized (i.e the Smith Waterman)? The extra money spent on GPUs could be used to buy multi-core processors (not to mention saving rack space). The vectorized implementation would have a greater impact on users than GPUs. I think SHRiMP has vectorized code embedded. This is on my wishlist for BFAST ahead of GPU support.

My 2 cents.

Makes sense. I am looking at using OpenCL rather than CUDA, hence still allowing it to take the path you have mentioned.