Trying to understand mpileup output format

[anderspitman]

We're working on a GPU-accelerated mpileup implementation. We have a generic algorithm working, and now we want to do actual variant calling on the results.

As an intermediate step, we would like to generate a VCF in the same format as bcftools mpileup, that we can pipe to bcftools call. This will let us leverage bcftools to verify our pileup engine.

I'm having trouble finding detailed information about the algorithms and data format output by bcftools mpileup. The output VCF header pointed me to the source code, but after spending a couple hours studying it, I think it would take quite a bit of time before I'm confident I understand it. I also spent some time searching and skimming papers and no luck so far.

Are there any other resources you could point me to?