Floating Point Exception on GPU

[schorlton]

Hi,

Thanks for great software. Running latest version of racon (commit b591b12c22539948782704446989893bde826a29) and hitting a floating point exception on GPU, but CPU works. Thanks for your help!

I'm attaching what I hope is a reproducible example. racon_debug.zip

root@a5698f05c7c3:/data/racon_trouble# racon -m 8 -x -6 -g -8 -w 500 --include-unpolished -t 4 --cudapoa-batches 1 --cudaaligner-batches 4 --cuda-banded-alignment filtered.fastq output.paf polished-input.fa
Using 1 GPU(s) to perform polishing
Initialize device 0
[CUDAPolisher] Constructed.
[racon::Polisher::initialize] loaded target sequences 0.000033 s
[racon::Polisher::initialize] loaded sequences 0.006921 s
[racon::Polisher::initialize] loaded overlaps 0.001669 s
GPU 0: Aligning with band width 68
[racon::CUDAPolisher::initialize] allocated memory on GPUs for alignment 0.989071 s
Alignment skipped by GPU: 415 / 921
[racon::Polisher::initialize] aligning overlaps [====================] 0.035475 s
[racon::Polisher::initialize] transformed data into windows 0.001138 s
[racon::CUDAPolisher::polish] allocated memory on GPUs for polishing 1.352416 s
Floating point exception (core dumped)

[rvaser]

Hi,
sorry for my later reply! As the GPU part was implemented by NVIDIA, I think you should raise the issue at https://github.com/clara-parabricks/GenomeWorks.

Best regards,
Robert

[schorlton]

Thanks! Done. 🤞