Chris Seymour

Hey, do you have the developer tools open in chrome at the same time as running chromote?

[cudapoa also outputs to stdout](https://github.com/clara-parabricks/GenomeWorks/blob/c84a30d1a743e913a4eb98bab15de230602c8982/cudapoa/src/cudapoa_batch.cuh#L559) which can go unseen and corrupt output formats _(fastx, sam, ect)_.

I'd also like to see a package on PyPI.

Sounds like a plan 👍

With delta zigzag as per #29 I see about a 40% improvement over my Python implementation with int32 arrays and 20% with int16 (the extra cast is expensive).

@jorj1988 thanks, I'll have a go at benchmarking your implementation and see what kind of performance improvements it yields. @aqrit I would certainly benefit from an AVX2 implementation if you're...

A look at the performance across types https://github.com/iiSeymour/pystreamvbyte/issues/1#issuecomment-492728720 I had a quick go at modifying the [tests/perf](https://github.com/iiSeymour/streamvbyte/blob/zigzag-overhead/tests/perf.c) utility to reproduce in C.  

My primary use case is working with int16 data and that means an expensive copy to cast up in Python before every encode/decode.

@aqrit here are some performance results against using `uint16` with pystreamvbyte *(looking good!)*. ``` $ seq 3 8 | xargs -I% ./tests/perf --dtype uint16 --size 1e% streamvbyte_encode encoding time 0.000878...

> There is a lot of "room for improvement" in the encoder. Encoding speed is more important than decoding for my use-case. Do you think AVX2 would result in a...