Peter Lindstrom

It is of course undesirable for zfp to crash if fed NaNs or infinities, which I suppose happens in the conversion from floating point to integer in the block-floating-point transform,...

Thanks for the suggestion. From IEEE 754-2019 Section 5.8 (Details of conversions from floating-point to integer formats), it seems clear enough that an exception would be generated: ``` When a...

I wasn't even aware there was a zfp PyPI package. You're likely looking for zfpy, the Python bindings to zfp: https://pypi.org/project/zfpy/.

I guess I had already forgotten as I did participate on that thread. :-) But #143 deals only with how to support compression programmatically--it says nothing about how to use...

@vicentebolea Not sure what these `ndarray` errors are. I'm busy over the next few days but will take a closer look as soon as I can.

`BIT_STREAM_WORD_TYPE` is a GNU make option for zfp (see https://zfp.readthedocs.io/en/release1.0.1/installation.html#c.BIT_STREAM_WORD_TYPE). We should make sure both or are accepted by CMake, or perhaps deprecate one of them.

This is sloppy documentation rather than bug in the code. While the "correct" way of invoking this compression mode via the [CLI](https://zfp.readthedocs.io/en/release1.0.1/zfpcmd.html#file-compressor) is to set `minbits = maxbits = 0`...

You are interpreting the precision correctly. That is, the relative error is at most $2^{9-p}$ for floats and $2^{12-p}$ for doubles. To verify this, I ran an experiment on a...

After some investigation, it turns out that there are two issues: 1. As I suspected, the error bound does not hold when the "lossy path" of the compression algorithm is...

Reversible (lossless) mode is not yet supported by the CUDA backend. See https://zfp.readthedocs.io/en/release1.0.1/execution.html#setting-the-execution-policy.