Peter Lindstrom

@alimanfoo, zfp does indeed decompress directly into a user-allocated buffer. Similarly, compression writes to a user-allocated buffer whose size zfp conservatively estimates for the user. Alternatively, the user can specify...

@alimanfoo, let me try to clarify. Whereas zarr prefers uncompressed chunks as large as 16 MB, zfp in its fixed-rate mode uses compressed chunks on the order of 8-128 bytes...

Regarding use cases, you'll have to excuse my ignorance of how zarr works and the use cases it was designed for, but here are a few that motivated the design...

@picca It is true that H5Z-ZFP requires `uint8` words. This is a hard requirement, not just one related to potentially failing tests. Meanwhile, zfp's tests have been written only for...

> Is total coverage actually necessary here or is just a sample along various dimensions of the configuration space sufficient? Well, for any given build configuration, you likely want to...

I'm not sufficiently familiar with HDF5 and how such flags might impact the average user, but I suspect many (most?) zfp binary installations--and there are lots of them in the...

> Do you mean when the HDF5 library `dlopen()`'s the filter plugin? Yes, exactly. That's when the zfp library is loaded, and even if things worked in the past, perhaps...

> For the 8bit stream case, no. There is nothing a user can do to "bypass" that check. If ZFP is not compiled correctly, it will fail any H5Dcreate() call...

@pkoosha FYI, the double precision HIP performance on the `staging` branch (soon to be merged into `develop`) has been substantially improved by reducing register spillage. Currently decode performance on an...

This is on the `staging` branch. The results are based on 3D fields of doubles (I believe one of the Miranda fields from [SDRBench](https://sdrbench.github.io)). You should not have to do...