Peter Lindstrom

Sam, thanks for fixing this. For some of the work we're doing at LLNL, we need "small" tolerances below `FLT_TRUE_MIN` that actually are *large* in relation to the data being...

We were seeing lots of outliers before the fix, which really hurt compression. Once we have some new results, I'll get back to you. Either way, it shouldn't matter if...

I can confirm that h5repack gives this error (with 8-bit word size, as needed). I've not patched my h5repack, but presumably one shouldn't have to with recent versions of HDF5--I,...

It produces the kind of output you might expect, with no warnings or errors. Output attached. [h5dump.txt.gz](https://github.com/LLNL/H5Z-ZFP/files/14425057/h5dump.txt.gz) [h5ls.txt.gz](https://github.com/LLNL/H5Z-ZFP/files/14425058/h5ls.txt.gz)

Could it be that `H5T_STD_U8LE` and `H5T_STD_I8LE` are not handled? zfp supports only `int32`, `int64`, `float`, and `double`. Here's part of the output I'm seeing: ``` h5repack -n -f UD=32013,0,4,3,0,3539053052,1062232653...

> What happens for you if you adjust the Makefile to read... Same errors as far as I can tell.

FWIW, the test passes if I leave out `HDF5_PLUGIN_PATH=../src/plugin`. Not sure what HDF5 does in this case when requesting zfp compression.

Some potential differences: I'm building H5Z-ZFP using GNU make and AppleClang 15.0.0. I'm using the `master` branch of both H5Z-ZFP and zfp (I had some vague recollection that there was...

So my mesh_repack.h5 does not contain any zfp magic numbers (`0x0570667a` in little endian), suggesting that compression is not even being invoked.

> Well, is the mesh_repack.h5 you are looking at one you generated by disabling setting of HDF5_PLUGIN_PATH as you mentioned https://github.com/LLNL/H5Z-ZFP/issues/137#issuecomment-1967545327 No, the path is correct and the plugin is...