Peter Lindstrom

@ben-e-whitney Any progress on this issue?

Can someone on the MGARD team please comment on the status of this issue? It appears unresolved in the latest 1.3 release.

Apple clang thinks they're different types. This code ``` #include #include template bool is_int64() { return std::is_same(); } int main() { std::cout

@william-silversmith It's a bit difficult for me to judge what is being done here. What does "test" do? In particular, how does it feed array dimensions to fpzip? What do...

Thanks for working on this. I'm still unclear on what nx, ny, nz, and nf are set to in each of the four cases that you report results for. This...

@william-silversmith Let me make sure I understand the above examples. The 'contiguous' and 'Fortran' specifications are applied to the numpy array, which potentially results in a change of layout (data...

Thanks, but I'm not sure this addresses the issue I was trying to get at. That's my fault, as I wasn't being very precise. First, I suspect that the affinity...

> NOTE: The version used is from a few weeks ago. The "latest" 0.5.2 (`3918121-dirty`) appears broken as it no longer recognizes `-o` yet won't run without `-o`. Never mind--it...

Some additional data points. When specifying the tolerance as an exact power of two using hexadecimal literals, e.g., `0x1p-149`, then we obtain the following tolerance and observed error: *pwe* =...

I think the issue is here: https://github.com/NCAR/SPERR/blob/3918121d0b192042a425cd3b21ff2f1d151b0952/src/SPERR.cpp#L162-L163 When *pwe* = 2-152, `max_t` &approx; 6.94e-46 < `FLT_TRUE_MIN` &approx; 1.40e-45. Hence, `m_max_threshold_f = m_threshold = 0`, which wreaks havoc with the subsequent...