Houjun Tang
Houjun Tang
> @roblatham00 , they can confirm, but I think this is something that I think the HDF team could use. Yes, I think a new hint indicating whether HDF5 should...
Hi @tpg2114 , seems like there is a bug with the HDF5 dataspace selection for the write, do you have reproducer codes that I run and debug?
@tpg2114 I'm able to write the data and generate the same plot with your modified codes, will start debugging.
@tpg2114 I believe I found the problem, it is due to the wrong offset/size to rank assignment when there are ranks with no box. Here is a patch with the...
@tpg2114 hmm... Any chance you can reproduce it with the HDF5Benchmark code? Here are two things to try: 1. Add a printf message (may need a fflush call to avoid...
If everything else is the same, independent and collective write should only be a performance difference. Since it works fine with independent but not collective, there could be issues either...
Sounds good, please let me know what you find with 1.10.8.
@tpg2114 could you print the following values before the hanging H5Dwrite: myProc, level, ch_offset[0], hs_procsize[0], hs_allprocsize[0]
@tpg2114 Thanks! These should be enough to reproduce the issue since everything seems to be fine until H5Dwrite. I have created a simple HDF5 program that uses the offset and...