Ehsan Totoni
Ehsan Totoni
Thanks for posting your solutions @samuel100 . Installing MPI and HDF5 and making sure HPAT picks them up properly seems challenging in a portable way. Maybe we need CMake.
If a parfor `p` has index variable `i` and array is accessed with index variable `j`, we need to make sure `j` doesn't depend on `i` before parallelizing `p`. This...
The result of index_test.jl is wrong I believe. All arrays should be SEQ but they are not. I update the test to properly check for this.
Thank you for the feedback, Tim! I have tried `SnoopCompile` before. Unfortunately, the performance problem is in type inference as I mentioned in person. Precompilation doesn't work because of the...
Yes, benchmarking time spent on inference of different functions is definitely useful.
For ParallelAccelerator and HPAT, inference is much larger (70% I think) because they are not numerical codes.
Could you give a Julia example code for this use case?
I agree; this is a general problem. Can we return an error when an escaping variable is not typed?
I see the same issue on all the benchmarks I have tested for HPAT. I'm working on testing OpenMP reduce on Cori now. I think we might have thread affinity...
OpenMP reduce is similar in performance seems like. I don't know where this performance difference comes from.