h5part AT lists.psi.ch
Subject: H5Part development and discussion
[H5part] H5Part field data meeting - August 15th, 2006
- From: Cristina Siegerist <cesiegerist AT lbl.gov>
- To: h5part AT lists.psi.ch
- Subject: [H5part] H5Part field data meeting - August 15th, 2006
- Date: Wed, 16 Aug 2006 09:05:04 -0700
- List-archive: <https://lists.web.psi.ch/pipermail/h5part/>
- List-id: H5Part development and discussion <h5part.lists.psi.ch>
- Organization: LBNL
H5Part update from Achim
- Added errror handling.
- Finished changes to the API (error handling, return values, data types).
- Tested it with BeamBeam3D in Horizon.
- Andreas started porting his code to the new API.
H5Part Field data API:
Each processor will declare the overall size of data array and they will select their local array to write to.
We will have a call to an H5Part my local size that will specify xmin, xmax, ymin, ymax, zmin, zmax (these include the ghost zones) that correspond to the IJK indeces. An MPI_ALLGather call will put together a Np x 6 vector of local sizes.
The ghost zones will be shaved if needed and also there will be a check for divisible by two for the ghost zone (equal number of cells to each side).
There will be an attribute IJK, KJI, IKJ to set the ordering of the dimensions. IJK is the default.
H5Part Next steps:
- implement compression
- tunning. chunking can be implemented once the logic is working correctly.
- conventions for coordinate systems. For the moment is uniform grid.
- Once the field data is in a prototype state, meet with Rob and Ji to talk about naming conventions, data types, etc.
- [H5part] H5Part field data meeting - August 15th, 2006, Cristina Siegerist, 08/16/2006
Archive powered by MHonArc 2.6.19.