Luke Shulenburger
Luke Shulenburger
For the qmcpack miniapp, it would be very useful to have gemv and ger. Typical problem sizes would be for matrix / vectors that have dimensionality of ~300 - 5000.
Not clear how to filter this to what would actually improve the code. Tons and tons of header include warnings for instance, but I can't figure what is wrong about...
Thanks, I'll dig more into the warning types.
"This code does store a copy of the coefficient matrix to compute rotated parameters, so it will only work with LCAO." This is a bit worrisome as the code that...
I like the goal of being able to change parallelization settings and getting at least equivalent results without having to change the input file. I think for the case of...
Let me make sure I understand the feature and its intent then. A user should set substeps so that all steps are decorrelated? If that works and is unaffected by...
The final state I would like to get to is having a user specify the tolerance with which a given thing is known and a bare minimum about the sampling...
My way of looking at this had the I/O frequency and hence the averaging granularity determined by the number of blocks. You appear to prefer letting it be governed by...
I agree with aborting if only blocks or steps is specified. My preference would be to allow only total_samples to be specified though.
I think the crux of the problem is that the synchronization is tied to the writing in the current model. We need to ensure that the code always writes with...