llyr-who / solvant Goto Github PK

Problem

std::array is quite limited, in terms of size. On my machine (relative high end XPS), anything beyond a 500x500 matrix causes a seg fault. I think this is because we went beyond what is allowed on the stack. Can we split it up? Can we do something smart? Or shall we resort to a raw memory? (vectors are bad, we are already wrapping them so we can provide some safety)

Suggested Solution

One SS could be to split this up into matrix and tiny_matrix. tiny_matrix will continue to use std::array<T,N>'s and matrix would use T*. Note that we can use std array instead of C-style array because std::array is identical in performance and nicer to use. So use it.

Acceptance Criteria.

• Careful well-reasoned/justified implementation
• testing
• benchmarking

Problem

There is no quick and easy way to get used of the solvant workflow. The only example available is very simple.

Suggested Solution

Add more examples to the examples folder in the root directory.

Acceptance Criteria

• A handful of executing working examples that "showcase" solvant

Problem

Guassian elimination is pretty weak as it is (complexity speaking). We need to look at the implementation and see if we can coax vectorisation. Or even get MPI/std::thread for sufficient problem sizes.

Acceptance Criteria

• clear benchmark improvements on original implemention

Problem

The way in which cmake has been used in solvant, georhiau and qsts (and some others) is very messy and undocumented.

Acceptance Criteria

• "Improve/modernise" workflow, then
• Embellish CMakeLists.txt's with more comments so that future devs understand the why's etc.

Problem

run make test and you will see.

AC

• reduce noise!