-
Hi @nullplay,
I wanted to start a discussion on Finch-MLIR MLIR tensors API. In https://github.com/pydata/sparse/tree/main/sparse/mlir_backend we have an initial Tensor class which provides constr…
-
SwitchVisitor and other callable visitors that are overloaded by looplets should become named functions, and named looplet functions should be given uniform names. also, clean up stepper code introdu…
-
Internal:
- [x] ctx should always be the first argument #323
- [x] all direct references to fields of the ctx should become getters/setters/etc #324
- [ ] virtual functions should mirror lowered …
-
Hi @willow-ahrens,
I wanted to adjust `similar` tests for `SingleList` and `SingleRLE` but I noticed that these levels can't be used in 2-dim tensors:
```julia
using Finch
arr = [0.0, 0.0, 2…
-
We decided to leave z3 out of it. Here's the final list of what's needed:
- [x] ScaleArray
- [x] ProductArray
- [x] singlerlelevels.jl
- [x] singlelistlevels.jl
- [x] continuous usage test case…
-
Hi @willow-ahrens,
I continue to work on tests for `similar` in the new constructors suite, and I came across such issue:
```julia
using Finch
fbr = Tensor(Dense(SparseBand(Element(0))), Mat…
-
I'm using Finch for counting sub-graphs where the edge_matrix is the adjacency matrix, and as part of this I produced the following kernel.
```
edge_matrix = Fiber!(SparseList(SparseList(Element(…
-
We should add looplets for stdlib sparse
-
It's too tricky to handle `cached` as a function, because we may want to simplify the lazy value differently than the eager one. Instead, it would be better to just keep a cache of symbol values in th…
-
Let's clean up the level formats a bit so that they can interoperate with other systems! A level is essentially an allocator for vectors, which allows us to represent e.g. matrices efficiently as vect…