Update documentation.

Author: RomeoV

docs/src/index.md

@@ -1,15 +1,23 @@
 # ThreadedDenseSparseMul.jl
 
-ThreadedDenseSparseMul.jl is a Julia package that provides a threaded implementation of dense-sparse matrix multiplication, built on top of `Polyester.jl`.
+`ThreadedDenseSparseMul.jl` is a Julia package that provides a threaded implementation of dense-sparse matrix multiplication, built on top of `Polyester.jl`.
 
-## Installation
+This package addresses the need for fast computation of `C ← C + D × S`, where `D` and `S` are dense and sparse matrices, respectively. It differs from existing solutions in the following ways:
 
-You can install ThreadedDenseSparseMul.jl using Julia's package manager:
+- SparseArrays.jl doesn't support threaded multiplication.
+- MKLSparse.jl doesn't support dense × sparsecsc multiplication directly.
+- ThreadedSparseCSR.jl and ThreadedSparseArrays.jl support only sparse × dense multiplication.
 
-```julia
-using Pkg
-Pkg.add("ThreadedDenseSparseMul")
-```
+ThreadedDenseSparseMul.jl shows significant performance improvements over base Julia implementations, especially for large matrices.
+
+## Performance
+#### [`fastdensesparsemul_threaded!`](@ref) outperforms `MKLSparse` by 2x:
+
+![`fastdensesparsemul!` outperforms MKLSparse by 2x.](assets/main.svg)
+
+#### [`fastdensesparsemul_outer_threaded!`](@ref) outperforms `SparseArrays` by 4x:
+
+![`fastdensesparsemulmul!` outperforms SparseArrays for outer product by 4x.](assets/main_outer.svg)
 
 ## Usage
 
@@ -32,23 +40,13 @@ fastdensesparsemul_outer_threaded!(buf, @view(A[:, 1]), B[1,:], 1, 0)
 The interface is adapted from the 5-parameter definition used by `mul!` and BLAS.
 
 ## API Reference
-
-- `fastdensesparsemul!(C, A, B, α, β)`: Computes `C = β*C + α*A*B`
-- `fastdensesparsemul_threaded!(C, A, B, α, β)`: Threaded version of `fastdensesparsemul!`
-- `fastdensesparsemul_outer!(C, a, b, α, β)`: Computes `C = β*C + α*a*b'`
-- `fastdensesparsemul_outer_threaded!(C, a, b, α, β)`: Threaded version of `fastdensesparsemul_outer!`
-
-## Rationale
-
-This package addresses the need for fast computation of `C ← C - D × S`, where `D` and `S` are dense and sparse matrices, respectively. It differs from existing solutions in the following ways:
-
-- SparseArrays.jl doesn't support threaded multiplication.
-- IntelMKL.jl doesn't support dense × sparsecsc multiplication directly.
-- ThreadedSparseCSR.jl and ThreadedSparseArrays.jl support only sparse × dense multiplication.
+```@autodocs
+Modules = [ThreadedDenseSparseMul]
+```
 
 ## Implementation
 
-The core implementation is quite simple, leveraging `Polyester.jl` for threading:
+The core implementation is quite simple, leveraging `Polyester.jl` for threading. The result is simply something similar to
 
 ```julia
 function fastdensesparsemul_threaded!(C::AbstractMatrix, A::AbstractMatrix, B::SparseMatrixCSC, α::Number, β::Number)
@@ -60,10 +58,6 @@ function fastdensesparsemul_threaded!(C::AbstractMatrix, A::AbstractMatrix, B::S
 end
 ```
 
-## Performance
-
-ThreadedDenseSparseMul.jl shows significant performance improvements over base Julia implementations, especially for large matrices. Benchmark results comparing against SparseArrays.jl and MKLSparse.jl are available in the repository.
-
 ## Contributing
 
 Contributions to ThreadedDenseSparseMul.jl are welcome! Please feel free to submit issues or pull requests on the GitHub repository.