Simplify mul! dispatch

[dkarrasch]

Guys, I saw the light, while I was struck by a thunder. Look what we can do! Let a generic mul! strip off potential AdjOrTrans wrappers, and peel off the actually data-storing array. With this design, it is crystal-clear what SparseArrays.jl, GPUArrays.jl and its surounding ecosystem need to overload. Not tons of mul!, but generic_matmatmul! with all three matrix arguments of their own(ed) kind: SparseMatrixCSC, CuArray, MtlArray etc. In this PR, we have 4(!) methods as follows:

julia> methods(LinearAlgebra.generic_matmatmul!, (Any, Any, Any, Any, Any, Any))
# 4 methods for generic function "generic_matmatmul!" from LinearAlgebra:
 [1] generic_matmatmul!(C::StridedMatrix{T}, tA, tB, A::StridedVecOrMat{T}, B::StridedVecOrMat{T}, _add::LinearAlgebra.MulAddMul) where T<:Union{Float32, Float64, ComplexF64, ComplexF32}
     @ ~/Documents/julia/usr/share/julia/stdlib/v1.10/LinearAlgebra/src/matmul.jl:349
 [2] generic_matmatmul!(C::StridedVecOrMat{Complex{T}}, tA, tB, A::StridedVecOrMat{Complex{T}}, B::StridedVecOrMat{T}, _add::MulAddMul=MulAddMul()) where {T<:BlasReal}
     @ ~/Documents/julia/usr/share/julia/stdlib/v1.10/LinearAlgebra/src/matmul.jl:411
 [3] generic_matmatmul!(C::AbstractMatrix, tA, tB, A::AbstractMatrix, B::AbstractMatrix, _add::LinearAlgebra.MulAddMul)
     @ ~/Documents/julia/usr/share/julia/stdlib/v1.10/LinearAlgebra/src/matmul.jl:780
 [4] generic_matmatmul!(C::AbstractVecOrMat, tA, tB, A::AbstractVecOrMat, B::AbstractVecOrMat, _add::LinearAlgebra.MulAddMul)
     @ ~/Documents/julia/usr/share/julia/stdlib/v1.10/LinearAlgebra/src/matmul.jl:798

That should be really helpful in terms of package load times.

[dkarrasch]

@nanosoldier runtests()

@nanosoldier renbenchmarks("linalg", vs=":master")

[dkarrasch]

@nanosoldier runbenchmarks("linalg", vs=":master")

[nanosoldier]

Your benchmark job has completed - no performance regressions were detected. A full report can be found here.

[nanosoldier]

Your package evaluation job has completed - possible new issues were detected.
A full report can be found here.

[maleadt]

Many tests were skipped because of PkgEval issue; let's try again:

@nanosoldier runtests()

[dkarrasch]

Aha, okay, out of the tested packages that failed I couldn't find a single related one. But some strange errors like undefined variables. Is that related to the issue you mentioned?

[nanosoldier]

Your package evaluation job has completed - possible new issues were detected.
A full report can be found here.

[dkarrasch]

There seem to be no related issues.

[maleadt]

Is that related to the issue you mentioned?

No, just the skipped packages with Disk quota exceeded.

[dkarrasch]

I wonder if this will help packages already, just because the LinearAlgebra dispatch "competitors" are gone?

I applied the same dispatch move to matvec multiplication, so to be safe, let's run pkgeval once more, and then merge. BTW, one can do something similar to HermOrSym mul! methods, but I'll do it in a separate PR. It's not so clear to me if that is worth it, but we'll see.

@nanosoldier runtests()

[nanosoldier]

Your package evaluation job has completed - possible new issues were detected.
A full report can be found here.