Bugged Sparse Matrix-Dense matrix multiplication, where dense matrix is transposed

[irhum]

Describe the bug
Given A and B are two matrices, where A is of type CuSparseMatrixCSC and B is a regular CuArray (for simplicity assume both are square)

A * B works
transpose(A) * B works
A * transpose(B) does not work
transpose(A) * transpose(B) does not work

The implementations of mul! are there on line 20 and line 22 of CuArrays/src/sparse/interfaces.jl, but instead of them being called properly, both of the failing multiplications throw a scalar getindex is disallowed, suggesting it may be defaulting to a generic method?

To Reproduce
The Minimal Working Example (MWE) for this bug:

using CuArrays
using SparseArrays
using LinearAlgebra

CuArrays.allowscalar(false)

randsparse_cpu = SparseMatrixCSC{Float32, Int32}(sprand(3000, 3000, 0.0008))
sparsecscmatrix = CUSPARSE.CuSparseMatrixCSC(randsparse)

densematrix = cu(rand(Float32, 3000, 3000))

sparsecscmatrix * densematrix
transpose(sparsecscmatrix) * densematrix
sparsecscmatrix * transpose(densematrix) # throws scalar getindex is disallowed, should throw different error
transpose(sparsecscmatrix) * transpose(densematrix) # throws scalar getindex is disallowed, should throw no error

Expected behavior
three of the multiplications should complete correctly, while sparsecscmatrix * transpose(densematrix) should throw a CUSPARSE error (CUSPARSE does not support 'N', 'T' multiplications for CSC matrices)

Build log

# post the output of Pkg.build()
# make sure the error still reproduces after that.
   Building TimeZones → `~/.julia/packages/TimeZones/OjMoF/deps/build.log`
   Building HDF5 ─────→ `~/.julia/packages/HDF5/pAi1D/deps/build.log`
   Building CodecZlib → `~/.julia/packages/CodecZlib/5t9zO/deps/build.log`
   Building NNlib ────→ `~/.julia/packages/NNlib/FAI3o/deps/build.log`
   Building Conda ────→ `~/.julia/packages/Conda/3rPhK/deps/build.log`
   Building PyCall ───→ `~/.julia/packages/PyCall/zqDXB/deps/build.log`
   Building WebIO ────→ `~/.julia/packages/WebIO/nTMDV/deps/build.log`

error still reproduces

Environment details (please complete this section)
Details on Julia:

# please post the output of:
versioninfo()
Julia Version 1.4.1
Commit 381693d3df* (2020-04-14 17:20 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Core(TM) i5-4570 CPU @ 3.20GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-8.0.1 (ORCJIT, haswell)
Environment:
  JULIA_NUM_THREADS = 4

Julia packages:

• CuArrays.jl:
• CUDAnative.jl:
• SparseArrays
• LinearAlgebra

CUDA: toolkit and driver version
10.2, 440.64

Additional context
Add any other context about the problem here.

[irhum]

with some further checking with the @which macro,

@which mul!(similar(densematrix, Float32, (size(sparsecscmatrix,1), size(densematrix,2))), sparsecscmatrix, densematrix)
# mul!(C::CuArray{T,2,P} where P, #A::Union{CuArrays.CUSPARSE.CuSparseMatrixBSR{T}, CuArrays.CUSPARSE.CuSparseMatrixCSC{T}, CuArrays.CUSPARSE.CuSparseMatrixCSR{T}, CuArrays.CUSPARSE.CuSparseMatrixHYB{T}}, B::CuArray{T,2,P} where P) where T at CuArrays/l0gXB/src/sparse/interfaces.jl:19

densematrix_T = transpose(densematrix)
@which mul!(similar(densematrix_T, Float32, (size(sparsecscmatrix,1), size(densematrix_T,2))), sparsecscmatrix, densematrix_T)
# mul!(C, A, B) at stdlib/v1.4/LinearAlgebra/src/matmul.jl:208

not quite sure why the latter isn't picking the appropriate method and falling back to the generic

[irhum]

Solved by #180, closing issue