Weird transpose timing

[PetrKryslUCSD]

Please visit https://discourse.julialang.org/t/what-is-the-fastest-way-of-realizing-transpose-of-a-csr-or-csc-matrix/136241/27?u=petrkryslucsd

The timing of the transpose is suspect. Is that real? Notice the huge allocation. And, of course, the time is way off.

pkrysl@samadira sparse_transpose_tests.jl % julia --project t.jl
[ Info: Matrix M = 20133 by N = 20133, sparsity = 0.0004966969651815428, nnz = 201916
(typeof(Ac), typeof(Ar)) = (SparseMatrixCSC{Float64, Int64}, SparseMatrixCSR{1, Float64, Int64})
(typeof(Gc), typeof(Gr)) = (GBMatrix{Float64, Float64}, GBMatrix{Float64, Float64})
norm(gbtranspose(Gc) - copy(transpose(Ac))) = 0.0
norm(gbtranspose(Gr) - copy(transpose(Ar))) = 0.0
[ Info: Benchmarking copy+transpose CSC
  624.167 μs (12 allocations: 3.24 MiB)
[ Info: Benchmarking copy+transpose CSR
  1.900 s (3 allocations: 3.02 GiB)      # <---------------- !?
[ Info: Benchmarking gbtranspose CSC
  758.167 μs (23 allocations: 3.70 MiB)
[ Info: Benchmarking gbtranspose CSR
  1.072 s (14 allocations: 3.02 GiB)
[ Info: Benchmarking csr_transpose
  684.417 μs (15 allocations: 3.54 MiB)
norm(csr_transpose(C) - copy(transpose(Ac))) = 0.0
[ Info: Benchmarking csr_transpose_2
  637.125 μs (9 allocations: 3.23 MiB)
norm(csr_transpose_2(C) - copy(transpose(Ac))) = 0.0

I believe that copy(Ar') results in a dense matrix here:

Ar = copy(sparsecsr(Ac))
copy(Ar')