decrement should always return a vector

[SobhanMP]

we only use it when it returns a vector, so i think it's good idea for AbstractMatrixCSCs that don't use Vectors (like Fixed)

[codecov-commenter]

Codecov Report

Merging #241 (d239e11) into main (43b4d01) will decrease coverage by 0.06%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main     #241      +/-   ##
==========================================
- Coverage   91.68%   91.62%   -0.07%     
==========================================
  Files          12       12              
  Lines        7302     7305       +3     
==========================================
- Hits         6695     6693       -2     
- Misses        607      612       +5     

Impacted Files	Coverage Δ
src/readonly.jl	72.22% <ø> (-1.47%)	⬇️
src/SparseArrays.jl	85.00% <100.00%> (-15.00%)	⬇️
src/solvers/umfpack.jl	87.70% <0.00%> (-0.41%)	⬇️

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[SobhanMP]

@Wimmerer do you think this is a good idea?

[rayegun]

I'm not super fond of it since most sparse matrices are not hypersparse.

But I think it is correct.

[SobhanMP]

this is an internal only used by CHOLMOD and lu! to convert to one based indexing rigth?

[rayegun]

Oh wrong issue, let me look at this

[SobhanMP]

The issue that for types that don't have a vector colptr, lu/qr wont work, this is one of the problematic parts

[rayegun]

A very quick and dirty benchmark shows that is ~1.3x slower. Do you see that as well?

[SobhanMP]

julia> dec_alt(x) = SparseArrays.decrement!(copy(x))
dec_alt (generic function with 1 method)

julia> dec_alt1(x) = SparseArrays.decrement!(Array(x))
dec_alt1 (generic function with 1 method)

julia> dec_alt2(x) = let y = Array{eltype(x),ndims(x)}(undef, size(x)...)
           y .= x .- oneunit(eltype(x))
           y
       end
dec_alt2 (generic function with 1 method)

julia> dec_alt3(x) = let y = Vector{eltype(x)}(undef, size(x)...)
           y .= x .- oneunit(eltype(x))
           y
       end
dec_alt3 (generic function with 1 method)

julia> dec_alt4(x) = let y = Vector(x)
           y .= y .- oneunit(eltype(x))
           y
       end
dec_alt4 (generic function with 1 method)

julia> dec_alt5(x) = let y = Vector(x)
           for i in eachindex(y, x)
               y[i] -= oneunit(eltype(x))
           end
           y
       end
dec_alt5 (generic function with 1 method)

julia> dec_alt6(x) = let y = Vector{eltype(x)}(undef, length(x))
           for i in eachindex(y, x)
               y[i] = x[i] - oneunit(eltype(x))
           end
           y
       end
dec_alt6 (generic function with 1 method)

julia> dec_alt7(x) = let y = Array{eltype(x),ndims(x)}(undef, size(x)...)
           for i in eachindex(y, x)
               y[i] = x[i] - oneunit(eltype(x))
           end
           y
       end
dec_alt7 (generic function with 1 method)

julia> dec_alt41(x) = let y = Array(x)
           y .= y .- oneunit(eltype(x))
           y
       end
dec_alt41 (generic function with 1 method)

julia> aa = [dec_alt,dec_alt1,dec_alt2,dec_alt3,dec_alt4,dec_alt41,dec_alt5,dec_alt6,dec_alt7]
9-element Vector{Function}:
 dec_alt (generic function with 1 method)
 dec_alt1 (generic function with 1 method)
 dec_alt2 (generic function with 1 method)
 dec_alt3 (generic function with 1 method)
 dec_alt4 (generic function with 1 method)
 dec_alt41 (generic function with 1 method)
 dec_alt5 (generic function with 1 method)
 dec_alt6 (generic function with 1 method)
 dec_alt7 (generic function with 1 method)

julia> for i in aa, j in aa
           @assert i(a) == j(a)
       end

julia> @benchmark dec_alt($a)
BenchmarkTools.Trial: 10000 samples with 6 evaluations.
 Range (min … max):  5.532 μs … 343.084 μs  ┊ GC (min … max):  0.00% … 92.03%
 Time  (median):     6.121 μs               ┊ GC (median):     0.00%
 Time  (mean ± σ):   9.884 μs ±  14.471 μs  ┊ GC (mean ± σ):  11.99% ±  8.91%

  █▅▄▃▃▃▂▂▁▁▁▁  ▁▄▂                                           ▁
  ██████████████████▇▇▇▇▆▆▆▆▆▆▆▄▅▃▅▅▅▄▄▅▃▅▁▁▁▃▁▃▁▁▁▁▁▃▁▁▁▁▃▁▄ █
  5.53 μs      Histogram: log(frequency) by time      63.1 μs <

 Memory estimate: 78.17 KiB, allocs estimate: 2.

julia> @benchmark dec_alt1($a)
BenchmarkTools.Trial: 10000 samples with 6 evaluations.
 Range (min … max):  5.533 μs … 239.264 μs  ┊ GC (min … max): 0.00% … 92.10%
 Time  (median):     5.878 μs               ┊ GC (median):    0.00%
 Time  (mean ± σ):   7.555 μs ±   9.059 μs  ┊ GC (mean ± σ):  9.94% ±  8.97%

  █▄▃▂▂▂▁▁ ▁▂▁▁                                               ▁
  ██████████████▇▆▃▄▅▇▆▇▃▁▁▃▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▃▄ █
  5.53 μs      Histogram: log(frequency) by time      52.8 μs <

 Memory estimate: 78.17 KiB, allocs estimate: 2.

julia> @benchmark dec_alt2($a)
BenchmarkTools.Trial: 10000 samples with 7 evaluations.
 Range (min … max):   4.380 μs … 442.384 μs  ┊ GC (min … max):  0.00% … 91.19%
 Time  (median):     13.540 μs               ┊ GC (median):     0.00%
 Time  (mean ± σ):   16.176 μs ±  24.474 μs  ┊ GC (mean ± σ):  13.95% ±  9.79%

  ▅█▁    ▁▁▁                                                    
  ███▆▅▆▇███▇▅▄▃▃▂▂▂▂▂▂▂▂▂▁▂▁▁▁▁▁▁▁▁▂▁▁▁▂▂▁▂▂▁▂▂▁▁▁▁▂▁▂▁▂▂▁▂▂▂ ▃
  4.38 μs         Histogram: frequency by time          106 μs <

 Memory estimate: 78.17 KiB, allocs estimate: 2.

julia> @benchmark dec_alt3($a)
BenchmarkTools.Trial: 10000 samples with 6 evaluations.
 Range (min … max):  3.812 μs … 272.141 μs  ┊ GC (min … max):  0.00% … 91.12%
 Time  (median):     5.442 μs               ┊ GC (median):     0.00%
 Time  (mean ± σ):   7.400 μs ±   9.344 μs  ┊ GC (mean ± σ):  11.08% ±  9.15%

  ▃█▅▄▄▄▃▃▂▂▁▁▁▁                                              ▁
  ███████████████▆▆▅▄▅▃▄▁▃▃▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▅ █
  3.81 μs      Histogram: log(frequency) by time      57.3 μs <

 Memory estimate: 78.17 KiB, allocs estimate: 2.

julia> @benchmark dec_alt4($a)
BenchmarkTools.Trial: 10000 samples with 8 evaluations.
 Range (min … max):  2.998 μs … 390.348 μs  ┊ GC (min … max):  0.00% … 86.35%
 Time  (median):     3.507 μs               ┊ GC (median):     0.00%
 Time  (mean ± σ):   6.478 μs ±  13.019 μs  ┊ GC (mean ± σ):  17.84% ± 10.62%

  █▅▅▄▄▃▃▂▃▃▂▂▂▁                                              ▂
  ████████████████▇▇▇▇▇▇▆▅▅▄▁▅▁▃▄▃▁▃▃▃▁▁▁▁▃▃▃▃▃▄▄▄▃▃▃▄▅▄▅▄▁▃▅ █
  3 μs         Histogram: log(frequency) by time      52.2 μs <

 Memory estimate: 78.17 KiB, allocs estimate: 2.

julia> @benchmark dec_alt5($a)
BenchmarkTools.Trial: 10000 samples with 6 evaluations.
 Range (min … max):   5.532 μs … 429.331 μs  ┊ GC (min … max):  0.00% … 94.89%
 Time  (median):      5.869 μs               ┊ GC (median):     0.00%
 Time  (mean ± σ):   10.103 μs ±  16.150 μs  ┊ GC (mean ± σ):  12.99% ±  8.92%

  █▅▃▂▁▂▁        ▄▄▂                                           ▁
  ███████▇▇▇▇▆▅▃▇███▇▇▆▆▅▅▅▅▆▅▄▅▄▄▄▄▄▄▁▅▄▄▃▃▃▃▄▁▁▁▁▃▃▁▁▁▃▁▁▁▃▄ █
  5.53 μs       Histogram: log(frequency) by time      63.5 μs <

 Memory estimate: 78.17 KiB, allocs estimate: 2.

julia> @benchmark dec_alt6($a)
BenchmarkTools.Trial: 10000 samples with 6 evaluations.
 Range (min … max):  4.325 μs … 238.120 μs  ┊ GC (min … max):  0.00% … 92.51%
 Time  (median):     5.529 μs               ┊ GC (median):     0.00%
 Time  (mean ± σ):   7.491 μs ±   9.086 μs  ┊ GC (mean ± σ):  10.81% ±  9.15%

  ▇█▄▅▄▃▃▂▁▁▁▁▁▁▁▁                                            ▂
  ████████████████▇▆▅▃▃▁▃▁▃▃▃▁▁▁▁▁▁▁▁▁▄▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▄▄ █
  4.32 μs      Histogram: log(frequency) by time      61.9 μs <

 Memory estimate: 78.17 KiB, allocs estimate: 2.

julia> @benchmark dec_alt7($a)
BenchmarkTools.Trial: 10000 samples with 6 evaluations.
 Range (min … max):  4.203 μs … 169.380 μs  ┊ GC (min … max):  0.00% … 89.03%
 Time  (median):     5.522 μs               ┊ GC (median):     0.00%
 Time  (mean ± σ):   7.203 μs ±   8.376 μs  ┊ GC (mean ± σ):  10.83% ±  9.11%

  ▅█▅▆▄▃▂▁▁▁                                                  ▁
  ████████████▇▅▅▇▆▁▄▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▅ █
  4.2 μs       Histogram: log(frequency) by time      59.9 μs <

 Memory estimate: 78.17 KiB, allocs estimate: 2.

julia> @benchmark dec_alt41($a)
BenchmarkTools.Trial: 10000 samples with 8 evaluations.
 Range (min … max):  2.998 μs … 358.718 μs  ┊ GC (min … max):  0.00% … 93.88%
 Time  (median):     6.672 μs               ┊ GC (median):     0.00%
 Time  (mean ± σ):   9.759 μs ±  19.404 μs  ┊ GC (mean ± σ):  18.92% ± 10.61%

  █                                                            
  █▄▄▄▄▅▅▄▃▃▃▂▂▂▂▂▂▂▂▂▂▂▂▂▁▂▂▂▂▂▂▁▂▂▂▁▁▂▂▂▁▁▂▂▂▁▁▁▁▂▁▁▁▁▂▁▁▁▂ ▂
  3 μs            Histogram: frequency by time        78.9 μs <

 Memory estimate: 78.17 KiB, allocs estimate: 2.

[SobhanMP]

i think version 4 is good, maybe there is no need to bother with weird indexes.

[SobhanMP]

@Wimmerer ping