`using TriangularSolve` fails on Julia 1.8.0-beta1

[cscherrer]

On this system

julia> versioninfo()
Julia Version 1.8.0-beta1
Commit 7b711ce699 (2022-02-23 15:09 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: 32 × AMD Ryzen Threadripper 2950X 16-Core Processor
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-13.0.1 (ORCJIT, znver1)
  Threads: 1 on 32 virtual cores
Environment:
  JULIA_EDITOR = code
  JULIA_NUM_THREADS = 

I get

julia> using TriangularSolve
[ Info: Precompiling TriangularSolve [d5829a12-d9aa-46ab-831f-fb7c9ab06edf]
ERROR: LoadError: DivideError: integer division error
Stacktrace:
  [1] vdiv
    @ ~/.julia/packages/VectorizationBase/yqq6V/src/llvm_intrin/binary_ops.jl:65 [inlined]
  [2] vdiv
    @ ~/.julia/packages/VectorizationBase/yqq6V/src/static.jl:56 [inlined]
  [3] vcld
    @ ~/.julia/packages/VectorizationBase/yqq6V/src/llvm_intrin/binary_ops.jl:79 [inlined]
  [4] rdiv_block_MandN!(spc::LayoutPointers.StridedPointer{Float64, 2, 1, 0, (1, 2), Tuple{Static.StaticInt{8}, Int64}, Tuple{Static.StaticInt{0}, Static.StaticInt{0}}}, spa::LayoutPointers.StridedPointer{Float64, 2, 1, 0, (1, 2), Tuple{Static.StaticInt{8}, Int64}, Tuple{Static.StaticInt{0}, Static.StaticInt{0}}}, spu::LayoutPointers.StridedPointer{Float64, 2, 1, 0, (1, 2), Tuple{Static.StaticInt{8}, Int64}, Tuple{Static.StaticInt{0}, Static.StaticInt{0}}}, M::Int64, N::Int64, #unused#::Val{false}, #unused#::Static.StaticInt{1})
    @ TriangularSolve ~/git/TriangularSolve.jl/src/TriangularSolve.jl:411
  [5] div_dispatch!(C::Matrix{Float64}, A::Matrix{Float64}, U::Matrix{Float64}, #unused#::Val{false}, #unused#::Val{false})
    @ TriangularSolve ~/git/TriangularSolve.jl/src/TriangularSolve.jl:312
  [6] rdiv!
    @ ~/git/TriangularSolve.jl/src/TriangularSolve.jl:323 [inlined]
  [7] top-level scope
    @ ~/git/TriangularSolve.jl/src/TriangularSolve.jl:542
  [8] include
    @ ./Base.jl:422 [inlined]
  [9] include_package_for_output(pkg::Base.PkgId, input::String, depot_path::Vector{String}, dl_load_path::Vector{String}, load_path::Vector{String}, concrete_deps::Vector{Pair{Base.PkgId, UInt64}}, source::Nothing)
    @ Base ./loading.jl:1399
 [10] top-level scope
    @ stdin:1

The problem comes in this section

@static if VERSION >= v"1.8.0-beta1"
  let
    while true
      A = rand(1, 1)
      B = rand(1, 1)
      res = similar(A)
      rdiv!(res, A, UpperTriangular(B))
      rdiv!(res, A, UnitUpperTriangular(B))
      rdiv!(res, A, UpperTriangular(B), Val(false))   # <----- THIS LINE
      rdiv!(res, A, UnitUpperTriangular(B), Val(false))

      __init__()
      ldiv!(res, LowerTriangular(B), A)
      ldiv!(res, UnitLowerTriangular(B), A)
      ldiv!(res, LowerTriangular(B), A, Val(false))
      ldiv!(res, UnitLowerTriangular(B), A, Val(false))
      break
    end
  end
end

I've tried commenting this out, loading the package, and then running the section manually. In this case I get a segfault.

[cscherrer]

(TriangularSolve) pkg> st
Project TriangularSolve v0.1.11
Status `~/git/TriangularSolve.jl/Project.toml`
  [fb6a15b2] CloseOpenIntervals
  [615f187c] IfElse
  [10f19ff3] LayoutPointers
  [bdcacae8] LoopVectorization
  [f517fe37] Polyester
  [aedffcd0] Static
  [3d5dd08c] VectorizationBase
  [37e2e46d] LinearAlgebra

[chriselrod]

Mind debugging it?
I.e., deving TriangularSolve, and figuring out which denominator is 0, and why?

TriangularSolve.jl/src/TriangularSolve.jl

Line 411 in 2ac1058

B_m = VectorizationBase.vcld(M, VectorizationBase.vcld(M, B)*WUF)*WUF

unroll_factor(W) shouldn't be <1, as there is a ifelse(lt(x,1), 1, x) that should lower bound it at 1.
W can't be 0, otherwise you'd have gotten the error a few lines earlier.
So it sounds like M or B are0, in which case, what are the inputs that make them 0?

[cscherrer]

I'm getting

(M, B, WUF) = (1, static(0), static(12))

so I guess the problem is in block_size?

[cscherrer]

Yep:

julia> function block_size(::Val{T}) where {T}
         elements_l2 = (VectorizationBase.cache_size(StaticInt(2))*StaticInt(19)) ÷ (VectorizationBase.static_sizeof(T)*StaticInt(60))
         Static.floortostaticint(sqrt(elements_l2))
       end
block_size

julia> block_size(Val(Float64))
static(0)

[cscherrer]

Ahh, I guess the problem is in VectorizationBase? This is pretty weird

julia> VectorizationBase.cache_size(StaticInt(2))
static(0)

[cscherrer]

This is turn is because of CpuId.cachesize:

julia> CpuId.cachesize.(1:4)
4-element Vector{Int64}:
 0
 0
 0
 0

[cscherrer]

Note that cachesize() does work

julia> cachesize()
(32768, 524288, 8388608)

[chriselrod]

I decided to change a couple things in CPUSummary, and also implement the workaround, so this should work now.
Mind updating and testing it?
Requires CPUSummary 0.1.20.

[cscherrer]

Looks good, tests are now passing! Thanks for the quick fix