@turbo on loop with sin functions gives error

[Chiil]

The following code crashes:

using BenchmarkTools
using LoopVectorization

itot = 384 
dx = 1. / itot
x = dx*collect(0:itot-1); y = dx*collect(0:itot-1); z = dx*collect(0:itot-1)
u = zeros(itot+8, itot+8, itot+8)

function assign!(u, x, y, z)
    @turbo for k in 1:length(z)
        for j in 1:length(y)
            for i in 1:length(x)
                u[i, j, k] = sin(x[i]) + sin(y[j]) + sin(z[k])
            end 
        end 
    end 
end

uv = @view u[5:5+itot-1, 5:5+itot-1, 5:5+itot-1]
@btime assign!(uv, x, y, z)

with errors:

  [1] fmap
    @ ~/.julia/packages/VectorizationBase/PFhof/src/vecunroll/fmap.jl:3 [inlined]
  [2] vtrunc
    @ ~/.julia/packages/VectorizationBase/PFhof/src/vecunroll/fmap.jl:54 [inlined]
  [3] trunc
    @ ~/.julia/packages/VectorizationBase/PFhof/src/base_defs.jl:47 [inlined]
  [4] sin_fast
    @ ~/.julia/packages/SLEEFPirates/F4zF8/src/trig.jl:200 [inlined]
  [5] macro expansion
    @ ~/.julia/dev/LoopVectorization/src/reconstruct_loopset.jl:706 [inlined]
  [6] _turbo_!
    @ ~/.julia/dev/LoopVectorization/src/reconstruct_loopset.jl:706 [inlined]
  [7] assign!(u::SubArray{Float64, 3, Array{Float64, 3}, Tuple{UnitRange{Int64}, UnitRange{Int64}, UnitRange{Int64}}, false}, x::Vector{Float64}, y::Vector{Float64}, z::Vector{Float64})
    @ Main ~/devel/stencilbuilder/julia/oink.jl:16
  [8] var"##core#263"()
    @ Main ~/.julia/packages/BenchmarkTools/tGTCy/src/execution.jl:479
  [9] var"##sample#264"(__params::BenchmarkTools.Parameters)
    @ Main ~/.julia/packages/BenchmarkTools/tGTCy/src/execution.jl:485
 [10] _run(b::BenchmarkTools.Benchmark, p::BenchmarkTools.Parameters; verbose::Bool, pad::String, kwargs::Base.Iterators.Pairs{Symbol, Integer, NTuple{4, Symbol}, NamedTuple{(:samples, :evals, :gctrial, :gcsample), Tuple{Int64, Int64, Bool, Bool}}})
    @ BenchmarkTools ~/.julia/packages/BenchmarkTools/tGTCy/src/execution.jl:98
 [11] #invokelatest#2
    @ ./essentials.jl:710 [inlined]
 [12] #run_result#45
    @ ~/.julia/packages/BenchmarkTools/tGTCy/src/execution.jl:33 [inlined]
 [13] run(b::BenchmarkTools.Benchmark, p::BenchmarkTools.Parameters; progressid::Nothing, nleaves::Float64, ndone::Float64, kwargs::Base.Iterators.Pairs{Symbol, Integer, NTuple{5, Symbol}, NamedTuple{(:verbose, :samples, :evals, :gctrial, :gcsample), Tuple{Bool, Int64, Int64, Bool, Bool}}})
    @ BenchmarkTools ~/.julia/packages/BenchmarkTools/tGTCy/src/execution.jl:116
 [14] #warmup#54
    @ ~/.julia/packages/BenchmarkTools/tGTCy/src/execution.jl:168 [inlined]
 [15] warmup(item::BenchmarkTools.Benchmark)
    @ BenchmarkTools ~/.julia/packages/BenchmarkTools/tGTCy/src/execution.jl:168
 [16] top-level scope
    @ ~/.julia/packages/BenchmarkTools/tGTCy/src/execution.jl:565
    ```

[mcabbott]

FWIW, I get a different error from the above code (which is from https://discourse.julialang.org/t/fast-performance-of-array-comprehension-without-allocations/65352/5.) M1 mac, either 1.8- native or 1.7- rosetta, either tagged LV or master:

julia> @btime assign!($uv, $x, $y, $z);
ERROR: MethodError: no method matching to_vecunrollscalar(::VectorizationBase.VecUnroll{1, 2, Float64, VectorizationBase.Vec{2, Float64}}, ::Static.StaticInt{3})
Closest candidates are:
  to_vecunrollscalar(::VectorizationBase.Vec{W, T}, ::Static.StaticInt{N}) where {N, W, T} at /Users/me/.julia/packages/SLEEFPirates/XGrnS/src/SLEEFPirates.jl:131
Stacktrace:
  [1] sin_fast
    @ ~/.julia/packages/SLEEFPirates/XGrnS/src/SLEEFPirates.jl:147 [inlined]
  [2] macro expansion
    @ ./none:0 [inlined]
  [3] _turbo_!
    @ ./none:0 [inlined]
  [4] assign!(u::SubArray{Float64, 3, Array{Float64, 3}, Tuple{UnitRange{Int64}, UnitRange{Int64}, UnitRange{Int64}}, false}, x::Vector{Float64}, y::Vector{Float64}, z::Vector{Float64})
    @ Main ./REPL[5]:2
  [5] var"##core#318"(uv#314::SubArray{Float64, 3, Array{Float64, 3}, Tuple{UnitRange{Int64}, UnitRange{Int64}, UnitRange{Int64}}, false}, x#315::Vector{Float64}, y#316::Vector{Float64}, z#317::Vector{Float64})
    @ Main ~/.julia/packages/BenchmarkTools/Rqvks/src/execution.jl:479
  [6] var"##sample#319"(__params::BenchmarkTools.Parameters)
    @ Main ~/.julia/packages/BenchmarkTools/Rqvks/src/execution.jl:485
  [7] _run(b::BenchmarkTools.Benchmark, p::BenchmarkTools.Parameters; verbose::Bool, pad::String, kwargs::Base.Pairs{Symbol, Integer, NTuple{4, Symbol}, NamedTuple{(:samples, :evals, :gctrial, :gcsample), Tuple{Int64, Int64, Bool, Bool}}})

[chriselrod]

I did add that example to the tests, but of course CI doesn't have the M1.
I fixed it on SLEEFPirates master. I'll tag a new release.

julia> @benchmark assign!($uv, $x, $y, $z)
BenchmarkTools.Trial: 146 samples with 1 evaluation.
 Range (min … max):  33.873 ms …  34.672 ms  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     34.279 ms               ┊ GC (median):    0.00%
 Time  (mean ± σ):   34.281 ms ± 103.310 μs  ┊ GC (mean ± σ):  0.00% ± 0.00%

                             ▄▆████▅▃
  ▅▁▁▅▁▇▅▁▅▁▁▁▁▁▁▅▁▁▁▁▁▁▁▁▁▅▅█████████▇▅▅▁▅▁▁▅▁▅▁▁▇▁▁▁▁▅▁▁▁▅▁▅ ▅
  33.9 ms       Histogram: log(frequency) by time      34.7 ms <

 Memory estimate: 0 bytes, allocs estimate: 0.

julia> versioninfo()
Julia Version 1.8.0-DEV.270
Commit 5f4c719d67* (2021-07-26 22:27 UTC)
Platform Info:
  OS: macOS (arm64-apple-darwin20.5.0)
  CPU: Apple M1