Switch CI to 1.11.

[maleadt]

No description provided.

[maleadt]

Looks like 1.11 is significantly slower; I bumped the timeouts.

[maleadt]

The hangs are legitimate. Subpackage tests only take up to 10 minutes on 1.10 (so the timeouts should be adjusted accordingly).

[maleadt]

Depends on new TensorOperations.jl release: Jutho/TensorOperations.jl#193 (comment)

[maleadt]

Looks like on 1.11 the cuTENSOR tests really do hang -- lots of compile time being spent on signatures like:

Base.Broadcast.materialize
Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{4}, Nothing, typeof(Base.:(+)), Tuple{Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{4}, Nothing, typeof(Base.:(*)), Tuple{Float16, Array{Float32, 4}}}, Array{Float16, 4}}}(style=Base.Broadcast.DefaultArrayStyle{4}(), f=typeof(Base.:(+))(), args=(Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{4}, Nothing, typeof(Base.:(*)), Tuple{Float16, Array{Float32, 4}}}(style=Base.Broadcast.DefaultArrayStyle{4}(), f=typeof(Base.:(*))(), args=(Float16(0x33f4), Array{Float32, 4}(dims=(3, 10, 12, 14), mem=Memory{Float32}(5040, 0x6425044cc8c0)[0.386422f, ..., 0.933269f, 0.919605f, 0.140272f, 0.979564f, 0.964575f, 0.393861f, 0.0486779f, 0.441696f, 0.375366f, 0.497105f])), axes=nothing), Array{Float16, 4}(dims=(3, 10, 12, 14), mem=Memory{Float16}(5040, 0x6424fe479940)[Float16(0x3bcd), ..., Float16(0x3967)])), axes=nothing)

Another one I observed was:

materialize(Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{2}, Nothing, typeof(Base.:(+)), Tuple{Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{2}, Nothing, typeof(Base.:(*)), Tuple{Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{2}, Nothing, typeof(Base.:(*)), Tuple{Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{2}, Nothing, typeof(Base.:(*)), Tuple{Float16, Array{Float32, 2}}}, Float16}}, Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{2}, Nothing, typeof(Base.conj), Tuple{Array{Float32, 2}}}}}, Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{2}, Nothing, typeof(Base.:(*)), Tuple{Float16, Array{Float16, 2}}}}}) from materialize(Base.Broadcast.Broadcasted{Style, Axes, F, Args} where Args<:Tuple where F where Axes where Style<:Union{Nothing, Base.Broadcast.BroadcastStyle})

This reproduces on the 1.11.2 backports branch too.

[maleadt]

Not a hang, 1.11 is just a lot slower:

1.11.1 --check-bounds=yes

elementwise binary |  352    352  3m58.3s
elementwise trinary |  252    252  6m34.7s
permutations  |   64     64  19.7s
reductions    |  200    200  1m40.8s

1.10 --check-bounds=yes

elementwise binary |  352    352  1m08.6s
elementwise trinary |  252    252  2m02.9s
permutations  |   64     64  6.8s
reductions    |  200    200  20.5s