Generalize `strides` for `ReinterpretArray` and `ReshapedArray` by N5N3 · Pull Request #44027 · JuliaLang/julia

N5N3 · 2022-02-03T07:15:36Z

strides for ReinterpretArray was genelized in #37414, but it works only for contiguous parent (So the check there is also not correct.)
This PR tries to extend it by using parent's strides rather than self's size during calculation.

Similar generalization is also applied to ReshapedArray.

Test added.

timholy · 2022-02-03T10:20:48Z

I haven't reviewed in detail, but when asking about strides and subarrays it's usually best to be sure to create some parents with dimensions non-commensurate with the step. E.g.,

A = reshape(1:21*4, 21, 4)
V = @view A[begin:4:end,:]

and check that all the conditions for strides (even spacing along each axis) are met for the values of V. To me it looks like all your tests use a parent size that's evenly-divisible by the step.

N5N3 · 2022-02-03T12:06:07Z

I add more test via testing a[i] === Base.unsafe_load(pointer(A, i)), as a[i] should be correct while pointer is strides-based.
I'm not sure is it OK to you @timholy. But with this there's no need to hardcode the experted output of strides.

Update abstractarray.jl

timholy

This mostly looks fantastic, nice work!

test/abstractarray.jl

base/reshapedarray.jl

we are in `Base`, so no need to `Base.`

test added

timholy · 2022-02-06T12:55:24Z

Thanks @N5N3! Really nice work.

mkitti · 2022-02-09T07:45:14Z

There might have been a regression here. HDF5.jl tests are showing this now fails: stride(fill(43, ()), 1)

plain: Error During Test at /home/runner/work/HDF5.jl/HDF5.jl/test/plain.jl:25
  Got exception outside of a @test
  ArgumentError: reducing with add_sum over an empty collection of element type Union{} is not allowed.
  You may be able to prevent this error by supplying an `init` value to the reducer.
  Stacktrace:
    [1] _empty_reduce_error(f::Any, T::Type)
      @ Base ./reduce.jl:307
    [2] reduce_empty(#unused#::typeof(Base.add_sum), #unused#::Core.TypeofBottom)
      @ Base ./reduce.jl:338
    [3] reduce_empty(op::Base.BottomRF{typeof(Base.add_sum)}, #unused#::Type{Union{}})
      @ Base ./reduce.jl:347
    [4] reduce_empty_iter
      @ ./reduce.jl:371 [inlined]
    [5] reduce_empty_iter
      @ ./reduce.jl:370 [inlined]
    [6] foldl_impl(op::Base.BottomRF{typeof(Base.add_sum)}, nt::Base._InitialValue, itr::Tuple{})
      @ Base ./reduce.jl:49
    [7] mapfoldl_impl(f::typeof(identity), op::typeof(Base.add_sum), nt::Base._InitialValue, itr::Tuple{})
      @ Base ./reduce.jl:44
    [8] mapfoldl(f::Function, op::Function, itr::Tuple{}; init::Base._InitialValue)
      @ Base ./reduce.jl:[162](https://github.com/JuliaIO/HDF5.jl/runs/5121315625?check_suite_focus=true#step:5:162)
    [9] mapfoldl
      @ ./reduce.jl:162 [inlined]
   [10] #mapreduce#264
      @ ./reduce.jl:294 [inlined]
   [11] mapreduce(f::Function, op::Function, itr::Tuple{})
      @ Base ./reduce.jl:294
   [12] sum(f::Function, a::Tuple{}; kw::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
      @ Base ./reduce.jl:520
   [13] sum(f::Function, a::Tuple{})
      @ Base ./reduce.jl:520
   [14] sum(a::Tuple{}; kw::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
      @ Base ./reduce.jl:549
   [15] sum(a::Tuple{})
      @ Base ./reduce.jl:549
   [16] stride(A::Array{Int32, 0}, k::Int64)
      @ Base ./abstractarray.jl:549
   [17] write_dataset(dataset::HDF5.Dataset, memtype::HDF5.Datatype, buf::Array{Int32, 0}, xfer::HDF5.DatasetTransferProperties) (repeats 2 times)
      @ HDF5 ~/work/HDF5.jl/HDF5.jl/src/HDF5.jl:1529
   [18] write_dataset(parent::HDF5.File, name::String, data::Array{Int32, 0}; pv::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
      @ HDF5 ~/work/HDF5.jl/HDF5.jl/src/HDF5.jl:1201
   [19] write_dataset(parent::HDF5.File, name::String, data::Array{Int32, 0})
      @ HDF5 ~/work/HDF5.jl/HDF5.jl/src/HDF5.jl:1[199](https://github.com/JuliaIO/HDF5.jl/runs/5121315625?check_suite_focus=true#step:5:199)
   [20] write(parent::HDF5.File, name::String, data::Array{Int32, 0}; pv::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
      @ HDF5 ~/work/HDF5.jl/HDF5.jl/src/HDF5.jl:1244
   [21] write(parent::HDF5.File, name::String, data::Array{Int32, 0})
      @ HDF5 ~/work/HDF5.jl/HDF5.jl/src/HDF5.jl:1244
   [22] setindex!(parent::HDF5.File, val::Array{Int32, 0}, path::String; pv::Base.Pairs{Symbol, Integer, Tuple{Symbol, Symbol}, NamedTuple{(:shuffle, :deflate), Tuple{Bool, Int64}}})
      @ HDF5 ~/work/HDF5.jl/HDF5.jl/src/HDF5.jl:649
   [23] macro expansion
      @ ~/work/HDF5.jl/HDF5.jl/test/plain.jl:37 [inlined]
   [24] macro expansion
      @ /opt/hostedtoolcache/julia/nightly/x64/share/julia/stdlib/v1.8/Test/src/Test.jl:1356 [inlined]
   [25] top-level scope
      @ ~/work/HDF5.jl/HDF5.jl/test/plain.jl:28
   [26] include(fname::String)
      @ Base.MainInclude ./client.jl:476
   [27] macro expansion
      @ ~/work/HDF5.jl/HDF5.jl/test/runtests.jl:18 [inlined]
   [28] macro expansion
      @ /opt/hostedtoolcache/julia/nightly/x64/share/julia/stdlib/v1.8/Test/src/Test.jl:1356 [inlined]
   [29] top-level scope
      @ ~/work/HDF5.jl/HDF5.jl/test/runtests.jl:17
   [30] include(fname::String)
      @ Base.MainInclude ./client.jl:476
   [31] top-level scope
      @ none:6
   [32] eval
      @ ./boot.jl:368 [inlined]
   [33] exec_options(opts::Base.JLOptions)
      @ Base ./client.jl:276
   [34] _start()
      @ Base ./client.jl:522
empty and 0-size arrays: Error During Test at /home/runner/work/HDF5.jl/HDF5.jl/test/plain.jl:604
  Got exception outside of a @test
  ArgumentError: reducing with add_sum over an empty collection of element type Union{} is not allowed.
  You may be able to prevent this error by supplying an `init` value to the reducer.
  Stacktrace:
    [1] _empty_reduce_error(f::Any, T::Type)
      @ Base ./reduce.jl:307
    [2] reduce_empty(#unused#::typeof(Base.add_sum), #unused#::Core.TypeofBottom)
      @ Base ./reduce.jl:338
    [3] reduce_empty(op::Base.BottomRF{typeof(Base.add_sum)}, #unused#::Type{Union{}})
      @ Base ./reduce.jl:347
    [4] reduce_empty_iter
      @ ./reduce.jl:371 [inlined]
    [5] reduce_empty_iter
      @ ./reduce.jl:370 [inlined]
    [6] foldl_impl(op::Base.BottomRF{typeof(Base.add_sum)}, nt::Base._InitialValue, itr::Tuple{})
      @ Base ./reduce.jl:49
    [7] mapfoldl_impl(f::typeof(identity), op::typeof(Base.add_sum), nt::Base._InitialValue, itr::Tuple{})
      @ Base ./reduce.jl:44
    [8] mapfoldl(f::Function, op::Function, itr::Tuple{}; init::Base._InitialValue)
      @ Base ./reduce.jl:162
    [9] mapfoldl
      @ ./reduce.jl:162 [inlined]
   [10] #mapreduce#264
      @ ./reduce.jl:294 [inlined]
   [11] mapreduce(f::Function, op::Function, itr::Tuple{})
      @ Base ./reduce.jl:294
   [12] sum(f::Function, a::Tuple{}; kw::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
      @ Base ./reduce.jl:520
   [13] sum(f::Function, a::Tuple{})
      @ Base ./reduce.jl:520
   [14] sum(a::Tuple{}; kw::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
      @ Base ./reduce.jl:549
   [15] sum(a::Tuple{})
      @ Base ./reduce.jl:549
   [16] stride(A::Array{Float64, 0}, k::Int64)
      @ Base ./abstractarray.jl:549
   [17] write_dataset(dataset::HDF5.Dataset, memtype::HDF5.Datatype, buf::Array{Float64, 0}, xfer::HDF5.DatasetTransferProperties) (repeats 2 times)
      @ HDF5 ~/work/HDF5.jl/HDF5.jl/src/HDF5.jl:1529
   [18] write_dataset(parent::HDF5.File, name::String, data::Array{Float64, 0}; pv::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
      @ HDF5 ~/work/HDF5.jl/HDF5.jl/src/HDF5.jl:1[201](https://github.com/JuliaIO/HDF5.jl/runs/5121315625?check_suite_focus=true#step:5:201)
   [19] write_dataset(parent::HDF5.File, name::String, data::Array{Float64, 0})
      @ HDF5 ~/work/HDF5.jl/HDF5.jl/src/HDF5.jl:1199
   [20] write(parent::HDF5.File, name::String, data::Array{Float64, 0}; pv::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
      @ HDF5 ~/work/HDF5.jl/HDF5.jl/src/HDF5.jl:1[244](https://github.com/JuliaIO/HDF5.jl/runs/5121315625?check_suite_focus=true#step:5:244)
   [21] write(parent::HDF5.File, name::String, data::Array{Float64, 0})
      @ HDF5 ~/work/HDF5.jl/HDF5.jl/src/HDF5.jl:1244
   [22] macro expansion
      @ ~/work/HDF5.jl/HDF5.jl/test/plain.jl:610 [inlined]
   [23] macro expansion
      @ /opt/hostedtoolcache/julia/nightly/x64/share/julia/stdlib/v1.8/Test/src/Test.jl:1356 [inlined]
   [24] top-level scope
      @ ~/work/HDF5.jl/HDF5.jl/test/plain.jl:605
   [25] include(fname::String)
      @ Base.MainInclude ./client.jl:476
   [26] macro expansion
      @ ~/work/HDF5.jl/HDF5.jl/test/runtests.jl:18 [inlined]
   [27] macro expansion
      @ /opt/hostedtoolcache/julia/nightly/x64/share/julia/stdlib/v1.8/Test/src/Test.jl:1356 [inlined]
   [28] top-level scope
      @ ~/work/HDF5.jl/HDF5.jl/test/runtests.jl:17
   [29] include(fname::String)
      @ Base.MainInclude ./client.jl:476
   [30] top-level scope
      @ none:6
   [31] eval
      @ ./boot.jl:368 [inlined]
   [32] exec_options(opts::Base.JLOptions)
      @ Base ./client.jl:[276](https://github.com/JuliaIO/HDF5.jl/runs/5121315625?check_suite_focus=true#step:5:276)
   [33] _start()
      @ Base ./client.jl:522

Working on confirming this.

mkitti · 2022-02-09T07:56:12Z

I just built master:

julia> stride( fill(43,()) , 1)
ERROR: ArgumentError: reducing with add_sum over an empty collection of element type Union{} is not allowed.
You may be able to prevent this error by supplying an `init` value to the reducer.
Stacktrace:
  [1] _empty_reduce_error(f::Any, T::Type)
    @ Base ./reduce.jl:307
  [2] reduce_empty(#unused#::typeof(Base.add_sum), #unused#::Core.TypeofBottom)
    @ Base ./reduce.jl:338
  [3] reduce_empty(op::Base.BottomRF{typeof(Base.add_sum)}, #unused#::Type{Union{}})
    @ Base ./reduce.jl:347
  [4] reduce_empty_iter
    @ ./reduce.jl:371 [inlined]
  [5] reduce_empty_iter
    @ ./reduce.jl:370 [inlined]
  [6] foldl_impl(op::Base.BottomRF{typeof(Base.add_sum)}, nt::Base._InitialValue, itr::Tuple{})
    @ Base ./reduce.jl:49
  [7] mapfoldl_impl(f::typeof(identity), op::typeof(Base.add_sum), nt::Base._InitialValue, itr::Tuple{})
    @ Base ./reduce.jl:44
  [8] mapfoldl(f::Function, op::Function, itr::Tuple{}; init::Base._InitialValue)
    @ Base ./reduce.jl:162
  [9] mapfoldl
    @ ./reduce.jl:162 [inlined]
 [10] #mapreduce#264
    @ ./reduce.jl:294 [inlined]
 [11] mapreduce(f::Function, op::Function, itr::Tuple{})
    @ Base ./reduce.jl:294
 [12] sum(f::Function, a::Tuple{}; kw::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ Base ./reduce.jl:520
 [13] sum(f::Function, a::Tuple{})
    @ Base ./reduce.jl:520
 [14] sum(a::Tuple{}; kw::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ Base ./reduce.jl:549
 [15] sum(a::Tuple{})
    @ Base ./reduce.jl:549
 [16] stride(A::Array{Int64, 0}, k::Int64)
    @ Base ./abstractarray.jl:549
 [17] top-level scope
    @ REPL[3]:1

N5N3 · 2022-02-09T08:08:19Z

The specialization for stride(A::Array, i) was removed in this PR. Seems our general fallback dislike 0-dim case.

function stride(A::AbstractArray, k::Integer)
    st = strides(A)
    k ≤ ndims(A) && return st[k]
    return sum(st .* size(A)) # here
end

I suggest to return length(A) instead of sum(st .* size(A)), as the value of stride(A, i) should be arbitrary if i > ndims(A)?

mkitti · 2022-02-09T08:15:32Z

Just add a keyword init = 1 to sum as the error suggests.

julia> function Base.stride(A::AbstractArray, k::Integer)
           st = strides(A)
           k ≤ ndims(A) && return st[k]
           return sum(st .* size(A); init = 1)
       end

N5N3 · 2022-02-09T08:25:52Z

We have a daily PkgEval (https://s3.amazonaws.com/julialang-reports/nanosoldier/pkgeval/by_date/2022-02/07/report.html).
Some other PKG seems also failed because of this PR.
I'll take a look to decide which value to use.

…aLang#44027)

It turns out `strides(a::StridedReinterpretArray)` won't call `Base.size_to_strides` anymore after JuliaLang#44027. As it is dispatched to `strides(::NonReshapedReinterpretArray)`/`strides(::ReshapedReinterpretArray)`. This commit fix that regression.

N5N3 added the arrays [a, r, r, a, y, s] label Feb 3, 2022

N5N3 force-pushed the strides-reinterp branch from 0035418 to 96a9cd6 Compare February 3, 2022 07:34

N5N3 added 2 commits February 4, 2022 00:11

Generalize strides for ReinterpretArray

13299f4

Generalize strides for ReshapedArray

b66a793

Update abstractarray.jl

N5N3 force-pushed the strides-reinterp branch from a262868 to b66a793 Compare February 3, 2022 16:37

vtjnash requested review from mbauman and timholy February 3, 2022 16:58

jishnub mentioned this pull request Feb 4, 2022

Optimize real matrix * complex vector by reinterpreting as real #44011

Merged

timholy reviewed Feb 4, 2022

View reviewed changes

test/abstractarray.jl Outdated Show resolved Hide resolved

base/reshapedarray.jl Show resolved Hide resolved

DenseArray might not be contiguous.

6e2fd5f

N5N3 force-pushed the strides-reinterp branch from 4831ce8 to 08c35e4 Compare February 5, 2022 07:37

N5N3 added 3 commits February 5, 2022 21:12

Some code clean

f58118a

we are in `Base`, so no need to `Base.`

Fix for non-integer-multipled reinterpret(T, a)

5bebfe2

test added

Move type-base strides test to inline.jl

7c64e45

N5N3 force-pushed the strides-reinterp branch from 08c35e4 to 7c64e45 Compare February 5, 2022 13:14

timholy approved these changes Feb 6, 2022

View reviewed changes

timholy merged commit e0a4b77 into JuliaLang:master Feb 6, 2022

N5N3 deleted the strides-reinterp branch February 6, 2022 13:41

mkitti mentioned this pull request Feb 9, 2022

Regression: stride errors on 0-dim array #44087

Closed

LilithHafner pushed a commit to LilithHafner/julia that referenced this pull request Feb 22, 2022

Generalize strides for ReinterpretArray and ReshapedArray (Juli…

62be2d8

…aLang#44027)

vtjnash mentioned this pull request Mar 8, 2022

Extend strides for ReshapedArray with strided parent. #44507

Merged

LilithHafner pushed a commit to LilithHafner/julia that referenced this pull request Mar 8, 2022

Generalize strides for ReinterpretArray and ReshapedArray (Juli…

56c630d

…aLang#44027)

palday mentioned this pull request Oct 18, 2022

Profile the objective JuliaStats/MixedModels.jl#639

Merged

3 tasks

moyner mentioned this pull request May 26, 2023

Upgrade to PartitionedArrays v0.3 fredrikekre/HYPRE.jl#16

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Generalize `strides` for `ReinterpretArray` and `ReshapedArray`#44027

Generalize `strides` for `ReinterpretArray` and `ReshapedArray`#44027
timholy merged 6 commits intoJuliaLang:masterfrom
N5N3:strides-reinterp

N5N3 commented Feb 3, 2022 •

edited

Loading

Uh oh!

timholy commented Feb 3, 2022

Uh oh!

N5N3 commented Feb 3, 2022

Uh oh!

timholy left a comment

Uh oh!

Uh oh!

Uh oh!

timholy commented Feb 6, 2022

Uh oh!

mkitti commented Feb 9, 2022

Uh oh!

mkitti commented Feb 9, 2022

Uh oh!

N5N3 commented Feb 9, 2022

Uh oh!

mkitti commented Feb 9, 2022

Uh oh!

N5N3 commented Feb 9, 2022 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

N5N3 commented Feb 3, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

timholy commented Feb 3, 2022

Uh oh!

N5N3 commented Feb 3, 2022

Uh oh!

timholy left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

timholy commented Feb 6, 2022

Uh oh!

mkitti commented Feb 9, 2022

Uh oh!

mkitti commented Feb 9, 2022

Uh oh!

N5N3 commented Feb 9, 2022

Uh oh!

mkitti commented Feb 9, 2022

Uh oh!

N5N3 commented Feb 9, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

N5N3 commented Feb 3, 2022 •

edited

Loading

N5N3 commented Feb 9, 2022 •

edited

Loading