https://github.com/JuliaLang/julia
Tip revision: 85d9f486893526abaeb7e045d22f096e4e3e4d11 authored by Matt Bauman on 22 August 2019, 22:37:27 UTC
better... use its own style but promote to Diagonal
better... use its own style but promote to Diagonal
Tip revision: 85d9f48
subarray.jl
# This file is a part of Julia. License is MIT: https://julialang.org/license
using Test, Random
######## Utilities ###########
# Generate an array similar to A[indx1, indx2, ...], but only call
# getindex with scalar-valued indices. This will be safe even if
# `getindex` someday calls `view`.
# The "nodrop" variant does not drop any dimensions (not even trailing ones)
function Agen_nodrop(A::AbstractArray, I...)
irep = replace_colon(A, I)
_Agen(A, ensure_iterable(irep)...)
end
# This drops scalar dimensions
function Agen_slice(A::AbstractArray, I...)
irep = replace_colon(A, I)
B = _Agen(A, ensure_iterable(irep)...)
sd = Int[]
for i = 1:length(I)
if isa(I[i], Real)
push!(sd, i)
end
end
dropdims(B, dims=sd)
end
_Agen(A, i1) = [A[j1] for j1 in i1]
_Agen(A, i1, i2) = [A[j1,j2] for j1 in i1, j2 in i2]
_Agen(A, i1, i2, i3) = [A[j1,j2,j3] for j1 in i1, j2 in i2, j3 in i3]
_Agen(A, i1, i2, i3, i4) = [A[j1,j2,j3,j4] for j1 in i1, j2 in i2, j3 in i3, j4 in i4]
_Agen(A, i1, i2, i3, i4, i5) = [A[j1,j2,j3,j4,j5] for j1 in i1, j2 in i2, j3 in i3, j4 in i4, j5 in i5]
_Agen(A, i1, i2, i3, i4, i5, i6) = [A[j1,j2,j3,j4,j5,j6] for j1 in i1, j2 in i2, j3 in i3, j4 in i4, j5 in i5, j6 in i6]
function replace_colon(A::AbstractArray, I)
Iout = Vector{Any}(undef, length(I))
I == (:,) && return (1:length(A),)
for d = 1:length(I)
Iout[d] = isa(I[d], Colon) ? (1:size(A,d)) : I[d]
end
(Iout...,)
end
ensure_iterable(::Tuple{}) = ()
ensure_iterable(t::Tuple{Union{Number, CartesianIndex}, Vararg{Any}}) = ((t[1],), ensure_iterable(Base.tail(t))...)
ensure_iterable(t::Tuple{Any, Vararg{Any}}) = (t[1], ensure_iterable(Base.tail(t))...)
index_ndims(t::Tuple) = tup2val(Base.index_ndims(t))
tup2val(::NTuple{N}) where {N} = Val(N)
# To avoid getting confused by manipulations that are implemented for SubArrays,
# it's good to copy the contents to an Array. This version protects against
# `similar` ever changing its meaning.
function copy_to_array(A::AbstractArray)
Ac = Array{eltype(A)}(undef, size(A))
copyto!(Ac, A)
end
# Discover the highest dimension along which the values in A are
# separated by a single increment. If A was extracted via getindex
# from reshape(1:N, ...), this is equivalent to finding the highest
# dimension of the SubArray consistent with a single stride in the
# parent array.
function single_stride_dim(A::Array)
ld = 0
while ld < ndims(A)
# Collapse all dimensions up to & including ld+1 into the first dimension
shp = [prod(size(A)[1:ld+1])]
for j = ld+2:ndims(A)
push!(shp, size(A,j))
end
Ar = reshape(A, shp...)
# Compute the diff along dimension 1
if size(Ar, 1) > 1
indices = map(d->1:size(Ar,d), [1:ndims(Ar);])
indicesp = copy(indices); indicesp[1] = 2:size(Ar,1)
indicesm = copy(indices); indicesm[1] = 1:size(Ar,1)-1
dA = Ar[indicesp...] - Ar[indicesm...]
ustride = unique(dA[:])
if length(ustride) == 1 # is it a single stride?
ld += 1
else
break
end
else
ld += 1
end
end
ld
end
single_stride_dim(@nospecialize(A)) = single_stride_dim(copy_to_array(A))
# Testing equality of AbstractArrays, using several different methods to access values
function test_cartesian(@nospecialize(A), @nospecialize(B))
isgood = true
for (IA, IB) in zip(eachindex(A), eachindex(B))
if A[IA] != B[IB]
@show IA IB A[IA] B[IB]
isgood = false
break
end
end
if !isgood
@show A
@show B
error("Mismatch")
end
end
function test_linear(@nospecialize(A), @nospecialize(B))
length(A) == length(B) || error("length mismatch")
isgood = true
for (iA, iB) in zip(1:length(A), 1:length(B))
if A[iA] != B[iB]
isgood = false
break
end
end
if !isgood
@show A
@show A.indices
@show B
error("Mismatch")
end
end
# "mixed" means 2 indices even for N-dimensional arrays
test_mixed(::AbstractArray{T,1}, ::Array) where {T} = nothing
test_mixed(::AbstractArray{T,2}, ::Array) where {T} = nothing
test_mixed(A, B::Array) = _test_mixed(A, reshape(B, size(A)))
function _test_mixed(@nospecialize(A), @nospecialize(B))
m = size(A, 1)
n = size(A, 2)
isgood = true
for J in CartesianIndices(size(A)[2:end]), i in 1:m
if A[i,J] != B[i,J]
isgood = false
break
end
end
if !isgood
@show A
@show B
error("Mismatch")
end
nothing
end
function test_bounds(@nospecialize(A))
@test_throws BoundsError A[0]
@test_throws BoundsError A[end+1]
trailing2 = ntuple(x->1, max(ndims(A)-2, 0))
trailing3 = ntuple(x->1, max(ndims(A)-3, 0))
@test_throws BoundsError A[1, 0, trailing2...]
@test_throws BoundsError A[1, end+1, trailing2...]
@test_throws BoundsError A[1, 1, 0, trailing3...]
@test_throws BoundsError A[1, 1, end+1, trailing3...]
@test_throws BoundsError A[0, 1, trailing2...]
@test_throws BoundsError A[end+1, 1, trailing2...]
@test_throws BoundsError A[0, 1, 1, trailing3...]
@test_throws BoundsError A[end+1, 1, 1, trailing3...]
@test_throws BoundsError A[1, 0, 1, trailing3...]
@test_throws BoundsError A[1, end+1, 1, trailing3...]
@test_throws BoundsError A[1, 0]
@test_throws BoundsError A[1, end+1]
@test_throws BoundsError A[1, 1, 0]
@test_throws BoundsError A[1, 1, end+1]
@test_throws BoundsError A[0, 1]
@test_throws BoundsError A[end+1, 1]
@test_throws BoundsError A[0, 1, 1]
@test_throws BoundsError A[end+1, 1, 1]
@test_throws BoundsError A[1, 0, 1]
@test_throws BoundsError A[1, end+1, 1]
end
function dim_break_linindex(I)
i = 1
while i <= length(I) && !isa(I[i], Vector{Int})
i += 1
end
i - 1
end
function runsubarraytests(A::Array, I...)
# Direct test of linear indexing inference
C = Agen_nodrop(A, I...)
ld = min(single_stride_dim(C), dim_break_linindex(I))
S = view(A, I...)
if Base.iscontiguous(S)
@test S.stride1 == 1
end
test_linear(S, C)
test_cartesian(S, C)
test_mixed(S, C)
end
function runsubarraytests(@nospecialize(A), I...)
# When A was created with view, we have to check bounds, since some
# of the "residual" dimensions have size 1. It's possible that we
# need dedicated tests for view.
for d = 1:length(I)-1
if !isa(I[d], Colon) && any(I[d] .> size(A,d))
return nothing
end
end
if !isa(I[end], Colon) && any(I[end] .> prod(size(A)[length(I):end]))
return nothing
end
AA = copy_to_array(A)
# Direct test of linear indexing inference
C = Agen_nodrop(AA, I...)
Cld = ld = min(single_stride_dim(C), dim_break_linindex(I))
Cdim = AIindex = 0
while Cdim <= Cld && AIindex < length(A.indices)
AIindex += 1
if isa(A.indices[AIindex], Real)
ld += 1
else
Cdim += 1
end
end
local S
try
S = view(A, I...)
catch
@show typeof(A)
@show A.indices
@show I
rethrow()
end
test_linear(S, C)
test_cartesian(S, C)
test_mixed(S, C)
end
# indexN is a cartesian index, indexNN is a linear index for 2 dimensions, and indexNNN is a linear index for 3 dimensions
function runviews(SB::AbstractArray, indexN, indexNN, indexNNN)
@assert ndims(SB) > 2
for i3 in indexN, i2 in indexN, i1 in indexN
runsubarraytests(SB, i1, i2, i3, ntuple(x->1, max(ndims(SB)-3, 0))...)
end
for i2 in indexN, i1 in indexN
runsubarraytests(SB, i1, i2, ntuple(x->1, max(ndims(SB)-2, 0))...)
end
for i1 in indexNNN
runsubarraytests(SB, i1)
end
end
function runviews(SB::AbstractArray{T, 3} where T, indexN, indexNN, indexNNN)
@assert ndims(SB) > 2
for i3 in indexN, i2 in indexN, i1 in indexN
runsubarraytests(SB, i1, i2, i3)
end
for i2 in indexN, i1 in indexN
runsubarraytests(SB, i1, i2, 1)
end
for i1 in indexNNN
runsubarraytests(SB, i1)
end
end
function runviews(SB::AbstractArray{T,2}, indexN, indexNN, indexNNN) where T
for i2 in indexN, i1 in indexN
runsubarraytests(SB, i1, i2)
end
for i1 in indexNN
runsubarraytests(SB, i1)
end
end
function runviews(SB::AbstractArray{T,1}, indexN, indexNN, indexNNN) where T
for i1 in indexN
runsubarraytests(SB, i1)
end
end
runviews(SB::AbstractArray{T,0}, indexN, indexNN, indexNNN) where {T} = nothing
######### Tests #########
testfull = Bool(parse(Int,(get(ENV, "JULIA_TESTFULL", "0"))))
### Views from Arrays ###
index5 = (1, :, 2:5, [4,1,5], reshape([2]), view(1:5,[2 3 4 1])) # all work with at least size 5
index25 = (3, :, 2:11, [19,9,7], reshape([10]), view(1:25,[19 15; 4 24]))
index125 = (113, :, 85:121, [99,14,103], reshape([72]), view(1:125,reshape([25,4,102,67], 1, 2, 2)))
if testfull
let A = copy(reshape(1:5*7*11, 11, 7, 5))
runviews(A, index5, index25, index125)
end
end
### Views from views ###
# "outer" indices create snips that have at least size 5 along each dimension,
# with the exception of Int-slicing
oindex = (:, 6, 3:7, reshape([12]), [8,4,6,12,5,7], [3:7 1:5 2:6 4:8 5:9])
_ndims(::AbstractArray{T,N}) where {T,N} = N
_ndims(x) = 1
if testfull
let B = copy(reshape(1:13^3, 13, 13, 13))
for o3 in oindex, o2 in oindex, o1 in oindex
viewB = view(B, o1, o2, o3)
runviews(viewB, index5, index25, index125)
end
end
end
if !testfull
let B = copy(reshape(1:13^3, 13, 13, 13))
for oind in ((:,:,:),
(:,:,6),
(:,6,:),
(6,:,:),
(:,3:7,:),
(3:7,:,:),
(3:7,6,:),
(3:7,6,0x6),
(6,UInt(3):UInt(7),3:7),
(13:-2:1,:,:),
([8,4,6,12,5,7],:,3:7),
(6,CartesianIndex.(6,[8,4,6,12,5,7])),
(CartesianIndex(13,6),[8,4,6,12,5,7]),
(1,:,view(1:13,[9,12,4,13,1])),
(view(1:13,[9,12,4,13,1]),2:6,4),
([1:5 2:6 3:7 4:8 5:9], :, 3))
runsubarraytests(B, oind...)
viewB = view(B, oind...)
runviews(viewB, index5, index25, index125)
end
end
end
# issue #11289
x11289 = randn(5,5)
@test isempty(view(x11289, Int[], :))
@test isempty(view(x11289, [2,5], Int[]))
@test isempty(view(x11289, Int[], 2))
####### "Classical" tests #######
# Tests where non-trailing dimensions are preserved
A = copy(reshape(1:120, 3, 5, 8))
sA = view(A, 2:2, 1:5, :)
@test @inferred(strides(sA)) == (1, 3, 15)
@test parent(sA) == A
@test parentindices(sA) == (2:2, 1:5, Base.Slice(1:8))
@test Base.parentdims(sA) == [1:3;]
@test size(sA) == (1, 5, 8)
@test axes(sA) === (Base.OneTo(1), Base.OneTo(5), Base.OneTo(8))
@test sA[1, 2, 1:8][:] == [5:15:120;]
sA[2:5:end] .= -1
@test all(sA[2:5:end] .== -1)
@test all(A[5:15:120] .== -1)
@test @inferred(strides(sA)) == (1,3,15)
@test stride(sA,3) == 15
@test stride(sA,4) == 120
test_bounds(sA)
sA = view(A, 1:3, 1:5, 5)
@test Base.parentdims(sA) == [1:2;]
sA[1:3,1:5] .= -2
@test all(A[:,:,5] .== -2)
fill!(sA, -3)
@test all(A[:,:,5] .== -3)
sA[:] .= 4
@test all(A[:,:,5] .== 4)
@test @inferred(strides(sA)) == (1,3)
test_bounds(sA)
sA = view(A, 1:3, 3:3, 2:5)
@test Base.parentdims(sA) == [1:3;]
@test size(sA) == (3,1,4)
@test axes(sA) === (Base.OneTo(3), Base.OneTo(1), Base.OneTo(4))
@test sA == A[1:3,3:3,2:5]
@test sA[:] == A[1:3,3,2:5][:]
test_bounds(sA)
sA = view(A, 1:2:3, 1:3:5, 1:2:8)
@test Base.parentdims(sA) == [1:3;]
@test @inferred(strides(sA)) == (2,9,30)
@test sA[:] == A[1:2:3, 1:3:5, 1:2:8][:]
# issue #8807
@test view(view([1:5;], 1:5), 1:5) == [1:5;]
# Test with mixed types
@test sA[:, Int16[1,2], big(2)] == [31 40; 33 42]
test_bounds(sA)
sA = view(A, 1:1, 1:5, [1 3; 4 2])
@test ndims(sA) == 4
@test axes(sA) === (Base.OneTo(1), Base.OneTo(5), Base.OneTo(2), Base.OneTo(2))
sA = view(A, 1:2, 3, [1 3; 4 2])
@test ndims(sA) == 3
@test axes(sA) === (Base.OneTo(2), Base.OneTo(2), Base.OneTo(2))
# logical indexing #4763
A = view([1:10;], 5:8)
@test A[A.<7] == view(A, A.<7) == [5, 6]
@test Base.unsafe_getindex(A, A.<7) == [5, 6]
B = reshape(1:16, 4, 4)
sB = view(B, 2:3, 2:3)
@test sB[sB.>8] == view(sB, sB.>8) == [10, 11]
@test Base.unsafe_getindex(sB, sB.>8) == [10, 11]
# Tests where dimensions are dropped
A = copy(reshape(1:120, 3, 5, 8))
sA = view(A, 2, :, 1:8)
@test parent(sA) == A
@test parentindices(sA) == (2, Base.Slice(1:5), 1:8)
@test Base.parentdims(sA) == [2:3;]
@test size(sA) == (5, 8)
@test axes(sA) === (Base.OneTo(5), Base.OneTo(8))
@test @inferred(strides(sA)) == (3,15)
@test sA[2, 1:8][:] == [5:15:120;]
@test sA[:,1] == [2:3:14;]
@test sA[2:5:end] == [5:15:110;]
sA[2:5:end] .= -1
@test all(sA[2:5:end] .== -1)
@test all(A[5:15:120] .== -1)
test_bounds(sA)
sA = view(A, 1:3, 1:5, 5)
@test Base.parentdims(sA) == [1:2;]
@test size(sA) == (3,5)
@test axes(sA) === (Base.OneTo(3),Base.OneTo(5))
@test @inferred(strides(sA)) == (1,3)
test_bounds(sA)
sA = view(A, 1:2:3, 3, 1:2:8)
@test Base.parentdims(sA) == [1,3]
@test size(sA) == (2,4)
@test axes(sA) === (Base.OneTo(2), Base.OneTo(4))
@test @inferred(strides(sA)) == (2,30)
@test sA[:] == A[sA.indices...][:]
test_bounds(sA)
let a = [5:8;]
@test parent(a) == a
@test parentindices(a) == (1:4,)
end
# issue #6218 - logical indexing
A = rand(2, 2, 3)
msk = fill(true, 2, 2)
msk[2,1] = false
sA = view(A, :, :, 1)
sA[msk] .= 1.0
@test sA[msk] == fill(1, count(msk))
# bounds checking upon construction; see #4044, #10296
@test_throws BoundsError view(1:10, 8:11)
A = reshape(1:20, 5, 4)
sA = view(A, 1:2, 1:3)
@test_throws BoundsError view(sA, 1:3, 1:3)
@test_throws BoundsError view(sA, 1:2, 1:4)
view(sA, 1:2, 1:2)
@test_throws BoundsError view(A, 17:23)
view(A, 17:20)
# Linear indexing by one multidimensional array:
A = reshape(1:120, 3, 5, 8)
sA = view(A, :, :, :)
@test sA[[72 17; 107 117]] == [72 17; 107 117]
@test sA[[99 38 119 14 76 81]] == [99 38 119 14 76 81]
@test sA[[fill(1, (2, 2, 2)); fill(2, (2, 2, 2))]] == [fill(1, (2, 2, 2)); fill(2, (2, 2, 2))]
sA = view(A, 1:2, 2:3, 3:4)
@test sA[(1:8)'] == [34 35 37 38 49 50 52 53]
@test sA[[1 2 4 4; 6 1 1 4]] == [34 35 38 38; 50 34 34 38]
# issue #11871
let a = fill(1., (2,2)),
b = view(a, 1:2, 1:2)
b[2] = 2
@test b[2] === 2.0
end
# issue #15138
let a = [1,2,3],
b = view(a, UInt(1):UInt(2))
@test b == view(a, UInt(1):UInt(2)) == view(view(a, :), UInt(1):UInt(2)) == [1,2]
end
let A = reshape(1:4, 2, 2),
B = view(A, :, :)
@test parent(B) === A
@test parent(view(B, 0x1, :)) === parent(view(B, 0x1, :)) === A
end
# issue #15168
let A = rand(10), sA = view(copy(A), :)
@test sA[Int16(1)] === sA[Int32(1)] === sA[Int64(1)] === A[1]
permute!(sA, Vector{Int16}(1:10))
@test A == sA
end
# the following segfaults with LLVM 3.8 on Windows, ref #15417
@test Array(view(view(reshape(1:13^3, 13, 13, 13), 3:7, 6:6, :), 1:2:5, :, 1:2:5)) ==
cat([68,70,72],[406,408,410],[744,746,748]; dims=3)
# tests @view (and replace_ref_end!)
X = reshape(1:24,2,3,4)
Y = 4:-1:1
@test isa(@view(X[1:3]), SubArray)
@test X[1:end] == @.(@view X[1:end]) # test compatibility of @. and @view
@test X[1:end-3] == @view X[1:end-3]
@test X[1:end,2,2] == @view X[1:end,2,2]
@test X[1,1:end-2,1] == @view X[1,1:end-2,1]
@test X[1,2,1:end-2] == @view X[1,2,1:end-2]
@test X[1,2,Y[2:end]] == @view X[1,2,Y[2:end]]
@test X[1:end,2,Y[2:end]] == @view X[1:end,2,Y[2:end]]
u = (1,2:3)
@test X[u...,2:end] == @view X[u...,2:end]
@test X[(1,)...,(2,)...,2:end] == @view X[(1,)...,(2,)...,2:end]
# test macro hygiene
let size=(x,y)-> error("should not happen"), Base=nothing
@test X[1:end,2,2] == @view X[1:end,2,2]
end
# test that side effects occur only once
let foo = [X]
@test X[2:end-1] == @view (push!(foo,X)[1])[2:end-1]
@test foo == [X, X]
end
# test @views macro
@views let f!(x) = x[1:end-1] .+= x[2:end].^2
x = [1,2,3,4]
f!(x)
@test x == [5,11,19,4]
@test x[1:3] isa SubArray
@test x[2] === 11
@test Dict((1:3) => 4)[1:3] === 4
x[1:2] .= 0
@test x == [0,0,19,4]
x[1:2] .= 5:6
@test x == [5,6,19,4]
f!(x[3:end])
@test x == [5,6,35,4]
x[Y[2:3]] .= 7:8
@test x == [5,8,7,4]
x[(3,)..., ()...] += 3
@test x == [5,8,10,4]
i = Int[]
# test that lhs expressions in update operations are evaluated only once:
x[push!(i,4)[1]] += 5
@test x == [5,8,10,9] && i == [4]
x[push!(i,3)[end]] += 2
@test x == [5,8,12,9] && i == [4,3]
@. x[3:end] = 0 # make sure @. works with end expressions in @views
@test x == [5,8,0,0]
end
@views @test isa(X[1:3], SubArray)
@test X[1:end] == @views X[1:end]
@test X[1:end-3] == @views X[1:end-3]
@test X[1:end,2,2] == @views X[1:end,2,2]
@test X[1,2,1:end-2] == @views X[1,2,1:end-2]
@test X[1,2,Y[2:end]] == @views X[1,2,Y[2:end]]
@test X[1:end,2,Y[2:end]] == @views X[1:end,2,Y[2:end]]
@test X[u...,2:end] == @views X[u...,2:end]
@test X[(1,)...,(2,)...,2:end] == @views X[(1,)...,(2,)...,2:end]
# test macro hygiene
let size=(x,y)-> error("should not happen"), Base=nothing
@test X[1:end,2,2] == @views X[1:end,2,2]
end
# issue #18034
# ensure that it is possible to create an isbits, IndexLinear view of an immutable Array
let
struct ImmutableTestArray{T, N} <: Base.DenseArray{T, N}
end
Base.size(::Union{ImmutableTestArray, Type{ImmutableTestArray}}) = (0, 0)
Base.IndexStyle(::Union{ImmutableTestArray, Type{ImmutableTestArray}}) = Base.IndexLinear()
a = ImmutableTestArray{Float64, 2}()
@test Base.IndexStyle(view(a, :, :)) == Base.IndexLinear()
@test isbits(view(a, :, :))
end
# ref issue #17351
@test @inferred(reverse(view([1 2; 3 4], :, 1), dims=1)) == [3, 1]
let
s = view(reshape(1:6, 2, 3), 1:2, 1:2)
@test @inferred(s[2,2,1]) === 4
end
# issue #18581: slices with OneTo axes can be linear
let
A18581 = rand(5, 5)
B18581 = view(A18581, :, axes(A18581,2))
@test IndexStyle(B18581) === IndexLinear()
end
@test sizeof(view(zeros(UInt8, 10), 1:4)) == 4
@test sizeof(view(zeros(UInt8, 10), 1:3)) == 3
@test sizeof(view(zeros(Float64, 10, 10), 1:3, 2:6)) == 120
# PR #25321
# checks that issue in type inference is resolved
A = rand(5,5,5,5)
V = view(A, 1:1 ,:, 1:3, :)
@test @inferred(strides(V)) == (1, 5, 25, 125)
# Issue #26263 — ensure that unaliascopy properly trims the array
A = rand(5,5,5,5)
V = view(A, 2:5, :, 2:5, 1:2:5)
@test @inferred(Base.unaliascopy(V)) == V == A[2:5, :, 2:5, 1:2:5]
@test @inferred(sum(Base.unaliascopy(V))) ≈ sum(V) ≈ sum(A[2:5, :, 2:5, 1:2:5])
# issue #27632
function _test_27632(A)
for J in CartesianIndices(size(A)[2:end])
A[1, J]
end
nothing
end
# check that this doesn't crash
_test_27632(view(ones(Int64, (1, 1, 1)), 1, 1, 1))
# issue #29608 - views of single values can be considered contiguous
@test Base.iscontiguous(view(ones(1), 1))