Make `RArray` a subtype of `DenseArray` #1696

albertomercurio · 2025-09-21T21:19:51Z

albertomercurio · 2025-09-22T08:42:29Z

Apparently all the tests are failing. I can try to figure out why, but the output is quite long to investigate.

To run the tests locally, should I run them with simply Pkg.test()?

albertomercurio · 2025-09-22T12:31:56Z

Making AnyTracedRArray a DenseArray subtype seems to fix a few examples, but it breaks many others. I don't know how to proceed.

giordano · 2025-09-22T14:51:15Z

I don't know how to proceed.

Honestly I don't think anyone does because no one has tried doing this. The most productive way is usually to look at one of the errors and start crafting a minimal reproducer, which hopefully would make it clearer what in the dispatch (presumably) is going haywire.

albertomercurio · 2025-09-22T16:42:55Z

Ok, just to give some informations.

There are 20 error when making only RArray a subtype of DenseArray
These errors become 70 or so when also making AnyTracedArray a subtype of DenseArray

Firs, does make sense to have AnyTracedArray <: DenseArray? Without this condition, the following code fails

sum_xxᵀ(x) = sum(x .* x')
x = rand(4,4)
x_ca = Reactant.to_rarray(x)
@jit(sum_xxᵀ(x_ca))

with error

ERROR: Scalar indexing is disallowed.
Invocation of getindex(::TracedRArray, ::Vararg{Int, N}) resulted in scalar indexing of a GPU array.
This is typically caused by calling an iterating implementation of a method.
Such implementations *do not* execute on the GPU, but very slowly on the CPU,
and therefore should be avoided.

If you want to allow scalar iteration, use `allowscalar` or `@allowscalar`
to enable scalar iteration globally or for the operations in question.
Stacktrace:
  [1] error(s::String)
    @ Base ./error.jl:35
  [2] errorscalar(op::String)
    @ GPUArraysCore ~/.julia/packages/GPUArraysCore/aNaXo/src/GPUArraysCore.jl:151
  [3] _assertscalar(op::String, behavior::GPUArraysCore.ScalarIndexing)
    @ GPUArraysCore ~/.julia/packages/GPUArraysCore/aNaXo/src/GPUArraysCore.jl:124
  [4] assertscalar(op::String)
    @ GPUArraysCore ~/.julia/packages/GPUArraysCore/aNaXo/src/GPUArraysCore.jl:112
  [5] getindex(::Reactant.TracedRArray{Float64, 2}, ::Int64, ::Int64)
    @ Reactant.TracedIndexing ~/.julia/dev/Reactant/src/Indexing.jl:40
  [6] getindex
    @ ~/.julia/dev/Reactant/src/Indexing.jl:48 [inlined]
  [7] _getindex
    @ ./abstractarray.jl:1341 [inlined]
  [8] getindex
    @ ./abstractarray.jl:1312 [inlined]
  [9] _permutedims!(P::PermutedDimsArray{…}, src::Reactant.TracedRArray{…}, R1::CartesianIndices{…}, R2::CartesianIndices{…}, R3::CartesianIndices{…}, ds::Int64, dp::Int64)
    @ Base.PermutedDimsArrays ./permuteddimsarray.jl:322
 [10] _copy!(P::PermutedDimsArray{Reactant.TracedRNumber{Float64}, 2, (2, 1), (2, 1), Reactant.TracedRArray{Float64, 2}}, src::Reactant.TracedRArray{Float64, 2})
    @ Base.PermutedDimsArrays ./permuteddimsarray.jl:311
 [11] permutedims!(dest::Reactant.TracedRArray{Float64, 2}, src::Reactant.TracedRArray{Float64, 2}, perm::Tuple{Int64, Int64})
    @ Base.PermutedDimsArrays ./permuteddimsarray.jl:287
 [12] permutedims(B::Reactant.TracedRArray{Float64, 2}, perm::Tuple{Int64, Int64})
    @ Base ./multidimensional.jl:1634
 [13] materialize_traced_array
    @ ~/.julia/dev/Reactant/src/stdlibs/LinearAlgebra.jl:48 [inlined]
 [14] materialize_traced_array(x::LinearAlgebra.Adjoint{Reactant.TracedRNumber{Float64}, Reactant.TracedRArray{Float64, 2}})
    @ Reactant.TracedLinearAlgebra ~/.julia/dev/Reactant/src/stdlibs/LinearAlgebra.jl:54
 [15] promote_to
    @ ~/.julia/dev/Reactant/src/TracedPromotion.jl:22 [inlined]
 [16] (::Nothing)(none::typeof(Reactant.promote_to), none::Type{…}, none::LinearAlgebra.Adjoint{…})
    @ Reactant ./<missing>:0
 [17] promote_to
    @ ~/.julia/dev/Reactant/src/TracedPromotion.jl:22 [inlined]
 [18] call_with_reactant(::typeof(Reactant.promote_to), ::Type{Reactant.TracedRArray{…}}, ::LinearAlgebra.Adjoint{Reactant.TracedRNumber{…}, Reactant.TracedRArray{…}})
    @ Reactant ~/.julia/dev/Reactant/src/utils.jl:0
 [19] promote_to
    @ ~/.julia/dev/Reactant/src/TracedPromotion.jl:8 [inlined]
 [20] broadcast_to_size
    @ ~/.julia/dev/Reactant/src/TracedPromotion.jl:94 [inlined]
 [21] broadcast_to_size
    @ ~/.julia/dev/Reactant/src/TracedPromotion.jl:112 [inlined]
 [22] (::Nothing)(none::typeof(Reactant.broadcast_to_size), none::Base.Broadcast.Extruded{LinearAlgebra.Adjoint{…}, Tuple{…}, Tuple{…}}, none::Tuple{Int64, Int64})
    @ Reactant ./<missing>:0
 [23] getproperty
    @ ./Base.jl:49 [inlined]
 [24] broadcast_to_size
    @ ~/.julia/dev/Reactant/src/TracedPromotion.jl:111 [inlined]
 [25] call_with_reactant(::typeof(Reactant.broadcast_to_size), ::Base.Broadcast.Extruded{LinearAlgebra.Adjoint{…}, Tuple{…}, Tuple{…}}, ::Tuple{Int64, Int64})
    @ Reactant ~/.julia/dev/Reactant/src/utils.jl:0
 [26] #15
    @ ./none:0 [inlined]
 [27] iterate
    @ ./generator.jl:48 [inlined]
 [28] _copyto!
    @ ~/.julia/dev/Reactant/src/TracedRArray.jl:502 [inlined]
 [29] (::Nothing)(none::typeof(Reactant.TracedRArrayOverrides._copyto!), none::Reactant.TracedRArray{…}, none::Base.Broadcast.Broadcasted{…})
    @ Reactant ./<missing>:0
 [30] getproperty
    @ ./Base.jl:49 [inlined]
 [31] size
    @ ~/.julia/dev/Reactant/src/TracedRArray.jl:248 [inlined]
 [32] axes
    @ ./abstractarray.jl:98 [inlined]
 [33] _copyto!
    @ ~/.julia/dev/Reactant/src/TracedRArray.jl:495 [inlined]
 [34] call_with_reactant(::typeof(Reactant.TracedRArrayOverrides._copyto!), ::Reactant.TracedRArray{…}, ::Base.Broadcast.Broadcasted{…})
    @ Reactant ~/.julia/dev/Reactant/src/utils.jl:0
 [35] copyto!
    @ ~/.julia/dev/Reactant/src/TracedRArray.jl:456 [inlined]
 [36] (::Nothing)(none::typeof(copyto!), none::Reactant.TracedRArray{Float64, 2}, none::Base.Broadcast.Broadcasted{Nothing, Tuple{…}, typeof(*), Tuple{…}})
    @ Reactant ./<missing>:0
 [37] copyto!
    @ ~/.julia/dev/Reactant/src/TracedRArray.jl:456 [inlined]
 [38] call_with_reactant(::typeof(copyto!), ::Reactant.TracedRArray{Float64, 2}, ::Base.Broadcast.Broadcasted{Nothing, Tuple{…}, typeof(*), Tuple{…}})
    @ Reactant ~/.julia/dev/Reactant/src/utils.jl:0
 [39] copyto!
    @ ./broadcast.jl:925 [inlined]
 [40] copy
    @ ~/.julia/dev/Reactant/src/TracedRArray.jl:447 [inlined]
 [41] (::Nothing)(none::typeof(copy), none::Base.Broadcast.Broadcasted{Reactant.TracedRArrayOverrides.AbstractReactantArrayStyle{…}, Tuple{…}, typeof(*), Tuple{…}})
    @ Reactant ./<missing>:0
 [42] call_with_reactant(::typeof(copy), ::Base.Broadcast.Broadcasted{Reactant.TracedRArrayOverrides.AbstractReactantArrayStyle{…}, Tuple{…}, typeof(*), Tuple{…}})
    @ Reactant ~/.julia/dev/Reactant/src/utils.jl:519
 [43] materialize
    @ ./broadcast.jl:872 [inlined]
 [44] sum_xxᵀ
    @ ./REPL[2]:1 [inlined]
 [45] (::Nothing)(none::typeof(sum_xxᵀ), none::Reactant.TracedRArray{Float64, 2})
    @ Reactant ./<missing>:0
 [46] Adjoint
    @ ~/.julia/juliaup/julia-1.11.7+0.x64.linux.gnu/share/julia/stdlib/v1.11/LinearAlgebra/src/adjtrans.jl:33 [inlined]
 [47] Adjoint
    @ ~/.julia/juliaup/julia-1.11.7+0.x64.linux.gnu/share/julia/stdlib/v1.11/LinearAlgebra/src/adjtrans.jl:64 [inlined]
 [48] adjoint
    @ ~/.julia/juliaup/julia-1.11.7+0.x64.linux.gnu/share/julia/stdlib/v1.11/LinearAlgebra/src/adjtrans.jl:189 [inlined]
 [49] sum_xxᵀ
    @ ./REPL[2]:1 [inlined]
 [50] call_with_reactant(::typeof(sum_xxᵀ), ::Reactant.TracedRArray{Float64, 2})
    @ Reactant ~/.julia/dev/Reactant/src/utils.jl:0
 [51] make_mlir_fn(f::typeof(sum_xxᵀ), args::Tuple{…}, kwargs::@NamedTuple{}, name::String, concretein::Bool; toscalar::Bool, return_dialect::Symbol, args_in_result::Symbol, construct_function_without_args::Bool, do_transpose::Bool, input_shardings::Nothing, output_shardings::Nothing, runtime::Val{…}, verify_arg_names::Nothing, argprefix::Symbol, resprefix::Symbol, resargprefix::Symbol, num_replicas::Int64, optimize_then_pad::Bool)
    @ Reactant.TracedUtils ~/.julia/dev/Reactant/src/TracedUtils.jl:348
 [52] compile_mlir!(mod::Reactant.MLIR.IR.Module, f::Function, args::Tuple{…}, compile_options::CompileOptions, callcache::Dict{…}, sdycache::Dict{…}; fn_kwargs::@NamedTuple{}, backend::String, runtime::Val{…}, legalize_stablehlo_to_mhlo::Bool, kwargs::@Kwargs{})
    @ Reactant.Compiler ~/.julia/dev/Reactant/src/Compiler.jl:1575
 [53] compile_mlir! (repeats 2 times)
    @ ~/.julia/dev/Reactant/src/Compiler.jl:1542 [inlined]
 [54] compile_xla(f::Function, args::Tuple{…}; before_xla_optimizations::Bool, client::Nothing, serializable::Bool, kwargs::@Kwargs{…})
    @ Reactant.Compiler ~/.julia/dev/Reactant/src/Compiler.jl:3464
 [55] compile_xla
    @ ~/.julia/dev/Reactant/src/Compiler.jl:3437 [inlined]
 [56] compile(f::Function, args::Tuple{…}; kwargs::@Kwargs{…})
    @ Reactant.Compiler ~/.julia/dev/Reactant/src/Compiler.jl:3536
Some type information was truncated. Use `show(err)` to see complete types.

If I make AnyTracedRArray{T,N} = DenseArray{TracedRNumber{T},N} it works. However, using a vector rather than a matrix doesn't work on both cases.

I will investigate more what is going on.

albertomercurio · 2025-09-22T21:48:20Z

Ok I should have managed to make it work.

First, I defined AnyTracedRArray{T,N} = AbstractArray{TracedRNumber{T},N} as before. If we need it we could do it in another PR.

I'm not sure if the definitions of Base.mightalias make sense or not. We probably need to be more specific I guess. But at least everything works.

codecov · 2025-09-22T22:17:40Z

Codecov Report

❌ Patch coverage is 37.50000% with 5 lines in your changes missing coverage. Please review.
✅ Project coverage is 69.75%. Comparing base (7d8866e) to head (68a6d96).
⚠️ Report is 2 commits behind head on main.

Files with missing lines	Patch %	Lines
src/Types.jl	0.00%	4 Missing ⚠️
src/TracedRArray.jl	75.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1696      +/-   ##
==========================================
+ Coverage   68.27%   69.75%   +1.47%     
==========================================
  Files         109      110       +1     
  Lines       11597    12129     +532     
==========================================
+ Hits         7918     8460     +542     
+ Misses       3679     3669      -10

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

wsmoses · 2025-09-22T23:47:57Z

src/TracedRArray.jl

+)
+for ArrayType1 in ArrayTypesAlias
+    for ArrayType2 in ArrayTypesAlias
+        @eval Base.mightalias(::$ArrayType1, ::$ArrayType2) = false


this definitely seems wrong since if they are equal they certainly may alias

I'm not familiar with the internal structure of TracedRArray. I could in principle see the implementation for Base arrays if I know where to look in TracedRArray.

wsmoses · 2025-09-22T23:48:19Z

src/TracedRArray.jl

+    :(Base.ReshapedArray{<:TracedRNumber{T},N,<:TracedRArray})
+    )
+    @eval function Base.permutedims(A::$ArrayType, perm) where {T,N}
+        return @opcall transpose(materialize_traced_array(A), Int64[perm...])


why doesn't anytacedrarray work?

Apparently SubArray and ReshapedArray are not subtypes of AnyTracedRArray

julia> SubArray{<:TracedRNumber,1,<:TracedRArray} <: AnyTracedRArray false julia> SubArray{<:TracedRNumber,1,<:TracedRArray} |> supertypes (SubArray{<:TracedRNumber, 1, <:TracedRArray}, AbstractVector{<:TracedRNumber}, Any)

Cc @avik-pal i thought anytracedrarray was just an abstractarray of a traced number?

Make RArray a subtype of DenseArray

3f715d1

Make AnyTracedRArray a DenseArray subtype

8988f3f

albertomercurio added 2 commits September 22, 2025 21:36

Fix permutedims errors

04e3d42

Fix errors

68a6d96

wsmoses reviewed Sep 22, 2025

View reviewed changes

wsmoses requested a review from avik-pal September 22, 2025 23:48

wsmoses mentioned this pull request Sep 29, 2025

FixedSizeArrays support (And Memory Support) #1669

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Make `RArray` a subtype of `DenseArray` #1696

Make `RArray` a subtype of `DenseArray` #1696

Uh oh!

albertomercurio commented Sep 21, 2025

Uh oh!

albertomercurio commented Sep 22, 2025

Uh oh!

albertomercurio commented Sep 22, 2025

Uh oh!

giordano commented Sep 22, 2025

Uh oh!

albertomercurio commented Sep 22, 2025

Uh oh!

albertomercurio commented Sep 22, 2025

Uh oh!

codecov bot commented Sep 22, 2025 •

edited

Loading

Uh oh!

wsmoses Sep 22, 2025

Uh oh!

albertomercurio Sep 23, 2025

Uh oh!

wsmoses Sep 22, 2025

Uh oh!

albertomercurio Sep 23, 2025

Uh oh!

wsmoses Sep 23, 2025

Uh oh!

Uh oh!

Make RArray a subtype of DenseArray #1696

Are you sure you want to change the base?

Make RArray a subtype of DenseArray #1696

Uh oh!

Conversation

albertomercurio commented Sep 21, 2025

Uh oh!

albertomercurio commented Sep 22, 2025

Uh oh!

albertomercurio commented Sep 22, 2025

Uh oh!

giordano commented Sep 22, 2025

Uh oh!

albertomercurio commented Sep 22, 2025

Uh oh!

albertomercurio commented Sep 22, 2025

Uh oh!

codecov bot commented Sep 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

wsmoses Sep 22, 2025

Choose a reason for hiding this comment

Uh oh!

albertomercurio Sep 23, 2025

Choose a reason for hiding this comment

Uh oh!

wsmoses Sep 22, 2025

Choose a reason for hiding this comment

Uh oh!

albertomercurio Sep 23, 2025

Choose a reason for hiding this comment

Uh oh!

wsmoses Sep 23, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Make `RArray` a subtype of `DenseArray` #1696

Make `RArray` a subtype of `DenseArray` #1696

codecov bot commented Sep 22, 2025 •

edited

Loading