Allow broadcast errors to surface later. (#491)

We are currently checking for `isconcretetype`, which is too limiting, as
e.g. CUDA.jl supports isbits-union arrays. Checking for `allocatedinline`
would be better, but let's just do away with the check entirely and have
the array constructor fail in the presence of unsupported element types.
This is better anyway, as some back-ends may not support isbits-unions.

We do however still check for Union{}, as that isn't allocated inline
so would fail array construction. By using Nothing there instead, we give
the GPU kernel (which is expected to throw an error) the chance to execute
and report an exeception dynamically. This also makes it possible to
trace a broadcast invocation under, e.g., Cthulhu.

Author: maleadt

.github/workflows/ci.yml

@@ -14,7 +14,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        version: ['1.6', '1.7', '1.8', '1.9', 'nightly']
+        version: ['1.8', '1.9', '1.10.0-beta2', 'nightly']
         os: [ubuntu-latest, macOS-latest, windows-latest]
         arch: [x64]
     steps:
@@ -50,7 +50,7 @@ jobs:
       - uses: actions/checkout@v3
       - uses: julia-actions/setup-julia@latest
         with:
-          version: '1.6'
+          version: '1.8'
       - name: Install dependencies
         run: julia --project=docs/ -e 'using Pkg; Pkg.develop(PackageSpec(path=pwd())); Pkg.instantiate()'
       - name: Build and deploy

Manifest.toml

@@ -1,178 +1,213 @@
 # This file is machine-generated - editing it directly is not advised
 
-[[Adapt]]
+manifest_format = "2.0"
+
+[[deps.Adapt]]
 deps = ["LinearAlgebra", "Requires"]
 git-tree-sha1 = "76289dc51920fdc6e0013c872ba9551d54961c24"
 uuid = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"
 version = "3.6.2"
 
-[[ArgTools]]
+[[deps.ArgTools]]
 uuid = "0dad84c5-d112-42e6-8d28-ef12dabb789f"
+version = "1.1.1"
 
-[[Artifacts]]
+[[deps.Artifacts]]
 uuid = "56f22d72-fd6d-98f1-02f0-08ddc0907c33"
 
-[[Base64]]
+[[deps.Base64]]
 uuid = "2a0f44e3-6c83-55bd-87e4-b1978d98bd5f"
 
-[[CEnum]]
+[[deps.CEnum]]
 git-tree-sha1 = "eb4cb44a499229b3b8426dcfb5dd85333951ff90"
 uuid = "fa961155-64e5-5f13-b03f-caf6b980ea82"
 version = "0.4.2"
 
-[[Dates]]
+[[deps.CompilerSupportLibraries_jll]]
+deps = ["Artifacts", "Libdl"]
+uuid = "e66e0078-7015-5450-92f7-15fbd957f2ae"
+version = "1.0.1+0"
+
+[[deps.Dates]]
 deps = ["Printf"]
 uuid = "ade2ca70-3891-5945-98fb-dc099432e06a"
 
-[[Downloads]]
-deps = ["ArgTools", "LibCURL", "NetworkOptions"]
+[[deps.Downloads]]
+deps = ["ArgTools", "FileWatching", "LibCURL", "NetworkOptions"]
 uuid = "f43a241f-c20a-4ad4-852c-f6b1247861c6"
+version = "1.6.0"
+
+[[deps.FileWatching]]
+uuid = "7b1f6079-737a-58dc-b8bc-7a2ca5c1b5ee"
 
-[[GPUArraysCore]]
+[[deps.GPUArraysCore]]
 deps = ["Adapt"]
 path = "lib/GPUArraysCore"
 uuid = "46192b85-c4d5-4398-a991-12ede77f4527"
 version = "0.1.5"
 
-[[InteractiveUtils]]
+[[deps.InteractiveUtils]]
 deps = ["Markdown"]
 uuid = "b77e0a4c-d291-57a0-90e8-8db25a27a240"
 
-[[JLLWrappers]]
+[[deps.JLLWrappers]]
 deps = ["Artifacts", "Preferences"]
 git-tree-sha1 = "7e5d6779a1e09a36db2a7b6cff50942a0a7d0fca"
 uuid = "692b3bcd-3c85-4b1f-b108-f13ce0eb3210"
 version = "1.5.0"
 
-[[LLVM]]
+[[deps.LLVM]]
 deps = ["CEnum", "LLVMExtra_jll", "Libdl", "Printf", "Unicode"]
 git-tree-sha1 = "8695a49bfe05a2dc0feeefd06b4ca6361a018729"
 uuid = "929cbde3-209d-540e-8aea-75f648917ca0"
 version = "6.1.0"
 
-[[LLVMExtra_jll]]
+[[deps.LLVMExtra_jll]]
 deps = ["Artifacts", "JLLWrappers", "LazyArtifacts", "Libdl", "TOML"]
 git-tree-sha1 = "c35203c1e1002747da220ffc3c0762ce7754b08c"
 uuid = "dad2f222-ce93-54a1-a47d-0025e8a3acab"
 version = "0.0.23+0"
 
-[[LazyArtifacts]]
+[[deps.LazyArtifacts]]
 deps = ["Artifacts", "Pkg"]
 uuid = "4af54fe1-eca0-43a8-85a7-787d91b784e3"
 
-[[LibCURL]]
+[[deps.LibCURL]]
 deps = ["LibCURL_jll", "MozillaCACerts_jll"]
 uuid = "b27032c2-a3e7-50c8-80cd-2d36dbcbfd21"
+version = "0.6.3"
 
-[[LibCURL_jll]]
+[[deps.LibCURL_jll]]
 deps = ["Artifacts", "LibSSH2_jll", "Libdl", "MbedTLS_jll", "Zlib_jll", "nghttp2_jll"]
 uuid = "deac9b47-8bc7-5906-a0fe-35ac56dc84c0"
+version = "7.84.0+0"
 
-[[LibGit2]]
+[[deps.LibGit2]]
 deps = ["Base64", "NetworkOptions", "Printf", "SHA"]
 uuid = "76f85450-5226-5b5a-8eaa-529ad045b433"
 
-[[LibSSH2_jll]]
+[[deps.LibSSH2_jll]]
 deps = ["Artifacts", "Libdl", "MbedTLS_jll"]
 uuid = "29816b5a-b9ab-546f-933c-edad1886dfa8"
+version = "1.10.2+0"
 
-[[Libdl]]
+[[deps.Libdl]]
 uuid = "8f399da3-3557-5675-b5ff-fb832c97cbdb"
 
-[[LinearAlgebra]]
-deps = ["Libdl"]
+[[deps.LinearAlgebra]]
+deps = ["Libdl", "libblastrampoline_jll"]
 uuid = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 
-[[Logging]]
+[[deps.Logging]]
 uuid = "56ddb016-857b-54e1-b83d-db4d58db5568"
 
-[[Markdown]]
+[[deps.Markdown]]
 deps = ["Base64"]
 uuid = "d6f4376e-aef5-505a-96c1-9c027394607a"
 
-[[MbedTLS_jll]]
+[[deps.MbedTLS_jll]]
 deps = ["Artifacts", "Libdl"]
 uuid = "c8ffd9c3-330d-5841-b78e-0817d7145fa1"
+version = "2.28.0+0"
 
-[[MozillaCACerts_jll]]
+[[deps.MozillaCACerts_jll]]
 uuid = "14a3606d-f60d-562e-9121-12d972cd8159"
+version = "2022.2.1"
 
-[[NetworkOptions]]
+[[deps.NetworkOptions]]
 uuid = "ca575930-c2e3-43a9-ace4-1e988b2c1908"
+version = "1.2.0"
+
+[[deps.OpenBLAS_jll]]
+deps = ["Artifacts", "CompilerSupportLibraries_jll", "Libdl"]
+uuid = "4536629a-c528-5b80-bd46-f80d51c5b363"
+version = "0.3.20+0"
 
-[[Pkg]]
+[[deps.Pkg]]
 deps = ["Artifacts", "Dates", "Downloads", "LibGit2", "Libdl", "Logging", "Markdown", "Printf", "REPL", "Random", "SHA", "Serialization", "TOML", "Tar", "UUIDs", "p7zip_jll"]
 uuid = "44cfe95a-1eb2-52ea-b672-e2afdf69b78f"
+version = "1.8.0"
 
-[[Preferences]]
+[[deps.Preferences]]
 deps = ["TOML"]
 git-tree-sha1 = "7eb1686b4f04b82f96ed7a4ea5890a4f0c7a09f1"
 uuid = "21216c6a-2e73-6563-6e65-726566657250"
 version = "1.4.0"
 
-[[Printf]]
+[[deps.Printf]]
 deps = ["Unicode"]
 uuid = "de0858da-6303-5e67-8744-51eddeeeb8d7"
 
-[[REPL]]
+[[deps.REPL]]
 deps = ["InteractiveUtils", "Markdown", "Sockets", "Unicode"]
 uuid = "3fa0cd96-eef1-5676-8a61-b3b8758bbffb"
 
-[[Random]]
-deps = ["Serialization"]
+[[deps.Random]]
+deps = ["SHA", "Serialization"]
 uuid = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 
-[[Reexport]]
+[[deps.Reexport]]
 git-tree-sha1 = "45e428421666073eab6f2da5c9d310d99bb12f9b"
 uuid = "189a3867-3050-52da-a836-e630ba90ab69"
 version = "1.2.2"
 
-[[Requires]]
+[[deps.Requires]]
 deps = ["UUIDs"]
 git-tree-sha1 = "838a3a4188e2ded87a4f9f184b4b0d78a1e91cb7"
 uuid = "ae029012-a4dd-5104-9daa-d747884805df"
 version = "1.3.0"
 
-[[SHA]]
+[[deps.SHA]]
 uuid = "ea8e919c-243c-51af-8825-aaa63cd721ce"
+version = "0.7.0"
 
-[[Serialization]]
+[[deps.Serialization]]
 uuid = "9e88b42a-f829-5b0c-bbe9-9e923198166b"
 
-[[Sockets]]
+[[deps.Sockets]]
 uuid = "6462fe0b-24de-5631-8697-dd941f90decc"
 
-[[SparseArrays]]
+[[deps.SparseArrays]]
 deps = ["LinearAlgebra", "Random"]
 uuid = "2f01184e-e22b-5df5-ae63-d93ebab69eaf"
 
-[[Statistics]]
+[[deps.Statistics]]
 deps = ["LinearAlgebra", "SparseArrays"]
 uuid = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 
-[[TOML]]
+[[deps.TOML]]
 deps = ["Dates"]
 uuid = "fa267f1f-6049-4f14-aa54-33bafae1ed76"
+version = "1.0.0"
 
-[[Tar]]
+[[deps.Tar]]
 deps = ["ArgTools", "SHA"]
 uuid = "a4e569a6-e804-4fa4-b0f3-eef7a1d5b13e"
+version = "1.10.1"
 
-[[UUIDs]]
+[[deps.UUIDs]]
 deps = ["Random", "SHA"]
 uuid = "cf7118a7-6976-5b1a-9a39-7adc72f591a4"
 
-[[Unicode]]
+[[deps.Unicode]]
 uuid = "4ec0a83e-493e-50e2-b9ac-8f72acf5a8f5"
 
-[[Zlib_jll]]
+[[deps.Zlib_jll]]
 deps = ["Libdl"]
 uuid = "83775a58-1f1d-513f-b197-d71354ab007a"
+version = "1.2.12+3"
+
+[[deps.libblastrampoline_jll]]
+deps = ["Artifacts", "Libdl", "OpenBLAS_jll"]
+uuid = "8e850b90-86db-534c-a0d3-1478176c7d93"
+version = "5.1.1+0"
 
-[[nghttp2_jll]]
+[[deps.nghttp2_jll]]
 deps = ["Artifacts", "Libdl"]
 uuid = "8e850ede-7688-5339-a07c-302acd2aaf8d"
+version = "1.48.0+0"
 
-[[p7zip_jll]]
+[[deps.p7zip_jll]]
 deps = ["Artifacts", "Libdl"]
 uuid = "3f19e933-33d8-53b3-aaab-bd5110c3b7a0"
+version = "17.4.0+0"

Project.toml

@@ -18,4 +18,4 @@ Adapt = "2.0, 3.0"
 GPUArraysCore = "= 0.1.5"
 LLVM = "3.9, 4, 5, 6"
 Reexport = "1"
-julia = "1.6"
+julia = "1.8"

lib/JLArrays/src/JLArrays.jl

@@ -144,7 +144,8 @@ struct JLArray{T, N} <: AbstractGPUArray{T, N}
     dims::Dims{N}
 
     function JLArray{T,N}(data::Array{T, N}, dims::Dims{N}) where {T,N}
-        @assert isbitstype(T) "JLArray only supports bits types"
+        isbitstype(T) || error("JLArray only supports bits types")
+        # when supporting isbits-union types, use `Base.allocatedinline` here.
         new(data, dims)
     end
 end

src/host/broadcast.jl

@@ -21,7 +21,11 @@ backend(::Type{Base.RefValue{AT}}) where {AT<:AbstractGPUArray} = backend(AT)
 # but make sure we don't dispatch to the optimized copy method that directly indexes
 function Broadcast.copy(bc::Broadcasted{<:AbstractGPUArrayStyle{0}})
     ElType = Broadcast.combine_eltypes(bc.f, bc.args)
-    isbitstype(ElType) || error("Cannot broadcast function returning non-isbits $ElType.")
+    if ElType == Union{}
+        # using a Union{} eltype would fail early, during GPU array construction,
+        # so use Nothing instead to give the error a chance to be thrown dynamically.
+        ElType = Nothing
+    end
     dest = copyto!(similar(bc, ElType), bc)
     return @allowscalar dest[CartesianIndex()]  # 0D broadcast needs to unwrap results
 end
@@ -30,9 +34,10 @@ end
 # iteration (see, e.g., CUDA.jl#145)
 @inline function Broadcast.copy(bc::Broadcasted{<:AbstractGPUArrayStyle})
     ElType = Broadcast.combine_eltypes(bc.f, bc.args)
-    if !Base.isconcretetype(ElType)
-        error("""GPU broadcast resulted in non-concrete element type $ElType.
-                 This probably means that the function you are broadcasting contains an error or type instability.""")
+    if ElType == Union{}
+        # using a Union{} eltype would fail early, during GPU array construction,
+        # so use Nothing instead to give the error a chance to be thrown dynamically.
+        ElType = Nothing
     end
     copyto!(similar(bc, ElType), bc)
 end

src/host/indexing.jl

@@ -156,20 +156,6 @@ function findminmax(binop, xs::AnyGPUArray; init, dims)
         return t1
     end
 
-    @static if VERSION < v"1.7.0-DEV.119"
-    # before JuliaLang/julia#35316, isless/isgreated did not order NaNs last
-    function reduction(t1::Tuple{<:AbstractFloat,<:Any}, t2::Tuple{<:AbstractFloat,<:Any})
-        (x, i), (y, j) = t1, t2
-
-        isnan(x) && return t1
-        isnan(y) && return t2
-
-        binop(x, y) && return t2
-        x == y && return (x, min(i, j))
-        return t1
-    end
-    end
-
     if dims == Colon()
         res = mapreduce(tuple, reduction, xs, indices; init = (init, dummy_index))