Split CUDA.jl into several packages (#3058)

Most CUDA.jl functionality has been moved into CUDACore.jl, and the cuBLAS/cuFFT/cuSPARSE/...
library submodules have become proper packages that will be released in lockstep with CUDACore.jl

CUDA.jl has become a shim package that imports and re-export all functionality it previously had,
so this change should not be breaking.

Author: maleadt

.buildkite/pipeline.yml

@@ -3,6 +3,38 @@ steps:
   - group: ":julia: Julia"
     key: "julia"
     steps:
+      # Julia 1.10 doesn't support [sources] in Project.toml,
+      # so we need to manually dev workspace packages before instantiating.
+      - label: "Julia 1.10"
+        plugins:
+          - JuliaCI/julia#v1:
+              version: "1.10"
+          - JuliaCI/julia-coverage#v1:
+              dirs:
+                - src
+                - lib
+                - examples
+        agents:
+          queue: "juliagpu"
+          cuda: "*"
+        commands: |
+          echo -e "[CUDA_Runtime_jll]\nlocal = \"true\"" >LocalPreferences.toml
+          julia --project=. -e '
+            using Pkg
+            Pkg.develop([PackageSpec(; path) for path in (
+              "CUDACore",
+              "lib/cublas", "lib/cusparse", "lib/cusolver", "lib/cufft", "lib/curand",
+              "lib/cudnn", "lib/cutensor", "lib/cutensornet", "lib/custatevec"
+            )])
+            Pkg.test(; test_args=["--quickfail"])'
+        if: |
+          build.message =~ /\[only tests\]/ ||
+          build.message =~ /\[only julia\]/ ||
+          build.message !~ /\[only/ &&
+            build.message !~ /\[skip tests\]/ &&
+            build.message !~ /\[skip julia\]/
+        timeout_in_minutes: 90
+
       - label: "Julia {{matrix.julia}}"
         plugins:
           - JuliaCI/julia#v1:
@@ -30,7 +62,6 @@ steps:
         matrix:
           setup:
             julia:
-              - "1.10"
               - "1.11"
               - "1.12"
               - "1.13"
@@ -48,9 +79,9 @@ steps:
       - label: "CUDA {{matrix.cuda}}"
         plugins:
           - JuliaCI/julia#v1:
-              version: "1.11"
+              version: "1.12"
           - JuliaCI/julia-test#v1:
-              test_args: "--quickfail core base libraries"
+              test_args: "--quickfail core libraries"
           - JuliaCI/julia-coverage#v1:
               dirs:
                 - src
@@ -65,7 +96,7 @@ steps:
           build.message !~ /\[only/ && !build.pull_request.draft &&
             build.message !~ /\[skip tests\]/ &&
             build.message !~ /\[skip cuda\]/
-        timeout_in_minutes: 60
+        timeout_in_minutes: 120
         matrix:
           setup:
             cuda:
@@ -85,43 +116,6 @@ steps:
           unset LD_LIBRARY_PATH
           echo -e "[CUDA_Runtime_jll]\nversion = \"{{matrix.cuda}}\"" >LocalPreferences.toml
 
-  - group: ":nesting_dolls: Subpackages"
-    depends_on: "cuda"
-    steps:
-      - label: "{{matrix.package}} on CUDA {{matrix.cuda}}"
-        matrix:
-          setup:
-            cuda:
-              - "12.0"
-              - "13.0"
-            package:
-              - "cudnn"
-              - "cutensor"
-              - "custatevec"
-              - "cutensornet"
-        plugins:
-          - JuliaCI/julia#v1:
-              version: "1.12"
-          - JuliaCI/julia-test#v1:
-              project: "lib/{{matrix.package}}"
-          - JuliaCI/julia-coverage#v1:
-              dirs:
-                - src
-                - lib
-                - examples
-        agents:
-          queue: "juliagpu"
-          cuda: "*"
-        if: |
-          build.message =~ /\[only tests\]/ ||
-          build.message =~ /\[only subpackages\]/ ||
-          build.message !~ /\[only/ && !build.pull_request.draft &&
-            build.message !~ /\[skip tests\]/ &&
-            build.message !~ /\[skip subpackages\]/
-        timeout_in_minutes: 30
-        commands: |
-          unset LD_LIBRARY_PATH
-
   - group: ":telescope: Downstream"
     depends_on: "cuda"
     steps:
@@ -185,7 +179,7 @@ steps:
       - label: "Enzyme.jl"
         plugins:
           - JuliaCI/julia#v1:
-              version: "1.10" # XXX: Enzyme.jl is broken on 1.11
+              version: "1.11"
           - JuliaCI/julia-coverage#v1:
               dirs:
                 - src
@@ -250,12 +244,12 @@ steps:
           julia --project -e '
             using CUDA
             @assert !CUDA.functional()
-            @assert !isdefined(CUDA, :libcudart)
+            @assert !isdefined(CUDACore, :libcudart)
             CUDA.set_runtime_version!(v"11.6")'
           julia --project -e '
             using CUDA
             @assert !CUDA.functional()
-            @assert isdefined(CUDA, :libcudart)'
+            @assert isdefined(CUDACore, :libcudart)'
         agents:
           queue: "juliagpu"
           intel: "*"
@@ -270,9 +264,9 @@ steps:
       - label: "Compute sanitizer"
         plugins:
           - JuliaCI/julia#v1:
-              version: "1.11"
+              version: "1.12"
           - JuliaCI/julia-test#v1:
-              test_args: "--sanitize core base"
+              test_args: "--sanitize core"
           - JuliaCI/julia-coverage#v1:
               dirs:
                 - src
@@ -294,9 +288,9 @@ steps:
       - label: "Legacy memory allocator"
         plugins:
           - JuliaCI/julia#v1:
-              version: "1.11"
+              version: "1.12"
           - JuliaCI/julia-test#v1:
-              test_args: "--quickfail core base"
+              test_args: "--quickfail core"
           - JuliaCI/julia-coverage#v1:
               dirs:
                 - src
@@ -320,9 +314,9 @@ steps:
       - label: "CuArray with {{matrix.memory}} memory"
         plugins:
           - JuliaCI/julia#v1:
-              version: "1.11"
+              version: "1.12"
           - JuliaCI/julia-test#v1:
-              test_args: "--quickfail core base libraries"
+              test_args: "--quickfail core"
           - JuliaCI/julia-coverage#v1:
               dirs:
                 - src
@@ -352,9 +346,9 @@ steps:
       - label: "MultiGPU"
         plugins:
           - JuliaCI/julia#v1:
-              version: "1.11"
+              version: "1.12"
           - JuliaCI/julia-test#v1:
-              test_args: "--gpu=0,1 --quickfail core base libraries"
+              test_args: "--gpu=0,1 --quickfail core libraries/cublas/xt"
           - JuliaCI/julia-coverage#v1:
               dirs:
                 - src
@@ -380,7 +374,7 @@ steps:
   - label: ":books: Documentation"
     plugins:
       - JuliaCI/julia#v1:
-          version: "1.11"
+          version: "1.12"
     commands: |
       unset LD_LIBRARY_PATH
       julia --project -e '

CUDACore/Project.toml

@@ -0,0 +1,84 @@
+name = "CUDACore"
+uuid = "bd0ed864-bdfe-4181-a5ed-ce625a5fdea2"
+version = "6.0.0"
+
+[deps]
+Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"
+BFloat16s = "ab4f0b2a-ad5b-11e8-123f-65d77653426b"
+CEnum = "fa961155-64e5-5f13-b03f-caf6b980ea82"
+CUDA_Compiler_jll = "d1e2174e-dfdc-576e-b43e-73b79eb1aca8"
+CUDA_Driver_jll = "4ee394cb-3365-5eb0-8335-949819d2adfc"
+CUDA_Runtime_Discovery = "1af6417a-86b4-443c-805f-a4643ffb695f"
+CUDA_Runtime_jll = "76a88914-d11a-5bdc-97e0-2f5a05c973a2"
+Crayons = "a8cc5b0e-0ffa-5ad4-8c14-923d3ee1735f"
+ExprTools = "e2ba6199-217a-4e67-a87a-7c52f15ade04"
+GPUArrays = "0c68f7d7-f131-5f86-a1c3-88cf8149b2d7"
+GPUCompiler = "61eb1bfa-7361-4325-ad38-22787b887f55"
+GPUToolbox = "096a3bc2-3ced-46d0-87f4-dd12716f4bfc"
+KernelAbstractions = "63c18a36-062a-441e-b654-da1e3ab1ce7c"
+LLVM = "929cbde3-209d-540e-8aea-75f648917ca0"
+LLVMLoopInfo = "8b046642-f1f6-4319-8d3c-209ddc03c586"
+LazyArtifacts = "4af54fe1-eca0-43a8-85a7-787d91b784e3"
+Libdl = "8f399da3-3557-5675-b5ff-fb832c97cbdb"
+LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
+Logging = "56ddb016-857b-54e1-b83d-db4d58db5568"
+NVTX = "5da4648a-3479-48b8-97b9-01cb529c0a1f"
+Preferences = "21216c6a-2e73-6563-6e65-726566657250"
+PrettyTables = "08abe8d2-0d0c-5749-adfa-8a2ac140af0d"
+Printf = "de0858da-6303-5e67-8744-51eddeeeb8d7"
+Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
+Random123 = "74087812-796a-5b5d-8853-05524746bad3"
+RandomNumbers = "e6cf234a-135c-5ec9-84dd-332b85af5143"
+StaticArrays = "90137ffa-7385-5640-81b9-e52037218182"
+Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
+demumble_jll = "1e29f10c-031c-5a83-9565-69cddfc27673"
+
+[weakdeps]
+ChainRulesCore = "d360d2e6-b24c-11e9-a2a3-2a2ae2dbcce4"
+EnzymeCore = "f151be2c-9106-41f4-ab19-57ee4f262869"
+SpecialFunctions = "276daf66-3868-5448-9aa4-cd146d93841b"
+
+[extensions]
+ChainRulesCoreExt = "ChainRulesCore"
+EnzymeCoreExt = "EnzymeCore"
+SpecialFunctionsExt = "SpecialFunctions"
+
+[compat]
+Adapt = "4.4"
+BFloat16s = "0.5, 0.6"
+CEnum = "0.2, 0.3, 0.4, 0.5"
+CUDA_Compiler_jll = "0.3, 0.4"
+CUDA_Driver_jll = "13"
+CUDA_Runtime_Discovery = "1"
+CUDA_Runtime_jll = "0.21"
+ChainRulesCore = "1"
+Crayons = "4"
+EnzymeCore = "0.8.2"
+ExprTools = "0.1"
+GPUArrays = "11.4.1"
+GPUCompiler = "1.4"
+GPUToolbox = "1.1"
+KernelAbstractions = "0.9.38"
+LLVM = "9.3.1"
+LLVMLoopInfo = "1"
+LazyArtifacts = "1"
+Libdl = "1"
+LinearAlgebra = "1"
+Logging = "1"
+NVTX = "1"
+Preferences = "1"
+PrettyTables = "3"
+Printf = "1"
+Random = "1"
+Random123 = "1.2"
+RandomNumbers = "1.5.3"
+SpecialFunctions = "1.3, 2"
+StaticArrays = "1.9.8"
+Statistics = "1"
+demumble_jll = "1.3"
+julia = "1.10"
+
+[extras]
+ChainRulesCore = "d360d2e6-b24c-11e9-a2a3-2a2ae2dbcce4"
+EnzymeCore = "f151be2c-9106-41f4-ab19-57ee4f262869"
+SpecialFunctions = "276daf66-3868-5448-9aa4-cd146d93841b"

ext/ChainRulesCoreExt.jl CUDACore/ext/ChainRulesCoreExt.jlext/ChainRulesCoreExt.jl renamed to CUDACore/ext/ChainRulesCoreExt.jl

@@ -2,7 +2,7 @@
 
 module ChainRulesCoreExt
 
-using CUDA: CuArray
+using CUDACore: CuArray
 
 isdefined(Base, :get_extension) ? (import ChainRulesCore) : (import ..ChainRulesCore)