Clean Up Loading

.githash

@@ -1 +1 @@
-5f235a255f7b040ede53f151d0652d6352e087e9
+b1468726ba4827ab3e5ebb6e96ad94b9df78aa46

.github/workflows/developer.yml

@@ -41,50 +41,29 @@ jobs:
     # runs-on: [self-hosted, linux, x64]
     runs-on: ubuntu-latest
     container:
-      # need nvcc 
-      image: nvidia/cuda:12.4.0-devel-ubuntu22.04
-    #   options: --gpus all
+      image: ubuntu:22.04
     defaults:
       run:
         shell: bash
     env:
         LEGATE_AUTO_CONFIG: 0
+        NO_CUDA: ON
     steps:
       - name: Check out the repo
         uses: actions/checkout@v4
       - name: bash
         run: source /etc/profile
-      - name: Get package spec
-        id: pkg
-        run: |
-          if [[ -n "${{ inputs.tag }}" ]]; then
-            echo "ref=${{ inputs.tag }}" >> $GITHUB_OUTPUT
-            echo "name=${{ inputs.tag }}" >> $GITHUB_OUTPUT
-          elif [[ "${{ github.ref_type }}" == "tag" ]]; then
-            echo "ref=${{ github.ref_name }}" >> $GITHUB_OUTPUT
-            echo "name=${{ github.ref_name }}" >> $GITHUB_OUTPUT
-          elif [[ "${{ github.event_name }}" == "pull_request" ]]; then
-            echo "ref=${{ github.sha }}" >> $GITHUB_OUTPUT
-            echo "name=pr-${{ github.event.pull_request.number }}" >> $GITHUB_OUTPUT
-          elif [[ "${{ github.ref_type }}" == "branch" ]]; then
-            echo "ref=${{ github.ref_name }}" >> $GITHUB_OUTPUT
-            echo "name=${{ github.ref_name }}" >> $GITHUB_OUTPUT
-          else
-            echo "ref=${{ github.sha }}" >> $GITHUB_OUTPUT
-            echo "name=dev" >> $GITHUB_OUTPUT
-          fi
 
-          VERSION=$(grep "^version = " Project.toml | cut -d'"' -f2)
-          echo "version=$VERSION" >> $GITHUB_OUTPUT
+      - name: Install basic utils
+        run: |
+          apt-get update && apt-get install -y wget curl git build-essential
+          curl -fsSL https://deb.nodesource.com/setup_20.x | bash -
+          apt-get install -y nodejs
 
       - name: Setup Julia
         uses: julia-actions/setup-julia@v2
         with:
           version: '1.11'
-
-      - name: Install basic utils
-        run: apt-get update && apt-get install -y wget curl git build-essential
-
 
       - name: Install CMake 3.30.7
         run: |
@@ -98,7 +77,7 @@ jobs:
         id: julia-cache
         uses: julia-actions/cache@v2
         with:
-            cache-name: julia-developer-ci-test-new
+            cache-name: julia-developer-ci
       - name: Cleanup old cuNumeric and Legate
         run: | 
           julia --color=yes -e 'using Pkg; Pkg.rm("cuNumeric")' || true # ensure clean env
@@ -113,16 +92,15 @@ jobs:
               using LegatePreferences; LegatePreferences.use_developer_mode();
               Pkg.add(PackageSpec(url = "https://github.com/JuliaLegate/Legate.jl", rev = "main"))
           '
-          julia --color=yes -e 'using Pkg; Pkg.build("Legate")'
 
           julia --color=yes -e '
               using Pkg;
-              Pkg.add(PackageSpec(url = "https://github.com/JuliaLegate/cuNumeric.jl", rev = "${{ steps.pkg.outputs.ref }}", subdir = "lib/CNPreferences"))  
+              Pkg.develop(PackageSpec(path = "lib/CNPreferences"))
               using CNPreferences; CNPreferences.use_developer_mode();
-              Pkg.add(PackageSpec(url = "https://github.com/JuliaLegate/cuNumeric.jl", rev = "${{ steps.pkg.outputs.ref }}"))
+              Pkg.develop(PackageSpec(path = "."))
           '
           julia --color=yes -e 'using Pkg; Pkg.build("cuNumeric")'
 
       - name: Perform Test
         run: |
-          julia --color=yes -e 'using Pkg; Pkg.test("cuNumeric")'
+          GPUTESTS=0 julia --color=yes -e 'using Pkg; Pkg.test("cuNumeric")'

.github/workflows/docs.yml

@@ -66,22 +66,6 @@ jobs:
           julia --color=yes -e 'using Pkg; Pkg.rm("cuNumeric")' || true # ensure clean env
           julia --color=yes -e 'using Pkg; Pkg.rm("Legate")' || true # ensure clean env
 
-      - name: Setup CUDA runtime version
-        run: |
-          julia --color=yes --project=docs -e '
-              using Pkg
-              Pkg.add("CUDA")
-              using CUDA
-              CUDA.set_runtime_version!(v"12.4")
-          '
-      - name: Set LD_LIBRARY_PATH for CUDA_Driver_jll
-        run: |
-          echo "LD_LIBRARY_PATH=$(julia -e 'using Pkg; \
-            Pkg.add(name = "CUDA_Driver_jll", version = "0.12.1"); \
-            using CUDA_Driver_jll; \
-            print(joinpath(CUDA_Driver_jll.artifact_dir, "lib"))' \
-          ):$LD_LIBRARY_PATH" >> $GITHUB_ENV
-
       - name: Clone cuNumeric at ref
         run: |
           REF=${{ steps.pkg.outputs.ref }}
@@ -104,7 +88,6 @@ jobs:
         run: |
           julia --color=yes --project=docs -e '
               using Pkg
-              Pkg.add(PackageSpec(url = "https://github.com/JuliaLegate/Legate.jl", rev = "main", subdir = "lib/LegatePreferences"))
               Pkg.add(PackageSpec(url = "https://github.com/JuliaLegate/Legate.jl", rev = "main"))
               Pkg.develop(path = joinpath(homedir(), ".julia", "dev", "cuNumeric/lib/CNPreferences"))
               Pkg.develop(path = joinpath(homedir(), ".julia", "dev", "cuNumeric"))

Project.toml

@@ -4,10 +4,6 @@ version = "0.1.0"
 
 [deps]
 CNPreferences = "3e078157-ea10-49d5-bf32-908f777cd46f"
-CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
-CUDA_Driver_jll = "4ee394cb-3365-5eb0-8335-949819d2adfc"
-CUDA_Runtime_jll = "76a88914-d11a-5bdc-97e0-2f5a05c973a2"
-CUTENSOR_jll = "35b6c64b-1ee1-5834-92a3-3f624899209a"
 CodecZlib = "944b1d66-785c-5afd-91f1-9de20f533193"
 CxxWrap = "1f15a43c-97ca-5a2a-ae31-89f07a497df4"
 Downloads = "f43a241f-c20a-4ad4-852c-f6b1247861c6"
@@ -18,7 +14,6 @@ Libdl = "8f399da3-3557-5675-b5ff-fb832c97cbdb"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 MacroTools = "1914dd2f-81c6-5fcd-8719-6d5c9610ff09"
 OpenBLAS32_jll = "656ef2d0-ae68-5445-9ca0-591084a874a2"
-OpenSSL_jll = "458c3c95-2e84-50aa-8efc-19380b2a3a95"
 Pkg = "44cfe95a-1eb2-52ea-b672-e2afdf69b78f"
 Preferences = "21216c6a-2e73-6563-6e65-726566657250"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
@@ -29,14 +24,20 @@ cunumeric_jl_wrapper_jll = "49048992-29d2-5fd1-994f-9cecf112d624"
 cupynumeric_jll = "2862d674-414d-5b0b-a494-b21f8deca547"
 libcxxwrap_julia_jll = "3eaa8342-bff7-56a5-9981-c04077f7cee7"
 
+[weakdeps]
+CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
+
 [compat]
-CUDA = "5.9"
 CNPreferences = "0.1.2"
 CxxWrap = "0.17"
 Legate = "0.1.0"
 LegatePreferences = "0.1.5"
 MacroTools = "0.5.16"
+OpenBLAS32_jll = "0.3"
 StatsBase = "0.34"
-cunumeric_jl_wrapper_jll = "25.10.0"
-cupynumeric_jll = "25.10.0"
+cunumeric_jl_wrapper_jll = "25.10.1"
+cupynumeric_jll = "25.10.1"
 julia = "1.10"
+
+[ext]
+CUDAExt = "CUDA"

deps/build.jl

@@ -114,7 +114,7 @@ function build_jlcxxwrap(repo_root)
 end
 
 function build_cpp_wrapper(
-    repo_root, cupynumeric_loc, legate_loc, blas_lib, install_root
+    repo_root, cupynumeric_loc, legate_loc, blas_loc, install_root
 )
     @info "libcunumeric_jl_wrapper: Building C++ Wrapper Library"
     if isdir(install_root)
@@ -126,7 +126,7 @@ function build_cpp_wrapper(
     build_cpp_wrapper = joinpath(repo_root, "scripts/build_cpp_wrapper.sh")
     nthreads = Threads.nthreads()
 
-    bld_command = `$build_cpp_wrapper $repo_root $cupynumeric_loc $legate_loc $blas_lib $install_root $nthreads`
+    bld_command = `$build_cpp_wrapper $repo_root $cupynumeric_loc $legate_loc $blas_loc $install_root $nthreads`
 
     # write out a bash script for debugging
     cmd_str = join(bld_command.exec, " ")
@@ -197,7 +197,7 @@ function build(mode)
             )
         end
         build_cpp_wrapper(
-            pkg_root, cupynumeric_root, up_dir(legate_lib), blas_lib,
+            pkg_root, cupynumeric_root, up_dir(legate_lib), up_dir(blas_lib),
             install_lib,
         )
     end

ext/CUDAExt/CUDAExt.jl

@@ -0,0 +1,20 @@
+module CUDAExt
+
+using Random
+using CUDA
+using cuNumeric
+
+include("cuda.jl")
+
+function __init__()
+    if CUDA.functional()
+        # in cuda.jl to notify /wrapper/src/cuda.cpp about CUDA.jl kernel state size
+        cuNumeric.set_kernel_state_size();
+        # in /wrapper/src/cuda.cpp
+        cuNumeric.register_tasks();
+    else
+        @warn "CUDA.jl is not functional; skipping CUDA kernel registration."
+    end
+end
+
+end # module CUDAExt

src/cuda.jl → ext/CUDAExt/cuda.jl

@@ -1,9 +1,8 @@
-using CUDA
-using Random
+export @cuda_task, @launch
 
 const KERNEL_OFFSET = sizeof(CUDA.KernelState)
 
-# cuNumeric.jl init will call this
+# __init__ will call this
 # kernel_state is the first ARG in the generated PTX in CUDA.jl
 function set_kernel_state_size()
     cuNumeric.register_kernel_state_size(UInt64(KERNEL_OFFSET))

scripts/build_cpp_wrapper.sh

@@ -2,13 +2,13 @@ set -e
 
 # Check if exactly one argument is provided
 if [[ $# -ne 6 ]]; then
-    echo "Usage: $0 <cunumeric-pkg> <cupynumeric-root> <legate-root> <blas-lib> <install-dir> <nthreads>"
+    echo "Usage: $0 <cunumeric-pkg> <cupynumeric-root> <legate-root> <blas-root> <install-dir> <nthreads>"
     exit 1
 fi
 CUNUMERICJL_ROOT_DIR=$1 # this is the repo root of cunumeric.jl
 CUPYNUMERIC_ROOT_DIR=$2
 LEGATE_ROOT_DIR=$3
-BLAS_LIB_DIR=$4
+BLAS_ROOT_DIR=$4
 INSTALL_DIR=$5
 NTHREADS=$6
 
@@ -42,10 +42,14 @@ fi
 
 echo $LEGATE_ROOT_DIR
 
+# Default to OFF (CUDA support enabled), but allow override via environment variable
+NO_CUDA=${NO_CUDA:-OFF}
+
 if [[ ! -f "$BUILD_DIR/CMakeCache.txt" ]]; then
     echo "Configuring project..."
     cmake -S "$CUNUMERIC_WRAPPER_SOURCE" -B "$BUILD_DIR" \
         -D BINARYBUILDER=OFF \
+        -D NOCUDA=$NO_CUDA \
         -D CMAKE_INSTALL_PREFIX="$INSTALL_DIR" \
         -D CMAKE_PREFIX_PATH="$CUPYNUMERIC_ROOT_DIR;$LEGATE_ROOT_DIR;" \
         -D CUPYNUMERIC_PATH="$CUPYNUMERIC_ROOT_DIR" \