Fix Python wheel installation issues (multiple subtlety linked items)

.github/workflows/publishing.yml

@@ -791,6 +791,46 @@ jobs:
             echo "Missing tests for nvidia-mgpu backend." && exit 1
           fi
 
+  wheel_validation:
+    name: Python wheel instructions validation
+    needs: cudaq_wheels
+
+    runs-on: linux-amd64-gpu-v100-latest-1
+
+    container:
+      image: ubuntu:22.04
+      options: --user root
+      env:
+        NVIDIA_VISIBLE_DEVICES: ${{ env.NVIDIA_VISIBLE_DEVICES }}
+        TERM: xterm
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.github_commit || needs.assets.outputs.github_commit }}
+
+      - name: Load wheel
+        uses: actions/download-artifact@v4
+        with:
+          name: x86_64-py3.10-wheels
+          path: /tmp/install
+
+      - name: Run simple validation
+        shell: bash
+        run: |
+          apt-get update && apt-get install -y --no-install-recommends ca-certificates vim wget openssh-client
+          cd /tmp
+          # Setup links for validate_wheel.sh script
+          ln -s $GITHUB_WORKSPACE/scripts/validate_wheel.sh .
+          ln -s $GITHUB_WORKSPACE/docs/sphinx/examples/python /tmp/examples
+          ln -s $GITHUB_WORKSPACE/docs/sphinx/snippets/python /tmp/snippets
+          ln -s $GITHUB_WORKSPACE/python/tests /tmp/tests
+          ln -s $GITHUB_WORKSPACE/python/README.md .
+          # Run the script
+          # FIXME - should we pass in "-q" to make it a quick test?
+          source validate_wheel.sh -w /tmp/install/cuda_quantum-*.whl -f /tmp -p 3.10
+
   image_validation:
     name: Docker image validation
     needs: [assets, config]

docs/sphinx/releases.rst

@@ -19,7 +19,7 @@ Check out our `documentation <https://nvidia.github.io/cuda-quantum/latest/using
 to get started with the new Python syntax support we have added, and `follow our blog <https://developer.nvidia.com/cuda-q>`__
 to learn more about the new setup and its performance benefits.
 
-- `Docker image <https://catalog.ngc.nvidia.com/orgs/nvidia/containers/cuda-quantum>`__
+- `Docker image <https://catalog.ngc.nvidia.com/orgs/nvidia/teams/nightly/containers/cuda-quantum/tags>`__ (use tag 0.7.0)
 - `Python wheel <https://pypi.org/project/cuda-quantum/>`__
 - `C++ installer <https://github.com/NVIDIA/cuda-quantum/releases>`__
 - `Documentation <https://nvidia.github.io/cuda-quantum/0.7.0>`__

python/README.md

@@ -41,7 +41,7 @@ Quantum with all its dependencies:
 conda create -y -n cuda-quantum python=3.10 pip
 conda install -y -n cuda-quantum -c "nvidia/label/cuda-11.8.0" cuda
 conda install -y -n cuda-quantum -c conda-forge mpi4py openmpi cxx-compiler
-conda env config vars set -n cuda-quantum LD_LIBRARY_PATH="$LD_LIBRARY_PATH:$CONDA_PREFIX/envs/cuda-quantum/lib"
+conda env config vars set -n cuda-quantum LD_LIBRARY_PATH="$CONDA_PREFIX/envs/cuda-quantum/lib:$LD_LIBRARY_PATH"
 conda env config vars set -n cuda-quantum MPI_PATH=$CONDA_PREFIX/envs/cuda-quantum
 conda run -n cuda-quantum pip install cuda-quantum
 conda activate cuda-quantum

python/cudaq/__init__.py

@@ -8,6 +8,24 @@
 
 import sys, os
 from ._packages import *
+
+# CUDAQ_DYNLIBS must be set before any other imports that would initialize
+# LinkedLibraryHolder.
+if not "CUDAQ_DYNLIBS" in os.environ:
+    try:
+        custatevec_libs = get_library_path("custatevec-cu11")
+        custatevec_path = os.path.join(custatevec_libs, "libcustatevec.so.1")
+
+        cutensornet_libs = get_library_path("cutensornet-cu11")
+        cutensornet_path = os.path.join(cutensornet_libs, "libcutensornet.so.2")
+
+        os.environ["CUDAQ_DYNLIBS"] = f"{custatevec_path}:{cutensornet_path}"
+    except:
+        import importlib.util
+        if not importlib.util.find_spec("cuda-quantum") is None:
+            print("Could not find a suitable cuQuantum Python package.")
+        pass
+
 from .kernel.kernel_decorator import kernel, PyKernelDecorator
 from .kernel.kernel_builder import make_kernel, QuakeValue, PyKernel
 from .kernel.ast_bridge import globalAstRegistry, globalKernelRegistry
@@ -98,21 +116,6 @@ def __clearKernelRegistries():
 from .kernels import uccsd
 from .dbg import ast
 
-if not "CUDAQ_DYNLIBS" in os.environ:
-    try:
-        custatevec_libs = get_library_path("custatevec-cu11")
-        custatevec_path = os.path.join(custatevec_libs, "libcustatevec.so.1")
-
-        cutensornet_libs = get_library_path("cutensornet-cu11")
-        cutensornet_path = os.path.join(cutensornet_libs, "libcutensornet.so.2")
-
-        os.environ["CUDAQ_DYNLIBS"] = f"{custatevec_path}:{cutensornet_path}"
-    except:
-        import importlib.util
-        if not importlib.util.find_spec("cuda-quantum") is None:
-            print("Could not find a suitable cuQuantum Python package.")
-        pass
-
 initKwargs = {}
 
 if '-target' in sys.argv:

scripts/validate_wheel.sh

@@ -15,14 +15,14 @@
 # Check the output for any tests that were skipped.
 
 # E.g. run the command 
-#   source validate.sh -w /tmp/cuda_quantum-*.whl -f /tmp -p 3.10 
+#   source validate_wheel.sh -w /tmp/cuda_quantum-*.whl -f /tmp -p 3.10 
 # in a container (with GPU support) defined by:
 #
 # ARG base_image=ubuntu:22.04
 # FROM ${base_image}
 # ARG cuda_quantum_wheel=cuda_quantum-0.6.0-cp310-cp310-manylinux_2_28_x86_64.whl
 # COPY $cuda_quantum_wheel /tmp/$cuda_quantum_wheel
-# COPY scripts/validate_wheel.sh validate.sh
+# COPY scripts/validate_wheel.sh validate_wheel.sh
 # COPY docs/sphinx/examples/python /tmp/examples/
 # COPY docs/sphinx/snippets/python /tmp/snippets/
 # COPY python/tests /tmp/tests/
@@ -32,12 +32,15 @@
 __optind__=$OPTIND
 OPTIND=1
 python_version=3.11
-while getopts ":f:p:w:" opt; do
+quick_test=false
+while getopts ":f:p:qw:" opt; do
   case $opt in
     f) root_folder="$OPTARG"
     ;;
     p) python_version="$OPTARG"
     ;;
+    q) quick_test=true
+    ;;
     w) cudaq_wheel="$OPTARG"
     ;;
     \?) echo "Invalid command line option -$OPTARG" >&2
@@ -85,6 +88,23 @@ while IFS= read -r line; do
 done <<< "$ompi_script"
 status_sum=0
 
+# Verify that the necessary GPU targets are installed and usable
+for tgt in nvidia nvidia-fp64 nvidia-mgpu tensornet; do
+    python3 -c "import cudaq; cudaq.set_target('${tgt}')"
+    if [ $? -ne 0 ]; then 
+        echo -e "\e[01;31mPython trivial test for target ${tgt} failed.\e[0m" >&2
+        status_sum=$((status_sum+1))
+    fi
+done
+
+# If this is a quick test, we return here.
+if $quick_test; then
+    if [ ! $status_sum -eq 0 ]; then
+        echo -e "\e[01;31mValidation produced errors.\e[0m" >&2
+    fi
+    (return 0 2>/dev/null) && return $status_sum || exit $status_sum
+fi
+
 # Run core tests
 echo "Running core tests."
 python3 -m pip install pytest numpy