Add C++ decomposition with scipy_openblas32 (#995)

### Before submitting Please complete the following checklist when submitting a PR: - [ ] All new features must include a unit test. If you've fixed a bug or added code that should be tested, add a test to the [`tests`](../tests) directory! - [ ] All new functions and code must be clearly commented and documented. If you do make documentation changes, make sure that the docs build and render correctly by running `make docs`. - [ ] Ensure that the test suite passes, by running `make test`. - [x] Add a new entry to the `.github/CHANGELOG.md` file, summarizing the change, and including a link back to the PR. - [x] Ensure that code is properly formatted by running `make format`. When all the above are checked, delete everything above the dashed line and fill in the pull request template. ------------------------------------------------------------------------------------------------------------ **Context:** [sc-78187] & [sc-53754] The `blas/lapack` support is a long standing issue for `lightning`. The main challenge is the wheels building for the `windows` platform, which does not support a `fortran` compiler natively. Recently, the `scipy` team release the standalone `scipy-openblas32` wheel for `linux`, `windows` and `macos` platform. This could solve the win support issue. There are a few technical discussions on how to add the `scipy-openblas32` support to `lightning`: - Dynamic loading vs static linking? Dynamic loading was chosen and the `relative path` to the `libscipy_openblas` is added to the `runpath` of the shared lib object of installations from wheel. In this case, `scipy-openblas32` is required to be installed in the same `site-packages` dir. However, for the direct C++ usage as well as `pip editable install`, the absolute path to `libscipy_openblas` is still needed to ensure the `libscipy_openblas` lib can be found and loaded. - As a result, `scipy-openblas32` is a dependency of `lightning`. - This implementation does not support `conda-forge`. This PR also brings a few other changes: - `BLASLib` loading is handled by a singleton class `BLASLibLoaderManager`, instead of the `compute_diagonalizing_gates` API. - The `openblas` support is tested against `macos`, `linux` and `windows` for the C++ tests on CIs and integration with `catalyst` locally. - An additional `cmake option`: `PY_INSTALL` is added to identify if the pip installation is editable or not. This `PY_INSTALL` is `ON` if `not self.editable_mode`. **Description of the Change:** **Benefits:** **Possible Drawbacks:** **Related GitHub Issues:** --------- Co-authored-by: ringo-but-quantum <[email protected]> Co-authored-by: lillian542 <[email protected]> Co-authored-by: Ali Asadi <[email protected]>
PennyLaneAI · Dec 4, 2024 · 705fe89 · 705fe89
1 parent f9e8f62
commit 705fe89
Show file tree

Hide file tree

Showing 42 changed files with 297 additions and 344 deletions.
diff --git a/.github/CHANGELOG.md b/.github/CHANGELOG.md
@@ -17,13 +17,13 @@
 
 ### Breaking changes
 
-* Removed all instances of `qml.QubitStateVector`.
-  [(#985)](https://github.com/PennyLaneAI/pennylane-lightning/pull/985)
-
 * Handling for the legacy operator arithmetic (the `Hamiltonian` and `Tensor` classes in PennyLane) is removed.
   [(#994)](https://github.com/PennyLaneAI/pennylane-lightning/pull/994)
   [(#997)](https://github.com/PennyLaneAI/pennylane-lightning/pull/997)
 
+* Removed all instances of `qml.QubitStateVector`.
+  [(#985)](https://github.com/PennyLaneAI/pennylane-lightning/pull/985)
+
 ### Improvements
 
 * Update the logic for enabling `grad_on_execution` during device execution.
@@ -38,9 +38,20 @@
 * Enable N-controlled gate and matrix support to `lightning.gpu` simulator for Catalyst.
   [(#1005)](https://github.com/PennyLaneAI/pennylane-lightning/pull/1005)
 
+* Generalize seeding mechanism for all measurements.
+  [(#1003)](https://github.com/PennyLaneAI/pennylane-lightning/pull/1003)
+
+* `lightning.qubit`, `lightning.gpu`, and `lightning.kokkos` now define
+  the `eval_jaxpr` method for integration with the experimental
+  capture project.
+  [(#1002)](https://github.com/PennyLaneAI/pennylane-lightning/pull/1002)
+
 * Update Kokkos version support to 4.4.1 and enable Lightning-Kokkos[CUDA] C++ tests on CI.
   [(#1000)](https://github.com/PennyLaneAI/pennylane-lightning/pull/1000)
 
+* Add native BLAS support to the C++ layer via dynamic `scipy-openblas32` loading.
+  [(#995)](https://github.com/PennyLaneAI/pennylane-lightning/pull/995)
+
 * Update installation instruction for Lightning-GPU-MPI to explicitly ask users to add `path/to/libmpi.so` to the `LD_LIBRARY_PATH`. Update the runtime error message to ensure users know how to fix.
   [(#993)](https://github.com/PennyLaneAI/pennylane-lightning/pull/993)
 
@@ -53,14 +64,6 @@
 * Update the `lightning.kokkos` CUDA backend for compatibility with Catalyst.
   [(#942)](https://github.com/PennyLaneAI/pennylane-lightning/pull/942)
 
-* `lightning.qubit`, `lightning.gpu`, and `lightning.kokkos` now define
-  the `eval_jaxpr` method for integration with the experimental
-  capture project.
-  [(#1002)](https://github.com/PennyLaneAI/pennylane-lightning/pull/1002)
-
-* Generalize seeding mechanism for all measurements.
-  [(#1003)](https://github.com/PennyLaneAI/pennylane-lightning/pull/1003)
-
 ### Documentation
 
 * Update Lightning-Tensor installation docs and usage suggestions.

diff --git a/.github/workflows/tests_gpu_cpp.yml b/.github/workflows/tests_gpu_cpp.yml
@@ -124,7 +124,7 @@ jobs:
 
       - name: Install required basic packages
         run: |
-          python -m pip install ninja cmake scipy
+          python -m pip install ninja cmake scipy-openblas32
 
       - name: Install required lightning_gpu only packages
         if: matrix.pl_backend == 'lightning_gpu'
@@ -144,6 +144,7 @@ jobs:
             -DBUILD_TESTS=ON \
             -DENABLE_PYTHON=OFF \
             -DPL_BACKEND=${{ matrix.pl_backend }} \
+            -DSCIPY_OPENBLAS=$(python -c "import scipy_openblas32; print(scipy_openblas32.get_lib_dir())") \
             -DCMAKE_CXX_COMPILER=$(which g++-$GCC_VERSION) \
             -DENABLE_COVERAGE=ON \
             -G Ninja

diff --git a/.github/workflows/tests_lgpumpi_cpp.yml b/.github/workflows/tests_lgpumpi_cpp.yml
@@ -129,6 +129,7 @@ jobs:
         run: |
           source /etc/profile.d/modules.sh && module use /opt/modules/ && module load ${{ matrix.mpilib }}/cuda-${{ matrix.cuda_version_maj }}.${{ matrix.cuda_version_min }}
           export CUQUANTUM_SDK=$(python -c "import site; print( f'{site.getsitepackages()[0]}/cuquantum')")
+          export SCIPY_OPENBLAS=$(python -c "import scipy_openblas32; print(scipy_openblas32.get_lib_dir())")
           cmake . -BBuild \
             -DPL_BACKEND=lightning_gpu \
             -DENABLE_PYTHON=OFF \
@@ -140,6 +141,7 @@ jobs:
             -DCMAKE_CUDA_COMPILER=$(which nvcc) \
             -DCMAKE_CUDA_ARCHITECTURES="86" \
             -DCUQUANTUM_SDK=${CUQUANTUM_SDK} \
+            -DSCIPY_OPENBLAS=${SCIPY_OPENBLAS} \
             -DPython_EXECUTABLE:FILE="${{ steps.python_path.outputs.python }}" \
             -G Ninja
           cmake --build ./Build

diff --git a/.github/workflows/tests_linux_cpp.yml b/.github/workflows/tests_linux_cpp.yml
@@ -46,12 +46,11 @@ jobs:
         pl_backend: ["lightning_qubit"]
         enable_kernel_omp: ["OFF", "ON"]
         enable_kernel_avx_streaming: ["OFF", "ON"]
-        enable_lapack: ["OFF", "ON"]
         exclude:
         - enable_kernel_omp: OFF
           enable_kernel_avx_streaming: ON
     timeout-minutes: 60
-    name: C++ Tests (${{ matrix.pl_backend }}, ENABLE_KERNEL_OMP=${{ matrix.enable_kernel_omp }}, ENABLE_KERNEL_AVX_STREAMING=${{ matrix.enable_kernel_avx_streaming }}), ENABLE_LAPACK=${{ matrix.enable_lapack }})
+    name: C++ Tests (${{ matrix.pl_backend }}, ENABLE_KERNEL_OMP=${{ matrix.enable_kernel_omp }}, ENABLE_KERNEL_AVX_STREAMING=${{ matrix.enable_kernel_avx_streaming }})
     runs-on: pl-4-core-large-runner
 
     steps:
@@ -66,15 +65,15 @@ jobs:
       - name: Install dependencies
         run: |
             echo ${{ github.event_name }} && sudo apt-get update && sudo apt-get -y -q install cmake gcc-$GCC_VERSION  g++-$GCC_VERSION ninja-build gcovr lcov
-            python -m pip install scipy
+            python -m pip install scipy-openblas32
 
       - name: Build and run unit tests
         run: |
             cmake . -BBuild -G Ninja \
               -DCMAKE_BUILD_TYPE=Debug \
               -DBUILD_TESTS=ON \
               -DENABLE_PYTHON=OFF \
-              -DENABLE_LAPACK=${{ matrix.enable_lapack }} \
+              -DSCIPY_OPENBLAS=$(python -c "import scipy_openblas32; print(scipy_openblas32.get_lib_dir())") \
               -DPL_BACKEND=${{ matrix.pl_backend }} \
               -DCMAKE_CXX_COMPILER=$(which g++-$GCC_VERSION) \
               -DENABLE_COVERAGE=ON \
@@ -88,13 +87,13 @@ jobs:
             for file in *runner ; do ./$file --order lex --reporter junit --out ./tests/results/report_$file.xml; done;
             lcov --directory . -b ../pennylane_lightning/core/src --capture --output-file coverage.info
             lcov --remove coverage.info '/usr/*' --output-file coverage.info
-            mv coverage.info coverage-${{ github.job }}-${{ matrix.pl_backend }}-${{ matrix.enable_kernel_avx_streaming }}-${{ matrix.enable_kernel_omp }}-${{ matrix.enable_lapack }}.info
+            mv coverage.info coverage-${{ github.job }}-${{ matrix.pl_backend }}-${{ matrix.enable_kernel_avx_streaming }}-${{ matrix.enable_kernel_omp }}.info
 
       - name: Upload test results
         uses: actions/upload-artifact@v4
         if: always()
         with:
-          name: ubuntu-tests-reports-${{ github.job }}-${{ matrix.pl_backend }}-${{ matrix.enable_kernel_avx_streaming }}-${{ matrix.enable_kernel_omp }}-${{ matrix.enable_lapack }}
+          name: ubuntu-tests-reports-${{ github.job }}-${{ matrix.pl_backend }}-${{ matrix.enable_kernel_avx_streaming }}-${{ matrix.enable_kernel_omp }}
           retention-days: 1
           include-hidden-files: true
           path: |
@@ -105,8 +104,8 @@ jobs:
       - name: Upload code coverage results
         uses: actions/upload-artifact@v4
         with:
-          name: ubuntu-codecov-results-${{ matrix.pl_backend }}-${{ matrix.enable_kernel_avx_streaming }}-${{ matrix.enable_kernel_omp }}-${{ matrix.enable_lapack }}
-          path: ./Build/coverage-${{ github.job }}-${{ matrix.pl_backend }}-${{ matrix.enable_kernel_avx_streaming }}-${{ matrix.enable_kernel_omp }}-${{ matrix.enable_lapack }}.info
+          name: ubuntu-codecov-results-${{ matrix.pl_backend }}-${{ matrix.enable_kernel_avx_streaming }}-${{ matrix.enable_kernel_omp }}
+          path: ./Build/coverage-${{ github.job }}-${{ matrix.pl_backend }}-${{ matrix.enable_kernel_avx_streaming }}-${{ matrix.enable_kernel_omp }}.info
           if-no-files-found: error
 
   cpptestswithOpenBLAS:
@@ -130,7 +129,7 @@ jobs:
       - name: Install dependencies
         run: |
           sudo apt-get update && sudo apt-get -y -q install cmake gcc-$GCC_VERSION  g++-$GCC_VERSION libopenblas-base libopenblas-dev ninja-build gcovr lcov
-          python -m pip install scipy
+          python -m pip install scipy-openblas32
 
       - name: Build and run unit tests
         run: |
@@ -139,6 +138,7 @@ jobs:
               -DENABLE_PYTHON=OFF \
               -DENABLE_BLAS=ON \
               -DPL_BACKEND=${{ matrix.pl_backend }} \
+              -DSCIPY_OPENBLAS=$(python -c "import scipy_openblas32; print(scipy_openblas32.get_lib_dir())") \
               -DBUILD_TESTS=ON \
               -DCMAKE_CXX_COMPILER=$(which g++-$GCC_VERSION) \
               -DENABLE_COVERAGE=ON
@@ -215,7 +215,7 @@ jobs:
       - name: Install dependencies
         run: |
           sudo apt-get update && sudo apt-get -y -q install cmake gcc-$GCC_VERSION  g++-$GCC_VERSION ninja-build gcovr lcov
-          python -m pip install scipy
+          python -m pip install scipy-openblas32
 
       - name: Build and run unit tests
         run: |
@@ -226,6 +226,7 @@ jobs:
               -DCMAKE_PREFIX_PATH=${{ github.workspace }}/Kokkos \
               -DPL_BACKEND=${{ matrix.pl_backend }} \
               -DCMAKE_CXX_COMPILER=$(which g++-$GCC_VERSION) \
+              -DSCIPY_OPENBLAS=$(python -c "import scipy_openblas32; print(scipy_openblas32.get_lib_dir())") \
               -DENABLE_COVERAGE=ON
             cmake --build ./Build
             cd ./Build
@@ -314,7 +315,7 @@ jobs:
       - name: Install dependencies
         run: |
           sudo apt-get update && sudo apt-get -y -q install cmake gcc-$GCC_VERSION  g++-$GCC_VERSION ninja-build gcovr lcov
-          python -m pip install scipy
+          python -m pip install scipy-openblas32
 
       - name: Build and run unit tests
         run: |
@@ -324,6 +325,7 @@ jobs:
               -DENABLE_PYTHON=OFF \
               -DCMAKE_PREFIX_PATH=${{ github.workspace }}/Kokkos \
               -DPL_BACKEND="lightning_qubit;lightning_kokkos" \
+              -DSCIPY_OPENBLAS=$(python -c "import scipy_openblas32; print(scipy_openblas32.get_lib_dir())") \
               -DCMAKE_CXX_COMPILER=$(which g++-$GCC_VERSION)
             cmake --build ./Build
             cd ./Build

diff --git a/.github/workflows/tests_lkcpu_python.yml b/.github/workflows/tests_lkcpu_python.yml
@@ -106,7 +106,7 @@ jobs:
       - name: Install dependencies
         run: |
           sudo apt-get update && sudo apt-get -y -q install cmake gcc-$GCC_VERSION  g++-$GCC_VERSION
-          python -m pip install scipy wheel
+          python -m pip install scipy-openblas32 wheel
 
       - name: Get required Python packages
         run: |

diff --git a/.github/workflows/tests_lkcuda_cpp.yml b/.github/workflows/tests_lkcuda_cpp.yml
@@ -182,7 +182,9 @@ jobs:
 
 
       - name: Install dependencies
-        run: sudo apt-get update && sudo apt-get -y -q install cmake gcc-$GCC_VERSION  g++-$GCC_VERSION  ninja-build gcovr lcov
+        run: |
+            sudo apt-get update && sudo apt-get -y -q install cmake gcc-$GCC_VERSION  g++-$GCC_VERSION  ninja-build gcovr lcov
+            python -m pip install scipy-openblas32
 
       - name: Build and run unit tests
         run: |
@@ -191,6 +193,8 @@ jobs:
             -DCMAKE_BUILD_TYPE=RelWithDebInfo \
             -DBUILD_TESTS=ON \
             -DENABLE_PYTHON=OFF \
+            -DSCIPY_OPENBLAS=$(python -c "import scipy_openblas32; print(scipy_openblas32.get_lib_dir())") \
+            -DCMAKE_PREFIX_PATH=${{ github.workspace }}/Kokkos \
             -DPL_BACKEND=${{ matrix.pl_backend }} \
             -DCMAKE_CXX_COMPILER=$(which g++-$GCC_VERSION) \
             -DENABLE_COVERAGE=ON

diff --git a/.github/workflows/tests_lqcpu_python.yml b/.github/workflows/tests_lqcpu_python.yml
@@ -78,7 +78,7 @@ jobs:
       - name: Install dependencies
         run: |
           sudo apt-get update && sudo apt-get -y -q install cmake gcc-$GCC_VERSION g++-$GCC_VERSION libopenblas-base libopenblas-dev
-          python -m pip install scipy wheel
+          python -m pip install scipy-openblas32 wheel
 
       - name: Get required Python packages
         run: |

diff --git a/.github/workflows/tests_windows_cpp.yml b/.github/workflows/tests_windows_cpp.yml
@@ -124,13 +124,19 @@ jobs:
 
       - name: Install dependencies
         run: |
-          python -m pip install cmake build ninja scipy
+          python -m pip install cmake build ninja scipy-openblas32
 
       - name: Setup OpenCppCoverage and add to PATH
         run: |
           choco install OpenCppCoverage -y
           echo "C:\Program Files\OpenCppCoverage" >> $env:GITHUB_PATH
 
+      - name: Add path to scipy-openblas32 to environment variable
+        run: |
+          $scipy_openblas32_path = python -c "import scipy_openblas32; print(scipy_openblas32.get_lib_dir())"
+          echo "SCIPY_OPENBLAS32=${scipy_openblas32_path}" | Out-File -Append -FilePath $env:GITHUB_ENV
+          Get-Content $env:GITHUB_ENV
+
       - name: Build and run unit tests for code coverage
         run: |
             cmake -BBuild `
@@ -202,7 +208,7 @@ jobs:
 
       - name: Install dependencies
         run: |
-          python -m pip install cmake build ninja
+          python -m pip install cmake build ninja scipy-openblas32
 
       - name: Checkout PennyLane-Lightning
         uses: actions/checkout@v4
@@ -221,6 +227,12 @@ jobs:
         run: |
           choco install OpenCppCoverage -y
           echo "C:\Program Files\OpenCppCoverage" >> $env:GITHUB_PATH
+      
+      - name: Add path to scipy-openblas32 to environment variable
+        run: |
+          $scipy_openblas32_path = python -c "import scipy_openblas32; print(scipy_openblas32.get_lib_dir())"
+          echo "SCIPY_OPENBLAS32=${scipy_openblas32_path}" | Out-File -Append -FilePath $env:GITHUB_ENV
+          Get-Content $env:GITHUB_ENV
 
       - name: Build and run unit tests for code coverage
         run: |

diff --git a/.github/workflows/tidy.yml b/.github/workflows/tidy.yml
@@ -51,6 +51,7 @@ jobs:
             -DCLANG_TIDY_BINARY=$(dirname $(which python))/clang-tidy \
             -DBUILD_TESTS=ON \
             -DENABLE_WARNINGS=ON \
+            -DSCIPY_OPENBLAS=$(python -c "import scipy_openblas32; print(scipy_openblas32.get_lib_dir())") \
             -DPL_BACKEND=${{ matrix.pl_backend }} \
             -DCMAKE_CXX_COMPILER="$(which g++-11)" \
             -DCMAKE_C_COMPILER="$(which gcc-11)"

diff --git a/.github/workflows/wheel_linux_x86_64.yml b/.github/workflows/wheel_linux_x86_64.yml
@@ -151,7 +151,7 @@ jobs:
           # Python build settings
           CIBW_BEFORE_BUILD: |
             cat /etc/dnf.conf | sed "s/\[main\]/\[main\]\ntimeout=5/g" > /etc/dnf.conf
-            python -m pip install ninja cmake~=3.27.0 scipy
+            python -m pip install ninja cmake~=3.27.0
             dnf clean all -y
             dnf install gcc-toolset-13 -y
             source /opt/rh/gcc-toolset-13/enable -y

diff --git a/.github/workflows/wheel_macos_arm64.yml b/.github/workflows/wheel_macos_arm64.yml
@@ -96,7 +96,7 @@ jobs:
 
           # Python build settings
           CIBW_BEFORE_BUILD: |
-            python -m pip install pybind11 ninja cmake~=3.27.0 setuptools scipy
+            python -m pip install pybind11 ninja cmake~=3.27.0 setuptools
 
           CIBW_ENVIRONMENT: |
             CMAKE_ARGS="-DCMAKE_CXX_COMPILER_TARGET=arm64-apple-macos11 -DCMAKE_SYSTEM_NAME=Darwin -DCMAKE_SYSTEM_PROCESSOR=ARM64 -DENABLE_OPENMP=OFF -DCMAKE_BUILD_TYPE=Release"

diff --git a/.github/workflows/wheel_macos_x86_64.yml b/.github/workflows/wheel_macos_x86_64.yml
@@ -145,7 +145,7 @@ jobs:
 
           # Python build settings
           CIBW_BEFORE_BUILD: |
-            python -m pip install pybind11 ninja cmake~=3.27.0 setuptools scipy
+            python -m pip install pybind11 ninja cmake~=3.27.0 setuptools
 
           CIBW_ENVIRONMENT: |
             CMAKE_ARGS="-DCMAKE_BUILD_TYPE=Release"

diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -47,8 +47,6 @@ option(ENABLE_WARNINGS "Enable warnings" ON)
 option(ENABLE_NATIVE "Enable native CPU build tuning" OFF)
 option(ENABLE_PYTHON "Enable compilation of the Python module" ON)
 
-option(ENABLE_LAPACK "Enable compilation with scipy/LAPACK" OFF)
-
 # OpenMP
 find_package(OpenMP)
 if (OpenMP_CXX_FOUND)
@@ -88,29 +86,6 @@ set(CMAKE_POLICY_DEFAULT_CMP0127 NEW) # To suppress pybind11 CMP0127 warning
 # Add pybind11
 include(FetchContent)
 
-if(ENABLE_LAPACK)
-    find_package(Python COMPONENTS Interpreter Development)
-    set(SCIPYLIBS ${Python_SITELIB})
-
-    if(CMAKE_SYSTEM_NAME MATCHES "Darwin")
-        set(SCIPYLIBS "/System/Library/Frameworks/Accelerate.framework/Versions/Current/Frameworks/vecLib.framework/libLAPACK.dylib")
-    elseif(CMAKE_SYSTEM_NAME MATCHES "Linux")
-        if(EXISTS ${SCIPYLIBS}/scipy.libs)
-            set(SCIPYLIBS ${SCIPYLIBS}/scipy.libs)
-        else()
-            # Fallback to the lib path of Python for `conda` support
-            set(SCIPYLIBS ${SCIPYLIBS}/../..)
-        endif()
-    elseif(CMAKE_SYSTEM_NAME MATCHES "Windows")
-    else()
-        message(FATAL_ERROR "Unsupported platform: ${CMAKE_SYSTEM_NAME}")
-    endif()
-
-    add_compile_definitions(SCIPY_LIBS_PATH="${SCIPYLIBS}")
-
-    message(STATUS "Python scipy-lib path: ${SCIPYLIBS}")
-endif()
-
 if(ENABLE_PYTHON)
     find_package(Python COMPONENTS Interpreter Development)
     FetchContent_Declare(pybind11
@@ -123,6 +98,23 @@ endif()
 # Print Python site-packages directory for reference
 message("Python site-packages directory: ${Python_SITELIB}")
 
+if(DEFINED PY_INSTALL)
+    # Note the following setting is only for pyenv and not for conda
+    # TODO: Add support for conda
+    set(SCIPY_OPENBLAS32_RUNTIME_LIB_PATH "$ORIGIN/../scipy_openblas32/lib")
+else()
+    include("${CMAKE_CURRENT_SOURCE_DIR}/cmake/FindSCIPY_OPENBLAS32.cmake")
+    if(NOT DEFINED SCIPY_OPENBLAS32_LIB_PATH AND NOT EXISTS ${SCIPY_OPENBLAS32_LIB_PATH})
+        set(SCIPY_OPENBLAS32_LIB_PATH "")
+        find_path_to_openblas(SCIPY_OPENBLAS32_LIB_PATH)
+        add_compile_definitions(SCIPY_OPENBLAS32_LIB="${SCIPY_OPENBLAS32_LIB_PATH}")
+        message(STATUS "SCIPY_OPENBLAS32_LIB_PATH: ${SCIPY_OPENBLAS32_LIB_PATH}")
+    else()
+        add_compile_definitions(SCIPY_OPENBLAS32_LIB="${SCIPY_OPENBLAS32_LIB_PATH}")
+    endif()
+    set(SCIPY_OPENBLAS32_RUNTIME_LIB_PATH "${SCIPY_OPENBLAS32_LIB_PATH}")
+endif()
+
 set (CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR})
 
 # All CMakeLists.txt in subdirectories use pennylane_lightning_compile_options and pennylane_lightning_external_libs
@@ -154,10 +146,13 @@ if(ENABLE_PYTHON)
     message(STATUS "ENABLE_PYTHON is ON.")
     pybind11_add_module("${PL_BACKEND}_ops" "pennylane_lightning/core/src/bindings/Bindings.cpp")
 
-    # Allow pip installation of cuQuantum & CUDA 12 libs to be accessible without setting LD_LIBRARY_PATH for lightning_gpu
+    set(CMAKE_BUILD_RPATH_USE_ORIGIN ON)
     if("${PL_BACKEND}" STREQUAL "lightning_gpu" OR "${PL_BACKEND}" STREQUAL "lightning_tensor")
-        set(CMAKE_BUILD_RPATH_USE_ORIGIN ON)
-        set_target_properties("${PL_BACKEND}_ops" PROPERTIES BUILD_RPATH "$ORIGIN/../cuquantum/lib:$ORIGIN/../nvidia/cuda_runtime/lib:$ORIGIN/../nvidia/cublas/lib:$ORIGIN/../nvidia/cusparse/lib:$ORIGIN")
+        # Allow pip installation of cuQuantum & CUDA 12 libs to be accessible without setting LD_LIBRARY_PATH for lightning_gpu
+        # BUILD_RPATH only works for the last call
+        set_target_properties("${PL_BACKEND}_ops" PROPERTIES BUILD_RPATH "$ORIGIN/../cuquantum/lib:$ORIGIN/../nvidia/cuda_runtime/lib:$ORIGIN/../nvidia/cublas/lib:$ORIGIN/../nvidia/cusparse/lib:${SCIPY_OPENBLAS32_RUNTIME_LIB_PATH}:$ORIGIN")
+    else()
+        set_target_properties("${PL_BACKEND}_ops" PROPERTIES BUILD_RPATH "${SCIPY_OPENBLAS32_RUNTIME_LIB_PATH}")
     endif()
 
     target_link_libraries("${PL_BACKEND}_ops" PRIVATE   lightning_compile_options