Update lightning_kokkos_catalyst CUDA target to work with Catalyst

.github/CHANGELOG.md

@@ -46,6 +46,9 @@
 
 ### Improvements
 
+* Update `lightning_kokkos_catalyst` CUDA target to work with Catalyst.
+  [(#942)](https://github.com/PennyLaneAI/pennylane-lightning/pull/942)
+
 *  Fix PTM stable-latest.
   [(#961)](https://github.com/PennyLaneAI/pennylane-lightning/pull/961)
 
@@ -69,7 +72,7 @@
 
 * Add zero-state initialization to both `StateVectorCudaManaged` and `StateVectorCudaMPI` constructors to remove the `reset_state` in the python layer ctor and refactor `setBasisState(state, wires)` in the C++ layer.
   [(#933)](https://github.com/PennyLaneAI/pennylane-lightning/pull/933)
-  
+
 * The `generate_samples` methods of lightning.{qubit/kokkos} can now take in a seed number to make the generated samples deterministic. This can be useful when, among other things, fixing flaky tests in CI.
   [(#927)](https://github.com/PennyLaneAI/pennylane-lightning/pull/927)
 

pennylane_lightning/core/_version.py

@@ -16,4 +16,4 @@
    Version number (major.minor.patch[-label])
 """
 
-__version__ = "0.39.0-dev51"
+__version__ = "0.39.0-dev52"

pennylane_lightning/core/src/simulators/lightning_kokkos/catalyst/CMakeLists.txt

@@ -50,8 +50,8 @@ else()
     endforeach()
 
     # Fetching include hpp headers
-    set(INCLUDE_HEADERS DataView.hpp
-                        Exception.hpp
+    set(INCLUDE_HEADERS Exception.hpp
+                        DataView.hpp
                         QuantumDevice.hpp
                         RuntimeCAPI.h
                         Types.h
@@ -73,19 +73,40 @@ else()
 
 endif()
 
+if(Kokkos_ENABLE_CUDA)
+    message(STATUS "Kokkos_ENABLE_CUDA is ON")
+
+    # KOKKOS_CUDA_OPTIONS:
+    # - disable_malloc_async: Disable asynchronous memory allocation to ensure
+    #   that memory operations complete before proceeding.
+    # - enable_lambda: Allow the use of lambda expressions in parallel
+    #   execution policies, used in LK kernel functors, and this flag is
+    #   required when calling those functors from an external library.
+    #   It's supported starting from NVCC v7.0.
+    # - enable_constexpr: Enable compile-time evaluations for constants.
+    #   There are some undefined behaviour when executing end-to-end Catalyst
+    #   programs related to the value of `constexpr`s in the definition of gate kernels.
+    #   This flag should be enforced for NVCC v7.5+.
+    if(NVCC_VERSION VERSION_LESS "7.0")
+        message(WARNING "Building lightning_kokkos_catalyst without lambda and constexpr support.")
+        target_compile_definitions(lightning_kokkos_catalyst PRIVATE KOKKOS_CUDA_OPTIONS="disable_malloc_async")
+    elseif(NVCC_VERSION VERSION_LESS "7.5")
+        message(WARNING "Building lightning_kokkos_catalyst without constexpr support.")
+        target_compile_definitions(lightning_kokkos_catalyst PRIVATE KOKKOS_CUDA_OPTIONS="disable_malloc_async enable_lambda")
+    else()
+        target_compile_definitions(lightning_kokkos_catalyst PRIVATE
+            KOKKOS_CUDA_OPTIONS="disable_malloc_async enable_lambda enable_constexpr"
+        )
+    endif()
+endif()
+
 target_include_directories(lightning_kokkos_catalyst INTERFACE  ${CMAKE_CURRENT_SOURCE_DIR})
 target_link_libraries(lightning_kokkos_catalyst PUBLIC  lightning_compile_options
-                                                        lightning_external_libs
-                                                        lightning_base
-                                                        lightning_gates
-                                                        lightning_utils
-                                                        lightning_kokkos
                                                         lightning_kokkos_algorithms
-                                                        lightning_kokkos_gates
                                                         lightning_kokkos_measurements
-                                                        lightning_kokkos_utils
 )
 
+
 if (BUILD_TESTS)
     enable_testing()
     add_subdirectory("tests")

pennylane_lightning/core/src/simulators/lightning_kokkos/catalyst/LightningKokkosSimulator.hpp

@@ -18,8 +18,6 @@
 
 #pragma once
 
-#define __device_lightning_kokkos
-
 #include <bitset>
 #include <cmath>
 #include <cstdint>
@@ -62,7 +60,8 @@ class LightningKokkosSimulator final : public Catalyst::Runtime::QuantumDevice {
     Catalyst::Runtime::CacheManager<Kokkos::complex<double>> cache_manager{};
     bool tape_recording{false};
 
-    std::size_t device_shots;
+    // set default to avoid C++ tests segfaults in analytic mode
+    std::size_t device_shots{0};
 
     std::mt19937 *gen{nullptr};
 
@@ -99,13 +98,14 @@ class LightningKokkosSimulator final : public Catalyst::Runtime::QuantumDevice {
     auto GenerateSamples(size_t shots) -> std::vector<size_t>;
 
   public:
-    explicit LightningKokkosSimulator(const std::string &kwargs = "{}") {
+    explicit LightningKokkosSimulator(
+        const std::string &kwargs = "{}") noexcept {
         auto &&args = Catalyst::Runtime::parse_kwargs(kwargs);
         device_shots = args.contains("shots")
                            ? static_cast<std::size_t>(std::stoll(args["shots"]))
                            : 0;
     }
-    ~LightningKokkosSimulator() = default;
+    ~LightningKokkosSimulator() noexcept = default;
 
     LightningKokkosSimulator(const LightningKokkosSimulator &) = delete;
     LightningKokkosSimulator &