Merge origin/main into docs-fern-native-api

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Signed-off-by: Ramakrishna Prabhu <ramakrishnap@nvidia.com>

Author: ramakrishnap-nv

build.sh

@@ -418,8 +418,7 @@ fi
 if buildAll || hasArg cuopt; then
     cd "${REPODIR}"/python/cuopt
 
-    # $EXTRA_CMAKE_ARGS gets concatenated into a string with [*] and then we find/replace spaces with semi-colons
-    SKBUILD_CMAKE_ARGS="-DCMAKE_PREFIX_PATH=${INSTALL_PREFIX};-DCMAKE_LIBRARY_PATH=${LIBCUOPT_BUILD_DIR};-DCMAKE_CUDA_ARCHITECTURES=${CUOPT_CMAKE_CUDA_ARCHITECTURES};${EXTRA_CMAKE_ARGS[*]// /;}" \
+    SKBUILD_CMAKE_ARGS="-DCMAKE_PREFIX_PATH=${INSTALL_PREFIX};-DCMAKE_LIBRARY_PATH=${LIBCUOPT_BUILD_DIR};-DCMAKE_CUDA_ARCHITECTURES=${CUOPT_CMAKE_CUDA_ARCHITECTURES};$(IFS=';'; echo "${EXTRA_CMAKE_ARGS[*]}")" \
         python "${PYTHON_ARGS_FOR_INSTALL[@]}" .
 fi
 

ci/build_wheel_libcuopt.sh

@@ -17,17 +17,34 @@ fi
 # Install Boost and TBB
 bash ci/utils/install_boost_tbb.sh
 
-# Install libuuid (needed by cuopt_grpc_server)
+# Install libuuid and LLVM's OpenMP runtime
 if command -v dnf &> /dev/null; then
-    dnf install -y libuuid-devel
+    # LLVM Toolset is distributed as a module on Rocky/RHEL 8.
+    dnf module install -y llvm-toolset
+    dnf install -y libuuid-devel libomp-devel
 elif command -v apt-get &> /dev/null; then
-    apt-get update && apt-get install -y uuid-dev
+    apt-get update
+    apt-get install -y uuid-dev libomp-dev
 fi
 
 # Install Protobuf + gRPC (protoc + grpc_cpp_plugin)
 bash ci/utils/install_protobuf_grpc.sh
 
-export SKBUILD_CMAKE_ARGS="-DCUOPT_BUILD_WHEELS=ON;-DDISABLE_DEPRECATION_WARNING=ON"
+# Compile with GCC, but use LLVM libomp as the OpenMP runtime bundled in the wheel. Resolve the
+# versioned ELF library rather than an unversioned linker script or compiler-toolset indirection.
+LIBOMP_LIBRARY="$(
+    ldconfig -p |
+        awk '$1 ~ /^libomp\.so(\.[0-9]+)*$/ && !library { library = $NF }
+             END { print library }'
+)"
+if [[ "${LIBOMP_LIBRARY}" != /* || ! -f "${LIBOMP_LIBRARY}" ]]; then
+    echo "Could not resolve the LLVM OpenMP runtime: '${LIBOMP_LIBRARY}'" >&2
+    exit 1
+fi
+
+echo "Using LLVM OpenMP runtime: ${LIBOMP_LIBRARY}"
+
+export SKBUILD_CMAKE_ARGS="-DOpenMP_gomp_LIBRARY:FILEPATH=${LIBOMP_LIBRARY}"
 
 # OpenSSL 3 hints for libcuopt's own find_package(OpenSSL).
 #

cpp/CMakeLists.txt

@@ -186,7 +186,6 @@ list(APPEND CUOPT_CUDA_FLAGS -Xfatbin=-compress-all)
 if (CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 12.9 AND CMAKE_CUDA_COMPILER_VERSION VERSION_LESS 13.0)
     list(APPEND CUOPT_CUDA_FLAGS -Xfatbin=--compress-level=3)
 endif ()
-list(APPEND CUOPT_CUDA_FLAGS -fopenmp)
 
 # Add jobserver flags for parallel compilation if PARALLEL_LEVEL is set
 if (PARALLEL_LEVEL AND NOT "${PARALLEL_LEVEL}" STREQUAL "")
@@ -198,13 +197,14 @@ if (PARALLEL_LEVEL AND NOT "${PARALLEL_LEVEL}" STREQUAL "")
     endif ()
 endif ()
 
-# The MIP solver requires OpenMP to work
-find_package(OpenMP REQUIRED)
+# The MIP solver requires OpenMP for both C++ and CUDA host code.
+find_package(OpenMP REQUIRED COMPONENTS CXX CUDA)
 message(VERBOSE "cuOpt: OpenMP found in ${OpenMP_CXX_INCLUDE_DIRS}")
 
-# MPS/QPS parser supports compressed inputs via bzip2 and zlib
+# MPS/QPS parser supports compressed inputs via bzip2, zlib and lz4
 option(CUOPT_PARSER_WITH_BZIP2 "Build MPS parser with bzip2 decompression" ON)
 option(CUOPT_PARSER_WITH_ZLIB "Build MPS parser with zlib decompression" ON)
+option(CUOPT_PARSER_WITH_LZ4 "Build experimental fast MPS parser with LZ4 decompression" ON)
 if (CUOPT_PARSER_WITH_BZIP2)
     find_package(BZip2 REQUIRED)
     add_compile_definitions(MPS_PARSER_WITH_BZIP2)
@@ -213,6 +213,10 @@ if (CUOPT_PARSER_WITH_ZLIB)
     find_package(ZLIB REQUIRED)
     add_compile_definitions(MPS_PARSER_WITH_ZLIB)
 endif ()
+if (CUOPT_PARSER_WITH_LZ4)
+    # No headers or link target needed; the experimental reader loads one liblz4 symbol at runtime.
+    add_compile_definitions(MPS_PARSER_WITH_LZ4)
+endif ()
 
 # Debug options
 if (CMAKE_BUILD_TYPE MATCHES Debug)
@@ -250,16 +254,30 @@ else ()
     find_package(RAFT REQUIRED)
 endif ()
 
+rapids_cpm_find(simde 0.8.2
+        CPM_ARGS
+        GIT_REPOSITORY https://github.com/simd-everywhere/simde.git
+        GIT_TAG v0.8.2
+        GIT_SHALLOW TRUE
+        DOWNLOAD_ONLY TRUE
+)
+
+if (NOT TARGET simde::simde)
+    add_library(simde::simde INTERFACE IMPORTED GLOBAL)
+    set_target_properties(simde::simde
+            PROPERTIES INTERFACE_INCLUDE_DIRECTORIES "${simde_SOURCE_DIR}")
+endif ()
+
 FetchContent_Declare(
         papilo
-        GIT_REPOSITORY "https://github.com/scipopt/papilo.git"
+        GIT_REPOSITORY "https://github.com/nguidotti/papilo.git"
         # We would want to get the main branch. However, the main branch
         # does not have some of the presolvers and settings that we need
         # Mainly, probing and clique merging.
         # This is the reason we are using the development branch
         # from Oct 12, 2025. Once these changes are merged into the main branch,
         #we can switch to the main branch.
-        GIT_TAG "741a2b9c8155b249d6df574d758b4d97d4417520"
+        GIT_TAG "0cfea20c5655249174dbd04f0fb7bd6b1f6e9a0c"
         GIT_PROGRESS TRUE
         EXCLUDE_FROM_ALL
         SYSTEM
@@ -293,13 +311,13 @@ set(BUILD_SHARED_LIBS ${BUILD_SHARED_LIBS_SAVED})
 # dejavu - header-only graph automorphism library for MIP symmetry detection
 # https://github.com/markusa4/dejavu (header-only, skip its CMakeLists.txt)
 FetchContent_Declare(
-  dejavu
-  GIT_REPOSITORY "https://github.com/markusa4/dejavu.git"
-  GIT_TAG "v2.1"
-  GIT_PROGRESS TRUE
-  EXCLUDE_FROM_ALL
-  SYSTEM
-  SOURCE_SUBDIR _nonexistent
+        dejavu
+        GIT_REPOSITORY "https://github.com/markusa4/dejavu.git"
+        GIT_TAG "v2.1"
+        GIT_PROGRESS TRUE
+        EXCLUDE_FROM_ALL
+        SYSTEM
+        SOURCE_SUBDIR _nonexistent
 )
 FetchContent_MakeAvailable(dejavu)
 message(STATUS "dejavu (graph automorphism): ${dejavu_SOURCE_DIR}")
@@ -369,7 +387,7 @@ if (NOT SKIP_GRPC_BUILD)
 
     # Proto search paths: manual protos in src/grpc, generated data proto in src/grpc/codegen/generated
     set(PROTO_PATH_MANUAL "${CMAKE_CURRENT_SOURCE_DIR}/src/grpc")
-    set(PROTO_PATH_GEN    "${CMAKE_CURRENT_SOURCE_DIR}/src/grpc/codegen/generated")
+    set(PROTO_PATH_GEN "${CMAKE_CURRENT_SOURCE_DIR}/src/grpc/codegen/generated")
 
     # Generate C++ code from cuopt_remote_data.proto (auto-generated data definitions)
     set(DATA_PROTO_FILE "${PROTO_PATH_GEN}/cuopt_remote_data.proto")
@@ -436,23 +454,34 @@ if (BUILD_TESTS)
 endif ()
 
 set(CUOPT_SRC_FILES)
+set(MPS_FAST_SRC_FILES)
 add_subdirectory(src)
 
 # nvcc 13.0.3 ICE (signal 11) compiling sliding_window.cu with 7 GPU architectures;
 # --split-compile breaks the codegen into per-arch sub-jobs to avoid the crash
 set_source_files_properties(
-    ${CMAKE_CURRENT_SOURCE_DIR}/src/routing/local_search/sliding_window.cu
-    PROPERTIES COMPILE_OPTIONS "--split-compile=0")
+        ${CMAKE_CURRENT_SOURCE_DIR}/src/routing/local_search/sliding_window.cu
+        PROPERTIES COMPILE_OPTIONS "--split-compile=0")
 
 if (HOST_LINEINFO)
-    set_source_files_properties(${CUOPT_SRC_FILES} DIRECTORY ${CMAKE_SOURCE_DIR} PROPERTIES COMPILE_OPTIONS "-g1")
+    set_source_files_properties(${CUOPT_SRC_FILES} DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR} PROPERTIES COMPILE_OPTIONS "-g1")
 endif ()
 
+# Needed for the fast MPS parser, available on all x86-64-v3 compliant x86 CPUs (essentially since Haswell ~2013)
+if (CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|AMD64|amd64)$" AND
+        CMAKE_CXX_COMPILER_ID MATCHES "^(GNU|Clang|AppleClang)$")
+    set_property(SOURCE ${MPS_FAST_SRC_FILES} DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
+            APPEND PROPERTY COMPILE_OPTIONS "-mbmi2;-mavx2;-msse4.2")
+endif ()
+
+# TODO: figure out a set of flags for ARM that fits the range of CPUs we wish to support (neoverse?)
+# NEON should be universal on aarch64 and enough for our purposes (parsing) though
+
 # Apply -UNDEBUG only to solver source files (not gRPC infrastructure).
 # Must happen before gRPC files are appended to CUOPT_SRC_FILES.
 # Uses APPEND to preserve any existing per-file options (e.g. -g1 from HOST_LINEINFO).
 if (DEFINE_ASSERT)
-    set_property(SOURCE ${CUOPT_SRC_FILES} DIRECTORY ${CMAKE_SOURCE_DIR}
+    set_property(SOURCE ${CUOPT_SRC_FILES} DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
             APPEND PROPERTY COMPILE_OPTIONS "-UNDEBUG")
 endif ()
 
@@ -479,7 +508,7 @@ if (NOT SKIP_GRPC_BUILD)
     # The conda-forge abseil shared library is built with NDEBUG and does not
     # export that symbol (abseil-cpp#1624).  Without this, Debug builds fail
     # at runtime with "undefined symbol: absl::…::Mutex::Dtor".
-    set_property(SOURCE ${GRPC_INFRA_FILES} DIRECTORY ${CMAKE_SOURCE_DIR}
+    set_property(SOURCE ${GRPC_INFRA_FILES} DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
             APPEND PROPERTY COMPILE_OPTIONS "-DNDEBUG")
 endif (NOT SKIP_GRPC_BUILD)
 
@@ -495,8 +524,8 @@ set_target_properties(cuopt
 )
 
 target_compile_definitions(cuopt
-  PUBLIC "CUOPT_LOG_ACTIVE_LEVEL=RAPIDS_LOGGER_LOG_LEVEL_${LIBCUOPT_LOGGING_LEVEL}"
-  PUBLIC CUSPARSE_ENABLE_EXPERIMENTAL_API
+        PUBLIC "CUOPT_LOG_ACTIVE_LEVEL=RAPIDS_LOGGER_LOG_LEVEL_${LIBCUOPT_LOGGING_LEVEL}"
+        PUBLIC CUSPARSE_ENABLE_EXPERIMENTAL_API
 )
 
 target_compile_options(cuopt
@@ -528,8 +557,8 @@ target_include_directories(cuopt PRIVATE
 )
 
 target_include_directories(cuopt SYSTEM PRIVATE
-  "${pslp_SOURCE_DIR}/include"
-  "${dejavu_SOURCE_DIR}"
+        "${pslp_SOURCE_DIR}/include"
+        "${dejavu_SOURCE_DIR}"
 )
 
 target_include_directories(cuopt
@@ -605,11 +634,13 @@ target_link_libraries(cuopt
         ${CUDSS_LIB_FILE}
         PRIVATE
         ${CUOPT_PRIVATE_CUDA_LIBS}
+        simde::simde
+        OpenMP::OpenMP_CXX
+        OpenMP::OpenMP_CUDA
         $<$<BOOL:${CUOPT_ENABLE_GRPC}>:protobuf::libprotobuf>
         $<$<BOOL:${CUOPT_ENABLE_GRPC}>:gRPC::grpc++>
 )
 
-
 # ##################################################################################################
 # - generate tests --------------------------------------------------------------------------------
 if (BUILD_TESTS)

cpp/cuopt_cli.cpp

@@ -90,11 +90,13 @@ inline cuopt::init_logger_t dummy_logger(
  *                  .mps/.qps and their .gz/.bz2 variants → MPS parser;
  *                  anything else is rejected.
  * @param initial_solution_file Path to initial solution file in SOL format
+ * @param mps_reader MPS reader implementation selected by the CLI
  * @param settings Merged solver settings (config file loaded in main, then CLI overrides applied)
  */
 int run_single_file(const std::string& file_path,
                     const std::string& initial_solution_file,
                     bool solve_relaxation,
+                    cuopt::mathematical_optimization::io::mps_reader_type_t mps_reader,
                     cuopt::mathematical_optimization::solver_settings_t<int, double>& settings)
 {
   cuopt::init_logger_t log(settings.get_parameter<std::string>(CUOPT_LOG_FILE),
@@ -108,7 +110,8 @@ int run_single_file(const std::string& file_path,
   {
     CUOPT_LOG_INFO("Reading file %s", base_filename.c_str());
     try {
-      mps_data_model = cuopt::mathematical_optimization::io::read<int, double>(file_path);
+      mps_data_model =
+        cuopt::mathematical_optimization::io::read<int, double>(file_path, mps_reader);
     } catch (const std::logic_error& e) {
       CUOPT_LOG_ERROR("Parser exception: %s", e.what());
       parsing_failed = true;
@@ -287,8 +290,8 @@ int main(int argc, char* argv[])
   program.add_argument("filename")
     .help(
       "input problem file; format dispatched by extension (case-insensitive). "
-      "Supported: .lp, .mps, .qps and their .gz / .bz2 compressed variants "
-      "(e.g. .lp.gz, .mps.bz2, .qps.gz)")
+      "Supported: .lp, .mps, .qps and their .gz / .bz2 / .lz4 compressed variants "
+      "(e.g. .lp.gz, .mps.bz2, .qps.lz4).")
     .nargs(1)
     .required();
 
@@ -306,6 +309,14 @@ int main(int argc, char* argv[])
     .help("path to parameter config file (key = value format, supports all parameters)")
     .default_value(std::string(""));
 
+  program.add_argument("--mps-reader")
+    .help(
+      "MPS reader implementation: default uses the production parser; experimental-fast uses the "
+      "experimental SIMD parser for free-format LP/MIP/QP/QCQP (SOCP) .mps/.qps files and their "
+      ".gz/.bz2/.lz4 compressed variants")
+    .default_value(std::string("default"))
+    .choices("default", "experimental-fast");
+
   program.add_argument("--dump-hyper-params")
     .help("print hyper-parameters only in config file format and exit")
     .default_value(false)
@@ -406,6 +417,12 @@ int main(int argc, char* argv[])
   const auto initial_solution_file = program.get<std::string>("--initial-solution");
   const auto solve_relaxation      = program.get<bool>("--relaxation");
   const auto params_file           = program.get<std::string>("--params-file");
+  const auto mps_reader_arg        = program.get<std::string>("--mps-reader");
+
+  auto mps_reader = cuopt::mathematical_optimization::io::mps_reader_type_t::default_reader;
+  if (mps_reader_arg == "experimental-fast") {
+    mps_reader = cuopt::mathematical_optimization::io::mps_reader_type_t::fast_experimental;
+  }
 
   cuopt::mathematical_optimization::solver_settings_t<int, double> settings;
   try {
@@ -435,5 +452,5 @@ int main(int argc, char* argv[])
     RAFT_CUDA_TRY(cudaSetDevice(0));
   }
 
-  return run_single_file(file_name, initial_solution_file, solve_relaxation, settings);
+  return run_single_file(file_name, initial_solution_file, solve_relaxation, mps_reader, settings);
 }

cpp/include/cuopt/mathematical_optimization/constants.h

@@ -75,6 +75,7 @@
 #define CUOPT_MIP_ZERO_HALF_CUTS                   "mip_zero_half_cuts"
 #define CUOPT_MIP_STRONG_CHVATAL_GOMORY_CUTS       "mip_strong_chvatal_gomory_cuts"
 #define CUOPT_MIP_REDUCED_COST_STRENGTHENING       "mip_reduced_cost_strengthening"
+#define CUOPT_MIP_RINS                             "mip_rins"
 #define CUOPT_MIP_OBJECTIVE_STEP                   "mip_objective_step"
 #define CUOPT_MIP_CUT_CHANGE_THRESHOLD             "mip_cut_change_threshold"
 #define CUOPT_MIP_CUT_MIN_ORTHOGONALITY            "mip_cut_min_orthogonality"
@@ -127,9 +128,19 @@
 #define CUOPT_MIP_HYPER_DIVING_NODE_LIMIT             "mip_hyper_diving_node_limit"
 #define CUOPT_MIP_HYPER_DIVING_ITERATION_LIMIT_FACTOR "mip_hyper_diving_iteration_limit_factor"
 #define CUOPT_MIP_HYPER_DIVING_BACKTRACK_LIMIT        "mip_hyper_diving_backtrack_limit"
-/* @brief Show per-strategy diving symbol in logs (true) instead of a generic 'D' */
+/* @brief Show per-strategy diving symbol in logs instead of a generic 'D' */
 #define CUOPT_MIP_HYPER_DIVING_SHOW_TYPE "mip_hyper_diving_show_type"
 
+/* @brief Recursive sub-MIP (RINS) hyper-parameters */
+#define CUOPT_MIP_HYPER_SUBMIP_BASE_TARGET_FIXRATE   "mip_hyper_submip_base_target_fixrate"
+#define CUOPT_MIP_HYPER_SUBMIP_MIN_FIXRATE           "mip_hyper_submip_min_fixrate"
+#define CUOPT_MIP_HYPER_SUBMIP_MIN_FIXRATE_CAP       "mip_hyper_submip_min_fixrate_cap"
+#define CUOPT_MIP_HYPER_SUBMIP_TARGET_MIP_GAP        "mip_hyper_submip_target_mip_gap"
+#define CUOPT_MIP_HYPER_SUBMIP_NODE_LIMIT_BASE       "mip_hyper_submip_node_limit_base"
+#define CUOPT_MIP_HYPER_SUBMIP_MAX_LEVEL             "mip_hyper_submip_max_level"
+#define CUOPT_MIP_HYPER_SUBMIP_ITERATION_LIMIT_RATIO "mip_hyper_submip_iteration_limit_ratio"
+#define CUOPT_MIP_HYPER_SUBMIP_ENABLE_CPUFJ          "mip_hyper_submip_enable_cpufj"
+
 /* @brief MIP determinism mode constants */
 #define CUOPT_MODE_OPPORTUNISTIC 0
 #define CUOPT_MODE_DETERMINISTIC 1