nvmath-python 0.9.0__cp314-cp314-win_amd64.whl

nvmath/bindings/cycublasMp.pxd

@@ -0,0 +1,183 @@
+# Copyright (c) 2025-2026, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
+#
+# SPDX-License-Identifier: Apache-2.0
+#
+# This code was automatically generated with version 0.8.0, generator version 0.3.1.dev1303+g031f1197f. Do not modify it directly.
+# This layer exposes the C header to Cython as-is.
+
+from libc.stdint cimport int64_t, uint32_t, uint64_t, intptr_t
+
+from .cycublas cimport (cublasComputeType_t, cublasOperation_t, cublasFillMode_t, cublasSideMode_t,
+                        cublasDiagType_t,)
+
+###############################################################################
+# Types (structs, enums, ...)
+###############################################################################
+
+# enums
+ctypedef enum cublasMpStatus_t "cublasMpStatus_t":
+    CUBLASMP_STATUS_SUCCESS "CUBLASMP_STATUS_SUCCESS" = 0
+    CUBLASMP_STATUS_NOT_INITIALIZED "CUBLASMP_STATUS_NOT_INITIALIZED" = 1
+    CUBLASMP_STATUS_ALLOCATION_FAILED "CUBLASMP_STATUS_ALLOCATION_FAILED" = 2
+    CUBLASMP_STATUS_INVALID_VALUE "CUBLASMP_STATUS_INVALID_VALUE" = 3
+    CUBLASMP_STATUS_ARCHITECTURE_MISMATCH "CUBLASMP_STATUS_ARCHITECTURE_MISMATCH" = 4
+    CUBLASMP_STATUS_EXECUTION_FAILED "CUBLASMP_STATUS_EXECUTION_FAILED" = 5
+    CUBLASMP_STATUS_INTERNAL_ERROR "CUBLASMP_STATUS_INTERNAL_ERROR" = 6
+    CUBLASMP_STATUS_NOT_SUPPORTED "CUBLASMP_STATUS_NOT_SUPPORTED" = 7
+    _CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR "_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR" = -42
+
+ctypedef enum cublasMpGridLayout_t "cublasMpGridLayout_t":
+    CUBLASMP_GRID_LAYOUT_COL_MAJOR "CUBLASMP_GRID_LAYOUT_COL_MAJOR" = 0
+    CUBLASMP_GRID_LAYOUT_ROW_MAJOR "CUBLASMP_GRID_LAYOUT_ROW_MAJOR" = 1
+
+ctypedef enum cublasMpMatmulDescriptorAttribute_t "cublasMpMatmulDescriptorAttribute_t":
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_TRANSA "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_TRANSA" = 0
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_TRANSB "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_TRANSB" = 1
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_COMPUTE_TYPE "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_COMPUTE_TYPE" = 2
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_ALGO_TYPE "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_ALGO_TYPE" = 3
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_COMMUNICATION_SM_COUNT "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_COMMUNICATION_SM_COUNT" = 4
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_EPILOGUE "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_EPILOGUE" = 5
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_BIAS_POINTER "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_BIAS_POINTER" = 6
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_BIAS_BATCH_STRIDE "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_BIAS_BATCH_STRIDE" = 7
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_BIAS_DATA_TYPE "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_BIAS_DATA_TYPE" = 8
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_EPILOGUE_AUX_POINTER "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_EPILOGUE_AUX_POINTER" = 9
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_EPILOGUE_AUX_LD "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_EPILOGUE_AUX_LD" = 10
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_EPILOGUE_AUX_BATCH_STRIDE "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_EPILOGUE_AUX_BATCH_STRIDE" = 11
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_EPILOGUE_AUX_DATA_TYPE "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_EPILOGUE_AUX_DATA_TYPE" = 12
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_EPILOGUE_AUX_SCALE_POINTER "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_EPILOGUE_AUX_SCALE_POINTER" = 13
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_EPILOGUE_AUX_AMAX_POINTER "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_EPILOGUE_AUX_AMAX_POINTER" = 14
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_EPILOGUE_AUX_SCALE_MODE "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_EPILOGUE_AUX_SCALE_MODE" = 15
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_A_SCALE_POINTER "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_A_SCALE_POINTER" = 16
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_A_SCALE_MODE "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_A_SCALE_MODE" = 17
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_B_SCALE_POINTER "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_B_SCALE_POINTER" = 18
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_B_SCALE_MODE "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_B_SCALE_MODE" = 19
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_C_SCALE_POINTER "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_C_SCALE_POINTER" = 20
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_C_SCALE_MODE "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_C_SCALE_MODE" = 21
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_D_SCALE_POINTER "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_D_SCALE_POINTER" = 22
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_D_SCALE_MODE "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_D_SCALE_MODE" = 23
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_AMAX_D_POINTER "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_AMAX_D_POINTER" = 24
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_D_OUT_SCALE_POINTER "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_D_OUT_SCALE_POINTER" = 25
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_D_OUT_SCALE_MODE "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_D_OUT_SCALE_MODE" = 26
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_COMMUNICATION_TYPE "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_COMMUNICATION_TYPE" = 27
+    CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_BIAS_RESULT_SCHEME "CUBLASMP_MATMUL_DESCRIPTOR_ATTRIBUTE_BIAS_RESULT_SCHEME" = 28
+
+ctypedef enum cublasMpMatmulAlgoType_t "cublasMpMatmulAlgoType_t":
+    CUBLASMP_MATMUL_ALGO_TYPE_DEFAULT "CUBLASMP_MATMUL_ALGO_TYPE_DEFAULT" = 0
+    CUBLASMP_MATMUL_ALGO_TYPE_SPLIT_P2P "CUBLASMP_MATMUL_ALGO_TYPE_SPLIT_P2P" = 1
+    CUBLASMP_MATMUL_ALGO_TYPE_SPLIT_MULTICAST "CUBLASMP_MATMUL_ALGO_TYPE_SPLIT_MULTICAST" = 2
+    CUBLASMP_MATMUL_ALGO_TYPE_ATOMIC_P2P "CUBLASMP_MATMUL_ALGO_TYPE_ATOMIC_P2P" = 3
+    CUBLASMP_MATMUL_ALGO_TYPE_ATOMIC_MULTICAST "CUBLASMP_MATMUL_ALGO_TYPE_ATOMIC_MULTICAST" = 4
+    CUBLASMP_MATMUL_ALGO_TYPE_NO_OVERLAP "CUBLASMP_MATMUL_ALGO_TYPE_NO_OVERLAP" = 5
+
+ctypedef enum cublasMpMatmulEpilogue_t "cublasMpMatmulEpilogue_t":
+    CUBLASMP_MATMUL_EPILOGUE_DEFAULT "CUBLASMP_MATMUL_EPILOGUE_DEFAULT" = 0
+    CUBLASMP_MATMUL_EPILOGUE_ALLREDUCE "CUBLASMP_MATMUL_EPILOGUE_ALLREDUCE" = 1
+    CUBLASMP_MATMUL_EPILOGUE_RELU "CUBLASMP_MATMUL_EPILOGUE_RELU" = 2
+    CUBLASMP_MATMUL_EPILOGUE_RELU_AUX "CUBLASMP_MATMUL_EPILOGUE_RELU_AUX" = (CUBLASMP_MATMUL_EPILOGUE_RELU | 128)
+    CUBLASMP_MATMUL_EPILOGUE_BIAS "CUBLASMP_MATMUL_EPILOGUE_BIAS" = 4
+    CUBLASMP_MATMUL_EPILOGUE_RELU_BIAS "CUBLASMP_MATMUL_EPILOGUE_RELU_BIAS" = (CUBLASMP_MATMUL_EPILOGUE_RELU | CUBLASMP_MATMUL_EPILOGUE_BIAS)
+    CUBLASMP_MATMUL_EPILOGUE_RELU_AUX_BIAS "CUBLASMP_MATMUL_EPILOGUE_RELU_AUX_BIAS" = (CUBLASMP_MATMUL_EPILOGUE_RELU_AUX | CUBLASMP_MATMUL_EPILOGUE_BIAS)
+    CUBLASMP_MATMUL_EPILOGUE_DRELU "CUBLASMP_MATMUL_EPILOGUE_DRELU" = (8 | 128)
+    CUBLASMP_MATMUL_EPILOGUE_DRELU_BGRAD "CUBLASMP_MATMUL_EPILOGUE_DRELU_BGRAD" = (CUBLASMP_MATMUL_EPILOGUE_DRELU | 16)
+    CUBLASMP_MATMUL_EPILOGUE_GELU "CUBLASMP_MATMUL_EPILOGUE_GELU" = 32
+    CUBLASMP_MATMUL_EPILOGUE_GELU_AUX "CUBLASMP_MATMUL_EPILOGUE_GELU_AUX" = (CUBLASMP_MATMUL_EPILOGUE_GELU | 128)
+    CUBLASMP_MATMUL_EPILOGUE_GELU_BIAS "CUBLASMP_MATMUL_EPILOGUE_GELU_BIAS" = (CUBLASMP_MATMUL_EPILOGUE_GELU | CUBLASMP_MATMUL_EPILOGUE_BIAS)
+    CUBLASMP_MATMUL_EPILOGUE_GELU_AUX_BIAS "CUBLASMP_MATMUL_EPILOGUE_GELU_AUX_BIAS" = (CUBLASMP_MATMUL_EPILOGUE_GELU_AUX | CUBLASMP_MATMUL_EPILOGUE_BIAS)
+    CUBLASMP_MATMUL_EPILOGUE_DGELU "CUBLASMP_MATMUL_EPILOGUE_DGELU" = (64 | 128)
+    CUBLASMP_MATMUL_EPILOGUE_DGELU_BGRAD "CUBLASMP_MATMUL_EPILOGUE_DGELU_BGRAD" = (CUBLASMP_MATMUL_EPILOGUE_DGELU | 16)
+    CUBLASMP_MATMUL_EPILOGUE_BGRADA "CUBLASMP_MATMUL_EPILOGUE_BGRADA" = 256
+    CUBLASMP_MATMUL_EPILOGUE_BGRADB "CUBLASMP_MATMUL_EPILOGUE_BGRADB" = 512
+
+ctypedef enum cublasMpMatmulMatrixScale_t "cublasMpMatmulMatrixScale_t":
+    CUBLASMP_MATMUL_MATRIX_SCALE_SCALAR_FP32 "CUBLASMP_MATMUL_MATRIX_SCALE_SCALAR_FP32" = 0
+    CUBLASMP_MATMUL_MATRIX_SCALE_VEC16_UE4M3 "CUBLASMP_MATMUL_MATRIX_SCALE_VEC16_UE4M3" = 1
+    CUBLASMP_MATMUL_MATRIX_SCALE_VEC32_UE8M0 "CUBLASMP_MATMUL_MATRIX_SCALE_VEC32_UE8M0" = 2
+    CUBLASMP_MATMUL_MATRIX_SCALE_OUTER_VEC_FP32 "CUBLASMP_MATMUL_MATRIX_SCALE_OUTER_VEC_FP32" = 3
+    CUBLASMP_MATMUL_MATRIX_SCALE_VEC128_FP32 "CUBLASMP_MATMUL_MATRIX_SCALE_VEC128_FP32" = 4
+    CUBLASMP_MATMUL_MATRIX_SCALE_BLK128x128_FP32 "CUBLASMP_MATMUL_MATRIX_SCALE_BLK128x128_FP32" = 5
+
+ctypedef enum cublasMpEmulationStrategy_t "cublasMpEmulationStrategy_t":
+    CUBLASMP_EMULATION_STRATEGY_DEFAULT "CUBLASMP_EMULATION_STRATEGY_DEFAULT" = 0
+    CUBLASMP_EMULATION_STRATEGY_PERFORMANT "CUBLASMP_EMULATION_STRATEGY_PERFORMANT" = 1
+    CUBLASMP_EMULATION_STRATEGY_EAGER "CUBLASMP_EMULATION_STRATEGY_EAGER" = 2
+
+ctypedef enum cublasMpResultScheme_t "cublasMpResultScheme_t":
+    CUBLASMP_RESULT_SCHEME_DEFAULT "CUBLASMP_RESULT_SCHEME_DEFAULT" = 0
+    CUBLASMP_RESULT_SCHEME_PARTIAL "CUBLASMP_RESULT_SCHEME_PARTIAL" = 1
+    CUBLASMP_RESULT_SCHEME_FULL "CUBLASMP_RESULT_SCHEME_FULL" = 2
+
+
+# types
+cdef extern from *:
+    """
+    #include <driver_types.h>
+    #include <library_types.h>
+    #include <cuComplex.h>
+    """
+    ctypedef void* cudaStream_t 'cudaStream_t'
+    ctypedef int cudaDataType_t 'cudaDataType_t'
+    ctypedef int cudaDataType 'cudaDataType'
+    ctypedef int libraryPropertyType_t 'libraryPropertyType_t'
+    ctypedef int libraryPropertyType 'libraryPropertyType'
+
+    ctypedef struct cuComplex:
+        pass
+    ctypedef struct cuDoubleComplex:
+        pass
+
+
+ctypedef void* ncclComm_t 'ncclComm_t'
+ctypedef void* cublasMpHandle_t 'cublasMpHandle_t'
+ctypedef void* cublasMpGrid_t 'cublasMpGrid_t'
+ctypedef void* cublasMpMatrixDescriptor_t 'cublasMpMatrixDescriptor_t'
+ctypedef void* cublasMpMatmulDescriptor_t 'cublasMpMatmulDescriptor_t'
+ctypedef void (*cublasMpLoggerCallback_t 'cublasMpLoggerCallback_t')(
+    int logLevel,
+    const char* functionName,
+    const char* message
+)
+
+
+###############################################################################
+# Functions
+###############################################################################
+
+cdef cublasMpStatus_t cublasMpCreate(cublasMpHandle_t* handle, cudaStream_t stream) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpDestroy(cublasMpHandle_t handle) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpSetStream(cublasMpHandle_t handle, cudaStream_t stream) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpGetStream(cublasMpHandle_t handle, cudaStream_t* stream) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpSetEmulationStrategy(cublasMpHandle_t handle, cublasMpEmulationStrategy_t emulationStrategy) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpGetEmulationStrategy(cublasMpHandle_t handle, cublasMpEmulationStrategy_t* emulationStrategy) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpGetVersion(int* version) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpGridCreate(int64_t nprow, int64_t npcol, cublasMpGridLayout_t layout, ncclComm_t comm, cublasMpGrid_t* grid) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpGridDestroy(cublasMpGrid_t grid) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpBufferRegister(cublasMpGrid_t grid, void* ptr, size_t size) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpBufferDeregister(cublasMpGrid_t grid, void* ptr) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpMatrixDescriptorCreate(int64_t m, int64_t n, int64_t mb, int64_t nb, int64_t rsrc, int64_t csrc, int64_t lld, cudaDataType_t type, cublasMpGrid_t grid, cublasMpMatrixDescriptor_t* desc) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpMatrixDescriptorDestroy(cublasMpMatrixDescriptor_t desc) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpMatrixDescriptorInit(int64_t m, int64_t n, int64_t mb, int64_t nb, int64_t rsrc, int64_t csrc, int64_t lld, cudaDataType_t type, cublasMpGrid_t grid, cublasMpMatrixDescriptor_t desc) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpMatmulDescriptorCreate(cublasMpMatmulDescriptor_t* matmulDesc, cublasComputeType_t computeType) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpMatmulDescriptorDestroy(cublasMpMatmulDescriptor_t matmulDesc) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpMatmulDescriptorInit(cublasMpMatmulDescriptor_t matmulDesc, cublasComputeType_t computeType) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpMatmulDescriptorSetAttribute(cublasMpMatmulDescriptor_t matmulDesc, cublasMpMatmulDescriptorAttribute_t attr, const void* buf, size_t sizeInBytes) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpMatmulDescriptorGetAttribute(cublasMpMatmulDescriptor_t matmulDesc, cublasMpMatmulDescriptorAttribute_t attr, void* buf, size_t sizeInBytes, size_t* sizeWritten) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpTrsm_bufferSize(cublasMpHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo, cublasOperation_t trans, cublasDiagType_t diag, int64_t m, int64_t n, const void* alpha, const void* a, int64_t ia, int64_t ja, cublasMpMatrixDescriptor_t descA, void* b, int64_t ib, int64_t jb, cublasMpMatrixDescriptor_t descB, cublasComputeType_t computeType, size_t* workspaceSizeInBytesOnDevice, size_t* workspaceSizeInBytesOnHost) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpTrsm(cublasMpHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo, cublasOperation_t trans, cublasDiagType_t diag, int64_t m, int64_t n, const void* alpha, const void* a, int64_t ia, int64_t ja, cublasMpMatrixDescriptor_t descA, void* b, int64_t ib, int64_t jb, cublasMpMatrixDescriptor_t descB, cublasComputeType_t computeType, void* d_work, size_t workspaceSizeInBytesOnDevice, void* h_work, size_t workspaceSizeInBytesOnHost) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpGemm_bufferSize(cublasMpHandle_t handle, cublasOperation_t transA, cublasOperation_t transB, int64_t m, int64_t n, int64_t k, const void* alpha, const void* a, int64_t ia, int64_t ja, cublasMpMatrixDescriptor_t descA, const void* b, int64_t ib, int64_t jb, cublasMpMatrixDescriptor_t descB, const void* beta, void* c, int64_t ic, int64_t jc, cublasMpMatrixDescriptor_t descC, cublasComputeType_t computeType, size_t* workspaceSizeInBytesOnDevice, size_t* workspaceSizeInBytesOnHost) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpGemm(cublasMpHandle_t handle, cublasOperation_t transA, cublasOperation_t transB, int64_t m, int64_t n, int64_t k, const void* alpha, const void* a, int64_t ia, int64_t ja, cublasMpMatrixDescriptor_t descA, const void* b, int64_t ib, int64_t jb, cublasMpMatrixDescriptor_t descB, const void* beta, void* c, int64_t ic, int64_t jc, cublasMpMatrixDescriptor_t descC, cublasComputeType_t computeType, void* d_work, size_t workspaceSizeInBytesOnDevice, void* h_work, size_t workspaceSizeInBytesOnHost) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpMatmul_bufferSize(cublasMpHandle_t handle, cublasMpMatmulDescriptor_t matmulDesc, int64_t m, int64_t n, int64_t k, const void* alpha, const void* a, int64_t ia, int64_t ja, cublasMpMatrixDescriptor_t descA, const void* b, int64_t ib, int64_t jb, cublasMpMatrixDescriptor_t descB, const void* beta, const void* c, int64_t ic, int64_t jc, cublasMpMatrixDescriptor_t descC, void* d, int64_t id, int64_t jd, cublasMpMatrixDescriptor_t descD, size_t* workspaceSizeInBytesOnDevice, size_t* workspaceSizeInBytesOnHost) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpMatmul(cublasMpHandle_t handle, cublasMpMatmulDescriptor_t matmulDesc, int64_t m, int64_t n, int64_t k, const void* alpha, const void* a, int64_t ia, int64_t ja, cublasMpMatrixDescriptor_t descA, const void* b, int64_t ib, int64_t jb, cublasMpMatrixDescriptor_t descB, const void* beta, const void* c, int64_t ic, int64_t jc, cublasMpMatrixDescriptor_t descC, void* d, int64_t id, int64_t jd, cublasMpMatrixDescriptor_t descD, void* d_work, size_t workspaceSizeInBytesOnDevice, void* h_work, size_t workspaceSizeInBytesOnHost) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpSyrk_bufferSize(cublasMpHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans, int64_t n, int64_t k, const void* alpha, const void* a, int64_t ia, int64_t ja, cublasMpMatrixDescriptor_t descA, const void* beta, void* c, int64_t ic, int64_t jc, cublasMpMatrixDescriptor_t descC, cublasComputeType_t computeType, size_t* workspaceSizeInBytesOnDevice, size_t* workspaceSizeInBytesOnHost) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpSyrk(cublasMpHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans, int64_t n, int64_t k, const void* alpha, const void* a, int64_t ia, int64_t ja, cublasMpMatrixDescriptor_t descA, const void* beta, void* c, int64_t ic, int64_t jc, cublasMpMatrixDescriptor_t descC, cublasComputeType_t computeType, void* d_work, size_t workspaceSizeInBytesOnDevice, void* h_work, size_t workspaceSizeInBytesOnHost) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef int64_t cublasMpNumroc(int64_t n, int64_t nb, uint32_t iproc, uint32_t isrcproc, uint32_t nprocs) except?-42 nogil
+cdef const char* cublasMpGetStatusString(cublasMpStatus_t status) except?NULL nogil
+cdef cublasMpStatus_t cublasMpGemr2D_bufferSize(cublasMpHandle_t handle, int64_t m, int64_t n, const void* a, int64_t ia, int64_t ja, cublasMpMatrixDescriptor_t descA, void* b, int64_t ib, int64_t jb, cublasMpMatrixDescriptor_t descB, size_t* workspaceSizeInBytesOnDevice, size_t* workspaceSizeInBytesOnHost, ncclComm_t global_comm) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpGemr2D(cublasMpHandle_t handle, int64_t m, int64_t n, const void* a, int64_t ia, int64_t ja, cublasMpMatrixDescriptor_t descA, void* b, int64_t ib, int64_t jb, cublasMpMatrixDescriptor_t descB, void* d_work, size_t workspaceSizeInBytesOnDevice, void* h_work, size_t workspaceSizeInBytesOnHost, ncclComm_t global_comm) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpTrmr2D_bufferSize(cublasMpHandle_t handle, cublasFillMode_t uplo, cublasDiagType_t diag, int64_t m, int64_t n, const void* a, int64_t ia, int64_t ja, cublasMpMatrixDescriptor_t descA, void* b, int64_t ib, int64_t jb, cublasMpMatrixDescriptor_t descB, size_t* workspaceSizeInBytesOnDevice, size_t* workspaceSizeInBytesOnHost, ncclComm_t global_comm) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpTrmr2D(cublasMpHandle_t handle, cublasFillMode_t uplo, cublasDiagType_t diag, int64_t m, int64_t n, const void* a, int64_t ia, int64_t ja, cublasMpMatrixDescriptor_t descA, void* b, int64_t ib, int64_t jb, cublasMpMatrixDescriptor_t descB, void* d_work, size_t workspaceSizeInBytesOnDevice, void* h_work, size_t workspaceSizeInBytesOnHost, ncclComm_t global_comm) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpGeadd_bufferSize(cublasMpHandle_t handle, cublasOperation_t trans, int64_t m, int64_t n, const void* alpha, const void* a, int64_t ia, int64_t ja, cublasMpMatrixDescriptor_t descA, const void* beta, void* c, int64_t ic, int64_t jc, cublasMpMatrixDescriptor_t descC, size_t* workspaceSizeInBytesOnDevice, size_t* workspaceSizeInBytesOnHost) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpGeadd(cublasMpHandle_t handle, cublasOperation_t trans, int64_t m, int64_t n, const void* alpha, const void* a, int64_t ia, int64_t ja, cublasMpMatrixDescriptor_t descA, const void* beta, void* c, int64_t ic, int64_t jc, cublasMpMatrixDescriptor_t descC, void* d_work, size_t workspaceSizeInBytesOnDevice, void* h_work, size_t workspaceSizeInBytesOnHost) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpTradd_bufferSize(cublasMpHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans, int64_t m, int64_t n, const void* alpha, const void* a, int64_t ia, int64_t ja, cublasMpMatrixDescriptor_t descA, const void* beta, void* c, int64_t ic, int64_t jc, cublasMpMatrixDescriptor_t descC, size_t* workspaceSizeInBytesOnDevice, size_t* workspaceSizeInBytesOnHost) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpTradd(cublasMpHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans, int64_t m, int64_t n, const void* alpha, const void* a, int64_t ia, int64_t ja, cublasMpMatrixDescriptor_t descA, const void* beta, void* c, int64_t ic, int64_t jc, cublasMpMatrixDescriptor_t descC, void* d_work, size_t workspaceSizeInBytesOnDevice, void* h_work, size_t workspaceSizeInBytesOnHost) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpMalloc(cublasMpGrid_t grid, void** ptr, size_t size) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cublasMpStatus_t cublasMpFree(cublasMpGrid_t grid, void* ptr) except?_CUBLASMPSTATUS_T_INTERNAL_LOADING_ERROR nogil

nvmath/bindings/cycudss.pxd

@@ -0,0 +1,224 @@
+# Copyright (c) 2025-2026, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
+#
+# SPDX-License-Identifier: Apache-2.0
+#
+# This code was automatically generated with version 0.7.0, generator version 0.3.1.dev1303+g031f1197f. Do not modify it directly.
+# This layer exposes the C header to Cython as-is.
+
+from libc.stdint cimport int64_t
+
+
+###############################################################################
+# Types (structs, enums, ...)
+###############################################################################
+
+# enums
+ctypedef enum cudssOpType_t "cudssOpType_t":
+    CUDSS_SUM "CUDSS_SUM"
+    CUDSS_MAX "CUDSS_MAX"
+    CUDSS_MIN "CUDSS_MIN"
+
+ctypedef enum cudssConfigParam_t "cudssConfigParam_t":
+    CUDSS_CONFIG_REORDERING_ALG "CUDSS_CONFIG_REORDERING_ALG"
+    CUDSS_CONFIG_FACTORIZATION_ALG "CUDSS_CONFIG_FACTORIZATION_ALG"
+    CUDSS_CONFIG_SOLVE_ALG "CUDSS_CONFIG_SOLVE_ALG"
+    CUDSS_CONFIG_USE_MATCHING "CUDSS_CONFIG_USE_MATCHING"
+    CUDSS_CONFIG_MATCHING_ALG "CUDSS_CONFIG_MATCHING_ALG"
+    CUDSS_CONFIG_SOLVE_MODE "CUDSS_CONFIG_SOLVE_MODE"
+    CUDSS_CONFIG_IR_N_STEPS "CUDSS_CONFIG_IR_N_STEPS"
+    CUDSS_CONFIG_IR_TOL "CUDSS_CONFIG_IR_TOL"
+    CUDSS_CONFIG_PIVOT_TYPE "CUDSS_CONFIG_PIVOT_TYPE"
+    CUDSS_CONFIG_PIVOT_THRESHOLD "CUDSS_CONFIG_PIVOT_THRESHOLD"
+    CUDSS_CONFIG_PIVOT_EPSILON "CUDSS_CONFIG_PIVOT_EPSILON"
+    CUDSS_CONFIG_MAX_LU_NNZ "CUDSS_CONFIG_MAX_LU_NNZ"
+    CUDSS_CONFIG_HYBRID_MODE "CUDSS_CONFIG_HYBRID_MODE"
+    CUDSS_CONFIG_HYBRID_DEVICE_MEMORY_LIMIT "CUDSS_CONFIG_HYBRID_DEVICE_MEMORY_LIMIT"
+    CUDSS_CONFIG_USE_CUDA_REGISTER_MEMORY "CUDSS_CONFIG_USE_CUDA_REGISTER_MEMORY"
+    CUDSS_CONFIG_HOST_NTHREADS "CUDSS_CONFIG_HOST_NTHREADS"
+    CUDSS_CONFIG_HYBRID_EXECUTE_MODE "CUDSS_CONFIG_HYBRID_EXECUTE_MODE"
+    CUDSS_CONFIG_PIVOT_EPSILON_ALG "CUDSS_CONFIG_PIVOT_EPSILON_ALG"
+    CUDSS_CONFIG_ND_NLEVELS "CUDSS_CONFIG_ND_NLEVELS"
+    CUDSS_CONFIG_UBATCH_SIZE "CUDSS_CONFIG_UBATCH_SIZE"
+    CUDSS_CONFIG_UBATCH_INDEX "CUDSS_CONFIG_UBATCH_INDEX"
+    CUDSS_CONFIG_USE_SUPERPANELS "CUDSS_CONFIG_USE_SUPERPANELS"
+    CUDSS_CONFIG_DEVICE_COUNT "CUDSS_CONFIG_DEVICE_COUNT"
+    CUDSS_CONFIG_DEVICE_INDICES "CUDSS_CONFIG_DEVICE_INDICES"
+    CUDSS_CONFIG_SCHUR_MODE "CUDSS_CONFIG_SCHUR_MODE"
+    CUDSS_CONFIG_DETERMINISTIC_MODE "CUDSS_CONFIG_DETERMINISTIC_MODE"
+
+ctypedef enum cudssDataParam_t "cudssDataParam_t":
+    CUDSS_DATA_INFO "CUDSS_DATA_INFO"
+    CUDSS_DATA_LU_NNZ "CUDSS_DATA_LU_NNZ"
+    CUDSS_DATA_NPIVOTS "CUDSS_DATA_NPIVOTS"
+    CUDSS_DATA_INERTIA "CUDSS_DATA_INERTIA"
+    CUDSS_DATA_PERM_REORDER_ROW "CUDSS_DATA_PERM_REORDER_ROW"
+    CUDSS_DATA_PERM_REORDER_COL "CUDSS_DATA_PERM_REORDER_COL"
+    CUDSS_DATA_PERM_ROW "CUDSS_DATA_PERM_ROW"
+    CUDSS_DATA_PERM_COL "CUDSS_DATA_PERM_COL"
+    CUDSS_DATA_DIAG "CUDSS_DATA_DIAG"
+    CUDSS_DATA_USER_PERM "CUDSS_DATA_USER_PERM"
+    CUDSS_DATA_HYBRID_DEVICE_MEMORY_MIN "CUDSS_DATA_HYBRID_DEVICE_MEMORY_MIN"
+    CUDSS_DATA_COMM "CUDSS_DATA_COMM"
+    CUDSS_DATA_MEMORY_ESTIMATES "CUDSS_DATA_MEMORY_ESTIMATES"
+    CUDSS_DATA_PERM_MATCHING "CUDSS_DATA_PERM_MATCHING"
+    CUDSS_DATA_SCALE_ROW "CUDSS_DATA_SCALE_ROW"
+    CUDSS_DATA_SCALE_COL "CUDSS_DATA_SCALE_COL"
+    CUDSS_DATA_NSUPERPANELS "CUDSS_DATA_NSUPERPANELS"
+    CUDSS_DATA_USER_SCHUR_INDICES "CUDSS_DATA_USER_SCHUR_INDICES"
+    CUDSS_DATA_SCHUR_SHAPE "CUDSS_DATA_SCHUR_SHAPE"
+    CUDSS_DATA_SCHUR_MATRIX "CUDSS_DATA_SCHUR_MATRIX"
+    CUDSS_DATA_USER_ELIMINATION_TREE "CUDSS_DATA_USER_ELIMINATION_TREE"
+    CUDSS_DATA_ELIMINATION_TREE "CUDSS_DATA_ELIMINATION_TREE"
+    CUDSS_DATA_USER_HOST_INTERRUPT "CUDSS_DATA_USER_HOST_INTERRUPT"
+
+ctypedef enum cudssPhase_t "cudssPhase_t":
+    CUDSS_PHASE_REORDERING "CUDSS_PHASE_REORDERING" = (1 << 0)
+    CUDSS_PHASE_SYMBOLIC_FACTORIZATION "CUDSS_PHASE_SYMBOLIC_FACTORIZATION" = (1 << 1)
+    CUDSS_PHASE_ANALYSIS "CUDSS_PHASE_ANALYSIS" = (CUDSS_PHASE_REORDERING | CUDSS_PHASE_SYMBOLIC_FACTORIZATION)
+    CUDSS_PHASE_FACTORIZATION "CUDSS_PHASE_FACTORIZATION" = (1 << 2)
+    CUDSS_PHASE_REFACTORIZATION "CUDSS_PHASE_REFACTORIZATION" = (1 << 3)
+    CUDSS_PHASE_SOLVE_FWD_PERM "CUDSS_PHASE_SOLVE_FWD_PERM" = (1 << 4)
+    CUDSS_PHASE_SOLVE_FWD "CUDSS_PHASE_SOLVE_FWD" = (1 << 5)
+    CUDSS_PHASE_SOLVE_DIAG "CUDSS_PHASE_SOLVE_DIAG" = (1 << 6)
+    CUDSS_PHASE_SOLVE_BWD "CUDSS_PHASE_SOLVE_BWD" = (1 << 7)
+    CUDSS_PHASE_SOLVE_BWD_PERM "CUDSS_PHASE_SOLVE_BWD_PERM" = (1 << 8)
+    CUDSS_PHASE_SOLVE_REFINEMENT "CUDSS_PHASE_SOLVE_REFINEMENT" = (1 << 9)
+    CUDSS_PHASE_SOLVE "CUDSS_PHASE_SOLVE" = (((((CUDSS_PHASE_SOLVE_FWD_PERM | CUDSS_PHASE_SOLVE_FWD) | CUDSS_PHASE_SOLVE_DIAG) | CUDSS_PHASE_SOLVE_BWD) | CUDSS_PHASE_SOLVE_BWD_PERM) | CUDSS_PHASE_SOLVE_REFINEMENT)
+
+ctypedef enum cudssStatus_t "cudssStatus_t":
+    CUDSS_STATUS_SUCCESS "CUDSS_STATUS_SUCCESS" = 0
+    CUDSS_STATUS_NOT_INITIALIZED "CUDSS_STATUS_NOT_INITIALIZED" = 1
+    CUDSS_STATUS_ALLOC_FAILED "CUDSS_STATUS_ALLOC_FAILED" = 2
+    CUDSS_STATUS_INVALID_VALUE "CUDSS_STATUS_INVALID_VALUE" = 3
+    CUDSS_STATUS_NOT_SUPPORTED "CUDSS_STATUS_NOT_SUPPORTED" = 4
+    CUDSS_STATUS_EXECUTION_FAILED "CUDSS_STATUS_EXECUTION_FAILED" = 5
+    CUDSS_STATUS_INTERNAL_ERROR "CUDSS_STATUS_INTERNAL_ERROR" = 6
+    _CUDSSSTATUS_T_INTERNAL_LOADING_ERROR "_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR" = -42
+
+ctypedef enum cudssMatrixType_t "cudssMatrixType_t":
+    CUDSS_MTYPE_GENERAL "CUDSS_MTYPE_GENERAL"
+    CUDSS_MTYPE_SYMMETRIC "CUDSS_MTYPE_SYMMETRIC"
+    CUDSS_MTYPE_HERMITIAN "CUDSS_MTYPE_HERMITIAN"
+    CUDSS_MTYPE_SPD "CUDSS_MTYPE_SPD"
+    CUDSS_MTYPE_HPD "CUDSS_MTYPE_HPD"
+
+ctypedef enum cudssMatrixViewType_t "cudssMatrixViewType_t":
+    CUDSS_MVIEW_FULL "CUDSS_MVIEW_FULL"
+    CUDSS_MVIEW_LOWER "CUDSS_MVIEW_LOWER"
+    CUDSS_MVIEW_UPPER "CUDSS_MVIEW_UPPER"
+
+ctypedef enum cudssIndexBase_t "cudssIndexBase_t":
+    CUDSS_BASE_ZERO "CUDSS_BASE_ZERO"
+    CUDSS_BASE_ONE "CUDSS_BASE_ONE"
+
+ctypedef enum cudssLayout_t "cudssLayout_t":
+    CUDSS_LAYOUT_COL_MAJOR "CUDSS_LAYOUT_COL_MAJOR"
+    CUDSS_LAYOUT_ROW_MAJOR "CUDSS_LAYOUT_ROW_MAJOR"
+
+ctypedef enum cudssAlgType_t "cudssAlgType_t":
+    CUDSS_ALG_DEFAULT "CUDSS_ALG_DEFAULT"
+    CUDSS_ALG_1 "CUDSS_ALG_1"
+    CUDSS_ALG_2 "CUDSS_ALG_2"
+    CUDSS_ALG_3 "CUDSS_ALG_3"
+    CUDSS_ALG_4 "CUDSS_ALG_4"
+    CUDSS_ALG_5 "CUDSS_ALG_5"
+
+ctypedef enum cudssPivotType_t "cudssPivotType_t":
+    CUDSS_PIVOT_COL "CUDSS_PIVOT_COL"
+    CUDSS_PIVOT_ROW "CUDSS_PIVOT_ROW"
+    CUDSS_PIVOT_NONE "CUDSS_PIVOT_NONE"
+
+ctypedef enum cudssMatrixFormat_t "cudssMatrixFormat_t":
+    CUDSS_MFORMAT_DENSE "CUDSS_MFORMAT_DENSE" = 1
+    CUDSS_MFORMAT_CSR "CUDSS_MFORMAT_CSR" = 2
+    CUDSS_MFORMAT_BATCH "CUDSS_MFORMAT_BATCH" = 4
+    CUDSS_MFORMAT_DISTRIBUTED "CUDSS_MFORMAT_DISTRIBUTED" = 8
+
+
+# types
+cdef extern from *:
+    """
+    #include <driver_types.h>
+    #include <library_types.h>
+    #include <cuComplex.h>
+    """
+    ctypedef void* cudaStream_t 'cudaStream_t'
+    ctypedef int cudaDataType_t 'cudaDataType_t'
+    ctypedef int cudaDataType 'cudaDataType'
+    ctypedef int libraryPropertyType_t 'libraryPropertyType_t'
+    ctypedef int libraryPropertyType 'libraryPropertyType'
+
+    ctypedef struct cuComplex:
+        float x
+        float y
+    ctypedef struct cuDoubleComplex:
+        double x
+        double y
+
+
+ctypedef void* cudssHandle_t 'cudssHandle_t'
+ctypedef void* cudssMatrix_t 'cudssMatrix_t'
+ctypedef void* cudssData_t 'cudssData_t'
+ctypedef void* cudssConfig_t 'cudssConfig_t'
+ctypedef struct cudssDistributedInterface_t 'cudssDistributedInterface_t':
+    int (*cudssCommRank)(void*, int*)
+    int (*cudssCommSize)(void*, int*)
+    int (*cudssSend)(const void*, int, cudaDataType_t, int, int, void*, cudaStream_t)
+    int (*cudssRecv)(void*, int, cudaDataType_t, int, int, void*, cudaStream_t)
+    int (*cudssBcast)(void*, int, cudaDataType_t, int, void*, cudaStream_t)
+    int (*cudssReduce)(const void*, void*, int, cudaDataType_t, cudssOpType_t, int, void*, cudaStream_t)
+    int (*cudssAllreduce)(const void*, void*, int, cudaDataType_t, cudssOpType_t, void*, cudaStream_t)
+    int (*cudssScatterv)(const void*, const int*, const int*, cudaDataType_t, void*, int, cudaDataType_t, int, void*, cudaStream_t)
+    int (*cudssCommSplit)(const void*, int, int, void*)
+    int (*cudssCommFree)(void*)
+
+ctypedef struct cudssThreadingInterface_t 'cudssThreadingInterface_t':
+    int (*cudssGetMaxThreads)()
+    void (*cudssParallelFor)(int, int, void*, cudss_thr_func_t)
+
+ctypedef struct cudssDeviceMemHandler_t 'cudssDeviceMemHandler_t':
+    void* ctx
+    int (*device_alloc)(void*, void**, size_t, cudaStream_t)
+    int (*device_free)(void*, void*, size_t, cudaStream_t)
+    char name[64]
+
+
+
+###############################################################################
+# Functions
+###############################################################################
+
+cdef cudssStatus_t cudssConfigSet(cudssConfig_t config, cudssConfigParam_t param, void* value, size_t sizeInBytes) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssConfigGet(cudssConfig_t config, cudssConfigParam_t param, void* value, size_t sizeInBytes, size_t* sizeWritten) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssDataSet(cudssHandle_t handle, cudssData_t data, cudssDataParam_t param, void* value, size_t sizeInBytes) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssDataGet(cudssHandle_t handle, cudssData_t data, cudssDataParam_t param, void* value, size_t sizeInBytes, size_t* sizeWritten) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssExecute(cudssHandle_t handle, int phase, cudssConfig_t solverConfig, cudssData_t solverData, cudssMatrix_t inputMatrix, cudssMatrix_t solution, cudssMatrix_t rhs) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssSetStream(cudssHandle_t handle, cudaStream_t stream) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssSetCommLayer(cudssHandle_t handle, const char* commLibFileName) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssSetThreadingLayer(cudssHandle_t handle, const char* thrLibFileName) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssConfigCreate(cudssConfig_t* solverConfig) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssConfigDestroy(cudssConfig_t solverConfig) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssDataCreate(cudssHandle_t handle, cudssData_t* solverData) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssDataDestroy(cudssHandle_t handle, cudssData_t solverData) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssCreate(cudssHandle_t* handle) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssCreateMg(cudssHandle_t* handle_pt, int device_count, int* device_indices) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssDestroy(cudssHandle_t handle) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssGetProperty(libraryPropertyType propertyType, int* value) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssMatrixCreateDn(cudssMatrix_t* matrix, int64_t nrows, int64_t ncols, int64_t ld, void* values, cudaDataType_t valueType, cudssLayout_t layout) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssMatrixCreateCsr(cudssMatrix_t* matrix, int64_t nrows, int64_t ncols, int64_t nnz, void* rowStart, void* rowEnd, void* colIndices, void* values, cudaDataType_t indexType, cudaDataType_t valueType, cudssMatrixType_t mtype, cudssMatrixViewType_t mview, cudssIndexBase_t indexBase) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssMatrixCreateBatchDn(cudssMatrix_t* matrix, int64_t batchCount, void* nrows, void* ncols, void* ld, void** values, cudaDataType_t indexType, cudaDataType_t valueType, cudssLayout_t layout) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssMatrixCreateBatchCsr(cudssMatrix_t* matrix, int64_t batchCount, void* nrows, void* ncols, void* nnz, void** rowStart, void** rowEnd, void** colIndices, void** values, cudaDataType_t indexType, cudaDataType_t valueType, cudssMatrixType_t mtype, cudssMatrixViewType_t mview, cudssIndexBase_t indexBase) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssMatrixDestroy(cudssMatrix_t matrix) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssMatrixGetDn(cudssMatrix_t matrix, int64_t* nrows, int64_t* ncols, int64_t* ld, void** values, cudaDataType_t* type, cudssLayout_t* layout) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssMatrixGetCsr(cudssMatrix_t matrix, int64_t* nrows, int64_t* ncols, int64_t* nnz, void** rowStart, void** rowEnd, void** colIndices, void** values, cudaDataType_t* indexType, cudaDataType_t* valueType, cudssMatrixType_t* mtype, cudssMatrixViewType_t* mview, cudssIndexBase_t* indexBase) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssMatrixSetValues(cudssMatrix_t matrix, void* values) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssMatrixSetCsrPointers(cudssMatrix_t matrix, void* rowOffsets, void* rowEnd, void* colIndices, void* values) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssMatrixGetBatchDn(cudssMatrix_t matrix, int64_t* batchCount, void** nrows, void** ncols, void** ld, void*** values, cudaDataType_t* indexType, cudaDataType_t* valueType, cudssLayout_t* layout) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssMatrixGetBatchCsr(cudssMatrix_t matrix, int64_t* batchCount, void** nrows, void** ncols, void** nnz, void*** rowStart, void*** rowEnd, void*** colIndices, void*** values, cudaDataType_t* indexType, cudaDataType_t* valueType, cudssMatrixType_t* mtype, cudssMatrixViewType_t* mview, cudssIndexBase_t* indexBase) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssMatrixSetBatchValues(cudssMatrix_t matrix, void** values) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssMatrixSetBatchCsrPointers(cudssMatrix_t matrix, void** rowOffsets, void** rowEnd, void** colIndices, void** values) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssMatrixGetFormat(cudssMatrix_t matrix, int* format) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssMatrixSetDistributionRow1d(cudssMatrix_t matrix, int64_t first_row, int64_t last_row) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssMatrixGetDistributionRow1d(cudssMatrix_t matrix, int64_t* first_row, int64_t* last_row) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssGetDeviceMemHandler(cudssHandle_t handle, cudssDeviceMemHandler_t* handler) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil
+cdef cudssStatus_t cudssSetDeviceMemHandler(cudssHandle_t handle, const cudssDeviceMemHandler_t* handler) except?_CUDSSSTATUS_T_INTERNAL_LOADING_ERROR nogil