nvmath-python 0.5.0__cp313-cp313-win_amd64.whl

nvmath/bindings/cycudss.pxd

@@ -0,0 +1,191 @@
+# Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
+#
+# SPDX-License-Identifier: Apache-2.0
+#
+# This code was automatically generated with version 0.5.0. 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_MATCHING_TYPE "CUDSS_CONFIG_MATCHING_TYPE"
+    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"
+
+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"
+
+ctypedef enum cudssPhase_t "cudssPhase_t":
+    CUDSS_PHASE_ANALYSIS "CUDSS_PHASE_ANALYSIS" = 1
+    CUDSS_PHASE_FACTORIZATION "CUDSS_PHASE_FACTORIZATION" = 2
+    CUDSS_PHASE_REFACTORIZATION "CUDSS_PHASE_REFACTORIZATION" = 4
+    CUDSS_PHASE_SOLVE "CUDSS_PHASE_SOLVE" = 8
+    CUDSS_PHASE_SOLVE_FWD "CUDSS_PHASE_SOLVE_FWD" = 16
+    CUDSS_PHASE_SOLVE_DIAG "CUDSS_PHASE_SOLVE_DIAG" = 32
+    CUDSS_PHASE_SOLVE_BWD "CUDSS_PHASE_SOLVE_BWD" = 64
+
+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"
+
+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
+
+
+# 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, cudssPhase_t 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 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 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

nvmath/bindings/cycufft.pxd

@@ -0,0 +1,320 @@
+# Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
+#
+# SPDX-License-Identifier: Apache-2.0
+#
+# This code was automatically generated across versions from 11.0.3 to 12.8.0. Do not modify it directly.
+# This layer exposes the C header to Cython as-is.
+
+###############################################################################
+# Types (structs, enums, ...)
+###############################################################################
+
+# enums
+ctypedef enum libFormat_t "libFormat_t":
+    LIB_FORMAT_CUFFT "LIB_FORMAT_CUFFT" = 0x0
+    LIB_FORMAT_UNDEFINED "LIB_FORMAT_UNDEFINED" = 0x1
+
+ctypedef enum cufftResult "cufftResult":
+    CUFFT_SUCCESS "CUFFT_SUCCESS" = 0x0
+    CUFFT_INVALID_PLAN "CUFFT_INVALID_PLAN" = 0x1
+    CUFFT_ALLOC_FAILED "CUFFT_ALLOC_FAILED" = 0x2
+    CUFFT_INVALID_TYPE "CUFFT_INVALID_TYPE" = 0x3
+    CUFFT_INVALID_VALUE "CUFFT_INVALID_VALUE" = 0x4
+    CUFFT_INTERNAL_ERROR "CUFFT_INTERNAL_ERROR" = 0x5
+    CUFFT_EXEC_FAILED "CUFFT_EXEC_FAILED" = 0x6
+    CUFFT_SETUP_FAILED "CUFFT_SETUP_FAILED" = 0x7
+    CUFFT_INVALID_SIZE "CUFFT_INVALID_SIZE" = 0x8
+    CUFFT_UNALIGNED_DATA "CUFFT_UNALIGNED_DATA" = 0x9
+    CUFFT_INCOMPLETE_PARAMETER_LIST "CUFFT_INCOMPLETE_PARAMETER_LIST" = 0xA
+    CUFFT_INVALID_DEVICE "CUFFT_INVALID_DEVICE" = 0xB
+    CUFFT_PARSE_ERROR "CUFFT_PARSE_ERROR" = 0xC
+    CUFFT_NO_WORKSPACE "CUFFT_NO_WORKSPACE" = 0xD
+    CUFFT_NOT_IMPLEMENTED "CUFFT_NOT_IMPLEMENTED" = 0xE
+    CUFFT_LICENSE_ERROR "CUFFT_LICENSE_ERROR" = 0x0F
+    CUFFT_NOT_SUPPORTED "CUFFT_NOT_SUPPORTED" = 0x10
+    _CUFFTRESULT_INTERNAL_LOADING_ERROR "_CUFFTRESULT_INTERNAL_LOADING_ERROR" = -42
+
+ctypedef enum cufftType "cufftType":
+    CUFFT_R2C "CUFFT_R2C" = 0x2a
+    CUFFT_C2R "CUFFT_C2R" = 0x2c
+    CUFFT_C2C "CUFFT_C2C" = 0x29
+    CUFFT_D2Z "CUFFT_D2Z" = 0x6a
+    CUFFT_Z2D "CUFFT_Z2D" = 0x6c
+    CUFFT_Z2Z "CUFFT_Z2Z" = 0x69
+
+ctypedef enum cufftCompatibility "cufftCompatibility":
+    CUFFT_COMPATIBILITY_FFTW_PADDING "CUFFT_COMPATIBILITY_FFTW_PADDING" = 0x01
+
+ctypedef enum cufftXtSubFormat "cufftXtSubFormat":
+    CUFFT_XT_FORMAT_INPUT "CUFFT_XT_FORMAT_INPUT" = 0x00
+    CUFFT_XT_FORMAT_OUTPUT "CUFFT_XT_FORMAT_OUTPUT" = 0x01
+    CUFFT_XT_FORMAT_INPLACE "CUFFT_XT_FORMAT_INPLACE" = 0x02
+    CUFFT_XT_FORMAT_INPLACE_SHUFFLED "CUFFT_XT_FORMAT_INPLACE_SHUFFLED" = 0x03
+    CUFFT_XT_FORMAT_1D_INPUT_SHUFFLED "CUFFT_XT_FORMAT_1D_INPUT_SHUFFLED" = 0x04
+    CUFFT_XT_FORMAT_DISTRIBUTED_INPUT "CUFFT_XT_FORMAT_DISTRIBUTED_INPUT" = 0x05
+    CUFFT_XT_FORMAT_DISTRIBUTED_OUTPUT "CUFFT_XT_FORMAT_DISTRIBUTED_OUTPUT" = 0x06
+    CUFFT_FORMAT_UNDEFINED "CUFFT_FORMAT_UNDEFINED" = 0x07
+
+ctypedef enum cufftXtCopyType "cufftXtCopyType":
+    CUFFT_COPY_HOST_TO_DEVICE "CUFFT_COPY_HOST_TO_DEVICE" = 0x00
+    CUFFT_COPY_DEVICE_TO_HOST "CUFFT_COPY_DEVICE_TO_HOST" = 0x01
+    CUFFT_COPY_DEVICE_TO_DEVICE "CUFFT_COPY_DEVICE_TO_DEVICE" = 0x02
+    CUFFT_COPY_UNDEFINED "CUFFT_COPY_UNDEFINED" = 0x03
+
+ctypedef enum cufftXtQueryType "cufftXtQueryType":
+    CUFFT_QUERY_1D_FACTORS "CUFFT_QUERY_1D_FACTORS" = 0x00
+    CUFFT_QUERY_UNDEFINED "CUFFT_QUERY_UNDEFINED" = 0x01
+
+ctypedef enum cufftXtWorkAreaPolicy "cufftXtWorkAreaPolicy":
+    CUFFT_WORKAREA_MINIMAL "CUFFT_WORKAREA_MINIMAL" = 0
+    CUFFT_WORKAREA_USER "CUFFT_WORKAREA_USER" = 1
+    CUFFT_WORKAREA_PERFORMANCE "CUFFT_WORKAREA_PERFORMANCE" = 2
+
+ctypedef enum cufftXtCallbackType "cufftXtCallbackType":
+    CUFFT_CB_LD_COMPLEX "CUFFT_CB_LD_COMPLEX" = 0x0
+    CUFFT_CB_LD_COMPLEX_DOUBLE "CUFFT_CB_LD_COMPLEX_DOUBLE" = 0x1
+    CUFFT_CB_LD_REAL "CUFFT_CB_LD_REAL" = 0x2
+    CUFFT_CB_LD_REAL_DOUBLE "CUFFT_CB_LD_REAL_DOUBLE" = 0x3
+    CUFFT_CB_ST_COMPLEX "CUFFT_CB_ST_COMPLEX" = 0x4
+    CUFFT_CB_ST_COMPLEX_DOUBLE "CUFFT_CB_ST_COMPLEX_DOUBLE" = 0x5
+    CUFFT_CB_ST_REAL "CUFFT_CB_ST_REAL" = 0x6
+    CUFFT_CB_ST_REAL_DOUBLE "CUFFT_CB_ST_REAL_DOUBLE" = 0x7
+    CUFFT_CB_UNDEFINED "CUFFT_CB_UNDEFINED" = 0x8
+
+ctypedef enum cufftProperty "cufftProperty":
+    NVFFT_PLAN_PROPERTY_INT64_PATIENT_JIT "NVFFT_PLAN_PROPERTY_INT64_PATIENT_JIT" = 0x1
+    NVFFT_PLAN_PROPERTY_INT64_MAX_NUM_HOST_THREADS "NVFFT_PLAN_PROPERTY_INT64_MAX_NUM_HOST_THREADS" = 0x2
+
+
+# types
+cdef extern from *:
+    """
+    #include <driver_types.h>
+    #include <library_types.h>
+    #include <cuComplex.h>
+
+    #define MAX_CUDA_DESCRIPTOR_GPUS 64
+    #define CUFFT_FORWARD -1
+    #define CUFFT_INVERSE  1
+    """
+    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
+
+    cdef const int MAX_CUDA_DESCRIPTOR_GPUS
+    cdef const int cuFFTFORWARD
+    cdef const int cuFFTINVERSE
+
+
+ctypedef float cufftReal 'cufftReal'
+ctypedef double cufftDoubleReal 'cufftDoubleReal'
+ctypedef cuComplex cufftComplex 'cufftComplex'
+ctypedef cuDoubleComplex cufftDoubleComplex 'cufftDoubleComplex'
+ctypedef int cufftHandle 'cufftHandle'
+ctypedef struct cufftXt1dFactors 'cufftXt1dFactors':
+    long long int size
+    long long int stringCount
+    long long int stringLength
+    long long int substringLength
+    long long int factor1
+    long long int factor2
+    long long int stringMask
+    long long int substringMask
+    long long int factor1Mask
+    long long int factor2Mask
+    int stringShift
+    int substringShift
+    int factor1Shift
+    int factor2Shift
+ctypedef cufftComplex (*cufftCallbackLoadC 'cufftCallbackLoadC')(
+    void* dataIn,
+    size_t offset,
+    void* callerInfo,
+    void* sharedPointer
+)
+ctypedef cufftDoubleComplex (*cufftCallbackLoadZ 'cufftCallbackLoadZ')(
+    void* dataIn,
+    size_t offset,
+    void* callerInfo,
+    void* sharedPointer
+)
+ctypedef cufftReal (*cufftCallbackLoadR 'cufftCallbackLoadR')(
+    void* dataIn,
+    size_t offset,
+    void* callerInfo,
+    void* sharedPointer
+)
+ctypedef cufftDoubleReal (*cufftCallbackLoadD 'cufftCallbackLoadD')(
+    void* dataIn,
+    size_t offset,
+    void* callerInfo,
+    void* sharedPointer
+)
+ctypedef cufftComplex (*cufftJITCallbackLoadC 'cufftJITCallbackLoadC')(
+    void* dataIn,
+    unsigned long long offset,
+    void* callerInfo,
+    void* sharedPointer
+)
+ctypedef cufftDoubleComplex (*cufftJITCallbackLoadZ 'cufftJITCallbackLoadZ')(
+    void* dataIn,
+    unsigned long long offset,
+    void* callerInfo,
+    void* sharedPointer
+)
+ctypedef cufftReal (*cufftJITCallbackLoadR 'cufftJITCallbackLoadR')(
+    void* dataIn,
+    unsigned long long offset,
+    void* callerInfo,
+    void* sharedPointer
+)
+ctypedef cufftDoubleReal (*cufftJITCallbackLoadD 'cufftJITCallbackLoadD')(
+    void* dataIn,
+    unsigned long long offset,
+    void* callerInfo,
+    void* sharedPointer
+)
+ctypedef void (*cufftCallbackStoreR 'cufftCallbackStoreR')(
+    void* dataOut,
+    size_t offset,
+    cufftReal element,
+    void* callerInfo,
+    void* sharedPointer
+)
+ctypedef void (*cufftJITCallbackStoreR 'cufftJITCallbackStoreR')(
+    void* dataOut,
+    unsigned long long offset,
+    cufftReal element,
+    void* callerInfo,
+    void* sharedPointer
+)
+ctypedef void (*cufftCallbackStoreD 'cufftCallbackStoreD')(
+    void* dataOut,
+    size_t offset,
+    cufftDoubleReal element,
+    void* callerInfo,
+    void* sharedPointer
+)
+ctypedef void (*cufftJITCallbackStoreD 'cufftJITCallbackStoreD')(
+    void* dataOut,
+    unsigned long long offset,
+    cufftDoubleReal element,
+    void* callerInfo,
+    void* sharedPointer
+)
+ctypedef void (*cufftCallbackStoreC 'cufftCallbackStoreC')(
+    void* dataOut,
+    size_t offset,
+    cufftComplex element,
+    void* callerInfo,
+    void* sharedPointer
+)
+ctypedef void (*cufftJITCallbackStoreC 'cufftJITCallbackStoreC')(
+    void* dataOut,
+    unsigned long long offset,
+    cufftComplex element,
+    void* callerInfo,
+    void* sharedPointer
+)
+ctypedef void (*cufftCallbackStoreZ 'cufftCallbackStoreZ')(
+    void* dataOut,
+    size_t offset,
+    cufftDoubleComplex element,
+    void* callerInfo,
+    void* sharedPointer
+)
+ctypedef void (*cufftJITCallbackStoreZ 'cufftJITCallbackStoreZ')(
+    void* dataOut,
+    unsigned long long offset,
+    cufftDoubleComplex element,
+    void* callerInfo,
+    void* sharedPointer
+)
+
+ctypedef struct cudaXtDesc_t 'cudaXtDesc_t':
+    int version "version"
+    int nGPUs "nGPUs"
+    int GPUs "GPUs" [MAX_CUDA_DESCRIPTOR_GPUS]
+    void* data "data" [MAX_CUDA_DESCRIPTOR_GPUS]
+    size_t size "size" [MAX_CUDA_DESCRIPTOR_GPUS]
+    void* cudaXtState "cudaXtState"
+
+ctypedef cudaXtDesc_t cudaXtDesc 'cudaXtDesc'
+
+ctypedef struct cudaLibXtDesc_t 'cudaLibXtDesc_t':
+    int version "version"
+    cudaXtDesc* descriptor "descriptor"
+    libFormat_t library "library"
+    int subFormat "subFormat"
+    void* libDescriptor "libDescriptor"
+
+ctypedef cudaLibXtDesc_t cudaLibXtDesc 'cudaLibXtDesc'
+
+
+###############################################################################
+# Functions
+###############################################################################
+
+cdef cufftResult cufftPlan1d(cufftHandle* plan, int nx, cufftType type, int batch) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftPlan2d(cufftHandle* plan, int nx, int ny, cufftType type) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftPlan3d(cufftHandle* plan, int nx, int ny, int nz, cufftType type) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftPlanMany(cufftHandle* plan, int rank, int* n, int* inembed, int istride, int idist, int* onembed, int ostride, int odist, cufftType type, int batch) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftMakePlan1d(cufftHandle plan, int nx, cufftType type, int batch, size_t* workSize) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftMakePlan2d(cufftHandle plan, int nx, int ny, cufftType type, size_t* workSize) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftMakePlan3d(cufftHandle plan, int nx, int ny, int nz, cufftType type, size_t* workSize) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftMakePlanMany(cufftHandle plan, int rank, int* n, int* inembed, int istride, int idist, int* onembed, int ostride, int odist, cufftType type, int batch, size_t* workSize) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftMakePlanMany64(cufftHandle plan, int rank, long long int* n, long long int* inembed, long long int istride, long long int idist, long long int* onembed, long long int ostride, long long int odist, cufftType type, long long int batch, size_t* workSize) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftGetSizeMany64(cufftHandle plan, int rank, long long int* n, long long int* inembed, long long int istride, long long int idist, long long int* onembed, long long int ostride, long long int odist, cufftType type, long long int batch, size_t* workSize) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftEstimate1d(int nx, cufftType type, int batch, size_t* workSize) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftEstimate2d(int nx, int ny, cufftType type, size_t* workSize) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftEstimate3d(int nx, int ny, int nz, cufftType type, size_t* workSize) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftEstimateMany(int rank, int* n, int* inembed, int istride, int idist, int* onembed, int ostride, int odist, cufftType type, int batch, size_t* workSize) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftCreate(cufftHandle* handle) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftGetSize1d(cufftHandle handle, int nx, cufftType type, int batch, size_t* workSize) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftGetSize2d(cufftHandle handle, int nx, int ny, cufftType type, size_t* workSize) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftGetSize3d(cufftHandle handle, int nx, int ny, int nz, cufftType type, size_t* workSize) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftGetSizeMany(cufftHandle handle, int rank, int* n, int* inembed, int istride, int idist, int* onembed, int ostride, int odist, cufftType type, int batch, size_t* workArea) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftGetSize(cufftHandle handle, size_t* workSize) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftSetWorkArea(cufftHandle plan, void* workArea) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftSetAutoAllocation(cufftHandle plan, int autoAllocate) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftExecC2C(cufftHandle plan, cufftComplex* idata, cufftComplex* odata, int direction) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftExecR2C(cufftHandle plan, cufftReal* idata, cufftComplex* odata) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftExecC2R(cufftHandle plan, cufftComplex* idata, cufftReal* odata) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftExecZ2Z(cufftHandle plan, cufftDoubleComplex* idata, cufftDoubleComplex* odata, int direction) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftExecD2Z(cufftHandle plan, cufftDoubleReal* idata, cufftDoubleComplex* odata) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftExecZ2D(cufftHandle plan, cufftDoubleComplex* idata, cufftDoubleReal* odata) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftSetStream(cufftHandle plan, cudaStream_t stream) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftDestroy(cufftHandle plan) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftGetVersion(int* version) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftGetProperty(libraryPropertyType type, int* value) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtSetGPUs(cufftHandle handle, int nGPUs, int* whichGPUs) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtMalloc(cufftHandle plan, cudaLibXtDesc** descriptor, cufftXtSubFormat format) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtMemcpy(cufftHandle plan, void* dstPointer, void* srcPointer, cufftXtCopyType type) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtFree(cudaLibXtDesc* descriptor) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtSetWorkArea(cufftHandle plan, void** workArea) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtExecDescriptorC2C(cufftHandle plan, cudaLibXtDesc* input, cudaLibXtDesc* output, int direction) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtExecDescriptorR2C(cufftHandle plan, cudaLibXtDesc* input, cudaLibXtDesc* output) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtExecDescriptorC2R(cufftHandle plan, cudaLibXtDesc* input, cudaLibXtDesc* output) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtExecDescriptorZ2Z(cufftHandle plan, cudaLibXtDesc* input, cudaLibXtDesc* output, int direction) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtExecDescriptorD2Z(cufftHandle plan, cudaLibXtDesc* input, cudaLibXtDesc* output) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtExecDescriptorZ2D(cufftHandle plan, cudaLibXtDesc* input, cudaLibXtDesc* output) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtQueryPlan(cufftHandle plan, void* queryStruct, cufftXtQueryType queryType) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtClearCallback(cufftHandle plan, cufftXtCallbackType cbType) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtSetCallbackSharedSize(cufftHandle plan, cufftXtCallbackType cbType, size_t sharedSize) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtMakePlanMany(cufftHandle plan, int rank, long long int* n, long long int* inembed, long long int istride, long long int idist, cudaDataType inputtype, long long int* onembed, long long int ostride, long long int odist, cudaDataType outputtype, long long int batch, size_t* workSize, cudaDataType executiontype) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtGetSizeMany(cufftHandle plan, int rank, long long int* n, long long int* inembed, long long int istride, long long int idist, cudaDataType inputtype, long long int* onembed, long long int ostride, long long int odist, cudaDataType outputtype, long long int batch, size_t* workSize, cudaDataType executiontype) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtExec(cufftHandle plan, void* input, void* output, int direction) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtExecDescriptor(cufftHandle plan, cudaLibXtDesc* input, cudaLibXtDesc* output, int direction) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtSetWorkAreaPolicy(cufftHandle plan, cufftXtWorkAreaPolicy policy, size_t* workSize) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtSetJITCallback(cufftHandle plan, const void* lto_callback_fatbin, size_t lto_callback_fatbin_size, cufftXtCallbackType type, void** caller_info) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftXtSetSubformatDefault(cufftHandle plan, cufftXtSubFormat subformat_forward, cufftXtSubFormat subformat_inverse) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftSetPlanPropertyInt64(cufftHandle plan, cufftProperty property, const long long int inputValueInt) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftGetPlanPropertyInt64(cufftHandle plan, cufftProperty property, long long int* returnPtrValue) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil
+cdef cufftResult cufftResetPlanProperty(cufftHandle plan, cufftProperty property) except?_CUFFTRESULT_INTERNAL_LOADING_ERROR nogil