stud_22-365/GLXEngine: GaLaXy Engine - пакеты компонентов GLScene & GXScene для Object Pascal и C++. Включает иерархию классов для создания 3D сцен с менеджерами поддержки физики Newton, звука BASS и шейдеров GLSL. Кросс-платформенный фреймворк с объектами на базе OpenGL и Vulkan.

Source/CUDA.Runtime.pas
785 строк37 KB
GLScene
Merged duplicate assets
28 дек 2022, 15:50
28 дек 2022, 15:50537f47c
//
// The graphics platform GLScene https://github.com/glscene
//
unit CUDA.Runtime;

(* GLScene CUDA Runtime *)

(*
 * Copyright 1993-2020 NVIDIA Corporation.  All rights reserved.
 *
 * NOTICE TO USER:
 *
 * This source code is subject to NVIDIA ownership rights under U.S. and
 * international Copyright laws.  Users and possessors of this source code
 * are hereby granted a nonexclusive, royalty-free license to use this code
 * in individual and commercial software.
 *
 * NVIDIA MAKES NO REPRESENTATION ABOUT THE SUITABILITY OF THIS SOURCE
 * CODE FOR ANY PURPOSE.  IT IS PROVIDED "AS IS" WITHOUT EXPRESS OR
 * IMPLIED WARRANTY OF ANY KIND.  NVIDIA DISCLAIMS ALL WARRANTIES WITH
 * REGARD TO THIS SOURCE CODE, INCLUDING ALL IMPLIED WARRANTIES OF
 * MERCHANTABILITY, NONINFRINGEMENT, AND FITNESS FOR A PARTICULAR PURPOSE.
 * IN NO EVENT SHALL NVIDIA BE LIABLE FOR ANY SPECIAL, INDIRECT, INCIDENTAL,
 * OR CONSEQUENTIAL DAMAGES, OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS
 * OF USE, DATA OR PROFITS,  WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
 * OR OTHER TORTIOUS ACTION,  ARISING OUT OF OR IN CONNECTION WITH THE USE
 * OR PERFORMANCE OF THIS SOURCE CODE.
 *
 * U.S. Government End Users.   This source code is a "commercial item" as
 * that term is defined at  48 C.F.R. 2.101 (OCT 1995), consisting  of
 * "commercial computer  software"  and "commercial computer software
 * documentation" as such terms are  used in 48 C.F.R. 12.212 (SEPT 1995)
 * and is provided to the U.S. Government only as a commercial end item.
 * Consistent with 48 C.F.R.12.212 and 48 C.F.R. 227.7202-1 through
 * 227.7202-4 (JUNE 1995), all U.S. Government End Users acquire the
 * source code with only those rights set forth herein.
 *
 * Any use of this source code in individual and commercial software must
 * include, in the user documentation and internal comments to the code,
 * the above Disclaimer and U.S. Government End Users Notice.
 *)

interface

{$I GLScene.inc}

uses
  Winapi.Windows,

  CUDA.Import,
  GLS.Logger;

const
{$IFDEF WIN32}
  CUDARTDLLNAMES: array [0 .. 9] of string = (
    'cudart32_42_9', 'cudart32_41_28',
    'cudart32_40_10', 'cudart32_32_16', 'cudart32_31_4',
    'cudart32_30_14', 'cudart32_30_9', 'cudart32_30_8', 'cudart32', 'cudart');
{$ENDIF}

{$IFDEF WIN64}
  CUDARTDLLNAMES: array [0 .. 7] of string = (
    'cudart64_42_9', 'cudart64_41_28',
    'cudart64_40_10', 'cudart64_32_16', 'cudart64_31_4',
    'cudart64_30_14', 'cudart64_30_9', 'cudart64_30_8');
{$ENDIF}

const
  // single precision constants
  CUDART_INF_F: Single = $7F800000;
  CUDART_NAN_F: Single = $7FFFFFFF;
  CUDART_MIN_DENORM_F: Single = $00000001;
  CUDART_MAX_NORMAL_F: Single = $7F7FFFFF;
  CUDART_NEG_ZERO_F: Single = $80000000;
  CUDART_ZERO_F = 0.0;
  CUDART_ONE_F = 1.0;
  CUDART_SQRT_HALF_F = 0.707106781;
  CUDART_SQRT_TWO_F = 1.414213562;
  CUDART_THIRD_F = 0.333333333;
  CUDART_PIO4_F = 0.785398163;
  CUDART_PIO2_F = 1.570796327;
  CUDART_3PIO4_F = 2.356194490;
  CUDART_2_OVER_PI_F = 0.636619772;
  CUDART_PI_F = 3.141592654;
  CUDART_L2E_F = 1.442695041;
  CUDART_L2T_F = 3.321928094;
  CUDART_LG2_F = 0.301029996;
  CUDART_LGE_F = 0.434294482;
  CUDART_LN2_F = 0.693147181;
  CUDART_LNT_F = 2.302585093;
  CUDART_LNPI_F = 1.144729886;
  CUDART_TWO_TO_M126_F = 1.175494351E-38;
  CUDART_TWO_TO_126_F = 8.507059173E37;
  CUDART_NORM_HUGE_F = 3.402823466E38;
  CUDART_TWO_TO_23_F = 8388608.0;
  CUDART_TWO_TO_24_F = 16777216.0;
  CUDART_TWO_TO_31_F = 2147483648.0;
  CUDART_TWO_TO_32_F = 4294967296.0;
  CUDART_REMQUO_BITS_F = 3;
  CUDART_REMQUO_MASK_F = CUDART_REMQUO_BITS_F;
  CUDART_TRIG_PLOSS_F = 48039.0;

  // double precision constants */
{$IFNDEF CUDA_NO_SM_13_DOUBLE_INTRINSICS}
  CUDART_INF: Double = $7FF0000000000000;
  CUDART_NAN: Double = $FFF8000000000000;
  CUDART_NEG_ZERO: Double = $8000000000000000;
  CUDART_MIN_DENORM: Double = $0000000000000001;
{$ELSE} // not CUDA_NO_SM_13_DOUBLE_INTRINSICS
  CUDART_INF: Double = $7FF0000000000000;
  CUDART_NAN: Double = $FFF8000000000000;
  CUDART_NEG_ZERO: Double = $8000000000000000;
  CUDART_MIN_DENORM: Double = $0000000000000001;
{$ENDIF}
  CUDART_ZERO = 0.0;
  CUDART_ONE = 1.0;
  CUDART_SQRT_TWO = 1.4142135623730951E+0;
  CUDART_SQRT_HALF = 7.0710678118654757E-1;
  CUDART_THIRD = 3.3333333333333333E-1;
  CUDART_TWOTHIRD = 6.6666666666666667E-1;
  CUDART_PIO4 = 7.8539816339744828E-1;
  CUDART_PIO4_HI = 7.8539816339744828E-1;
  CUDART_PIO4_LO = 3.0616169978683830E-17;
  CUDART_PIO2 = 1.5707963267948966E+0;
  CUDART_PIO2_HI = 1.5707963267948966E+0;
  CUDART_PIO2_LO = 6.1232339957367660E-17;
  CUDART_3PIO4 = 2.3561944901923448E+0;
  CUDART_2_OVER_PI = 6.3661977236758138E-1;
  CUDART_PI = 3.1415926535897931E+0;
  CUDART_PI_HI = 3.1415926535897931E+0;
  CUDART_PI_LO = 1.2246467991473532E-16;
  CUDART_SQRT_2PI_HI = 2.5066282746310007E+0;
  CUDART_SQRT_2PI_LO = -1.8328579980459167E-16;
  CUDART_SQRT_PIO2_HI = 1.2533141373155003E+0;
  CUDART_SQRT_PIO2_LO = -9.1642899902295834E-17;
  CUDART_L2E = 1.4426950408889634E+0;
  CUDART_L2E_HI = 1.4426950408889634E+0;
  CUDART_L2E_LO = 2.0355273740931033E-17;
  CUDART_L2T = 3.3219280948873622E+0;
  CUDART_LG2 = 3.0102999566398120E-1;
  CUDART_LG2_HI = 3.0102999566398120E-1;
  CUDART_LG2_LO = -2.8037281277851704E-18;
  CUDART_LGE = 4.3429448190325182E-1;
  CUDART_LGE_HI = 4.3429448190325182E-1;
  CUDART_LGE_LO = 1.09831965021676510E-17;
  CUDART_LN2 = 6.9314718055994529E-1;
  CUDART_LN2_HI = 6.9314718055994529E-1;
  CUDART_LN2_LO = 2.3190468138462996E-17;
  CUDART_LNT = 2.3025850929940459E+0;
  CUDART_LNT_HI = 2.3025850929940459E+0;
  CUDART_LNT_LO = -2.1707562233822494E-16;
  CUDART_LNPI = 1.1447298858494002E+0;
  CUDART_LN2_X_1024 = 7.0978271289338397E+2;
  CUDART_LN2_X_1025 = 7.1047586007394398E+2;
  CUDART_LN2_X_1075 = 7.4513321910194122E+2;
  CUDART_LG2_X_1024 = 3.0825471555991675E+2;
  CUDART_LG2_X_1075 = 3.2360724533877976E+2;
  CUDART_TWO_TO_23 = 8388608.0;
  CUDART_TWO_TO_52 = 4503599627370496.0;
  CUDART_TWO_TO_54 = 18014398509481984.0;
  CUDART_TWO_TO_M54 = 5.5511151231257827E-17;
  CUDART_TWO_TO_M1022 = 2.22507385850720140E-308;
  CUDART_TRIG_PLOSS = 2147483648.0;

type
  TcudaError = (cudaSuccess, cudaErrorMissingConfiguration,
    cudaErrorMemoryAllocation, cudaErrorInitializationError,
    cudaErrorLaunchFailure, cudaErrorPriorLaunchFailure, cudaErrorLaunchTimeout,
    cudaErrorLaunchOutOfResources, cudaErrorInvalidDeviceFunction,
    cudaErrorInvalidConfiguration, cudaErrorInvalidDevice,
    cudaErrorInvalidValue, cudaErrorInvalidPitchValue, cudaErrorInvalidSymbol,
    cudaErrorMapBufferObjectFailed, cudaErrorUnmapBufferObjectFailed,
    cudaErrorInvalidHostPointer, cudaErrorInvalidDevicePointer,
    cudaErrorInvalidTexture, cudaErrorInvalidTextureBinding,
    cudaErrorInvalidChannelDescriptor, cudaErrorInvalidMemcpyDirection,
    cudaErrorAddressOfConstant, cudaErrorTextureFetchFailed,
    cudaErrorTextureNotBound, cudaErrorSynchronizationError,
    cudaErrorInvalidFilterSetting, cudaErrorInvalidNormSetting,
    cudaErrorMixedDeviceExecution, cudaErrorCudartUnloading, cudaErrorUnknown,
    cudaErrorNotYetImplemented, cudaErrorMemoryValueTooLarge,
    cudaErrorInvalidResourceHandle, cudaErrorNotReady, cudaErrorStartupFailure,
    cudaErrorApiFailureBase);

  { +//DEVICE_BUILTIN*/ }
  TCudaChannelFormatKind = (cudaChannelFormatKindSigned,
    cudaChannelFormatKindUnsigned, cudaChannelFormatKindFloat);

  TCudaGLMapFlags = (cudaGLMapFlagsNone,
    /// < Default; Assume resource can be read/written
    cudaGLMapFlagsReadOnly,
    /// < CUDA kernels will not write to this resource
    cudaGLMapFlagsWriteDiscard);
  /// < CUDA kernels will only write to and will not read from this resource

  { +//DEVICE_BUILTIN*/ }
  PcudaChannelFormatDesc = ^TCudaChannelFormatDesc;

  TCudaChannelFormatDesc = record
    x: Integer;
    y: Integer;
    z: Integer;
    w: Integer;
    f: TCudaChannelFormatKind;
  end;

  { +//DEVICE_BUILTIN*/ }
  TcudaArray = record
  end; // !ATTENTION foreward Declaration?)

  { +//DEVICE_BUILTIN*/ }
  TcudaMemcpyKind = (cudaMemcpyHostToHost { = 0 } , cudaMemcpyHostToDevice,
    cudaMemcpyDeviceToHost, cudaMemcpyDeviceToDevice);

  { +//DEVICE_BUILTIN*/ }
  TcudaPitchedPtr = record
    ptr: Pointer;
    pitch: NativeUInt;
    xsize: NativeUInt;
    ysize: NativeUInt;
  end;

  { +//DEVICE_BUILTIN*/ }
  TcudaExtent = record
    width: NativeUInt;
    height: NativeUInt;
    depth: NativeUInt;
  end;

  { +//DEVICE_BUILTIN*/ }
  TcudaPos = record
    x: NativeUInt;
    y: NativeUInt;
    z: NativeUInt;
  end;

  { +//DEVICE_BUILTIN*/ }
  TcudaMemcpy3DParms = record
    srcArray: Pointer;
    srcPos: TcudaPos;
    srcPtr: TcudaPitchedPtr;
    dstArray: Pointer;
    dstPos: TcudaPos;
    dstPtr: TcudaPitchedPtr;
    extent: TcudaExtent;
    kind: TcudaMemcpyKind;
  end;

  { +//DEVICE_BUILTIN*/ }
  PCudaDeviceProp = ^TCudaDeviceProp;

  TCudaDeviceProp = record
    name: array [0 .. 256 - 1] of AnsiChar;
    totalGlobalMem: NativeUInt;
    sharedMemPerBlock: NativeUInt;
    regsPerBlock: Integer;
    warpSize: Integer;
    memPitch: NativeUInt;
    maxThreadsPerBlock: Integer;
    maxThreadsDim: array [0 .. 3 - 1] of Integer;
    maxGridSize: array [0 .. 3 - 1] of Integer;
    clockRate: Integer;
    totalConstMem: NativeUInt;
    major: Integer;
    minor: Integer;
    textureAlignment: NativeUInt;
    deviceOverlap: Integer;
    multiProcessorCount: Integer;
    // Specified whether there is a run time limit on kernels
    kernelExecTimeoutEnabled: Integer;
    // Device is egrated as opposed to discrete
    egrated: Integer;
    // Device can map host memory with cudaHostAlloc/cudaHostGetDevicePoer
    canMapHostMemory: Integer;
    // Compute mode (See ::cudaComputeMode)
    computeMode: Integer;
    // Maximum 1D texture size
    maxTexture1D: Integer;
    // Maximum 2D texture dimensions
    maxTexture2D: array[0..1] of Integer;
    // Maximum 3D texture dimensions
    maxTexture3D: array[0..2] of Integer;
    // Maximum 2D texture array dimensions
    maxTexture2DArray: array[0..2] of Integer;
    // Alignment requirements for surfaces
    surfaceAlignment: NativeUInt;
     // Device can possibly execute multiple kernels concurrently
    concurrentKernels: Integer;
    // Device has ECC support enabled
    ECCEnabled: Integer;
    // PCI bus ID of the device
    pciBusID: Integer;
    // PCI device ID of the device
    pciDeviceID: Integer;
    // 1 if device is a Tesla device using TCC driver, 0 otherwise
    tccDriver: Integer;
    __cudaReserved: array [0 .. 20] of Integer;
  end;

  TcudaTextureAddressMode = (cudaAddressModeWrap, cudaAddressModeClamp, cudaAddressModeMirror);

  TcudaTextureFilterMode = (cudaFilterModePoint, cudaFilterModeLinear);

  TcudaTextureReadMode = (cudaReadModeElementType, cudaReadModeNormalizedFloat);

  PTextureReference = ^TTextureReference;

  TTextureReference = record
    normalized: Integer;
    filterMode: TcudaTextureFilterMode;
    addressMode: array [0 .. 2] of TcudaTextureAddressMode;
    channelDesc: TCudaChannelFormatDesc;
    __cudaReserved: array [0 .. 15] of Integer;
  end;

  PcudaArray = ^TcudaArray;

  { +//****************************************************************************** }
  { -** }
  { -* SHORTHAND TYPE DEFINITION USED BY RUNTIME API* }
  { -** }
  { =*******************************************************************************/ }

  { +//DEVICE_BUILTIN*/ }
  cudaError_t = TcudaError;
  { +//DEVICE_BUILTIN*/ }
  cudaStream_t = Integer;
  { +//DEVICE_BUILTIN*/ }
  cudaEvent_t = Integer;

(*******************************************************************************)

var

cudaBindTexture: function(var offset: NativeUInt; const texref: PTextureReference;
    var devPtr: Pointer; var desc: TCudaChannelFormatDesc; size: NativeUInt): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaBindTexture2D: function(var offset: NativeUInt; const texref: PTextureReference; const devPtr: Pointer;
    var desc: TCudaChannelFormatDesc; width, height, pitch: NativeUInt): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaBindTextureToArray: function(const texref: PTextureReference; const cudaArray: PcudaArray): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaUnbindTexture: function(const texref: PTextureReference): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGetTextureAlignmentOffset: function(offset: NativeUInt; const texref: PTextureReference): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGetTextureReference: function(const texref: PTextureReference; const symbol: PAnsiChar): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGetChannelDesc: function(var desc: TCudaChannelFormatDesc; const array_: Pointer): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaCreateChannelDesc: function(x, y, z, w: Integer; f: TCudaChannelFormatKind): TCudaChannelFormatDesc;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
(* ******************************************************************************
  *                                                                              *
  *                                                                              *
  *                                                                              *
  ****************************************************************************** *)

cudaMalloc3D: function(var pitchedDevPtr: TcudaPitchedPtr; extent: TcudaExtent): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMalloc3DArray: function(var arrayPtr: PcudaArray; const desc: TCudaChannelFormatDesc; extent: TcudaExtent; flags: Cardinal): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemset3D: function(pitchedDevPtr: TcudaPitchedPtr; value: Integer; extent: TcudaExtent): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemcpy3D: function(const p: TcudaMemcpy3DParms): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemcpy3DAsync: function(const p: TcudaMemcpy3DParms; stream: cudaStream_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMalloc: function(var devPtr; size: NativeUInt): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMallocHost: function(var ptr: Pointer; size: NativeUInt): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
  cudaMallocPitch: function(var devPtr; var pitch: NativeUInt; width: NativeUInt; height: NativeUInt): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMallocArray: function(var aarray: Pointer; var desc: TCudaChannelFormatDesc; width: NativeUInt; height: NativeUInt): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaFree: function(devPtr: Pointer): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaFreeHost: function(ptr: Pointer): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaFreeArray: function(const aarray: Pointer): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaHostAlloc: function(var pHost: Pointer; bytes: NativeUInt; flags: Cardinal): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaHostGetDevicePointer: function(var pDevice: Pointer; pHost: Pointer; flags: Cardinal): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaHostGetFlags: function(var pFlags: Cardinal; pHost: Pointer): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemGetInfo: function(var free: NativeUInt; var total: NativeUInt): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemcpy: function(dst: Pointer; src: Pointer; count: NativeUInt; kind: TcudaMemcpyKind): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemcpyToArray: function(var dst: PcudaArray; wOffset: NativeUInt; hOffset: NativeUInt; var src; count: NativeUInt; kind: TcudaMemcpyKind): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemcpyFromArray: function(var dst; const src: PcudaArray; wOffset: NativeUInt; hOffset: NativeUInt; count: NativeUInt; kind: TcudaMemcpyKind): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemcpyArrayToArray: function(dst: PcudaArray; wOffsetDst: NativeUInt; hOffsetDst: NativeUInt; const src: PcudaArray; wOffsetSrc: NativeUInt;
    hOffsetSrc: NativeUInt; count: NativeUInt; const kind: TcudaMemcpyKind = cudaMemcpyDeviceToDevice): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemcpy2D: function(var dst; dpitch: NativeUInt; var src; spitch: NativeUInt;
    width: NativeUInt; height: NativeUInt; kind: TcudaMemcpyKind): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemcpy2DToArray: function(dst: PcudaArray; wOffset: NativeUInt;
  hOffset: NativeUInt; var src; spitch: NativeUInt; width: NativeUInt; height: NativeUInt; kind: TcudaMemcpyKind): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemcpy2DFromArray: function(var dst; dpitch: NativeUInt; src: PcudaArray; wOffset: NativeUInt; hOffset: NativeUInt; width: NativeUInt; height: NativeUInt;
    kind: TcudaMemcpyKind): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemcpy2DArrayToArray: function(dst: PcudaArray; wOffsetDst: NativeUInt;
    hOffsetDst: NativeUInt; src: PcudaArray; wOffsetSrc: NativeUInt; hOffsetSrc: NativeUInt;
    width: NativeUInt; height: NativeUInt; const kind: TcudaMemcpyKind = cudaMemcpyDeviceToDevice): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemcpyToSymbol: function(symbol: PAnsiChar; var src; count: NativeUInt; const offset: NativeUInt = 0;
    const kind: TcudaMemcpyKind = cudaMemcpyHostToDevice): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemcpyFromSymbol: function(var dst; symbol: PAnsiChar; count: NativeUInt; const offset: NativeUInt = 0;
    const kind: TcudaMemcpyKind = cudaMemcpyDeviceToHost): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}

{ +//*************************************************************************** }
{ -** }
{ -** }
{ -** }
{ =***************************************************************************** }

cudaMemcpyAsync: function(var dst; const src; count: NativeUInt; kind: TcudaMemcpyKind; stream: cudaStream_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemcpyToArrayAsync: function(dst: PcudaArray; wOffset: NativeUInt; hOffset: NativeUInt; const src; count: NativeUInt; kind: TcudaMemcpyKind;
    stream: cudaStream_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemcpyFromArrayAsync: function(var dst; const src: PcudaArray;
    wOffset: NativeUInt; hOffset: NativeUInt; count: NativeUInt; kind: TcudaMemcpyKind; stream: cudaStream_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemcpy2DAsync: function(var dst; dpitch: NativeUInt; const src;
    spitch: NativeUInt; width: NativeUInt; height: NativeUInt; kind: TcudaMemcpyKind; stream: cudaStream_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemcpy2DToArrayAsync: function(dst: PcudaArray; wOffset: NativeUInt;
    hOffset: NativeUInt; const src; spitch: NativeUInt; width: NativeUInt; height: NativeUInt;
    kind: TcudaMemcpyKind; stream: cudaStream_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemcpy2DFromArrayAsync: function(var dst; dpitch: NativeUInt;
    const src: PcudaArray; wOffset: NativeUInt; hOffset: NativeUInt; width: NativeUInt;
    height: NativeUInt; kind: TcudaMemcpyKind; stream: cudaStream_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemcpyToSymbolAsync: function(const symbol: PAnsiChar; const src;
    count: NativeUInt; offset: NativeUInt; kind: TcudaMemcpyKind; stream: cudaStream_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemcpyFromSymbolAsync: function(var dst; const symbol: PAnsiChar;
  count: NativeUInt; offset: NativeUInt; kind: TcudaMemcpyKind; stream: cudaStream_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}

(******************************************************************************
 *                                                                            *
 *                                                                            *
 *                                                                            *
 *****************************************************************************)

cudaMemset: function(var devPtr; value: Integer; count: NativeUInt): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaMemset2D: function(var devPtr; pitch: NativeUInt; value: Integer;
    width: NativeUInt; height: NativeUInt): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
(*****************************************************************************
 *                                                                           *
 *                                                                           *
 *                                                                           *
 *****************************************************************************)

cudaGetSymbolAddress: function(var devPtr: Pointer; const symbol: PAnsiChar): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGetSymbolSize: function(var size: NativeUInt; const symbol: PAnsiChar): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}

{ +//*************************************************************************** }
{ -** }
{ -** }
{ -** }
{ =***************************************************************************** }

cudaGetDeviceCount: function(var count: Integer): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGetDeviceProperties: function(var prop: TCudaDeviceProp; device: Integer): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaChooseDevice: function(var device: Integer; const prop: PCudaDeviceProp): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaSetDevice: function(device: Integer): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGetDevice: function(var device: Integer): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaSetDeviceFlags: function(flags: Integer): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaSetValidDevices: function(device_arr: PInteger; len: Integer): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
(******************************************************************************
 *
 *
 *
 *****************************************************************************)

cudaConfigureCall: function(gridDim, blockDim: TDim3; sharedMem: NativeUInt; stream: cudaStream_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaSetupArgument: function(const arg: Pointer; size: NativeUInt; offset: NativeUInt): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaFuncSetCacheConfig: function(const func: PAnsiChar; cacheConfig: TcudaFuncCache): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaLaunch: function(const entry: PAnsiChar): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaFuncGetAttributes: function(var attr: TcudaFuncAttributes; const func: PAnsiChar): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
  { +//****************************************************************************** }
  { -** }
  { -** }
  { -** }
  { =*******************************************************************************/ }

cudaGetLastError: function: cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
  { +//****************************************************************************** }
  { -** }
  { -** }
  { -** }
  { =*******************************************************************************/ }
cudaGLSetGLDevice: function(device: Integer): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGLRegisterBufferObject: function(bufObj: Cardinal): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGraphicsGLRegisterImage: function(const resource: PCUgraphicsResource; image: Cardinal; target: Cardinal; flags: Cardinal): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGraphicsGLRegisterBuffer: function(const resource: PCUgraphicsResource; buffer: Cardinal; flags: Cardinal): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGLMapBufferObject: function(devPtr: Pointer; bufObj: Cardinal): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGLUnmapBufferObject: function(bufObj: Cardinal): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGLUnregisterBufferObject: function(bufObj: Cardinal): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGLSetBufferObjectMapFlags: function(bufObj: Cardinal; flags: TCudaGLMapFlags): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGLMapBufferObjectAsync: function(var devPtr: Pointer; bufObj: Cardinal; stream: cudaStream_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGLUnmapBufferObjectAsync: function(bufObj: Cardinal; stream: cudaStream_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGraphicsUnregisterResource: function(resource: PCUgraphicsResource): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGraphicsResourceSetMapFlags: function(resource: PCUgraphicsResource; flags: Cardinal): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGraphicsMapResources: function(count: Integer; const resources: PCUgraphicsResource; stream: cudaStream_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGraphicsUnmapResources: function(count: Integer; const resources: PCUgraphicsResource; stream: cudaStream_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGraphicsResourceGetMappedPointer: function(var pDevPtr: TCUdeviceptr;
    var pSize: Cardinal; resource: PCUgraphicsResource): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGraphicsSubResourceGetMappedArray: function(var pArray: PCUarray;
    resource: PCUgraphicsResource; arrayIndex: Cardinal; mipLevel: Cardinal): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaGetErrorString: function(error: cudaError_t): PAnsiChar;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaDriverGetVersion: function(out driverVersion: Integer): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaRuntimeGetVersion: function(out runtimeVersion: Integer): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
(* ******************************************************************************
 *                                                                              *
 *                                                                              *
 *                                                                              *
 ****************************************************************************** *)

cudaSetDoubleForDevice: function(var d: Double): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaSetDoubleForHost: function(var d: Double): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
(* ******************************************************************************
 *                                                                              *
 *                                                                              *
 *                                                                              *
 ****************************************************************************** *)

cudaStreamCreate: function(var pStream: cudaStream_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaStreamDestroy: function(stream: cudaStream_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaStreamSynchronize: function(stream: cudaStream_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaStreamQuery: function(stream: cudaStream_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
(* ******************************************************************************
 *                                                                              *
 *                                                                              *
 *                                                                              *
 ****************************************************************************** *)

cudaEventCreate: function(var event: cudaEvent_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaEventCreateWithFlags: function(var event: cudaEvent_t; flags: Integer): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaEventRecord: function(event: cudaEvent_t; stream: cudaStream_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaEventQuery: function(event: cudaEvent_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaEventSynchronize: function(event: cudaEvent_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaEventDestroy: function(event: cudaEvent_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaEventElapsedTime: function(var ms: Single; start: cudaEvent_t; ending: cudaEvent_t): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaWGLGetDevice: function(var device: Integer; hGpu: HGPUNV): cudaError_t;
{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
(* ******************************************************************************
 *                                                                              *
 *                                                                              *
 *                                                                              *
 ****************************************************************************** *)

cudaThreadExit: function(): cudaError_t;{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaThreadSynchronize: function(): cudaError_t;{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaThreadSetLimit: function(limit: TcudaLimit; value: NativeUInt): cudaError_t;{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
cudaThreadGetLimit: function(var value: NativeUInt; limit: TcudaLimit): cudaError_t;{$IFDEF MSWINDOWS}stdcall;{$ELSE}cdecl;{$ENDIF}
function cudaGetLastErrorString: string;
function InitCUDART: Boolean;
procedure CloseCUDART;
function InitCUDARTFromLibrary(const LibName: WideString): Boolean;
function IsCUDARTInitialized: Boolean;

//--------------------------------------------
implementation
//--------------------------------------------

function cudaGetLastErrorString: string;
begin
  Result := string(cudaGetErrorString(cudaGetLastError));
end;

const
  INVALID_MODULEHANDLE = 0;

var
{$IFDEF MSWINDOWS}
  CUDARTHandle: HINST = INVALID_MODULEHANDLE;
{$ELSE}
  CUDARTHandle: TLibHandle = INVALID_MODULEHANDLE;
{$ENDIF}

function CUDARTGetProcAddress(ProcName: PAnsiChar): Pointer;
begin
  Result := GetProcAddress(CUDARTHandle, ProcName);
end;

function InitCUDART: Boolean;
var
  I: Integer;
begin
  Result := True;
  if CUDARTHandle = INVALID_MODULEHANDLE then
  begin
    for I := 0 to High(CUDARTDLLNAMES) do
    begin
      if InitCUDARTFromLibrary(CUDARTDLLNAMES[I] + '.dll') then
        Exit;
    end;
  end;
  Result := False;
end;

procedure CloseCUDART;
begin
  if CUDARTHandle <> INVALID_MODULEHANDLE then
  begin
    FreeLibrary(CUDARTHandle);
    CUDARTHandle := INVALID_MODULEHANDLE;
  end;
end;

function InitCUDARTFromLibrary(const LibName: WideString): Boolean;
var
  V: Integer;
begin
  CloseCUDART;
  CUDARTHandle := GetModuleHandleW(PWideChar(LibName));
  if CUDARTHandle = INVALID_MODULEHANDLE then
    CUDARTHandle := LoadLibraryW(PWideChar(LibName));

  if CUDARTHandle = INVALID_MODULEHANDLE then
    Exit(False);

  cudaFreeHost := CUDARTGetProcAddress('cudaFreeHost');
  cudaFuncGetAttributes := CUDARTGetProcAddress('cudaFuncGetAttributes');
  cudaGetChannelDesc := CUDARTGetProcAddress('cudaGetChannelDesc');
  cudaGetDevice := CUDARTGetProcAddress('cudaGetDevice');
  cudaGetDeviceCount := CUDARTGetProcAddress('cudaGetDeviceCount');
  cudaGetDeviceProperties := CUDARTGetProcAddress('cudaGetDeviceProperties');
  cudaGetErrorString := CUDARTGetProcAddress('cudaGetErrorString');
  cudaGetLastError := CUDARTGetProcAddress('cudaGetLastError');
  cudaGetSymbolAddress := CUDARTGetProcAddress('cudaGetSymbolAddress');
  cudaGetSymbolSize := CUDARTGetProcAddress('cudaGetSymbolSize');
  cudaGetTextureAlignmentOffset := CUDARTGetProcAddress('cudaGetTextureAlignmentOffset');
  cudaGetTextureReference := CUDARTGetProcAddress('cudaGetTextureReference');
  cudaGLMapBufferObject := CUDARTGetProcAddress('cudaGLMapBufferObject');
  cudaGLMapBufferObjectAsync := CUDARTGetProcAddress('cudaGLMapBufferObjectAsync');
  cudaGLRegisterBufferObject := CUDARTGetProcAddress('cudaGLRegisterBufferObject');
  cudaGLSetBufferObjectMapFlags := CUDARTGetProcAddress('cudaGLSetBufferObjectMapFlags');
  cudaGLSetGLDevice := CUDARTGetProcAddress('cudaGLSetGLDevice');
  cudaGLUnmapBufferObject := CUDARTGetProcAddress('cudaGLUnmapBufferObject');
  cudaGLUnmapBufferObjectAsync := CUDARTGetProcAddress('cudaGLUnmapBufferObjectAsync');
  cudaGLUnregisterBufferObject := CUDARTGetProcAddress('cudaGLUnregisterBufferObject');
  cudaGraphicsGLRegisterBuffer := CUDARTGetProcAddress('cudaGraphicsGLRegisterBuffer');
  cudaGraphicsGLRegisterImage := CUDARTGetProcAddress('cudaGraphicsGLRegisterImage');
  cudaGraphicsMapResources := CUDARTGetProcAddress('cudaGraphicsMapResources');
  cudaGraphicsResourceGetMappedPointer := CUDARTGetProcAddress('cudaGraphicsResourceGetMappedPointer');
  cudaGraphicsResourceSetMapFlags := CUDARTGetProcAddress('cudaGraphicsResourceSetMapFlags');
  cudaGraphicsSubResourceGetMappedArray := CUDARTGetProcAddress('cudaGraphicsSubResourceGetMappedArray');
  cudaGraphicsUnmapResources := CUDARTGetProcAddress('cudaGraphicsUnmapResources');
  cudaGraphicsUnregisterResource := CUDARTGetProcAddress('cudaGraphicsUnregisterResource');
  cudaHostAlloc := CUDARTGetProcAddress('cudaHostAlloc');
  cudaHostGetDevicePointer := CUDARTGetProcAddress('cudaHostGetDevicePointer');
  cudaHostGetFlags := CUDARTGetProcAddress('cudaHostGetFlags');
  cudaLaunch := CUDARTGetProcAddress('cudaLaunch');
  cudaMalloc := CUDARTGetProcAddress('cudaMalloc');
  cudaMalloc3D := CUDARTGetProcAddress('cudaMalloc3D');
  cudaMalloc3DArray := CUDARTGetProcAddress('cudaMalloc3DArray');
  cudaMallocArray := CUDARTGetProcAddress('cudaMallocArray');
  cudaMallocHost := CUDARTGetProcAddress('cudaMallocHost');
  cudaMallocPitch := CUDARTGetProcAddress('cudaMallocPitch');
  cudaMemcpy := CUDARTGetProcAddress('cudaMemcpy');
  cudaMemcpy2D := CUDARTGetProcAddress('cudaMemcpy2D');
  cudaMemcpy2DArrayToArray := CUDARTGetProcAddress('cudaMemcpy2DArrayToArray');
  cudaMemcpy2DAsync := CUDARTGetProcAddress('cudaMemcpy2DAsync');
  cudaMemcpy2DFromArray := CUDARTGetProcAddress('cudaMemcpy2DFromArray');
  cudaMemcpy2DFromArrayAsync := CUDARTGetProcAddress('cudaMemcpy2DFromArrayAsync');
  cudaMemcpy2DToArray := CUDARTGetProcAddress('cudaMemcpy2DToArray');
  cudaMemcpy2DToArrayAsync := CUDARTGetProcAddress('cudaMemcpy2DToArrayAsync');
  cudaMemcpy3D := CUDARTGetProcAddress('cudaMemcpy3D');
  cudaMemcpy3DAsync := CUDARTGetProcAddress('cudaMemcpy3DAsync');
  cudaMemcpyArrayToArray := CUDARTGetProcAddress('cudaMemcpyArrayToArray');
  cudaMemcpyAsync := CUDARTGetProcAddress('cudaMemcpyAsync');
  cudaMemcpyFromArray := CUDARTGetProcAddress('cudaMemcpyFromArray');
  cudaMemcpyFromArrayAsync := CUDARTGetProcAddress('cudaMemcpyFromArrayAsync');
  cudaMemcpyFromSymbol := CUDARTGetProcAddress('cudaMemcpyFromSymbol');
  cudaMemcpyFromSymbolAsync := CUDARTGetProcAddress('cudaMemcpyFromSymbolAsync');
  cudaMemcpyToArray := CUDARTGetProcAddress('cudaMemcpyToArray');
  cudaMemcpyToArrayAsync := CUDARTGetProcAddress('cudaMemcpyToArrayAsync');
  cudaMemcpyToSymbol := CUDARTGetProcAddress('cudaMemcpyToSymbol');
  cudaMemcpyToSymbolAsync := CUDARTGetProcAddress('cudaMemcpyToSymbolAsync');
  cudaMemGetInfo := CUDARTGetProcAddress('cudaMemGetInfo');
  cudaMemset := CUDARTGetProcAddress('cudaMemset');
  cudaMemset2D := CUDARTGetProcAddress('cudaMemset2D');
  cudaMemset3D := CUDARTGetProcAddress('cudaMemset3D');
  cudaRuntimeGetVersion := CUDARTGetProcAddress('cudaRuntimeGetVersion');
  cudaSetDevice := CUDARTGetProcAddress('cudaSetDevice');
  cudaSetDeviceFlags := CUDARTGetProcAddress('cudaSetDeviceFlags');
  cudaSetDoubleForDevice := CUDARTGetProcAddress('cudaSetDoubleForDevice');
  cudaSetDoubleForHost := CUDARTGetProcAddress('cudaSetDoubleForHost');
  cudaSetupArgument := CUDARTGetProcAddress('cudaSetupArgument');
  cudaSetValidDevices := CUDARTGetProcAddress('cudaSetValidDevices');
  cudaStreamCreate := CUDARTGetProcAddress('cudaStreamCreate');
  cudaStreamDestroy := CUDARTGetProcAddress('cudaStreamDestroy');
  cudaStreamQuery := CUDARTGetProcAddress('cudaStreamQuery');
  cudaStreamSynchronize := CUDARTGetProcAddress('cudaStreamSynchronize');
  cudaThreadExit := CUDARTGetProcAddress('cudaThreadExit');
  cudaThreadSynchronize := CUDARTGetProcAddress('cudaThreadSynchronize');
  cudaThreadSetLimit := CUDARTGetProcAddress('cudaThreadSetLimit');
  cudaThreadGetLimit := CUDARTGetProcAddress('cudaThreadGetLimit');
  cudaUnbindTexture := CUDARTGetProcAddress('cudaUnbindTexture');
  cudaWGLGetDevice := CUDARTGetProcAddress('cudaWGLGetDevice');

  cudaRuntimeGetVersion(V);
  GLSLogger.LogInfoFmt('%s version %d is loaded', [LibName, V]);
  Result := True;
end;

function IsCUDARTInitialized: Boolean;
begin
  Result := (CUDARTHandle <> INVALID_MODULEHANDLE);
end;

//-----------------------------------------------
initialization
//-----------------------------------------------

//-----------------------------------------------
finalization
//-----------------------------------------------

CloseCUDART;

end.