Blender: util_cuda.cpp Source File

Blender V2.61 - r43446
00001 /*
00002  * Copyright 2011, Blender Foundation.
00003  *
00004  * This program is free software; you can redistribute it and/or
00005  * modify it under the terms of the GNU General Public License
00006  * as published by the Free Software Foundation; either version 2
00007  * of the License, or (at your option) any later version.
00008  *
00009  * This program is distributed in the hope that it will be useful,
00010  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00011  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
00012  * GNU General Public License for more details.
00013  *
00014  * You should have received a copy of the GNU General Public License
00015  * along with this program; if not, write to the Free Software Foundation,
00016  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
00017  */
00018 
00019 #include <stdlib.h>
00020 
00021 #include "util_cuda.h"
00022 #include "util_debug.h"
00023 #include "util_dynlib.h"
00024 #include "util_path.h"
00025 #include "util_string.h"
00026 
00027 /* function defininitions */
00028 
00029 tcuInit *cuInit;
00030 tcuDriverGetVersion *cuDriverGetVersion;
00031 tcuDeviceGet *cuDeviceGet;
00032 tcuDeviceGetCount *cuDeviceGetCount;
00033 tcuDeviceGetName *cuDeviceGetName;
00034 tcuDeviceComputeCapability *cuDeviceComputeCapability;
00035 tcuDeviceTotalMem *cuDeviceTotalMem;
00036 tcuDeviceGetProperties *cuDeviceGetProperties;
00037 tcuDeviceGetAttribute *cuDeviceGetAttribute;
00038 tcuCtxCreate *cuCtxCreate;
00039 tcuCtxDestroy *cuCtxDestroy;
00040 tcuCtxAttach *cuCtxAttach;
00041 tcuCtxDetach *cuCtxDetach;
00042 tcuCtxPushCurrent *cuCtxPushCurrent;
00043 tcuCtxPopCurrent *cuCtxPopCurrent;
00044 tcuCtxGetDevice *cuCtxGetDevice;
00045 tcuCtxSynchronize *cuCtxSynchronize;
00046 tcuModuleLoad *cuModuleLoad;
00047 tcuModuleLoadData *cuModuleLoadData;
00048 tcuModuleLoadDataEx *cuModuleLoadDataEx;
00049 tcuModuleLoadFatBinary *cuModuleLoadFatBinary;
00050 tcuModuleUnload *cuModuleUnload;
00051 tcuModuleGetFunction *cuModuleGetFunction;
00052 tcuModuleGetGlobal *cuModuleGetGlobal;
00053 tcuModuleGetTexRef *cuModuleGetTexRef;
00054 tcuModuleGetSurfRef *cuModuleGetSurfRef;
00055 tcuMemGetInfo *cuMemGetInfo;
00056 tcuMemAlloc *cuMemAlloc;
00057 tcuMemAllocPitch *cuMemAllocPitch;
00058 tcuMemFree *cuMemFree;
00059 tcuMemGetAddressRange *cuMemGetAddressRange;
00060 tcuMemAllocHost *cuMemAllocHost;
00061 tcuMemFreeHost *cuMemFreeHost;
00062 tcuMemHostAlloc *cuMemHostAlloc;
00063 tcuMemHostGetDevicePointer *cuMemHostGetDevicePointer;
00064 tcuMemHostGetFlags *cuMemHostGetFlags;
00065 tcuMemcpyHtoD *cuMemcpyHtoD;
00066 tcuMemcpyDtoH *cuMemcpyDtoH;
00067 tcuMemcpyDtoD *cuMemcpyDtoD;
00068 tcuMemcpyDtoA *cuMemcpyDtoA;
00069 tcuMemcpyAtoD *cuMemcpyAtoD;
00070 tcuMemcpyHtoA *cuMemcpyHtoA;
00071 tcuMemcpyAtoH *cuMemcpyAtoH;
00072 tcuMemcpyAtoA *cuMemcpyAtoA;
00073 tcuMemcpy2D *cuMemcpy2D;
00074 tcuMemcpy2DUnaligned *cuMemcpy2DUnaligned;
00075 tcuMemcpy3D *cuMemcpy3D;
00076 tcuMemcpyHtoDAsync *cuMemcpyHtoDAsync;
00077 tcuMemcpyDtoHAsync *cuMemcpyDtoHAsync;
00078 tcuMemcpyDtoDAsync *cuMemcpyDtoDAsync;
00079 tcuMemcpyHtoAAsync *cuMemcpyHtoAAsync;
00080 tcuMemcpyAtoHAsync *cuMemcpyAtoHAsync;
00081 tcuMemcpy2DAsync *cuMemcpy2DAsync;
00082 tcuMemcpy3DAsync *cuMemcpy3DAsync;
00083 tcuMemsetD8 *cuMemsetD8;
00084 tcuMemsetD16 *cuMemsetD16;
00085 tcuMemsetD32 *cuMemsetD32;
00086 tcuMemsetD2D8 *cuMemsetD2D8;
00087 tcuMemsetD2D16 *cuMemsetD2D16;
00088 tcuMemsetD2D32 *cuMemsetD2D32;
00089 tcuFuncSetBlockShape *cuFuncSetBlockShape;
00090 tcuFuncSetSharedSize *cuFuncSetSharedSize;
00091 tcuFuncGetAttribute *cuFuncGetAttribute;
00092 tcuFuncSetCacheConfig *cuFuncSetCacheConfig;
00093 tcuArrayCreate *cuArrayCreate;
00094 tcuArrayGetDescriptor *cuArrayGetDescriptor;
00095 tcuArrayDestroy *cuArrayDestroy;
00096 tcuArray3DCreate *cuArray3DCreate;
00097 tcuArray3DGetDescriptor *cuArray3DGetDescriptor;
00098 tcuTexRefCreate *cuTexRefCreate;
00099 tcuTexRefDestroy *cuTexRefDestroy;
00100 tcuTexRefSetArray *cuTexRefSetArray;
00101 tcuTexRefSetAddress *cuTexRefSetAddress;
00102 tcuTexRefSetAddress2D *cuTexRefSetAddress2D;
00103 tcuTexRefSetFormat *cuTexRefSetFormat;
00104 tcuTexRefSetAddressMode *cuTexRefSetAddressMode;
00105 tcuTexRefSetFilterMode *cuTexRefSetFilterMode;
00106 tcuTexRefSetFlags *cuTexRefSetFlags;
00107 tcuTexRefGetAddress *cuTexRefGetAddress;
00108 tcuTexRefGetArray *cuTexRefGetArray;
00109 tcuTexRefGetAddressMode *cuTexRefGetAddressMode;
00110 tcuTexRefGetFilterMode *cuTexRefGetFilterMode;
00111 tcuTexRefGetFormat *cuTexRefGetFormat;
00112 tcuTexRefGetFlags *cuTexRefGetFlags;
00113 tcuSurfRefSetArray *cuSurfRefSetArray;
00114 tcuSurfRefGetArray *cuSurfRefGetArray;
00115 tcuParamSetSize *cuParamSetSize;
00116 tcuParamSeti *cuParamSeti;
00117 tcuParamSetf *cuParamSetf;
00118 tcuParamSetv *cuParamSetv;
00119 tcuParamSetTexRef *cuParamSetTexRef;
00120 tcuLaunch *cuLaunch;
00121 tcuLaunchGrid *cuLaunchGrid;
00122 tcuLaunchGridAsync *cuLaunchGridAsync;
00123 tcuEventCreate *cuEventCreate;
00124 tcuEventRecord *cuEventRecord;
00125 tcuEventQuery *cuEventQuery;
00126 tcuEventSynchronize *cuEventSynchronize;
00127 tcuEventDestroy *cuEventDestroy;
00128 tcuEventElapsedTime *cuEventElapsedTime;
00129 tcuStreamCreate *cuStreamCreate;
00130 tcuStreamQuery *cuStreamQuery;
00131 tcuStreamSynchronize *cuStreamSynchronize;
00132 tcuStreamDestroy *cuStreamDestroy;
00133 tcuGraphicsUnregisterResource *cuGraphicsUnregisterResource;
00134 tcuGraphicsSubResourceGetMappedArray *cuGraphicsSubResourceGetMappedArray;
00135 tcuGraphicsResourceGetMappedPointer *cuGraphicsResourceGetMappedPointer;
00136 tcuGraphicsResourceSetMapFlags *cuGraphicsResourceSetMapFlags;
00137 tcuGraphicsMapResources *cuGraphicsMapResources;
00138 tcuGraphicsUnmapResources *cuGraphicsUnmapResources;
00139 tcuGetExportTable *cuGetExportTable;
00140 tcuCtxSetLimit *cuCtxSetLimit;
00141 tcuCtxGetLimit *cuCtxGetLimit;
00142 tcuGLCtxCreate *cuGLCtxCreate;
00143 tcuGraphicsGLRegisterBuffer *cuGraphicsGLRegisterBuffer;
00144 tcuGraphicsGLRegisterImage *cuGraphicsGLRegisterImage;
00145 tcuCtxSetCurrent *cuCtxSetCurrent;
00146 
00147 CCL_NAMESPACE_BEGIN
00148 
00149 /* utility macros */
00150 #define CUDA_LIBRARY_FIND_CHECKED(name) \
00151     name = (t##name*)dynamic_library_find(lib, #name);
00152 
00153 #define CUDA_LIBRARY_FIND(name) \
00154     name = (t##name*)dynamic_library_find(lib, #name); \
00155     assert(name);
00156 
00157 #define CUDA_LIBRARY_FIND_V2(name) \
00158     name = (t##name*)dynamic_library_find(lib, #name "_v2"); \
00159     assert(name);
00160 
00161 /* initialization function */
00162 
00163 bool cuLibraryInit()
00164 {
00165     static bool initialized = false;
00166     static bool result = false;
00167 
00168     if(initialized)
00169         return result;
00170     
00171     initialized = true;
00172 
00173     /* library paths */
00174 #ifdef _WIN32
00175     /* expected in c:/windows/system or similar, no path needed */
00176     const char *path = "nvcuda.dll";
00177 #elif defined(__APPLE__)
00178     /* default installation path */
00179     const char *path = "/usr/local/cuda/lib/libcuda.dylib";
00180 #else
00181     const char *path = "libcuda.so";
00182 #endif
00183 
00184     /* load library */
00185     DynamicLibrary *lib = dynamic_library_open(path);
00186 
00187     if(lib == NULL)
00188         return false;
00189 
00190     /* detect driver version */
00191     int driver_version = 1000;
00192 
00193     CUDA_LIBRARY_FIND_CHECKED(cuDriverGetVersion);
00194     if(cuDriverGetVersion)
00195         cuDriverGetVersion(&driver_version);
00196 
00197     /* we require version 4.0 */
00198     if(driver_version < 4000)
00199         return false;
00200 
00201     /* fetch all function pointers */
00202     CUDA_LIBRARY_FIND(cuInit);
00203     CUDA_LIBRARY_FIND(cuDeviceGet);
00204     CUDA_LIBRARY_FIND(cuDeviceGetCount);
00205     CUDA_LIBRARY_FIND(cuDeviceGetName);
00206     CUDA_LIBRARY_FIND(cuDeviceComputeCapability);
00207     CUDA_LIBRARY_FIND(cuDeviceTotalMem);
00208     CUDA_LIBRARY_FIND(cuDeviceGetProperties);
00209     CUDA_LIBRARY_FIND(cuDeviceGetAttribute);
00210     CUDA_LIBRARY_FIND(cuCtxCreate);
00211     CUDA_LIBRARY_FIND(cuCtxDestroy);
00212     CUDA_LIBRARY_FIND(cuCtxAttach);
00213     CUDA_LIBRARY_FIND(cuCtxDetach);
00214     CUDA_LIBRARY_FIND(cuCtxPushCurrent);
00215     CUDA_LIBRARY_FIND(cuCtxPopCurrent);
00216     CUDA_LIBRARY_FIND(cuCtxGetDevice);
00217     CUDA_LIBRARY_FIND(cuCtxSynchronize);
00218     CUDA_LIBRARY_FIND(cuModuleLoad);
00219     CUDA_LIBRARY_FIND(cuModuleLoadData);
00220     CUDA_LIBRARY_FIND(cuModuleUnload);
00221     CUDA_LIBRARY_FIND(cuModuleGetFunction);
00222     CUDA_LIBRARY_FIND(cuModuleGetGlobal);
00223     CUDA_LIBRARY_FIND(cuModuleGetTexRef);
00224     CUDA_LIBRARY_FIND(cuMemGetInfo);
00225     CUDA_LIBRARY_FIND(cuMemAlloc);
00226     CUDA_LIBRARY_FIND(cuMemAllocPitch);
00227     CUDA_LIBRARY_FIND(cuMemFree);
00228     CUDA_LIBRARY_FIND(cuMemGetAddressRange);
00229     CUDA_LIBRARY_FIND(cuMemAllocHost);
00230     CUDA_LIBRARY_FIND(cuMemFreeHost);
00231     CUDA_LIBRARY_FIND(cuMemHostAlloc);
00232     CUDA_LIBRARY_FIND(cuMemHostGetDevicePointer);
00233     CUDA_LIBRARY_FIND(cuMemcpyHtoD);
00234     CUDA_LIBRARY_FIND(cuMemcpyDtoH);
00235     CUDA_LIBRARY_FIND(cuMemcpyDtoD);
00236     CUDA_LIBRARY_FIND(cuMemcpyDtoA);
00237     CUDA_LIBRARY_FIND(cuMemcpyAtoD);
00238     CUDA_LIBRARY_FIND(cuMemcpyHtoA);
00239     CUDA_LIBRARY_FIND(cuMemcpyAtoH);
00240     CUDA_LIBRARY_FIND(cuMemcpyAtoA);
00241     CUDA_LIBRARY_FIND(cuMemcpy2D);
00242     CUDA_LIBRARY_FIND(cuMemcpy2DUnaligned);
00243     CUDA_LIBRARY_FIND(cuMemcpy3D);
00244     CUDA_LIBRARY_FIND(cuMemcpyHtoDAsync);
00245     CUDA_LIBRARY_FIND(cuMemcpyDtoHAsync);
00246     CUDA_LIBRARY_FIND(cuMemcpyHtoAAsync);
00247     CUDA_LIBRARY_FIND(cuMemcpyAtoHAsync);
00248     CUDA_LIBRARY_FIND(cuMemcpy2DAsync);
00249     CUDA_LIBRARY_FIND(cuMemcpy3DAsync);
00250     CUDA_LIBRARY_FIND(cuMemsetD8);
00251     CUDA_LIBRARY_FIND(cuMemsetD16);
00252     CUDA_LIBRARY_FIND(cuMemsetD32);
00253     CUDA_LIBRARY_FIND(cuMemsetD2D8);
00254     CUDA_LIBRARY_FIND(cuMemsetD2D16);
00255     CUDA_LIBRARY_FIND(cuMemsetD2D32);
00256     CUDA_LIBRARY_FIND(cuFuncSetBlockShape);
00257     CUDA_LIBRARY_FIND(cuFuncSetSharedSize);
00258     CUDA_LIBRARY_FIND(cuFuncGetAttribute);
00259     CUDA_LIBRARY_FIND(cuArrayCreate);
00260     CUDA_LIBRARY_FIND(cuArrayGetDescriptor);
00261     CUDA_LIBRARY_FIND(cuArrayDestroy);
00262     CUDA_LIBRARY_FIND(cuArray3DCreate);
00263     CUDA_LIBRARY_FIND(cuArray3DGetDescriptor);
00264     CUDA_LIBRARY_FIND(cuTexRefCreate);
00265     CUDA_LIBRARY_FIND(cuTexRefDestroy);
00266     CUDA_LIBRARY_FIND(cuTexRefSetArray);
00267     CUDA_LIBRARY_FIND(cuTexRefSetAddress);
00268     CUDA_LIBRARY_FIND(cuTexRefSetAddress2D);
00269     CUDA_LIBRARY_FIND(cuTexRefSetFormat);
00270     CUDA_LIBRARY_FIND(cuTexRefSetAddressMode);
00271     CUDA_LIBRARY_FIND(cuTexRefSetFilterMode);
00272     CUDA_LIBRARY_FIND(cuTexRefSetFlags);
00273     CUDA_LIBRARY_FIND(cuTexRefGetAddress);
00274     CUDA_LIBRARY_FIND(cuTexRefGetArray);
00275     CUDA_LIBRARY_FIND(cuTexRefGetAddressMode);
00276     CUDA_LIBRARY_FIND(cuTexRefGetFilterMode);
00277     CUDA_LIBRARY_FIND(cuTexRefGetFormat);
00278     CUDA_LIBRARY_FIND(cuTexRefGetFlags);
00279     CUDA_LIBRARY_FIND(cuParamSetSize);
00280     CUDA_LIBRARY_FIND(cuParamSeti);
00281     CUDA_LIBRARY_FIND(cuParamSetf);
00282     CUDA_LIBRARY_FIND(cuParamSetv);
00283     CUDA_LIBRARY_FIND(cuParamSetTexRef);
00284     CUDA_LIBRARY_FIND(cuLaunch);
00285     CUDA_LIBRARY_FIND(cuLaunchGrid);
00286     CUDA_LIBRARY_FIND(cuLaunchGridAsync);
00287     CUDA_LIBRARY_FIND(cuEventCreate);
00288     CUDA_LIBRARY_FIND(cuEventRecord);
00289     CUDA_LIBRARY_FIND(cuEventQuery);
00290     CUDA_LIBRARY_FIND(cuEventSynchronize);
00291     CUDA_LIBRARY_FIND(cuEventDestroy);
00292     CUDA_LIBRARY_FIND(cuEventElapsedTime);
00293     CUDA_LIBRARY_FIND(cuStreamCreate);
00294     CUDA_LIBRARY_FIND(cuStreamQuery);
00295     CUDA_LIBRARY_FIND(cuStreamSynchronize);
00296     CUDA_LIBRARY_FIND(cuStreamDestroy);
00297 
00298     /* cuda 2.1 */
00299     CUDA_LIBRARY_FIND(cuModuleLoadDataEx);
00300     CUDA_LIBRARY_FIND(cuModuleLoadFatBinary);
00301     CUDA_LIBRARY_FIND(cuGLCtxCreate);
00302     CUDA_LIBRARY_FIND(cuGraphicsGLRegisterBuffer);
00303     CUDA_LIBRARY_FIND(cuGraphicsGLRegisterImage);
00304 
00305     /* cuda 2.3 */
00306     CUDA_LIBRARY_FIND(cuMemHostGetFlags);
00307     CUDA_LIBRARY_FIND(cuGraphicsGLRegisterBuffer);
00308     CUDA_LIBRARY_FIND(cuGraphicsGLRegisterImage);
00309 
00310     /* cuda 3.0 */
00311     CUDA_LIBRARY_FIND(cuMemcpyDtoDAsync);
00312     CUDA_LIBRARY_FIND(cuFuncSetCacheConfig);
00313     CUDA_LIBRARY_FIND(cuGraphicsUnregisterResource);
00314     CUDA_LIBRARY_FIND(cuGraphicsSubResourceGetMappedArray);
00315     CUDA_LIBRARY_FIND(cuGraphicsResourceGetMappedPointer);
00316     CUDA_LIBRARY_FIND(cuGraphicsResourceSetMapFlags);
00317     CUDA_LIBRARY_FIND(cuGraphicsMapResources);
00318     CUDA_LIBRARY_FIND(cuGraphicsUnmapResources);
00319     CUDA_LIBRARY_FIND(cuGetExportTable);
00320 
00321     /* cuda 3.1 */
00322     CUDA_LIBRARY_FIND(cuModuleGetSurfRef);
00323     CUDA_LIBRARY_FIND(cuSurfRefSetArray);
00324     CUDA_LIBRARY_FIND(cuSurfRefGetArray);
00325     CUDA_LIBRARY_FIND(cuCtxSetLimit);
00326     CUDA_LIBRARY_FIND(cuCtxGetLimit);
00327 
00328     /* functions which changed 3.1 -> 3.2 for 64 bit stuff, the cuda library
00329        has both the old ones for compatibility and new ones with _v2 postfix,
00330        we load the _v2 ones here. */
00331     CUDA_LIBRARY_FIND_V2(cuDeviceTotalMem);
00332     CUDA_LIBRARY_FIND_V2(cuCtxCreate);
00333     CUDA_LIBRARY_FIND_V2(cuModuleGetGlobal);
00334     CUDA_LIBRARY_FIND_V2(cuMemGetInfo);
00335     CUDA_LIBRARY_FIND_V2(cuMemAlloc);
00336     CUDA_LIBRARY_FIND_V2(cuMemAllocPitch);
00337     CUDA_LIBRARY_FIND_V2(cuMemFree);
00338     CUDA_LIBRARY_FIND_V2(cuMemGetAddressRange);
00339     CUDA_LIBRARY_FIND_V2(cuMemAllocHost);
00340     CUDA_LIBRARY_FIND_V2(cuMemHostGetDevicePointer);
00341     CUDA_LIBRARY_FIND_V2(cuMemcpyHtoD);
00342     CUDA_LIBRARY_FIND_V2(cuMemcpyDtoH);
00343     CUDA_LIBRARY_FIND_V2(cuMemcpyDtoD);
00344     CUDA_LIBRARY_FIND_V2(cuMemcpyDtoA);
00345     CUDA_LIBRARY_FIND_V2(cuMemcpyAtoD);
00346     CUDA_LIBRARY_FIND_V2(cuMemcpyHtoA);
00347     CUDA_LIBRARY_FIND_V2(cuMemcpyAtoH);
00348     CUDA_LIBRARY_FIND_V2(cuMemcpyAtoA);
00349     CUDA_LIBRARY_FIND_V2(cuMemcpyHtoAAsync);
00350     CUDA_LIBRARY_FIND_V2(cuMemcpyAtoHAsync);
00351     CUDA_LIBRARY_FIND_V2(cuMemcpy2D);
00352     CUDA_LIBRARY_FIND_V2(cuMemcpy2DUnaligned);
00353     CUDA_LIBRARY_FIND_V2(cuMemcpy3D);
00354     CUDA_LIBRARY_FIND_V2(cuMemcpyHtoDAsync);
00355     CUDA_LIBRARY_FIND_V2(cuMemcpyDtoHAsync);
00356     CUDA_LIBRARY_FIND_V2(cuMemcpyDtoDAsync);
00357     CUDA_LIBRARY_FIND_V2(cuMemcpy2DAsync);
00358     CUDA_LIBRARY_FIND_V2(cuMemcpy3DAsync);
00359     CUDA_LIBRARY_FIND_V2(cuMemsetD8);
00360     CUDA_LIBRARY_FIND_V2(cuMemsetD16);
00361     CUDA_LIBRARY_FIND_V2(cuMemsetD32);
00362     CUDA_LIBRARY_FIND_V2(cuMemsetD2D8);
00363     CUDA_LIBRARY_FIND_V2(cuMemsetD2D16);
00364     CUDA_LIBRARY_FIND_V2(cuMemsetD2D32);
00365     CUDA_LIBRARY_FIND_V2(cuArrayCreate);
00366     CUDA_LIBRARY_FIND_V2(cuArrayGetDescriptor);
00367     CUDA_LIBRARY_FIND_V2(cuArray3DCreate);
00368     CUDA_LIBRARY_FIND_V2(cuArray3DGetDescriptor);
00369     CUDA_LIBRARY_FIND_V2(cuTexRefSetAddress);
00370     CUDA_LIBRARY_FIND_V2(cuTexRefSetAddress2D);
00371     CUDA_LIBRARY_FIND_V2(cuTexRefGetAddress);
00372     CUDA_LIBRARY_FIND_V2(cuGraphicsResourceGetMappedPointer);
00373     CUDA_LIBRARY_FIND_V2(cuGLCtxCreate);
00374 
00375     /* cuda 4.0 */
00376     CUDA_LIBRARY_FIND(cuCtxSetCurrent);
00377 
00378 #ifndef WITH_CUDA_BINARIES
00379 #ifdef _WIN32
00380     return false; /* runtime build doesn't work at the moment */
00381 #else
00382     if(cuCompilerPath() == "")
00383         return false;
00384 #endif
00385 #endif
00386 
00387     /* success */
00388     result = true;
00389 
00390     return result;
00391 }
00392 
00393 string cuCompilerPath()
00394 {
00395 #ifdef _WIN32
00396     const char *defaultpath = "C:/CUDA/bin";
00397     const char *executable = "nvcc.exe";
00398 #else
00399     const char *defaultpath = "/usr/local/cuda/bin";
00400     const char *executable = "nvcc";
00401 #endif
00402 
00403     const char *binpath = getenv("CUDA_BIN_PATH");
00404 
00405     string nvcc;
00406 
00407     if(binpath)
00408         nvcc = path_join(binpath, executable);
00409     else
00410         nvcc = path_join(defaultpath, executable);
00411 
00412     if(path_exists(nvcc))
00413         return nvcc;
00414 
00415 #ifndef _WIN32
00416     if(system("which nvcc") == 0)
00417         return "nvcc";
00418 #endif
00419 
00420     return "";
00421 }
00422 
00423 CCL_NAMESPACE_END
00424