Mercurial > hg > Gears > GearsAgda

comparison src/parallel_execution/helper_cuda.h @ 304:9755206813cb

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
helper_string.h for ANSI C
author Shinji KONO <kono@ie.u-ryukyu.ac.jp>
date Tue, 14 Feb 2017 11:36:41 +0900
parents 1dbaef86593b
children ec0a5b4fba05
comparison
equal deleted inserted replaced
303:1dbaef86593b 304:9755206813cb
286 return "cudaErrorNvlinkUncorrectable"; 286 return "cudaErrorNvlinkUncorrectable";
287 } 287 }
288 288
289 return "<unknown>"; 289 return "<unknown>";
290 } 290 }
291 #endif 291 #else
292
293 #ifdef __cuda_cuda_h__
294 // CUDA Driver API errors 292 // CUDA Driver API errors
295 const char *_cudaGetErrorEnum(CUresult error) 293 const char *_cudaGetErrorEnum(CUresult error)
296 { 294 {
297 switch (error) 295 switch (error)
298 { 296 {
979 #ifndef DEVICE_RESET 977 #ifndef DEVICE_RESET
980 #define DEVICE_RESET 978 #define DEVICE_RESET
981 #endif 979 #endif
982 #endif 980 #endif
983 981
984 template< typename T > 982 #ifndef __DRIVER_TYPES_H__
985 void check(T result, char const *const func, const char *const file, int const line) 983 static inline void check(CUresult result, char const *const func, const char *const file, int const line)
986 { 984 {
987 if (result) 985 if (result)
988 { 986 {
989 fprintf(stderr, "CUDA error at %s:%d code=%d(%s) \"%s\" \n", 987 fprintf(stderr, "CUDA error at %s:%d code=%d(%s) \"%s\" \n",
990 file, line, static_cast<unsigned int>(result), _cudaGetErrorEnum(result), func); 988 file, line, (unsigned int)(result), _cudaGetErrorEnum(result), func);
991 DEVICE_RESET 989 DEVICE_RESET
992 // Make sure we call CUDA Device Reset before exiting 990 // Make sure we call CUDA Device Reset before exiting
993 exit(EXIT_FAILURE); 991 exit(EXIT_FAILURE);
994 } 992 }
995 } 993 }
994 #else
995 static inline void check(cudaError_t result, char const *const func, const char *const file, int const line)
996 {
997 if (result)
998 {
999 fprintf(stderr, "CUDA error at %s:%d code=%d(%s) \"%s\" \n",
1000 file, line, (unsigned int)(result), _cudaGetErrorEnum(result), func);
1001 DEVICE_RESET
1002 // Make sure we call CUDA Device Reset before exiting
1003 exit(EXIT_FAILURE);
1004 }
1005 }
1006 #endif
996 1007
997 #ifdef __DRIVER_TYPES_H__ 1008 #ifdef __DRIVER_TYPES_H__
998 // This will output the proper CUDA error strings in the event that a CUDA host call returns an error 1009 // This will output the proper CUDA error strings in the event that a CUDA host call returns an error
999 #define checkCudaErrors(val) check ( (val), #val, __FILE__, __LINE__ ) 1010 #define checkCudaErrors(val) check ( (val), #val, __FILE__, __LINE__ )
1000 1011
1027 1038
1028 // Beginning of GPU Architecture definitions 1039 // Beginning of GPU Architecture definitions
1029 inline int _ConvertSMVer2Cores(int major, int minor) 1040 inline int _ConvertSMVer2Cores(int major, int minor)
1030 { 1041 {
1031 // Defines for GPU Architecture types (using the SM version to determine the # of cores per SM 1042 // Defines for GPU Architecture types (using the SM version to determine the # of cores per SM
1032 typedef struct 1043 typedef struct sSMtoCores
1033 { 1044 {
1034 int SM; // 0xMm (hexidecimal notation), M = SM Major version, and m = SM minor version 1045 int SM; // 0xMm (hexidecimal notation), M = SM Major version, and m = SM minor version
1035 int Cores; 1046 int Cores;
1036 } sSMtoCores; 1047 } sSMtoCores;
1037 1048
1095 fprintf(stderr, ">> gpuDeviceInit (-device=%d) is not a valid GPU device. <<\n", devID); 1106 fprintf(stderr, ">> gpuDeviceInit (-device=%d) is not a valid GPU device. <<\n", devID);
1096 fprintf(stderr, "\n"); 1107 fprintf(stderr, "\n");
1097 return -devID; 1108 return -devID;
1098 } 1109 }
1099 1110
1100 cudaDeviceProp deviceProp; 1111 struct cudaDeviceProp deviceProp;
1101 checkCudaErrors(cudaGetDeviceProperties(&deviceProp, devID)); 1112 checkCudaErrors(cudaGetDeviceProperties(&deviceProp, devID));
1102 1113
1103 if (deviceProp.computeMode == cudaComputeModeProhibited) 1114 if (deviceProp.computeMode == cudaComputeModeProhibited)
1104 { 1115 {
1105 fprintf(stderr, "Error: device is running in <Compute Mode Prohibited>, no threads can use ::cudaSetDevice().\n"); 1116 fprintf(stderr, "Error: device is running in <Compute Mode Prohibited>, no threads can use ::cudaSetDevice().\n");
1125 int max_perf_device = 0; 1136 int max_perf_device = 0;
1126 int device_count = 0, best_SM_arch = 0; 1137 int device_count = 0, best_SM_arch = 0;
1127 int devices_prohibited = 0; 1138 int devices_prohibited = 0;
1128 1139
1129 unsigned long long max_compute_perf = 0; 1140 unsigned long long max_compute_perf = 0;
1130 cudaDeviceProp deviceProp; 1141 struct cudaDeviceProp deviceProp;
1131 cudaGetDeviceCount(&device_count); 1142 cudaGetDeviceCount(&device_count);
1132 1143
1133 checkCudaErrors(cudaGetDeviceCount(&device_count)); 1144 checkCudaErrors(cudaGetDeviceCount(&device_count));
1134 1145
1135 if (device_count == 0) 1146 if (device_count == 0)
1214 1225
1215 1226
1216 // Initialization code to find the best CUDA Device 1227 // Initialization code to find the best CUDA Device
1217 inline int findCudaDevice(int argc, const char **argv) 1228 inline int findCudaDevice(int argc, const char **argv)
1218 { 1229 {
1219 cudaDeviceProp deviceProp; 1230 struct cudaDeviceProp deviceProp;
1220 int devID = 0; 1231 int devID = 0;
1221 1232
1222 // If the command-line has a device number specified, use it 1233 // If the command-line has a device number specified, use it
1223 if (checkCmdLineFlag(argc, argv, "device")) 1234 if (checkCmdLineFlag(argc, argv, "device"))
1224 { 1235 {
1253 } 1264 }
1254 1265
1255 // General check for CUDA GPU SM Capabilities 1266 // General check for CUDA GPU SM Capabilities
1256 inline bool checkCudaCapabilities(int major_version, int minor_version) 1267 inline bool checkCudaCapabilities(int major_version, int minor_version)
1257 { 1268 {
1258 cudaDeviceProp deviceProp; 1269 struct cudaDeviceProp deviceProp;
1259 deviceProp.major = 0; 1270 deviceProp.major = 0;
1260 deviceProp.minor = 0; 1271 deviceProp.minor = 0;
1261 int dev; 1272 int dev;
1262 1273
1263 checkCudaErrors(cudaGetDevice(&dev)); 1274 checkCudaErrors(cudaGetDevice(&dev));