| 204 | |
| 205 | |
| 206 | template <typename ElemType> nano_time_t |
| 207 | ScalPerformanceTest<ElemType>::clblasPerfSingle(void) |
| 208 | { |
| 209 | nano_time_t time; |
| 210 | cl_event event; |
| 211 | cl_int status; |
| 212 | cl_command_queue queue = base_->commandQueues()[0]; |
| 213 | bool is_css_zds = (params_.K == 1)? true: false; // K indicates csscal/zdscal |
| 214 | |
| 215 | status = clEnqueueWriteBuffer(queue, mobjX_, CL_TRUE, 0, |
| 216 | (lengthX + params_.offBX) * sizeof(ElemType), backX_, 0, NULL, &event); |
| 217 | if (status != CL_SUCCESS) { |
| 218 | cerr << "Matrix A buffer object enqueuing error, status = " << |
| 219 | status << endl; |
| 220 | |
| 221 | return NANOTIME_ERR; |
| 222 | } |
| 223 | |
| 224 | status = clWaitForEvents(1, &event); |
| 225 | if (status != CL_SUCCESS) { |
| 226 | cout << "Wait on event failed, status = " << |
| 227 | status << endl; |
| 228 | |
| 229 | return NANOTIME_ERR; |
| 230 | } |
| 231 | |
| 232 | event = NULL; |
| 233 | time = getCurrentTime(); |
| 234 | |
| 235 | #define TIMING |
| 236 | #ifdef TIMING |
| 237 | clFinish( queue); |
| 238 | int iter = 50; |
| 239 | for ( int i=1; i <= iter; i++) |
| 240 | { |
| 241 | #endif |
| 242 | |
| 243 | status = (cl_int)clMath::clblas::scal(is_css_zds, params_.N, alpha_, mobjX_, params_.offBX, params_.incx, |
| 244 | 1, &queue, 0, NULL, &event); |
| 245 | if (status != CL_SUCCESS) { |
| 246 | cerr << "The CLBLAS SCAL function failed, status = " << |
| 247 | status << endl; |
| 248 | |
| 249 | return NANOTIME_ERR; |
| 250 | } |
| 251 | #ifdef TIMING |
| 252 | } // iter loop |
| 253 | clFinish( queue); |
| 254 | time = getCurrentTime() - time; |
| 255 | time /= iter; |
| 256 | #else |
| 257 | |
| 258 | status = flushAll(1, &queue); |
| 259 | if (status != CL_SUCCESS) { |
| 260 | cerr << "clFlush() failed, status = " << status << endl; |
| 261 | return NANOTIME_ERR; |
| 262 | } |
| 263 |
nothing calls this directly
no test coverage detected