API tools faq
paste
Advertisement
Guest User

Untitled

a guest
Jul 29th, 2014
387
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 13.79 KB | None | 0 0
raw download clone embed print report
  1.  
  2. $ su
  3. Password:
  4. # nvidia-smi -i 0 -c 3
  5. Set compute mode to EXCLUSIVE_PROCESS for GPU 0000:02:00.0.
  6. All done.
  7. # exit
  8. exit
  9. $ cat t349.cu
  10. #include <stdio.h>
  11. #include <stdlib.h>
  12. #include <pthread.h>
  13.  
  14. #define ARR_SIZE 10
  15. #define NUM_DEVICE 1
  16. #define NUM_THR 4
  17.  
  18. typedef struct {
  19. int *arr;
  20. int *dev_arr;
  21. int *dev_result;
  22. int *result;
  23. int dev_num;
  24. int thr_num;
  25. } cuda_st;
  26.  
  27. __global__ void kernel_fc(int *dev_arr, int *dev_result)
  28. {
  29. int idx = threadIdx.x;
  30. printf("dev_arr[%d] = %d\n", idx, dev_arr[idx]);
  31. atomicAdd(dev_result, dev_arr[idx]);
  32. }
  33.  
  34. void *thread_func(void* struc)
  35. {
  36. cuda_st * data = (cuda_st*)struc;
  37. printf("thread %d func start\n", data->thr_num);
  38. printf("arr %d = ", data->dev_num);
  39. for(int i=0; i<10; i++) {
  40. printf("%d ", data->arr[i]);
  41. }
  42. printf("\n");
  43. cudaSetDevice(data->dev_num);
  44. cudaMemcpy(data->dev_arr, data->arr, sizeof(int)*ARR_SIZE, cudaMemcpyHostToDevice);
  45. kernel_fc<<<1,ARR_SIZE>>>(data->dev_arr, data->dev_result);
  46. cudaMemcpy(data->result, data->dev_result, sizeof(int), cudaMemcpyDeviceToHost);
  47. printf("thread %d func exit\n", data->thr_num);
  48. return NULL;
  49. }
  50.  
  51. int main(void)
  52. {
  53. // Make object
  54. cuda_st cuda[NUM_DEVICE][NUM_THR];
  55.  
  56. // Make thread
  57. pthread_t pthread[NUM_DEVICE*NUM_THR];
  58.  
  59. // Host array memory allocation
  60. int *arr[NUM_DEVICE];
  61. for(int i=0; i<NUM_DEVICE; i++) {
  62. arr[i] = (int*)malloc(sizeof(int)*ARR_SIZE);
  63. }
  64.  
  65. // Fill this host array up with specified data
  66. for(int i=0; i<NUM_DEVICE; i++) {
  67. for(int j=0; j<ARR_SIZE; j++) {
  68. arr[i][j] = i*ARR_SIZE+j;
  69. }
  70. }
  71.  
  72. // To confirm host array data
  73. for(int i=0; i<NUM_DEVICE; i++) {
  74. printf("arr[%d] = ", i);
  75. for(int j=0; j<ARR_SIZE; j++) {
  76. printf("%d ", arr[i][j]);
  77. }
  78. printf("\n");
  79. }
  80.  
  81. // Result memory allocation
  82. int *result[NUM_DEVICE];
  83. for(int i=0; i<NUM_DEVICE; i++) {
  84. result[i] = (int*)malloc(sizeof(int));
  85. memset(result[i], 0, sizeof(int));
  86. }
  87.  
  88. // Device array memory allocation
  89. int *dev_arr[NUM_DEVICE];
  90. for(int i=0; i<NUM_DEVICE; i++) {
  91. cudaSetDevice(i);
  92. cudaMalloc(&dev_arr[i], sizeof(int)*ARR_SIZE);
  93. }
  94.  
  95. // Device result memory allocation
  96. int *dev_result[NUM_DEVICE];
  97. for(int i=0; i<NUM_DEVICE; i++) {
  98. cudaSetDevice(i);
  99. cudaMalloc(&dev_result[i], sizeof(int));
  100. cudaMemset(dev_result[i], 0, sizeof(int));
  101. }
  102.  
  103. // Connect these pointers with object
  104. for (int i=0; i<NUM_DEVICE; i++)
  105. for (int j=0; j<NUM_THR; j++) {
  106. cuda[i][j].arr = arr[i];
  107. cuda[i][j].dev_arr = dev_arr[i];
  108. cuda[i][j].result = result[i];
  109. cuda[i][j].dev_result = dev_result[i];
  110. cuda[i][j].dev_num = i;
  111. cuda[i][j].thr_num = j;
  112. }
  113.  
  114. // Create and excute pthread
  115. for(int i=0; i<NUM_DEVICE; i++)
  116. for (int j=0; j<NUM_THR; j++) {
  117. pthread_create(&pthread[(i*NUM_THR)+j], NULL, thread_func, (void*)&cuda[i][j]);
  118. }
  119.  
  120. // Join pthread
  121. for(int i=0; i<NUM_DEVICE*NUM_THR; i++) {
  122. pthread_join(pthread[i], NULL);
  123. }
  124.  
  125. for(int i=0; i<NUM_DEVICE; i++)
  126. for (int j=0; j < NUM_THR; j++) {
  127. printf("result[%d][%d] = %d\n", i,j, (*cuda[i][j].result));
  128. }
  129.  
  130. return 0;
  131. }
  132.  
  133. $ nvcc -arch=sm_20 -o t349 t349.cu -lpthread
  134. $ cuda-memcheck ./t349
  135. ========= CUDA-MEMCHECK
  136. arr[0] = 0 1 2 3 4 5 6 7 8 9
  137. thread 0 func start
  138. arr 0 = 0 1 2 3 4 5 6 7 8 9
  139. thread 3 func start
  140. arr 0 = thread 1 func start
  141. arr 0 = 0 1 2 3 4 5 6 7 8 9
  142. 0 1 2 3 4 5 6 7 8 9
  143. thread 2 func start
  144. arr 0 = 0 1 2 3 4 5 6 7 8 9
  145. dev_arr[0] = 0
  146. dev_arr[1] = 1
  147. dev_arr[2] = 2
  148. dev_arr[3] = 3
  149. dev_arr[4] = 4
  150. dev_arr[5] = 5
  151. dev_arr[6] = 6
  152. dev_arr[7] = 7
  153. dev_arr[8] = 8
  154. dev_arr[9] = 9
  155. thread 0 func exit
  156. dev_arr[0] = 0
  157. dev_arr[1] = 1
  158. dev_arr[2] = 2
  159. dev_arr[3] = 3
  160. dev_arr[4] = 4
  161. dev_arr[5] = 5
  162. dev_arr[6] = 6
  163. dev_arr[7] = 7
  164. dev_arr[8] = 8
  165. dev_arr[9] = 9
  166. dev_arr[0] = 0
  167. dev_arr[1] = 1
  168. dev_arr[2] = 2
  169. dev_arr[3] = 3
  170. dev_arr[4] = 4
  171. dev_arr[5] = 5
  172. dev_arr[6] = 6
  173. dev_arr[7] = 7
  174. dev_arr[8] = 8
  175. dev_arr[9] = 9
  176. thread 3 func exit
  177. thread 1 func exit
  178. dev_arr[0] = 0
  179. dev_arr[1] = 1
  180. dev_arr[2] = 2
  181. dev_arr[3] = 3
  182. dev_arr[4] = 4
  183. dev_arr[5] = 5
  184. dev_arr[6] = 6
  185. dev_arr[7] = 7
  186. dev_arr[8] = 8
  187. dev_arr[9] = 9
  188. thread 2 func exit
  189. result[0][0] = 180
  190. result[0][1] = 180
  191. result[0][2] = 180
  192. result[0][3] = 180
  193. ========= ERROR SUMMARY: 0 errors
  194. $ su
  195. Password:
  196. # nvidia-smi -i 0 -c 1
  197. Set compute mode to EXCLUSIVE_THREAD for GPU 0000:02:00.0.
  198. All done.
  199. # exit
  200. exit
  201. $ cuda-memcheck ./t349
  202. ========= CUDA-MEMCHECK
  203. arr[0] = 0 1 2 3 4 5 6 7 8 9
  204. thread 0 func start
  205. arr 0 = 0 1 2 3 4 5 6 7 8 9
  206. thread 1 func start
  207. arr 0 = 0 1 2 3 4 5 6 7 8 9
  208. ========= Program hit cudaErrorDeviceAlreadyInUse (error 54) due to "exclusive-thread device already in use by a different thread" on CUDA API call to cudaSetDevice.
  209. ========= Saved host backtrace up to driver entry point at error
  210. ========= Host Frame:/usr/lib64/libcuda.so.1 [0x2ee943]
  211. ========= Host Frame:./t349 [0x347d0]
  212. ========= Host Frame:./t349 [0x2b4e]
  213. ========= Host Frame:/lib64/libpthread.so.0 [0x673d]
  214. ========= Host Frame:/lib64/libc.so.6 (clone + 0x6d) [0xd3d1d]
  215. =========
  216. thread 2 func start
  217. arr 0 = 0 1 2 3 4 5 6 7 8 9
  218. thread 3 func start
  219. arr 0 = 0 1 2 3 4 5 6 7 8 9
  220. ========= Program hit cudaErrorDeviceAlreadyInUse (error 54) due to "exclusive-thread device already in use by a different thread" on CUDA API call to cudaSetDevice.
  221. ========= Saved host backtrace up to driver entry point at error
  222. ========= Host Frame:/usr/lib64/libcuda.so.1 [0x2ee943]
  223. ========= Host Frame:./t349 [0x347d0]
  224. ========= Host Frame:./t349 [0x2b4e]
  225. ========= Host Frame:/lib64/libpthread.so.0 [0x673d]
  226. ========= Host Frame:/lib64/libc.so.6 (clone + 0x6d) [0xd3d1d]
  227. =========
  228. ========= Program hit cudaErrorDeviceAlreadyInUse (error 54) due to "exclusive-thread device already in use by a different thread" on CUDA API call to cudaSetDevice.
  229. ========= Saved host backtrace up to driver entry point at error
  230. ========= Host Frame:/usr/lib64/libcuda.so.1 [0x2ee943]
  231. ========= Host Frame:./t349 [0x347d0]
  232. ========= Host Frame:./t349 [0x2b4e]
  233. ========= Host Frame:/lib64/libpthread.so.0 [0x673d]
  234. ========= Host Frame:/lib64/libc.so.6 (clone + 0x6d) [0xd3d1d]
  235. =========
  236. ========= Program hit cudaErrorDeviceAlreadyInUse (error 54) due to "exclusive-thread device already in use by a different thread" on CUDA API call to cudaSetDevice.
  237. ========= Saved host backtrace up to driver entry point at error
  238. ========= Host Frame:/usr/lib64/libcuda.so.1 [0x2ee943]
  239. ========= Host Frame:./t349 [0x347d0]
  240. ========= Host Frame:./t349 [0x2b4e]
  241. ========= Host Frame:/lib64/libpthread.so.0 [0x673d]
  242. ========= Host Frame:/lib64/libc.so.6 (clone + 0x6d) [0xd3d1d]
  243. =========
  244. ========= Program hit cudaErrorDevicesUnavailable (error 46) due to "all CUDA-capable devices are busy or unavailable" on CUDA API call to cudaMemcpy.
  245. ========= Saved host backtrace up to driver entry point at error
  246. ========= Host Frame:/usr/lib64/libcuda.so.1 [0x2ee943]
  247. ========= Host Frame:./t349 [0x3831f]
  248. ========= Host Frame:./t349 [0x2b6c]
  249. ========= Host Frame:/lib64/libpthread.so.0 [0x673d]
  250. ========= Host Frame:/lib64/libc.so.6 (clone + 0x6d) [0xd3d1d]
  251. =========
  252. ========= Program hit cudaErrorDevicesUnavailable (error 46) due to "all CUDA-capable devices are busy or unavailable" on CUDA API call to cudaMemcpy.
  253. ========= Saved host backtrace up to driver entry point at error
  254. ========= Host Frame:/usr/lib64/libcuda.so.1 [0x2ee943]
  255. ========= Host Frame:./t349 [0x3831f]
  256. ========= Host Frame:./t349 [0x2b6c]
  257. ========= Host Frame:/lib64/libpthread.so.0 [0x673d]
  258. ========= Host Frame:/lib64/libc.so.6 (clone + 0x6d) [0xd3d1d]
  259. =========
  260. ========= Program hit cudaErrorDevicesUnavailable (error 46) due to "all CUDA-capable devices are busy or unavailable" on CUDA API call to cudaMemcpy.
  261. ========= Saved host backtrace up to driver entry point at error
  262. ========= Host Frame:/usr/lib64/libcuda.so.1 [0x2ee943]
  263. ========= Host Frame:./t349 [0x3831f]
  264. ========= Host Frame:./t349 [0x2b6c]
  265. ========= Host Frame:/lib64/libpthread.so.0 [0x673d]
  266. ========= Host Frame:/lib64/libc.so.6 (clone + 0x6d) [0xd3d1d]
  267. =========
  268. ========= Program hit cudaErrorDevicesUnavailable (error 46) due to "all CUDA-capable devices are busy or unavailable" on CUDA API call to cudaMemcpy.
  269. ========= Saved host backtrace up to driver entry point at error
  270. ========= Host Frame:/usr/lib64/libcuda.so.1 [0x2ee943]
  271. ========= Host Frame:./t349 [0x3831f]
  272. ========= Host Frame:./t349 [0x2b6c]
  273. ========= Host Frame:/lib64/libpthread.so.0 [0x673d]
  274. ========= Host Frame:/lib64/libc.so.6 (clone + 0x6d) [0xd3d1d]
  275. =========
  276. ========= Program hit cudaErrorDevicesUnavailable (error 46) due to "all CUDA-capable devices are busy or unavailable" on CUDA API call to cudaLaunch.
  277. ========= Saved host backtrace up to driver entry point at error
  278. ========= Host Frame:/usr/lib64/libcuda.so.1 [0x2ee943]
  279. ========= Host Frame:./t349 [0x3b47e]
  280. ========= Host Frame:./t349 [0x2c55]
  281. ========= Host Frame:./t349 (__gxx_personality_v0 + 0x306) [0x25be]
  282. ========= Host Frame:./t349 (__gxx_personality_v0 + 0x325) [0x25dd]
  283. ========= Host Frame:./t349 [0x2bd4]
  284. ========= Host Frame:/lib64/libpthread.so.0 [0x673d]
  285. ========= Host Frame:/lib64/libc.so.6 (clone + 0x6d) [0xd3d1d]
  286. =========
  287. ========= Program hit cudaErrorDevicesUnavailable (error 46) due to "all CUDA-capable devices are busy or unavailable" on CUDA API call to cudaLaunch.
  288. ========= Saved host backtrace up to driver entry point at error
  289. ========= Host Frame:/usr/lib64/libcuda.so.1 [0x2ee943]
  290. ========= Host Frame:./t349 [0x3b47e]
  291. ========= Host Frame:./t349 [0x2c55]
  292. ========= Host Frame:./t349 (__gxx_personality_v0 + 0x306) [0x25be]
  293. ========= Host Frame:./t349 (__gxx_personality_v0 + 0x325) [0x25dd]
  294. ========= Host Frame:./t349 [0x2bd4]
  295. ========= Host Frame:/lib64/libpthread.so.0 [0x673d]
  296. ========= Host Frame:/lib64/libc.so.6 (clone + 0x6d) [0xd3d1d]
  297. =========
  298. ========= Program hit cudaErrorDevicesUnavailable (error 46) due to "all CUDA-capable devices are busy or unavailable" on CUDA API call to cudaLaunch.
  299. ========= Saved host backtrace up to driver entry point at error
  300. ========= Host Frame:/usr/lib64/libcuda.so.1 [0x2ee943]
  301. ========= Host Frame:./t349 [0x3b47e]
  302. ========= Host Frame:./t349 [0x2c55]
  303. ========= Host Frame:./t349 (__gxx_personality_v0 + 0x306) [0x25be]
  304. ========= Host Frame:./t349 (__gxx_personality_v0 + 0x325) [0x25dd]
  305. ========= Host Frame:./t349 [0x2bd4]
  306. ========= Host Frame:/lib64/libpthread.so.0 [0x673d]
  307. ========= Host Frame:/lib64/libc.so.6 (clone + 0x6d) [0xd3d1d]
  308. =========
  309. ========= Program hit cudaErrorDevicesUnavailable (error 46) due to "all CUDA-capable devices are busy or unavailable" on CUDA API call to cudaMemcpy.
  310. ========= Saved host backtrace up to driver entry point at error
  311. ========= Host Frame:/usr/lib64/libcuda.so.1 [0x2ee943]
  312. ========= Host Frame:./t349 [0x3831f]
  313. ========= Host Frame:./t349 [0x2bf3]
  314. thread 1 func exit
  315. ========= Host Frame:/lib64/libpthread.so.0 [0x673d]
  316. ========= Host Frame:/lib64/libc.so.6 (clone + 0x6d) [0xd3d1d]
  317. =========
  318. ========= Program hit cudaErrorDevicesUnavailable (error 46) due to "all CUDA-capable devices are busy or unavailable" on CUDA API call to cudaLaunch.
  319. ========= Saved host backtrace up to driver entry point at error
  320. ========= Host Frame:/usr/lib64/libcuda.so.1 [0x2ee943]
  321. ========= Host Frame:./t349 [0x3b47e]
  322. ========= Host Frame:./t349 [0x2c55]
  323. ========= Host Frame:./t349 (__gxx_personality_v0 + 0x306) [0x25be]
  324. ========= Host Frame:./t349 (__gxx_personality_v0 + 0x325) [0x25dd]
  325. ========= Host Frame:./t349 [0x2bd4]
  326. ========= Host Frame:/lib64/libpthread.so.0 [0x673d]
  327. ========= Host Frame:/lib64/libc.so.6 (clone + 0x6d) [0xd3d1d]
  328. =========
  329. thread 0 func exit
  330. ========= Program hit cudaErrorDevicesUnavailable (error 46) due to "all CUDA-capable devices are busy or unavailable" on CUDA API call to cudaMemcpy.
  331. ========= Saved host backtrace up to driver entry point at error
  332. ========= Host Frame:/usr/lib64/libcuda.so.1 [0x2ee943]
  333. ========= Host Frame:./t349 [0x3831f]
  334. ========= Host Frame:./t349 [0x2bf3]
  335. ========= Host Frame:/lib64/libpthread.so.0 [0x673d]
  336. ========= Host Frame:/lib64/libc.so.6 (clone + 0x6d) [0xd3d1d]
  337. =========
  338. ========= Program hit cudaErrorDevicesUnavailable (error 46) due to "all CUDA-capable devices are busy or unavailable" on CUDA API call to cudaMemcpy.
  339. thread 2 func exit
  340. ========= Saved host backtrace up to driver entry point at error
  341. ========= Host Frame:/usr/lib64/libcuda.so.1 [0x2ee943]
  342. ========= Host Frame:./t349 [0x3831f]
  343. ========= Host Frame:./t349 [0x2bf3]
  344. ========= Host Frame:/lib64/libpthread.so.0 [0x673d]
  345. ========= Host Frame:/lib64/libc.so.6 (clone + 0x6d) [0xd3d1d]
  346. =========
  347. ========= Program hit cudaErrorDevicesUnavailable (error 46) due to "all CUDA-capable devices are busy or unavailable" on CUDA API call to cudaMemcpy.
  348. thread 3 func exit
  349. ========= Saved host backtrace up to driver entry point at error
  350. ========= Host Frame:/usr/lib64/libcuda.so.1 [0x2ee943]
  351. ========= Host Frame:./t349 [0x3831f]
  352. ========= Host Frame:./t349 [0x2bf3]
  353. ========= Host Frame:/lib64/libpthread.so.0 [0x673d]
  354. result[0][0] = 0
  355. ========= Host Frame:/lib64/libc.so.6 (clone + 0x6d) [0xd3d1d]
  356. result[0][1] = 0
  357. =========
  358. result[0][2] = 0
  359. result[0][3] = 0
  360. ========= ERROR SUMMARY: 16 errors
  361. $
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement
Public Pastes
Advertisement
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!