csr_interface.c 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428
  1. /* StarPU --- Runtime system for heterogeneous multicore architectures.
  2. *
  3. * Copyright (C) 2008-2020 Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
  4. * Copyright (C) 2010 Mehdi Juhoor
  5. *
  6. * StarPU is free software; you can redistribute it and/or modify
  7. * it under the terms of the GNU Lesser General Public License as published by
  8. * the Free Software Foundation; either version 2.1 of the License, or (at
  9. * your option) any later version.
  10. *
  11. * StarPU is distributed in the hope that it will be useful, but
  12. * WITHOUT ANY WARRANTY; without even the implied warranty of
  13. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  14. *
  15. * See the GNU Lesser General Public License in COPYING.LGPL for more details.
  16. */
  17. #include <starpu.h>
  18. #ifdef BUILDING_STARPU
  19. #include <datawizard/memory_nodes.h>
  20. #endif
  21. static int copy_any_to_any(void *src_interface, unsigned src_node, void *dst_interface, unsigned dst_node, void *async_data);
  22. static const struct starpu_data_copy_methods csr_copy_data_methods_s =
  23. {
  24. .any_to_any = copy_any_to_any,
  25. };
  26. static void register_csr_handle(starpu_data_handle_t handle, unsigned home_node, void *data_interface);
  27. static int csr_pointer_is_inside(void *data_interface, unsigned node, void *ptr);
  28. static starpu_ssize_t allocate_csr_buffer_on_node(void *data_interface_, unsigned dst_node);
  29. static void free_csr_buffer_on_node(void *data_interface, unsigned node);
  30. static size_t csr_interface_get_size(starpu_data_handle_t handle);
  31. static int csr_compare(void *data_interface_a, void *data_interface_b);
  32. static uint32_t footprint_csr_interface_crc32(starpu_data_handle_t handle);
  33. static starpu_ssize_t describe(void *data_interface, char *buf, size_t size);
  34. static int pack_data(starpu_data_handle_t handle, unsigned node, void **ptr, starpu_ssize_t *count);
  35. static int unpack_data(starpu_data_handle_t handle, unsigned node, void *ptr, size_t count);
  36. struct starpu_data_interface_ops starpu_interface_csr_ops =
  37. {
  38. .register_data_handle = register_csr_handle,
  39. .allocate_data_on_node = allocate_csr_buffer_on_node,
  40. .free_data_on_node = free_csr_buffer_on_node,
  41. .copy_methods = &csr_copy_data_methods_s,
  42. .get_size = csr_interface_get_size,
  43. .interfaceid = STARPU_CSR_INTERFACE_ID,
  44. .interface_size = sizeof(struct starpu_csr_interface),
  45. .footprint = footprint_csr_interface_crc32,
  46. .compare = csr_compare,
  47. .describe = describe,
  48. .pointer_is_inside = csr_pointer_is_inside,
  49. .name = "STARPU_CSR_INTERFACE",
  50. .pack_data = pack_data,
  51. .unpack_data = unpack_data
  52. };
  53. static int csr_pointer_is_inside(void *data_interface, unsigned node, void *ptr)
  54. {
  55. (void) node;
  56. struct starpu_csr_interface *csr_interface = data_interface;
  57. return ((char*) ptr >= (char*) csr_interface->nzval &&
  58. (char*) ptr < (char*) csr_interface->nzval + csr_interface->nnz*csr_interface->elemsize)
  59. || ((char*) ptr >= (char*) csr_interface->colind &&
  60. (char*) ptr < (char*) csr_interface->colind + csr_interface->nnz*sizeof(uint32_t))
  61. || ((char*) ptr >= (char*) csr_interface->rowptr &&
  62. (char*) ptr < (char*) csr_interface->rowptr + (csr_interface->nrow+1)*sizeof(uint32_t));
  63. }
  64. static void register_csr_handle(starpu_data_handle_t handle, unsigned home_node, void *data_interface)
  65. {
  66. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *) data_interface;
  67. unsigned node;
  68. for (node = 0; node < STARPU_MAXNODES; node++)
  69. {
  70. struct starpu_csr_interface *local_interface = (struct starpu_csr_interface *)
  71. starpu_data_get_interface_on_node(handle, node);
  72. if (node == home_node)
  73. {
  74. local_interface->nzval = csr_interface->nzval;
  75. local_interface->colind = csr_interface->colind;
  76. }
  77. else
  78. {
  79. local_interface->nzval = 0;
  80. local_interface->colind = NULL;
  81. }
  82. local_interface->id = csr_interface->id;
  83. local_interface->rowptr = csr_interface->rowptr;
  84. local_interface->nnz = csr_interface->nnz;
  85. local_interface->nrow = csr_interface->nrow;
  86. local_interface->firstentry = csr_interface->firstentry;
  87. local_interface->elemsize = csr_interface->elemsize;
  88. }
  89. }
  90. /* declare a new data with the BLAS interface */
  91. void starpu_csr_data_register(starpu_data_handle_t *handleptr, int home_node,
  92. uint32_t nnz, uint32_t nrow, uintptr_t nzval, uint32_t *colind, uint32_t *rowptr, uint32_t firstentry, size_t elemsize)
  93. {
  94. struct starpu_csr_interface csr_interface =
  95. {
  96. .id = STARPU_CSR_INTERFACE_ID,
  97. .nnz = nnz,
  98. .nrow = nrow,
  99. .nzval = nzval,
  100. .colind = colind,
  101. .rowptr = rowptr,
  102. .firstentry = firstentry,
  103. .elemsize = elemsize
  104. };
  105. #ifndef STARPU_SIMGRID
  106. if (home_node >= 0 && starpu_node_get_kind(home_node) == STARPU_CPU_RAM)
  107. {
  108. if (nnz)
  109. {
  110. if (elemsize)
  111. {
  112. STARPU_ASSERT_ACCESSIBLE(nzval);
  113. STARPU_ASSERT_ACCESSIBLE(nzval + nnz*elemsize - 1);
  114. }
  115. STARPU_ASSERT_ACCESSIBLE(colind);
  116. STARPU_ASSERT_ACCESSIBLE((uintptr_t) colind + nnz*sizeof(uint32_t) - 1);
  117. }
  118. STARPU_ASSERT_ACCESSIBLE(rowptr);
  119. STARPU_ASSERT_ACCESSIBLE((uintptr_t) rowptr + (nrow+1)*sizeof(uint32_t) - 1);
  120. }
  121. #endif
  122. starpu_data_register(handleptr, home_node, &csr_interface, &starpu_interface_csr_ops);
  123. }
  124. static uint32_t footprint_csr_interface_crc32(starpu_data_handle_t handle)
  125. {
  126. return starpu_hash_crc32c_be(starpu_csr_get_nnz(handle), 0);
  127. }
  128. static int csr_compare(void *data_interface_a, void *data_interface_b)
  129. {
  130. struct starpu_csr_interface *csr_a = (struct starpu_csr_interface *) data_interface_a;
  131. struct starpu_csr_interface *csr_b = (struct starpu_csr_interface *) data_interface_b;
  132. /* Two matricess are considered compatible if they have the same size */
  133. return (csr_a->nnz == csr_b->nnz)
  134. && (csr_a->nrow == csr_b->nrow)
  135. && (csr_a->elemsize == csr_b->elemsize);
  136. }
  137. /* offer an access to the data parameters */
  138. uint32_t starpu_csr_get_nnz(starpu_data_handle_t handle)
  139. {
  140. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *)
  141. starpu_data_get_interface_on_node(handle, STARPU_MAIN_RAM);
  142. #ifdef STARPU_DEBUG
  143. STARPU_ASSERT_MSG(csr_interface->id == STARPU_CSR_INTERFACE_ID, "Error. The given data is not a csr.");
  144. #endif
  145. return csr_interface->nnz;
  146. }
  147. uint32_t starpu_csr_get_nrow(starpu_data_handle_t handle)
  148. {
  149. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *)
  150. starpu_data_get_interface_on_node(handle, STARPU_MAIN_RAM);
  151. #ifdef STARPU_DEBUG
  152. STARPU_ASSERT_MSG(csr_interface->id == STARPU_CSR_INTERFACE_ID, "Error. The given data is not a csr.");
  153. #endif
  154. return csr_interface->nrow;
  155. }
  156. uint32_t starpu_csr_get_firstentry(starpu_data_handle_t handle)
  157. {
  158. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *)
  159. starpu_data_get_interface_on_node(handle, STARPU_MAIN_RAM);
  160. #ifdef STARPU_DEBUG
  161. STARPU_ASSERT_MSG(csr_interface->id == STARPU_CSR_INTERFACE_ID, "Error. The given data is not a csr.");
  162. #endif
  163. return csr_interface->firstentry;
  164. }
  165. size_t starpu_csr_get_elemsize(starpu_data_handle_t handle)
  166. {
  167. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *)
  168. starpu_data_get_interface_on_node(handle, STARPU_MAIN_RAM);
  169. #ifdef STARPU_DEBUG
  170. STARPU_ASSERT_MSG(csr_interface->id == STARPU_CSR_INTERFACE_ID, "Error. The given data is not a csr.");
  171. #endif
  172. return csr_interface->elemsize;
  173. }
  174. uintptr_t starpu_csr_get_local_nzval(starpu_data_handle_t handle)
  175. {
  176. unsigned node;
  177. node = starpu_worker_get_local_memory_node();
  178. STARPU_ASSERT(starpu_data_test_if_allocated_on_node(handle, node));
  179. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *)
  180. starpu_data_get_interface_on_node(handle, node);
  181. #ifdef STARPU_DEBUG
  182. STARPU_ASSERT_MSG(csr_interface->id == STARPU_CSR_INTERFACE_ID, "Error. The given data is not a csr.");
  183. #endif
  184. return csr_interface->nzval;
  185. }
  186. uint32_t *starpu_csr_get_local_colind(starpu_data_handle_t handle)
  187. {
  188. unsigned node;
  189. node = starpu_worker_get_local_memory_node();
  190. STARPU_ASSERT(starpu_data_test_if_allocated_on_node(handle, node));
  191. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *)
  192. starpu_data_get_interface_on_node(handle, node);
  193. #ifdef STARPU_DEBUG
  194. STARPU_ASSERT_MSG(csr_interface->id == STARPU_CSR_INTERFACE_ID, "Error. The given data is not a csr.");
  195. #endif
  196. return csr_interface->colind;
  197. }
  198. uint32_t *starpu_csr_get_local_rowptr(starpu_data_handle_t handle)
  199. {
  200. unsigned node;
  201. node = starpu_worker_get_local_memory_node();
  202. STARPU_ASSERT(starpu_data_test_if_allocated_on_node(handle, node));
  203. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *)
  204. starpu_data_get_interface_on_node(handle, node);
  205. #ifdef STARPU_DEBUG
  206. STARPU_ASSERT_MSG(csr_interface->id == STARPU_CSR_INTERFACE_ID, "Error. The given data is not a csr.");
  207. #endif
  208. return csr_interface->rowptr;
  209. }
  210. static size_t csr_interface_get_size(starpu_data_handle_t handle)
  211. {
  212. size_t size;
  213. uint32_t nnz = starpu_csr_get_nnz(handle);
  214. uint32_t nrow = starpu_csr_get_nrow(handle);
  215. size_t elemsize = starpu_csr_get_elemsize(handle);
  216. size = nnz*elemsize + nnz*sizeof(uint32_t) + (nrow+1)*sizeof(uint32_t);
  217. return size;
  218. }
  219. /* memory allocation/deallocation primitives for the BLAS interface */
  220. /* returns the size of the allocated area */
  221. static starpu_ssize_t allocate_csr_buffer_on_node(void *data_interface_, unsigned dst_node)
  222. {
  223. uintptr_t addr_nzval = 0;
  224. uint32_t *addr_colind = NULL, *addr_rowptr = NULL;
  225. starpu_ssize_t allocated_memory;
  226. /* we need the 3 arrays to be allocated */
  227. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *) data_interface_;
  228. uint32_t nnz = csr_interface->nnz;
  229. uint32_t nrow = csr_interface->nrow;
  230. size_t elemsize = csr_interface->elemsize;
  231. if (nnz)
  232. {
  233. addr_nzval = starpu_malloc_on_node(dst_node, nnz*elemsize);
  234. if (!addr_nzval)
  235. goto fail_nzval;
  236. addr_colind = (uint32_t*) starpu_malloc_on_node(dst_node, nnz*sizeof(uint32_t));
  237. if (!addr_colind)
  238. goto fail_colind;
  239. }
  240. else
  241. {
  242. addr_nzval = 0;
  243. addr_colind = NULL;
  244. }
  245. addr_rowptr = (uint32_t*) starpu_malloc_on_node(dst_node, (nrow+1)*sizeof(uint32_t));
  246. if (!addr_rowptr)
  247. goto fail_rowptr;
  248. /* allocation succeeded */
  249. allocated_memory =
  250. nnz*elemsize + nnz*sizeof(uint32_t) + (nrow+1)*sizeof(uint32_t);
  251. /* update the data properly in consequence */
  252. csr_interface->nzval = addr_nzval;
  253. csr_interface->colind = addr_colind;
  254. csr_interface->rowptr = addr_rowptr;
  255. return allocated_memory;
  256. fail_rowptr:
  257. if (nnz)
  258. starpu_free_on_node(dst_node, (uintptr_t) addr_colind, nnz*sizeof(uint32_t));
  259. fail_colind:
  260. if (nnz)
  261. starpu_free_on_node(dst_node, addr_nzval, nnz*elemsize);
  262. fail_nzval:
  263. /* allocation failed */
  264. return -ENOMEM;
  265. }
  266. static void free_csr_buffer_on_node(void *data_interface, unsigned node)
  267. {
  268. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *) data_interface;
  269. uint32_t nnz = csr_interface->nnz;
  270. uint32_t nrow = csr_interface->nrow;
  271. size_t elemsize = csr_interface->elemsize;
  272. if (nnz)
  273. {
  274. starpu_free_on_node(node, csr_interface->nzval, nnz*elemsize);
  275. starpu_free_on_node(node, (uintptr_t) csr_interface->colind, nnz*sizeof(uint32_t));
  276. }
  277. starpu_free_on_node(node, (uintptr_t) csr_interface->rowptr, (nrow+1)*sizeof(uint32_t));
  278. }
  279. /* as not all platform easily have a BLAS lib installed ... */
  280. static int copy_any_to_any(void *src_interface, unsigned src_node, void *dst_interface, unsigned dst_node, void *async_data)
  281. {
  282. struct starpu_csr_interface *src_csr = (struct starpu_csr_interface *) src_interface;
  283. struct starpu_csr_interface *dst_csr = (struct starpu_csr_interface *) dst_interface;
  284. uint32_t nnz = src_csr->nnz;
  285. uint32_t nrow = src_csr->nrow;
  286. size_t elemsize = src_csr->elemsize;
  287. int ret = 0;
  288. if (nnz)
  289. {
  290. if (starpu_interface_copy(src_csr->nzval, 0, src_node, dst_csr->nzval, 0, dst_node, nnz*elemsize, async_data))
  291. ret = -EAGAIN;
  292. if (starpu_interface_copy((uintptr_t)src_csr->colind, 0, src_node, (uintptr_t)dst_csr->colind, 0, dst_node, nnz*sizeof(uint32_t), async_data))
  293. ret = -EAGAIN;
  294. }
  295. if (starpu_interface_copy((uintptr_t)src_csr->rowptr, 0, src_node, (uintptr_t)dst_csr->rowptr, 0, dst_node, (nrow+1)*sizeof(uint32_t), async_data))
  296. ret = -EAGAIN;
  297. starpu_interface_data_copy(src_node, dst_node, nnz*elemsize + (nnz+nrow+1)*sizeof(uint32_t));
  298. return ret;
  299. }
  300. static starpu_ssize_t describe(void *data_interface, char *buf, size_t size)
  301. {
  302. struct starpu_csr_interface *csr = (struct starpu_csr_interface *) data_interface;
  303. return snprintf(buf, size, "C%ux%ux%u",
  304. (unsigned) csr->nnz,
  305. (unsigned) csr->nrow,
  306. (unsigned) csr->elemsize);
  307. }
  308. static int pack_data(starpu_data_handle_t handle, unsigned node, void **ptr, starpu_ssize_t *count)
  309. {
  310. STARPU_ASSERT(starpu_data_test_if_allocated_on_node(handle, node));
  311. struct starpu_csr_interface *csr = (struct starpu_csr_interface *) starpu_data_get_interface_on_node(handle, node);
  312. // We first pack colind
  313. *count = csr->nnz * sizeof(csr->colind[0]);
  314. // Then rowptr
  315. *count += (csr->nrow + 1) * sizeof(csr->rowptr[0]);
  316. // Then nnzval
  317. *count += csr->nnz * csr->elemsize;
  318. if (ptr != NULL)
  319. {
  320. *ptr = (void *)starpu_malloc_on_node_flags(node, *count, 0);
  321. char *tmp = *ptr;
  322. if (csr->nnz)
  323. {
  324. memcpy(tmp, (void*)csr->colind, csr->nnz * sizeof(csr->colind[0]));
  325. tmp += csr->nnz * sizeof(csr->colind[0]);
  326. memcpy(tmp, (void*)csr->rowptr, (csr->nrow + 1) * sizeof(csr->rowptr[0]));
  327. tmp += (csr->nrow + 1) * sizeof(csr->rowptr[0]);
  328. }
  329. memcpy(tmp, (void*)csr->nzval, csr->nnz * csr->elemsize);
  330. }
  331. return 0;
  332. }
  333. static int unpack_data(starpu_data_handle_t handle, unsigned node, void *ptr, size_t count)
  334. {
  335. STARPU_ASSERT(starpu_data_test_if_allocated_on_node(handle, node));
  336. struct starpu_csr_interface *csr = (struct starpu_csr_interface *) starpu_data_get_interface_on_node(handle, node);
  337. STARPU_ASSERT(count == (csr->nnz * sizeof(csr->colind[0]))+((csr->nrow + 1) * sizeof(csr->rowptr[0]))+(csr->nnz * csr->elemsize));
  338. char *tmp = ptr;
  339. if (csr->nnz)
  340. {
  341. memcpy((void*)csr->colind, tmp, csr->nnz * sizeof(csr->colind[0]));
  342. tmp += csr->nnz * sizeof(csr->colind[0]);
  343. memcpy((void*)csr->rowptr, tmp, (csr->nrow + 1) * sizeof(csr->rowptr[0]));
  344. tmp += (csr->nrow + 1) * sizeof(csr->rowptr[0]);
  345. }
  346. memcpy((void*)csr->nzval, tmp, csr->nnz * csr->elemsize);
  347. starpu_free_on_node_flags(node, (uintptr_t)ptr, count, 0);
  348. return 0;
  349. }