bcsr_interface.c 10 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325
  1. /* StarPU --- Runtime system for heterogeneous multicore architectures.
  2. *
  3. * Copyright (C) 2009-2013 Université de Bordeaux 1
  4. * Copyright (C) 2010, 2011, 2012, 2013 Centre National de la Recherche Scientifique
  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. #include <common/config.h>
  19. #include <datawizard/coherency.h>
  20. #include <datawizard/copy_driver.h>
  21. #include <datawizard/filters.h>
  22. #include <starpu_hash.h>
  23. #include <starpu_cuda.h>
  24. #include <starpu_opencl.h>
  25. #include <drivers/opencl/driver_opencl.h>
  26. /*
  27. * BCSR : blocked CSR, we use blocks of size (r x c)
  28. */
  29. static int copy_any_to_any(void *src_interface, unsigned src_node, void *dst_interface, unsigned dst_node, void *async_data);
  30. static const struct starpu_data_copy_methods bcsr_copy_data_methods_s =
  31. {
  32. .any_to_any = copy_any_to_any,
  33. };
  34. static void register_bcsr_handle(starpu_data_handle_t handle, unsigned home_node, void *data_interface);
  35. static starpu_ssize_t allocate_bcsr_buffer_on_node(void *data_interface, unsigned dst_node);
  36. static void free_bcsr_buffer_on_node(void *data_interface, unsigned node);
  37. static size_t bcsr_interface_get_size(starpu_data_handle_t handle);
  38. static int bcsr_compare(void *data_interface_a, void *data_interface_b);
  39. static uint32_t footprint_bcsr_interface_crc32(starpu_data_handle_t handle);
  40. static struct starpu_data_interface_ops interface_bcsr_ops =
  41. {
  42. .register_data_handle = register_bcsr_handle,
  43. .allocate_data_on_node = allocate_bcsr_buffer_on_node,
  44. .free_data_on_node = free_bcsr_buffer_on_node,
  45. .copy_methods = &bcsr_copy_data_methods_s,
  46. .get_size = bcsr_interface_get_size,
  47. .interfaceid = STARPU_BCSR_INTERFACE_ID,
  48. .interface_size = sizeof(struct starpu_bcsr_interface),
  49. .footprint = footprint_bcsr_interface_crc32,
  50. .compare = bcsr_compare
  51. };
  52. static void register_bcsr_handle(starpu_data_handle_t handle, unsigned home_node, void *data_interface)
  53. {
  54. struct starpu_bcsr_interface *bcsr_interface = (struct starpu_bcsr_interface *) data_interface;
  55. unsigned node;
  56. for (node = 0; node < STARPU_MAXNODES; node++)
  57. {
  58. struct starpu_bcsr_interface *local_interface = (struct starpu_bcsr_interface *)
  59. starpu_data_get_interface_on_node(handle, node);
  60. if (node == home_node)
  61. {
  62. local_interface->nzval = bcsr_interface->nzval;
  63. local_interface->colind = bcsr_interface->colind;
  64. local_interface->rowptr = bcsr_interface->rowptr;
  65. }
  66. else
  67. {
  68. local_interface->nzval = 0;
  69. local_interface->colind = NULL;
  70. local_interface->rowptr = NULL;
  71. }
  72. local_interface->nnz = bcsr_interface->nnz;
  73. local_interface->nrow = bcsr_interface->nrow;
  74. local_interface->firstentry = bcsr_interface->firstentry;
  75. local_interface->r = bcsr_interface->r;
  76. local_interface->c = bcsr_interface->c;
  77. local_interface->elemsize = bcsr_interface->elemsize;
  78. }
  79. }
  80. void starpu_bcsr_data_register(starpu_data_handle_t *handleptr, unsigned home_node,
  81. uint32_t nnz, uint32_t nrow, uintptr_t nzval, uint32_t *colind,
  82. uint32_t *rowptr, uint32_t firstentry,
  83. uint32_t r, uint32_t c, size_t elemsize)
  84. {
  85. struct starpu_bcsr_interface bcsr_interface =
  86. {
  87. .nzval = nzval,
  88. .colind = colind,
  89. .rowptr = rowptr,
  90. .nnz = nnz,
  91. .nrow = nrow,
  92. .firstentry = firstentry,
  93. .r = r,
  94. .c = c,
  95. .elemsize = elemsize
  96. };
  97. starpu_data_register(handleptr, home_node, &bcsr_interface, &interface_bcsr_ops);
  98. }
  99. static uint32_t footprint_bcsr_interface_crc32(starpu_data_handle_t handle)
  100. {
  101. uint32_t hash;
  102. hash = starpu_crc32_be(starpu_bcsr_get_nnz(handle), 0);
  103. hash = starpu_crc32_be(starpu_bcsr_get_c(handle), hash);
  104. hash = starpu_crc32_be(starpu_bcsr_get_r(handle), hash);
  105. return hash;
  106. }
  107. static int bcsr_compare(void *data_interface_a, void *data_interface_b)
  108. {
  109. struct starpu_bcsr_interface *bcsr_a = (struct starpu_bcsr_interface *) data_interface_a;
  110. struct starpu_bcsr_interface *bcsr_b = (struct starpu_bcsr_interface *) data_interface_b;
  111. /* Two matricess are considered compatible if they have the same size */
  112. return ((bcsr_a->nnz == bcsr_b->nnz)
  113. && (bcsr_a->nrow == bcsr_b->nrow)
  114. && (bcsr_a->r == bcsr_b->r)
  115. && (bcsr_a->c == bcsr_b->c)
  116. && (bcsr_a->elemsize == bcsr_b->elemsize));
  117. }
  118. /* offer an access to the data parameters */
  119. uint32_t starpu_bcsr_get_nnz(starpu_data_handle_t handle)
  120. {
  121. struct starpu_bcsr_interface *data_interface = (struct starpu_bcsr_interface *)
  122. starpu_data_get_interface_on_node(handle, 0);
  123. return data_interface->nnz;
  124. }
  125. uint32_t starpu_bcsr_get_nrow(starpu_data_handle_t handle)
  126. {
  127. struct starpu_bcsr_interface *data_interface = (struct starpu_bcsr_interface *)
  128. starpu_data_get_interface_on_node(handle, 0);
  129. return data_interface->nrow;
  130. }
  131. uint32_t starpu_bcsr_get_firstentry(starpu_data_handle_t handle)
  132. {
  133. struct starpu_bcsr_interface *data_interface = (struct starpu_bcsr_interface *)
  134. starpu_data_get_interface_on_node(handle, 0);
  135. return data_interface->firstentry;
  136. }
  137. uint32_t starpu_bcsr_get_r(starpu_data_handle_t handle)
  138. {
  139. struct starpu_bcsr_interface *data_interface = (struct starpu_bcsr_interface *)
  140. starpu_data_get_interface_on_node(handle, 0);
  141. return data_interface->r;
  142. }
  143. uint32_t starpu_bcsr_get_c(starpu_data_handle_t handle)
  144. {
  145. struct starpu_bcsr_interface *data_interface = (struct starpu_bcsr_interface *)
  146. starpu_data_get_interface_on_node(handle, 0);
  147. return data_interface->c;
  148. }
  149. size_t starpu_bcsr_get_elemsize(starpu_data_handle_t handle)
  150. {
  151. struct starpu_bcsr_interface *data_interface = (struct starpu_bcsr_interface *)
  152. starpu_data_get_interface_on_node(handle, 0);
  153. return data_interface->elemsize;
  154. }
  155. uintptr_t starpu_bcsr_get_local_nzval(starpu_data_handle_t handle)
  156. {
  157. unsigned node;
  158. node = _starpu_memory_node_get_local_key();
  159. STARPU_ASSERT(starpu_data_test_if_allocated_on_node(handle, node));
  160. struct starpu_bcsr_interface *data_interface = (struct starpu_bcsr_interface *)
  161. starpu_data_get_interface_on_node(handle, node);
  162. return data_interface->nzval;
  163. }
  164. uint32_t *starpu_bcsr_get_local_colind(starpu_data_handle_t handle)
  165. {
  166. /* XXX 0 */
  167. struct starpu_bcsr_interface *data_interface = (struct starpu_bcsr_interface *)
  168. starpu_data_get_interface_on_node(handle, 0);
  169. return data_interface->colind;
  170. }
  171. uint32_t *starpu_bcsr_get_local_rowptr(starpu_data_handle_t handle)
  172. {
  173. /* XXX 0 */
  174. struct starpu_bcsr_interface *data_interface = (struct starpu_bcsr_interface *)
  175. starpu_data_get_interface_on_node(handle, 0);
  176. return data_interface->rowptr;
  177. }
  178. static size_t bcsr_interface_get_size(starpu_data_handle_t handle)
  179. {
  180. size_t size;
  181. uint32_t nnz = starpu_bcsr_get_nnz(handle);
  182. uint32_t nrow = starpu_bcsr_get_nrow(handle);
  183. uint32_t r = starpu_bcsr_get_r(handle);
  184. uint32_t c = starpu_bcsr_get_c(handle);
  185. size_t elemsize = starpu_bcsr_get_elemsize(handle);
  186. size = nnz*r*c*elemsize + nnz*sizeof(uint32_t) + (nrow+1)*sizeof(uint32_t);
  187. return size;
  188. }
  189. /* memory allocation/deallocation primitives for the BLAS interface */
  190. /* returns the size of the allocated area */
  191. static starpu_ssize_t allocate_bcsr_buffer_on_node(void *data_interface_, unsigned dst_node)
  192. {
  193. uintptr_t addr_nzval, addr_colind, addr_rowptr;
  194. starpu_ssize_t allocated_memory;
  195. /* we need the 3 arrays to be allocated */
  196. struct starpu_bcsr_interface *bcsr_interface = (struct starpu_bcsr_interface *) data_interface_;
  197. uint32_t nnz = bcsr_interface->nnz;
  198. uint32_t nrow = bcsr_interface->nrow;
  199. size_t elemsize = bcsr_interface->elemsize;
  200. uint32_t r = bcsr_interface->r;
  201. uint32_t c = bcsr_interface->c;
  202. addr_nzval = starpu_malloc_on_node(dst_node, nnz*r*c*elemsize);
  203. if (!addr_nzval)
  204. goto fail_nzval;
  205. addr_colind = starpu_malloc_on_node(dst_node, nnz*sizeof(uint32_t));
  206. if (!addr_colind)
  207. goto fail_colind;
  208. addr_rowptr = starpu_malloc_on_node(dst_node, (nrow+1)*sizeof(uint32_t));
  209. if (!addr_rowptr)
  210. goto fail_rowptr;
  211. /* allocation succeeded */
  212. allocated_memory =
  213. nnz*r*c*elemsize + nnz*sizeof(uint32_t) + (nrow+1)*sizeof(uint32_t);
  214. /* update the data properly in consequence */
  215. bcsr_interface->nzval = addr_nzval;
  216. bcsr_interface->colind = (uint32_t*) addr_colind;
  217. bcsr_interface->rowptr = (uint32_t*) addr_rowptr;
  218. return allocated_memory;
  219. fail_rowptr:
  220. starpu_free_on_node(dst_node, addr_colind, nnz*sizeof(uint32_t));
  221. fail_colind:
  222. starpu_free_on_node(dst_node, addr_nzval, nnz*r*c*elemsize);
  223. fail_nzval:
  224. /* allocation failed */
  225. return -ENOMEM;
  226. }
  227. static void free_bcsr_buffer_on_node(void *data_interface, unsigned node)
  228. {
  229. struct starpu_bcsr_interface *bcsr_interface = (struct starpu_bcsr_interface *) data_interface;
  230. uint32_t nnz = bcsr_interface->nnz;
  231. uint32_t nrow = bcsr_interface->nrow;
  232. size_t elemsize = bcsr_interface->elemsize;
  233. uint32_t r = bcsr_interface->r;
  234. uint32_t c = bcsr_interface->c;
  235. starpu_free_on_node(node, bcsr_interface->nzval, nnz*r*c*elemsize);
  236. starpu_free_on_node(node, (uintptr_t) bcsr_interface->colind, nnz*sizeof(uint32_t));
  237. starpu_free_on_node(node, (uintptr_t) bcsr_interface->rowptr, (nrow+1)*sizeof(uint32_t));
  238. }
  239. static int copy_any_to_any(void *src_interface, unsigned src_node, void *dst_interface, unsigned dst_node, void *async_data)
  240. {
  241. struct starpu_bcsr_interface *src_bcsr = (struct starpu_bcsr_interface *) src_interface;
  242. struct starpu_bcsr_interface *dst_bcsr = (struct starpu_bcsr_interface *) dst_interface;
  243. uint32_t nnz = src_bcsr->nnz;
  244. uint32_t nrow = src_bcsr->nrow;
  245. size_t elemsize = src_bcsr->elemsize;
  246. uint32_t r = src_bcsr->r;
  247. uint32_t c = src_bcsr->c;
  248. int ret = 0;
  249. if (starpu_interface_copy(src_bcsr->nzval, 0, src_node, dst_bcsr->nzval, 0, dst_node, nnz*elemsize*r*c, async_data))
  250. ret = -EAGAIN;
  251. if (starpu_interface_copy((uintptr_t)src_bcsr->colind, 0, src_node, (uintptr_t)dst_bcsr->colind, 0, dst_node, nnz*sizeof(uint32_t), async_data))
  252. ret = -EAGAIN;
  253. if (starpu_interface_copy((uintptr_t)src_bcsr->rowptr, 0, src_node, (uintptr_t)dst_bcsr->rowptr, 0, dst_node, (nrow+1)*sizeof(uint32_t), async_data))
  254. ret = -EAGAIN;
  255. _STARPU_TRACE_DATA_COPY(src_node, dst_node, nnz*elemsize*r*c + (nnz+nrow+1)*sizeof(uint32_t));
  256. return ret;
  257. }