graph.cpp 13 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391
  1. /*****************************************************************************
  2. * dcc project CFG related functions
  3. * (C) Cristina Cifuentes
  4. ****************************************************************************/
  5. #include "dcc.h"
  6. #include <string.h>
  7. #if __BORLAND__
  8. #include <alloc.h>
  9. #else
  10. #include <malloc.h> /* For free() */
  11. #endif
  12. #include "graph.h"
  13. static BB * rmJMP(Function * pProc, Int marker, BB * pBB);
  14. static void mergeFallThrough(Function * pProc, BB * pBB);
  15. static void dfsNumbering(BB * pBB, std::vector<BB*> &dfsLast, Int *first, Int *last);
  16. /*****************************************************************************
  17. * createCFG - Create the basic control flow graph
  18. ****************************************************************************/
  19. void Function::createCFG()
  20. {
  21. /* Splits Icode associated with the procedure into Basic Blocks.
  22. * The links between BBs represent the control flow graph of the
  23. * procedure.
  24. * A Basic Block is defined to end on one of the following instructions:
  25. * 1) Conditional and unconditional jumps
  26. * 2) CALL(F)
  27. * 3) RET(F)
  28. * 4) On the instruction before a join (a flagged TARGET)
  29. * 5) Repeated string instructions
  30. * 6) End of procedure
  31. */
  32. Int i;
  33. Int ip, start;
  34. BB * psBB;
  35. BB * pBB;
  36. ICODE * pIcode = Icode.GetFirstIcode();
  37. stats.numBBbef = stats.numBBaft = 0;
  38. for (ip = start = 0; Icode.IsValid(pIcode); ip++, pIcode++)
  39. {
  40. /* Stick a NOWHERE_NODE on the end if we terminate
  41. * with anything other than a ret, jump or terminate */
  42. if (ip + 1 == Icode.GetNumIcodes() &&
  43. ! (pIcode->ic.ll.flg & TERMINATES) &&
  44. pIcode->ic.ll.opcode != iJMP && pIcode->ic.ll.opcode != iJMPF &&
  45. pIcode->ic.ll.opcode != iRET && pIcode->ic.ll.opcode != iRETF)
  46. {
  47. pBB=BB::Create(start, ip, NOWHERE_NODE, 0, this);
  48. }
  49. /* Only process icodes that have valid instructions */
  50. else if ((pIcode->ic.ll.flg & NO_CODE) != NO_CODE)
  51. {
  52. switch (pIcode->ic.ll.opcode) {
  53. case iJB: case iJBE: case iJAE: case iJA:
  54. case iJL: case iJLE: case iJGE: case iJG:
  55. case iJE: case iJNE: case iJS: case iJNS:
  56. case iJO: case iJNO: case iJP: case iJNP:
  57. case iJCXZ:
  58. pBB = BB::Create(start, ip, TWO_BRANCH, 2, this);
  59. CondJumps:
  60. start = ip + 1;
  61. pBB->edges[0].ip = (dword)start;
  62. /* This is for jumps off into nowhere */
  63. if (pIcode->ic.ll.flg & NO_LABEL)
  64. {
  65. pBB->numOutEdges--;
  66. pBB->edges.pop_back();
  67. }
  68. else
  69. pBB->edges[1].ip = pIcode->ic.ll.immed.op;
  70. break;
  71. case iLOOP: case iLOOPE: case iLOOPNE:
  72. pBB = BB::Create(start, ip, LOOP_NODE, 2, this);
  73. goto CondJumps;
  74. case iJMPF: case iJMP:
  75. if (pIcode->ic.ll.flg & SWITCH)
  76. {
  77. pBB = BB::Create(start, ip, MULTI_BRANCH, pIcode->ic.ll.caseTbl.numEntries, this);
  78. for (i = 0; i < pIcode->ic.ll.caseTbl.numEntries; i++)
  79. pBB->edges[i].ip = pIcode->ic.ll.caseTbl.entries[i];
  80. hasCase = TRUE;
  81. }
  82. else if ((pIcode->ic.ll.flg & (I | NO_LABEL)) == I)
  83. {
  84. pBB = BB::Create(start, ip, ONE_BRANCH, 1, this);
  85. pBB->edges[0].ip = pIcode->ic.ll.immed.op;
  86. }
  87. else
  88. BB::Create(start, ip, NOWHERE_NODE, 0, this);
  89. start = ip + 1;
  90. break;
  91. case iCALLF: case iCALL:
  92. {
  93. Function * p = pIcode->ic.ll.immed.proc.proc;
  94. if (p)
  95. i = ((p->flg) & TERMINATES) ? 0 : 1;
  96. else
  97. i = 1;
  98. pBB = BB::Create(start, ip, CALL_NODE, i, this);
  99. start = ip + 1;
  100. if (i)
  101. pBB->edges[0].ip = (dword)start;
  102. }
  103. break;
  104. case iRET: case iRETF:
  105. BB::Create(start, ip, RETURN_NODE, 0, this);
  106. start = ip + 1;
  107. break;
  108. default:
  109. /* Check for exit to DOS */
  110. if (pIcode->ic.ll.flg & TERMINATES)
  111. {
  112. pBB = BB::Create(start, ip, TERMINATE_NODE, 0, this);
  113. start = ip + 1;
  114. }
  115. /* Check for a fall through */
  116. else if (Icode.GetFirstIcode()[ip + 1].ic.ll.flg & (TARGET | CASE))
  117. {
  118. pBB = BB::Create(start, ip, FALL_NODE, 1, this);
  119. start = ip + 1;
  120. pBB->edges[0].ip = (dword)start;
  121. }
  122. break;
  123. }
  124. }
  125. }
  126. std::vector<BB *>::iterator iter=heldBBs.begin();
  127. /* Convert list of BBs into a graph */
  128. for (; iter!=heldBBs.end(); ++iter)
  129. {
  130. pBB = *iter;
  131. for (i = 0; i < pBB->numOutEdges; i++)
  132. {
  133. ip = pBB->edges[i].ip;
  134. if (ip >= SYNTHESIZED_MIN)
  135. {
  136. fatalError (INVALID_SYNTHETIC_BB);
  137. return ;
  138. }
  139. auto iter2=std::find_if(heldBBs.begin(),heldBBs.end(),
  140. [ip](BB *psBB)->bool {return psBB->begin()==ip;});
  141. if(iter2==heldBBs.end())
  142. fatalError(NO_BB, ip, name);
  143. psBB = *iter2;
  144. pBB->edges[i].BBptr = psBB;
  145. psBB->inEdges.push_back(0);
  146. }
  147. }
  148. }
  149. void Function::markImpure()
  150. {
  151. SYM * psym;
  152. for (int i = 0; i < Icode.GetNumIcodes(); i++)
  153. {
  154. if (Icode.GetLlFlag(i) & (SYM_USE | SYM_DEF))
  155. {
  156. psym = &symtab.sym[Icode.GetIcode(i)->ic.ll.caseTbl.numEntries];
  157. for (int c = (Int)psym->label; c < (Int)psym->label+psym->size; c++)
  158. {
  159. if (BITMAP(c, BM_CODE))
  160. {
  161. Icode.SetLlFlag(i, IMPURE);
  162. flg |= IMPURE;
  163. break;
  164. }
  165. }
  166. }
  167. }
  168. }
  169. /*****************************************************************************
  170. * newBB - Allocate new BB and link to end of list
  171. *****************************************************************************/
  172. /*****************************************************************************
  173. * freeCFG - Deallocates a cfg
  174. ****************************************************************************/
  175. void Function::freeCFG()
  176. {
  177. std::for_each(heldBBs.begin(),heldBBs.end(),[](BB *p)->void {delete p;});
  178. }
  179. /*****************************************************************************
  180. * compressCFG - Remove redundancies and add in-edge information
  181. ****************************************************************************/
  182. void Function::compressCFG()
  183. {
  184. BB * pBB, *pNxt;
  185. Int ip, first=0, last, i;
  186. /* First pass over BB list removes redundant jumps of the form
  187. * (Un)Conditional -> Unconditional jump */
  188. std::vector<BB*>::iterator iter=cfg.begin();
  189. for (;iter!=cfg.end(); ++iter)
  190. {
  191. pBB = *iter;
  192. if(pBB->inEdges.empty() || (pBB->nodeType != ONE_BRANCH && pBB->nodeType != TWO_BRANCH))
  193. continue;
  194. for (i = 0; i < pBB->numOutEdges; i++)
  195. {
  196. ip = pBB->rbegin();
  197. pNxt = rmJMP(this, ip, pBB->edges[i].BBptr);
  198. if (pBB->numOutEdges) /* Might have been clobbered */
  199. {
  200. pBB->edges[i].BBptr = pNxt;
  201. Icode.SetImmediateOp(ip, (dword)pNxt->start);
  202. }
  203. }
  204. }
  205. /* Next is a depth-first traversal merging any FALL_NODE or
  206. * ONE_BRANCH that fall through to a node with that as their only
  207. * in-edge. */
  208. this->cfg.front()->mergeFallThrough(Icode);
  209. /* Remove redundant BBs created by the above compressions
  210. * and allocate in-edge arrays as required. */
  211. stats.numBBaft = stats.numBBbef;
  212. for(auto iter=cfg.begin(); iter!=cfg.end(); ++iter)
  213. {
  214. pBB = *iter;
  215. if (pBB->inEdges.empty())
  216. {
  217. if (iter == cfg.begin()) /* Init it misses out on */
  218. pBB->index = UN_INIT;
  219. else
  220. {
  221. if (pBB->numOutEdges)
  222. pBB->edges.clear();
  223. delete pBB;
  224. stats.numBBaft--;
  225. }
  226. }
  227. else
  228. {
  229. pBB->inEdgeCount = pBB->inEdges.size();
  230. }
  231. }
  232. /* Allocate storage for dfsLast[] array */
  233. numBBs = stats.numBBaft;
  234. dfsLast.resize(numBBs,0); // = (BB **)allocMem(numBBs * sizeof(BB *))
  235. /* Now do a dfs numbering traversal and fill in the inEdges[] array */
  236. last = numBBs - 1;
  237. cfg.front()->dfsNumbering(dfsLast, &first, &last);
  238. }
  239. /****************************************************************************
  240. * rmJMP - If BB addressed is just a JMP it is replaced with its target
  241. ***************************************************************************/
  242. static BB * rmJMP(Function * pProc, Int marker, BB * pBB)
  243. {
  244. marker += DFS_JMP;
  245. while (pBB->nodeType == ONE_BRANCH && pBB->length == 1)
  246. {
  247. if (pBB->traversed != marker)
  248. {
  249. pBB->traversed = marker;
  250. pBB->inEdges.pop_back();
  251. if (not pBB->inEdges.empty())
  252. {
  253. pBB->edges[0].BBptr->inEdges.push_back(0);
  254. }
  255. else
  256. {
  257. pProc->Icode.SetLlFlag(pBB->start, NO_CODE);
  258. pProc->Icode.SetLlInvalid(pBB->start, TRUE);
  259. }
  260. pBB = pBB->edges[0].BBptr;
  261. }
  262. else { /* We are going around in circles */
  263. pBB->nodeType = NOWHERE_NODE;
  264. pProc->Icode.GetIcode(pBB->start)->ic.ll.immed.op = (dword)pBB->start;
  265. pProc->Icode.SetImmediateOp(pBB->start, (dword)pBB->start);
  266. do {
  267. pBB = pBB->edges[0].BBptr;
  268. pBB->inEdges.pop_back(); // was --numInedges
  269. if (! pBB->inEdges.empty())
  270. {
  271. pProc->Icode.SetLlFlag(pBB->start, NO_CODE);
  272. pProc->Icode.SetLlInvalid(pBB->start, TRUE);
  273. }
  274. } while (pBB->nodeType != NOWHERE_NODE);
  275. pBB->edges.clear();
  276. pBB->numOutEdges = 0;
  277. }
  278. }
  279. return pBB;
  280. }
  281. /*****************************************************************************
  282. * mergeFallThrough
  283. ****************************************************************************/
  284. void BB::mergeFallThrough( CIcodeRec &Icode)
  285. {
  286. BB * pChild;
  287. Int i, _ip;
  288. if (!this)
  289. {
  290. printf("mergeFallThrough on empty BB!\n");
  291. }
  292. while (nodeType == FALL_NODE || nodeType == ONE_BRANCH)
  293. {
  294. pChild = edges[0].BBptr;
  295. /* Jump to next instruction can always be removed */
  296. if (nodeType == ONE_BRANCH)
  297. {
  298. _ip = start + length;
  299. for (i = _ip; i < pChild->start && (Icode.GetLlFlag(i) & NO_CODE); i++);
  300. if (i != pChild->start)
  301. break;
  302. Icode.SetLlFlag(_ip - 1, NO_CODE);
  303. Icode.SetLlInvalid(_ip - 1, TRUE);
  304. nodeType = FALL_NODE;
  305. length--;
  306. }
  307. /* If there's no other edges into child can merge */
  308. if (pChild->inEdges.size() != 1)
  309. break;
  310. nodeType = pChild->nodeType;
  311. length = pChild->start + pChild->length - start;
  312. Icode.ClearLlFlag(pChild->start, TARGET);
  313. numOutEdges = pChild->numOutEdges;
  314. edges.swap(pChild->edges);
  315. pChild->numOutEdges = 0;
  316. pChild->inEdges.clear();
  317. pChild->edges.clear();
  318. }
  319. traversed = DFS_MERGE;
  320. /* Process all out edges recursively */
  321. for (i = 0; i < numOutEdges; i++)
  322. if (edges[i].BBptr->traversed != DFS_MERGE)
  323. edges[i].BBptr->mergeFallThrough(Icode);
  324. }
  325. /*****************************************************************************
  326. * dfsNumbering - Numbers nodes during first and last visits and determine
  327. * in-edges
  328. ****************************************************************************/
  329. void BB::dfsNumbering(std::vector<BB *> &dfsLast, Int *first, Int *last)
  330. {
  331. BB * pChild;
  332. byte i;
  333. traversed = DFS_NUM;
  334. dfsFirstNum = (*first)++;
  335. /* index is being used as an index to inEdges[]. */
  336. for (i = 0; i < numOutEdges; i++)
  337. {
  338. pChild = edges[i].BBptr;
  339. pChild->inEdges[pChild->index++] = this;
  340. /* Is this the last visit? */
  341. if (pChild->index == pChild->inEdges.size())
  342. pChild->index = UN_INIT;
  343. if (pChild->traversed != DFS_NUM)
  344. pChild->dfsNumbering(dfsLast, first, last);
  345. }
  346. dfsLastNum = *last;
  347. dfsLast[(*last)--] = this;
  348. }