nsz Git - libfirm/blob - ir/opt/opt_inline.c

   1 /*
   2  * Copyright (C) 1995-2011 University of Karlsruhe.  All right reserved.
   3  *
   4  * This file is part of libFirm.
   5  *
   6  * This file may be distributed and/or modified under the terms of the
   7  * GNU General Public License version 2 as published by the Free Software
   8  * Foundation and appearing in the file LICENSE.GPL included in the
   9  * packaging of this file.
  10  *
  11  * Licensees holding valid libFirm Professional Edition licenses may use
  12  * this file in accordance with the libFirm Commercial License.
  13  * Agreement provided with the Software.
  14  *
  15  * This file is provided AS IS with NO WARRANTY OF ANY KIND, INCLUDING THE
  16  * WARRANTY OF DESIGN, MERCHANTABILITY AND FITNESS FOR A PARTICULAR
  17  * PURPOSE.
  18  */
  19
  20 /**
  21  * @file
  22  * @brief    Dead node elimination and Procedure Inlining.
  23  * @author   Michael Beck, Goetz Lindenmaier
  24  */
  25 #include "config.h"
  26
  27 #include <limits.h>
  28 #include <stdbool.h>
  29 #include <assert.h>
  30
  31 #include "irnode_t.h"
  32 #include "irgraph_t.h"
  33 #include "irprog_t.h"
  34
  35 #include "iroptimize.h"
  36 #include "ircons_t.h"
  37 #include "iropt_t.h"
  38 #include "irgopt.h"
  39 #include "irgmod.h"
  40 #include "irgwalk.h"
  41
  42 #include "array_t.h"
  43 #include "list.h"
  44 #include "pset.h"
  45 #include "pmap.h"
  46 #include "pdeq.h"
  47 #include "xmalloc.h"
  48 #include "pqueue.h"
  49
  50 #include "irouts.h"
  51 #include "irloop_t.h"
  52 #include "irbackedge_t.h"
  53 #include "opt_init.h"
  54 #include "cgana.h"
  55 #include "trouts.h"
  56 #include "error.h"
  57
  58 #include "analyze_irg_args.h"
  59 #include "iredges_t.h"
  60 #include "irflag_t.h"
  61 #include "irhooks.h"
  62 #include "irtools.h"
  63 #include "iropt_dbg.h"
  64 #include "irpass_t.h"
  65 #include "irnodemap.h"
  66
  67 DEBUG_ONLY(static firm_dbg_module_t *dbg;)
  68
  69 /*------------------------------------------------------------------*/
  70 /* Routines for dead node elimination / copying garbage collection  */
  71 /* of the obstack.                                                  */
  72 /*------------------------------------------------------------------*/
  73
  74 /**
  75  * Remember the new node in the old node by using a field all nodes have.
  76  */
  77 static void set_new_node(ir_node *node, ir_node *new_node)
  78 {
  79         set_irn_link(node, new_node);
  80 }
  81
  82 /**
  83  * Get this new node, before the old node is forgotten.
  84  */
  85 static inline ir_node *get_new_node(ir_node *old_node)
  86 {
  87         assert(irn_visited(old_node));
  88         return (ir_node*) get_irn_link(old_node);
  89 }
  90
  91 /*--------------------------------------------------------------------*/
  92 /*  Functionality for inlining                                         */
  93 /*--------------------------------------------------------------------*/
  94
  95 /**
  96  * Copy node for inlineing.  Updates attributes that change when
  97  * inlineing but not for dead node elimination.
  98  *
  99  * Copies the node by calling copy_node() and then updates the entity if
 100  * it's a local one.  env must be a pointer of the frame type of the
 101  * inlined procedure. The new entities must be in the link field of
 102  * the entities.
 103  */
 104 static void copy_node_inline(ir_node *node, void *env)
 105 {
 106         ir_graph *new_irg  = (ir_graph*) env;
 107         ir_node  *new_node = irn_copy_into_irg(node, new_irg);
 108
 109         set_new_node(node, new_node);
 110         if (is_Sel(node)) {
 111                 ir_graph  *old_irg        = get_irn_irg(node);
 112                 ir_type   *old_frame_type = get_irg_frame_type(old_irg);
 113                 ir_entity *old_entity     = get_Sel_entity(node);
 114                 assert(is_Sel(new_node));
 115                 /* use copied entities from the new frame */
 116                 if (get_entity_owner(old_entity) == old_frame_type) {
 117                         ir_entity *new_entity = (ir_entity*)get_entity_link(old_entity);
 118                         assert(new_entity != NULL);
 119                         set_Sel_entity(new_node, new_entity);
 120                 }
 121         } else if (is_Block(new_node)) {
 122                 new_node->attr.block.irg.irg = new_irg;
 123         }
 124 }
 125
 126 static void set_preds_inline(ir_node *node, void *env)
 127 {
 128         ir_node *new_node;
 129
 130         irn_rewire_inputs(node);
 131
 132         /* move constants into start block */
 133         new_node = get_new_node(node);
 134         if (is_irn_constlike(new_node)) {
 135                 ir_graph *new_irg     = (ir_graph *) env;
 136                 ir_node  *start_block = get_irg_start_block(new_irg);
 137                 set_nodes_block(new_node, start_block);
 138         }
 139 }
 140
 141 /**
 142  * Walker: checks if P_value_arg_base is used.
 143  */
 144 static void find_addr(ir_node *node, void *env)
 145 {
 146         bool *allow_inline = (bool*)env;
 147
 148         if (is_Block(node) && get_Block_entity(node)) {
 149                 /**
 150                  * Currently we can't handle blocks whose address was taken correctly
 151                  * when inlining
 152                  */
 153                 *allow_inline = false;
 154         } else if (is_Sel(node)) {
 155                 ir_graph *irg = current_ir_graph;
 156                 if (get_Sel_ptr(node) == get_irg_frame(irg)) {
 157                         /* access to frame */
 158                         ir_entity *ent = get_Sel_entity(node);
 159                         if (get_entity_owner(ent) != get_irg_frame_type(irg)) {
 160                                 /* access to value_type */
 161                                 *allow_inline = false;
 162                         }
 163                         if (is_parameter_entity(ent)) {
 164                                 *allow_inline = false;
 165                         }
 166                 }
 167         } else if (is_Alloc(node) && get_Alloc_where(node) == stack_alloc) {
 168                 /* From GCC:
 169                  * Refuse to inline alloca call unless user explicitly forced so as this
 170                  * may change program's memory overhead drastically when the function
 171                  * using alloca is called in loop.  In GCC present in SPEC2000 inlining
 172                  * into schedule_block cause it to require 2GB of ram instead of 256MB.
 173                  *
 174                  * Sorrily this is true with our implementation also.
 175                  * Moreover, we cannot differentiate between alloca() and VLA yet, so
 176                  * this disables inlining of functions using VLA (which are completely
 177                  * save).
 178                  *
 179                  * 2 Solutions:
 180                  * - add a flag to the Alloc node for "real" alloca() calls
 181                  * - add a new Stack-Restore node at the end of a function using
 182                  *   alloca()
 183                  */
 184                 *allow_inline = false;
 185         }
 186 }
 187
 188 /**
 189  * Check if we can inline a given call.
 190  * Currently, we cannot inline two cases:
 191  * - call with compound arguments
 192  * - graphs that take the address of a parameter
 193  *
 194  * check these conditions here
 195  */
 196 static bool can_inline(ir_node *call, ir_graph *called_graph)
 197 {
 198         ir_entity          *called      = get_irg_entity(called_graph);
 199         ir_type            *called_type = get_entity_type(called);
 200         ir_type            *call_type   = get_Call_type(call);
 201         size_t              n_params    = get_method_n_params(called_type);
 202         size_t              n_arguments = get_method_n_params(call_type);
 203         size_t              n_res       = get_method_n_ress(called_type);
 204         irg_inline_property prop        = get_irg_inline_property(called_graph);
 205         size_t              i;
 206         bool                res;
 207
 208         if (prop == irg_inline_forbidden)
 209                 return false;
 210
 211         if (n_arguments != n_params) {
 212                 /* this is a bad feature of C: without a prototype, we can
 213                  * call a function with less parameters than needed. Currently
 214                  * we don't support this, although we could use Unknown than. */
 215                 return false;
 216         }
 217         if (n_res != get_method_n_ress(call_type)) {
 218                 return false;
 219         }
 220
 221         /* Argh, compiling C has some bad consequences:
 222          * It is implementation dependent what happens in that case.
 223          * We support inlining, if the bitsize of the types matches AND
 224          * the same arithmetic is used. */
 225         for (i = 0; i < n_params; ++i) {
 226                 ir_type *param_tp = get_method_param_type(called_type, i);
 227                 ir_type *arg_tp   = get_method_param_type(call_type, i);
 228
 229                 if (param_tp != arg_tp) {
 230                         ir_mode *pmode = get_type_mode(param_tp);
 231                         ir_mode *amode = get_type_mode(arg_tp);
 232
 233                         if (pmode == NULL || amode == NULL)
 234                                 return false;
 235                         if (get_mode_size_bits(pmode) != get_mode_size_bits(amode))
 236                                 return false;
 237                         if (get_mode_arithmetic(pmode) != get_mode_arithmetic(amode))
 238                                 return false;
 239                         /* otherwise we can simply "reinterpret" the bits */
 240                 }
 241         }
 242         for (i = 0; i < n_res; ++i) {
 243                 ir_type *decl_res_tp = get_method_res_type(called_type, i);
 244                 ir_type *used_res_tp = get_method_res_type(call_type, i);
 245
 246                 if (decl_res_tp != used_res_tp) {
 247                         ir_mode *decl_mode = get_type_mode(decl_res_tp);
 248                         ir_mode *used_mode = get_type_mode(used_res_tp);
 249                         if (decl_mode == NULL || used_mode == NULL)
 250                                 return false;
 251                         if (get_mode_size_bits(decl_mode) != get_mode_size_bits(used_mode))
 252                                 return false;
 253                         if (get_mode_arithmetic(decl_mode) != get_mode_arithmetic(used_mode))
 254                                 return false;
 255                         /* otherwise we can "reinterpret" the bits */
 256                 }
 257         }
 258
 259         /* check parameters for compound arguments */
 260         for (i = 0; i < n_params; ++i) {
 261                 ir_type *p_type = get_method_param_type(call_type, i);
 262
 263                 if (is_compound_type(p_type))
 264                         return false;
 265         }
 266
 267         /* check results for compound arguments */
 268         for (i = 0; i < n_res; ++i) {
 269                 ir_type *r_type = get_method_res_type(call_type, i);
 270
 271                 if (is_compound_type(r_type))
 272                         return false;
 273         }
 274
 275         res = true;
 276         irg_walk_graph(called_graph, find_addr, NULL, &res);
 277
 278         return res;
 279 }
 280
 281 enum exc_mode {
 282         exc_handler,    /**< There is a handler. */
 283         exc_no_handler  /**< Exception handling not represented. */
 284 };
 285
 286 /**
 287  * copy all entities on the stack frame on 1 irg to the stackframe of another.
 288  * Sets entity links of the old entities to the copies
 289  */
 290 static void copy_frame_entities(ir_graph *from, ir_graph *to)
 291 {
 292         ir_type *from_frame = get_irg_frame_type(from);
 293         ir_type *to_frame   = get_irg_frame_type(to);
 294         size_t   n_members  = get_class_n_members(from_frame);
 295         size_t   i;
 296         assert(from_frame != to_frame);
 297
 298         for (i = 0; i < n_members; ++i) {
 299                 ir_entity *old_ent = get_class_member(from_frame, i);
 300                 ir_entity *new_ent = copy_entity_own(old_ent, to_frame);
 301                 set_entity_link(old_ent, new_ent);
 302                 assert (!is_parameter_entity(old_ent));
 303         }
 304 }
 305
 306 /* Inlines a method at the given call site. */
 307 int inline_method(ir_node *call, ir_graph *called_graph)
 308 {
 309         ir_node       *pre_call;
 310         ir_node       *post_call, *post_bl;
 311         ir_node       *in[pn_Start_max+1];
 312         ir_node       *end, *end_bl, *block;
 313         ir_node       **res_pred;
 314         ir_node       **cf_pred;
 315         ir_node       **args_in;
 316         ir_node       *ret, *phi;
 317         int           arity, n_ret, n_exc, n_res, i, j, rem_opt;
 318         int           irn_arity, n_params;
 319         int           n_mem_phi;
 320         enum exc_mode exc_handling;
 321         ir_type       *mtp;
 322         ir_type       *ctp;
 323         ir_entity     *ent;
 324         ir_graph      *rem;
 325         ir_graph      *irg = get_irn_irg(call);
 326
 327         /* we cannot inline some types of calls */
 328         if (! can_inline(call, called_graph))
 329                 return 0;
 330
 331         /* We cannot inline a recursive call. The graph must be copied before
 332          * the call the inline_method() using create_irg_copy(). */
 333         if (called_graph == irg)
 334                 return 0;
 335
 336         ent      = get_irg_entity(called_graph);
 337         mtp      = get_entity_type(ent);
 338         ctp      = get_Call_type(call);
 339         n_params = get_method_n_params(mtp);
 340         n_res    = get_method_n_ress(mtp);
 341
 342         rem = current_ir_graph;
 343         current_ir_graph = irg;
 344
 345         DB((dbg, LEVEL_1, "Inlining %+F(%+F) into %+F\n", call, called_graph, irg));
 346
 347         /* optimizations can cause problems when allocating new nodes */
 348         rem_opt = get_opt_optimize();
 349         set_optimize(0);
 350
 351         /* Handle graph state */
 352         assert(get_irg_phase_state(irg) != phase_building);
 353         assert(get_irg_pinned(irg) == op_pin_state_pinned);
 354         assert(get_irg_pinned(called_graph) == op_pin_state_pinned);
 355         clear_irg_state(irg, IR_GRAPH_STATE_CONSISTENT_DOMINANCE
 356                            | IR_GRAPH_STATE_CONSISTENT_ENTITY_USAGE);
 357         set_irg_callee_info_state(irg, irg_callee_info_inconsistent);
 358         clear_irg_state(irg, IR_GRAPH_STATE_CONSISTENT_ENTITY_USAGE);
 359         edges_deactivate(irg);
 360
 361         /* here we know we WILL inline, so inform the statistics */
 362         hook_inline(call, called_graph);
 363
 364         /* -- Decide how to handle exception control flow: Is there a handler
 365            for the Call node, or do we branch directly to End on an exception?
 366            exc_handling:
 367            0 There is a handler.
 368            2 Exception handling not represented in Firm. -- */
 369         {
 370                 ir_node *Xproj = NULL;
 371                 ir_node *proj;
 372                 for (proj = (ir_node*)get_irn_link(call); proj != NULL;
 373                      proj = (ir_node*)get_irn_link(proj)) {
 374                         long proj_nr = get_Proj_proj(proj);
 375                         if (proj_nr == pn_Call_X_except) Xproj = proj;
 376                 }
 377                 exc_handling = Xproj != NULL ? exc_handler : exc_no_handler;
 378         }
 379
 380         /* create the argument tuple */
 381         args_in = ALLOCAN(ir_node*, n_params);
 382
 383         block = get_nodes_block(call);
 384         for (i = n_params - 1; i >= 0; --i) {
 385                 ir_node *arg      = get_Call_param(call, i);
 386                 ir_type *param_tp = get_method_param_type(mtp, i);
 387                 ir_mode *mode     = get_type_mode(param_tp);
 388
 389                 if (mode != get_irn_mode(arg)) {
 390                         arg = new_r_Conv(block, arg, mode);
 391                 }
 392                 args_in[i] = arg;
 393         }
 394
 395         /* the procedure and later replaces the Start node of the called graph.
 396          * Post_call is the old Call node and collects the results of the called
 397          * graph. Both will end up being a tuple. */
 398         post_bl = get_nodes_block(call);
 399         /* XxMxPxPxPxT of Start + parameter of Call */
 400         in[pn_Start_M]              = get_Call_mem(call);
 401         in[pn_Start_X_initial_exec] = new_r_Jmp(post_bl);
 402         in[pn_Start_P_frame_base]   = get_irg_frame(irg);
 403         in[pn_Start_T_args]         = new_r_Tuple(post_bl, n_params, args_in);
 404         pre_call = new_r_Tuple(post_bl, pn_Start_max+1, in);
 405         post_call = call;
 406
 407         /* --
 408            The new block gets the ins of the old block, pre_call and all its
 409            predecessors and all Phi nodes. -- */
 410         part_block(pre_call);
 411
 412         /* increment visited flag for later walk */
 413         inc_irg_visited(called_graph);
 414
 415         /* link some nodes to nodes in the current graph so instead of copying
 416          * the linked nodes will get used.
 417          * So the copier will use the created Tuple instead of copying the start
 418          * node, similar for singleton nodes like NoMem and Bad.
 419          * Note: this will prohibit predecessors to be copied - only do it for
 420          *       nodes without predecessors */
 421         {
 422                 ir_node *start_block;
 423                 ir_node *start;
 424                 ir_node *nomem;
 425
 426                 start_block = get_irg_start_block(called_graph);
 427                 set_new_node(start_block, get_nodes_block(pre_call));
 428                 mark_irn_visited(start_block);
 429
 430                 start = get_irg_start(called_graph);
 431                 set_new_node(start, pre_call);
 432                 mark_irn_visited(start);
 433
 434                 nomem = get_irg_no_mem(called_graph);
 435                 set_new_node(nomem, get_irg_no_mem(irg));
 436                 mark_irn_visited(nomem);
 437         }
 438
 439         /* entitiy link is used to link entities on old stackframe to the
 440          * new stackframe */
 441         irp_reserve_resources(irp, IRP_RESOURCE_ENTITY_LINK);
 442
 443         /* copy entities and nodes */
 444         assert(!irn_visited(get_irg_end(called_graph)));
 445         copy_frame_entities(called_graph, irg);
 446         irg_walk_core(get_irg_end(called_graph), copy_node_inline, set_preds_inline,
 447                       irg);
 448
 449         irp_free_resources(irp, IRP_RESOURCE_ENTITY_LINK);
 450
 451         /* -- Merge the end of the inlined procedure with the call site -- */
 452         /* We will turn the old Call node into a Tuple with the following
 453            predecessors:
 454            -1:  Block of Tuple.
 455            0: Phi of all Memories of Return statements.
 456            1: Jmp from new Block that merges the control flow from all exception
 457            predecessors of the old end block.
 458            2: Tuple of all arguments.
 459            3: Phi of Exception memories.
 460            In case the old Call directly branches to End on an exception we don't
 461            need the block merging all exceptions nor the Phi of the exception
 462            memories.
 463         */
 464
 465         /* Precompute some values */
 466         end_bl = get_new_node(get_irg_end_block(called_graph));
 467         end    = get_new_node(get_irg_end(called_graph));
 468         arity  = get_Block_n_cfgpreds(end_bl);    /* arity = n_exc + n_ret  */
 469         n_res  = get_method_n_ress(get_Call_type(call));
 470
 471         res_pred = XMALLOCN(ir_node*, n_res);
 472         cf_pred  = XMALLOCN(ir_node*, arity);
 473
 474         /* archive keepalives */
 475         irn_arity = get_irn_arity(end);
 476         for (i = 0; i < irn_arity; i++) {
 477                 ir_node *ka = get_End_keepalive(end, i);
 478                 if (! is_Bad(ka))
 479                         add_End_keepalive(get_irg_end(irg), ka);
 480         }
 481
 482         /* replace Return nodes by Jump nodes */
 483         n_ret = 0;
 484         for (i = 0; i < arity; i++) {
 485                 ir_node *ret;
 486                 ret = get_Block_cfgpred(end_bl, i);
 487                 if (is_Return(ret)) {
 488                         ir_node *block = get_nodes_block(ret);
 489                         cf_pred[n_ret] = new_r_Jmp(block);
 490                         n_ret++;
 491                 }
 492         }
 493         set_irn_in(post_bl, n_ret, cf_pred);
 494
 495         /* build a Tuple for all results of the method.
 496          * add Phi node if there was more than one Return. */
 497         turn_into_tuple(post_call, pn_Call_max+1);
 498         /* First the Memory-Phi */
 499         n_mem_phi = 0;
 500         for (i = 0; i < arity; i++) {
 501                 ret = get_Block_cfgpred(end_bl, i);
 502                 if (is_Return(ret)) {
 503                         cf_pred[n_mem_phi++] = get_Return_mem(ret);
 504                 }
 505                 /* memory output for some exceptions is directly connected to End */
 506                 if (is_Call(ret)) {
 507                         cf_pred[n_mem_phi++] = new_r_Proj(ret, mode_M, 3);
 508                 } else if (is_fragile_op(ret)) {
 509                         /* We rely that all cfops have the memory output at the same position. */
 510                         cf_pred[n_mem_phi++] = new_r_Proj(ret, mode_M, 0);
 511                 } else if (is_Raise(ret)) {
 512                         cf_pred[n_mem_phi++] = new_r_Proj(ret, mode_M, 1);
 513                 }
 514         }
 515         phi = new_r_Phi(post_bl, n_mem_phi, cf_pred, mode_M);
 516         set_Tuple_pred(call, pn_Call_M, phi);
 517         /* Conserve Phi-list for further inlinings -- but might be optimized */
 518         if (get_nodes_block(phi) == post_bl) {
 519                 set_irn_link(phi, get_irn_link(post_bl));
 520                 set_irn_link(post_bl, phi);
 521         }
 522         /* Now the real results */
 523         if (n_res > 0) {
 524                 ir_node *result_tuple;
 525                 for (j = 0; j < n_res; j++) {
 526                         ir_type *res_type = get_method_res_type(ctp, j);
 527                         ir_mode *res_mode = get_type_mode(res_type);
 528                         n_ret = 0;
 529                         for (i = 0; i < arity; i++) {
 530                                 ret = get_Block_cfgpred(end_bl, i);
 531                                 if (is_Return(ret)) {
 532                                         ir_node *res = get_Return_res(ret, j);
 533                                         if (get_irn_mode(res) != res_mode) {
 534                                                 ir_node *block = get_nodes_block(res);
 535                                                 res = new_r_Conv(block, res, res_mode);
 536                                         }
 537                                         cf_pred[n_ret] = res;
 538                                         n_ret++;
 539                                 }
 540                         }
 541                         if (n_ret > 0) {
 542                                 phi = new_r_Phi(post_bl, n_ret, cf_pred, res_mode);
 543                         } else {
 544                                 phi = new_r_Bad(irg, res_mode);
 545                         }
 546                         res_pred[j] = phi;
 547                         /* Conserve Phi-list for further inlinings -- but might be optimized */
 548                         if (get_nodes_block(phi) == post_bl) {
 549                                 set_Phi_next(phi, get_Block_phis(post_bl));
 550                                 set_Block_phis(post_bl, phi);
 551                         }
 552                 }
 553                 result_tuple = new_r_Tuple(post_bl, n_res, res_pred);
 554                 set_Tuple_pred(call, pn_Call_T_result, result_tuple);
 555         } else {
 556                 set_Tuple_pred(call, pn_Call_T_result, new_r_Bad(irg, mode_T));
 557         }
 558         /* handle the regular call */
 559         set_Tuple_pred(call, pn_Call_X_regular, new_r_Jmp(post_bl));
 560
 561         /* Finally the exception control flow.
 562            We have two possible situations:
 563            First if the Call branches to an exception handler:
 564            We need to add a Phi node to
 565            collect the memory containing the exception objects.  Further we need
 566            to add another block to get a correct representation of this Phi.  To
 567            this block we add a Jmp that resolves into the X output of the Call
 568            when the Call is turned into a tuple.
 569            Second: There is no exception edge. Just add all inlined exception
 570            branches to the End node.
 571          */
 572         if (exc_handling == exc_handler) {
 573                 n_exc = 0;
 574                 for (i = 0; i < arity; i++) {
 575                         ir_node *ret, *irn;
 576                         ret = get_Block_cfgpred(end_bl, i);
 577                         irn = skip_Proj(ret);
 578                         if (is_fragile_op(irn) || is_Raise(irn)) {
 579                                 cf_pred[n_exc] = ret;
 580                                 ++n_exc;
 581                         }
 582                 }
 583                 if (n_exc > 0) {
 584                         if (n_exc == 1) {
 585                                 /* simple fix */
 586                                 set_Tuple_pred(call, pn_Call_X_except, cf_pred[0]);
 587                         } else {
 588                                 ir_node *block = new_r_Block(irg, n_exc, cf_pred);
 589                                 set_Tuple_pred(call, pn_Call_X_except, new_r_Jmp(block));
 590                         }
 591                 } else {
 592                         set_Tuple_pred(call, pn_Call_X_except, new_r_Bad(irg, mode_X));
 593                 }
 594         } else {
 595                 ir_node *main_end_bl;
 596                 int main_end_bl_arity;
 597                 ir_node **end_preds;
 598
 599                 /* assert(exc_handling == 1 || no exceptions. ) */
 600                 n_exc = 0;
 601                 for (i = 0; i < arity; i++) {
 602                         ir_node *ret = get_Block_cfgpred(end_bl, i);
 603                         ir_node *irn = skip_Proj(ret);
 604
 605                         if (is_fragile_op(irn) || is_Raise(irn)) {
 606                                 cf_pred[n_exc] = ret;
 607                                 n_exc++;
 608                         }
 609                 }
 610                 main_end_bl       = get_irg_end_block(irg);
 611                 main_end_bl_arity = get_irn_arity(main_end_bl);
 612                 end_preds         = XMALLOCN(ir_node*, n_exc + main_end_bl_arity);
 613
 614                 for (i = 0; i < main_end_bl_arity; ++i)
 615                         end_preds[i] = get_irn_n(main_end_bl, i);
 616                 for (i = 0; i < n_exc; ++i)
 617                         end_preds[main_end_bl_arity + i] = cf_pred[i];
 618                 set_irn_in(main_end_bl, n_exc + main_end_bl_arity, end_preds);
 619                 set_Tuple_pred(call, pn_Call_X_except, new_r_Bad(irg, mode_X));
 620                 free(end_preds);
 621         }
 622         free(res_pred);
 623         free(cf_pred);
 624
 625         /* --  Turn CSE back on. -- */
 626         set_optimize(rem_opt);
 627         current_ir_graph = rem;
 628
 629         return 1;
 630 }
 631
 632 /********************************************************************/
 633 /* Apply inlining to small methods.                                 */
 634 /********************************************************************/
 635
 636 static struct obstack  temp_obst;
 637
 638 /** Represents a possible inlinable call in a graph. */
 639 typedef struct call_entry {
 640         ir_node    *call;       /**< The Call node. */
 641         ir_graph   *callee;     /**< The callee IR-graph. */
 642         list_head  list;        /**< List head for linking the next one. */
 643         int        loop_depth;  /**< The loop depth of this call. */
 644         int        benefice;    /**< The calculated benefice of this call. */
 645         unsigned   local_adr:1; /**< Set if this call gets an address of a local variable. */
 646         unsigned   all_const:1; /**< Set if this call has only constant parameters. */
 647 } call_entry;
 648
 649 /**
 650  * environment for inlining small irgs
 651  */
 652 typedef struct inline_env_t {
 653         struct obstack obst;  /**< An obstack where call_entries are allocated on. */
 654         list_head      calls; /**< The call entry list. */
 655 } inline_env_t;
 656
 657 /**
 658  * Returns the irg called from a Call node. If the irg is not
 659  * known, NULL is returned.
 660  *
 661  * @param call  the call node
 662  */
 663 static ir_graph *get_call_called_irg(ir_node *call)
 664 {
 665         ir_node *addr;
 666
 667         addr = get_Call_ptr(call);
 668         if (is_SymConst_addr_ent(addr)) {
 669                 ir_entity *ent = get_SymConst_entity(addr);
 670                 /* we don't know which function gets finally bound to a weak symbol */
 671                 if (get_entity_linkage(ent) & IR_LINKAGE_WEAK)
 672                         return NULL;
 673
 674                 return get_entity_irg(ent);
 675         }
 676
 677         return NULL;
 678 }
 679
 680 /**
 681  * Walker: Collect all calls to known graphs inside a graph.
 682  */
 683 static void collect_calls(ir_node *call, void *env)
 684 {
 685         (void) env;
 686         if (is_Call(call)) {
 687                 ir_graph *called_irg = get_call_called_irg(call);
 688
 689                 if (called_irg != NULL) {
 690                         /* The Call node calls a locally defined method.  Remember to inline. */
 691                         inline_env_t *ienv  = (inline_env_t*)env;
 692                         call_entry   *entry = OALLOC(&ienv->obst, call_entry);
 693                         entry->call       = call;
 694                         entry->callee     = called_irg;
 695                         entry->loop_depth = 0;
 696                         entry->benefice   = 0;
 697                         entry->local_adr  = 0;
 698                         entry->all_const  = 0;
 699
 700                         list_add_tail(&entry->list, &ienv->calls);
 701                 }
 702         }
 703 }
 704
 705 /**
 706  * Inlines all small methods at call sites where the called address comes
 707  * from a Const node that references the entity representing the called
 708  * method.
 709  * The size argument is a rough measure for the code size of the method:
 710  * Methods where the obstack containing the firm graph is smaller than
 711  * size are inlined.
 712  */
 713 void inline_small_irgs(ir_graph *irg, int size)
 714 {
 715         ir_graph *rem = current_ir_graph;
 716         inline_env_t env;
 717         call_entry *entry;
 718
 719         current_ir_graph = irg;
 720         /* Handle graph state */
 721         assert(get_irg_phase_state(irg) != phase_building);
 722         free_callee_info(irg);
 723
 724         /* Find Call nodes to inline.
 725            (We can not inline during a walk of the graph, as inlining the same
 726            method several times changes the visited flag of the walked graph:
 727            after the first inlining visited of the callee equals visited of
 728            the caller.  With the next inlining both are increased.) */
 729         obstack_init(&env.obst);
 730         INIT_LIST_HEAD(&env.calls);
 731         irg_walk_graph(irg, NULL, collect_calls, &env);
 732
 733         if (! list_empty(&env.calls)) {
 734                 /* There are calls to inline */
 735                 ir_reserve_resources(irg, IR_RESOURCE_IRN_LINK|IR_RESOURCE_PHI_LIST);
 736                 collect_phiprojs(irg);
 737
 738                 list_for_each_entry(call_entry, entry, &env.calls, list) {
 739                         ir_graph            *callee = entry->callee;
 740                         irg_inline_property prop    = get_irg_inline_property(callee);
 741
 742                         if (prop == irg_inline_forbidden) {
 743                                 continue;
 744                         }
 745
 746                         if (prop >= irg_inline_forced ||
 747                             _obstack_memory_used(callee->obst) - (int)obstack_room(callee->obst) < size) {
 748                                 inline_method(entry->call, callee);
 749                         }
 750                 }
 751                 ir_free_resources(irg, IR_RESOURCE_IRN_LINK|IR_RESOURCE_PHI_LIST);
 752         }
 753         obstack_free(&env.obst, NULL);
 754         current_ir_graph = rem;
 755 }
 756
 757 typedef struct inline_small_irgs_pass_t {
 758         ir_graph_pass_t pass;
 759         int            size;
 760 } inline_small_irgs_pass_t;
 761
 762 /**
 763  * Wrapper to run inline_small_irgs() as a pass.
 764  */
 765 static int inline_small_irgs_wrapper(ir_graph *irg, void *context)
 766 {
 767         inline_small_irgs_pass_t *pass = (inline_small_irgs_pass_t*)context;
 768
 769         inline_small_irgs(irg, pass->size);
 770         return 0;
 771 }
 772
 773 /* create a pass for inline_small_irgs() */
 774 ir_graph_pass_t *inline_small_irgs_pass(const char *name, int size)
 775 {
 776         inline_small_irgs_pass_t *pass = XMALLOCZ(inline_small_irgs_pass_t);
 777
 778         pass->size = size;
 779         return def_graph_pass_constructor(
 780                 &pass->pass, name ? name : "inline_small_irgs", inline_small_irgs_wrapper);
 781 }
 782
 783 /**
 784  * Environment for inlining irgs.
 785  */
 786 typedef struct {
 787         list_head calls;             /**< List of of all call nodes in this graph. */
 788         unsigned  *local_weights;    /**< Once allocated, the beneficial weight for transmitting local addresses. */
 789         unsigned  n_nodes;           /**< Number of nodes in graph except Id, Tuple, Proj, Start, End. */
 790         unsigned  n_blocks;          /**< Number of Blocks in graph without Start and End block. */
 791         unsigned  n_nodes_orig;      /**< for statistics */
 792         unsigned  n_call_nodes;      /**< Number of Call nodes in the graph. */
 793         unsigned  n_call_nodes_orig; /**< for statistics */
 794         unsigned  n_callers;         /**< Number of known graphs that call this graphs. */
 795         unsigned  n_callers_orig;    /**< for statistics */
 796         unsigned  got_inline:1;      /**< Set, if at least one call inside this graph was inlined. */
 797         unsigned  recursive:1;       /**< Set, if this function is self recursive. */
 798 } inline_irg_env;
 799
 800 /**
 801  * Allocate a new environment for inlining.
 802  */
 803 static inline_irg_env *alloc_inline_irg_env(void)
 804 {
 805         inline_irg_env *env    = OALLOC(&temp_obst, inline_irg_env);
 806         INIT_LIST_HEAD(&env->calls);
 807         env->local_weights     = NULL;
 808         env->n_nodes           = -2; /* do not count count Start, End */
 809         env->n_blocks          = -2; /* do not count count Start, End Block */
 810         env->n_nodes_orig      = -2; /* do not count Start, End */
 811         env->n_call_nodes      = 0;
 812         env->n_call_nodes_orig = 0;
 813         env->n_callers         = 0;
 814         env->n_callers_orig    = 0;
 815         env->got_inline        = 0;
 816         env->recursive         = 0;
 817         return env;
 818 }
 819
 820 typedef struct walker_env {
 821         inline_irg_env *x;     /**< the inline environment */
 822         char ignore_runtime;   /**< the ignore runtime flag */
 823         char ignore_callers;   /**< if set, do change callers data */
 824 } wenv_t;
 825
 826 /**
 827  * post-walker: collect all calls in the inline-environment
 828  * of a graph and sum some statistics.
 829  */
 830 static void collect_calls2(ir_node *call, void *ctx)
 831 {
 832         wenv_t         *env = (wenv_t*)ctx;
 833         inline_irg_env *x = env->x;
 834         unsigned        code = get_irn_opcode(call);
 835         ir_graph       *callee;
 836         call_entry     *entry;
 837
 838         /* count meaningful nodes in irg */
 839         if (code != iro_Proj && code != iro_Tuple && code != iro_Sync) {
 840                 if (code != iro_Block) {
 841                         ++x->n_nodes;
 842                         ++x->n_nodes_orig;
 843                 } else {
 844                         ++x->n_blocks;
 845                 }
 846         }
 847
 848         if (code != iro_Call) return;
 849
 850         /* check, if it's a runtime call */
 851         if (env->ignore_runtime) {
 852                 ir_node *symc = get_Call_ptr(call);
 853
 854                 if (is_SymConst_addr_ent(symc)) {
 855                         ir_entity *ent = get_SymConst_entity(symc);
 856
 857                         if (get_entity_additional_properties(ent) & mtp_property_runtime)
 858                                 return;
 859                 }
 860         }
 861
 862         /* collect all call nodes */
 863         ++x->n_call_nodes;
 864         ++x->n_call_nodes_orig;
 865
 866         callee = get_call_called_irg(call);
 867         if (callee != NULL) {
 868                 if (! env->ignore_callers) {
 869                         inline_irg_env *callee_env = (inline_irg_env*)get_irg_link(callee);
 870                         /* count all static callers */
 871                         ++callee_env->n_callers;
 872                         ++callee_env->n_callers_orig;
 873                 }
 874                 if (callee == current_ir_graph)
 875                         x->recursive = 1;
 876
 877                 /* link it in the list of possible inlinable entries */
 878                 entry = OALLOC(&temp_obst, call_entry);
 879                 entry->call       = call;
 880                 entry->callee     = callee;
 881                 entry->loop_depth = get_irn_loop(get_nodes_block(call))->depth;
 882                 entry->benefice   = 0;
 883                 entry->local_adr  = 0;
 884                 entry->all_const  = 0;
 885
 886                 list_add_tail(&entry->list, &x->calls);
 887         }
 888 }
 889
 890 /**
 891  * Returns TRUE if the number of callers is 0 in the irg's environment,
 892  * hence this irg is a leaf.
 893  */
 894 inline static int is_leaf(ir_graph *irg)
 895 {
 896         inline_irg_env *env = (inline_irg_env*)get_irg_link(irg);
 897         return env->n_call_nodes == 0;
 898 }
 899
 900 /**
 901  * Returns TRUE if the number of nodes in the callee is
 902  * smaller then size in the irg's environment.
 903  */
 904 inline static int is_smaller(ir_graph *callee, unsigned size)
 905 {
 906         inline_irg_env *env = (inline_irg_env*)get_irg_link(callee);
 907         return env->n_nodes < size;
 908 }
 909
 910 /**
 911  * Duplicate a call entry.
 912  *
 913  * @param entry     the original entry to duplicate
 914  * @param new_call  the new call node
 915  * @param loop_depth_delta
 916  *                  delta value for the loop depth
 917  */
 918 static call_entry *duplicate_call_entry(const call_entry *entry,
 919                                         ir_node *new_call, int loop_depth_delta)
 920 {
 921         call_entry *nentry = OALLOC(&temp_obst, call_entry);
 922         nentry->call       = new_call;
 923         nentry->callee     = entry->callee;
 924         nentry->benefice   = entry->benefice;
 925         nentry->loop_depth = entry->loop_depth + loop_depth_delta;
 926         nentry->local_adr  = entry->local_adr;
 927         nentry->all_const  = entry->all_const;
 928
 929         return nentry;
 930 }
 931
 932 /**
 933  * Append all call nodes of the source environment to the nodes of in the destination
 934  * environment.
 935  *
 936  * @param dst         destination environment
 937  * @param src         source environment
 938  * @param loop_depth  the loop depth of the call that is replaced by the src list
 939  */
 940 static void append_call_list(inline_irg_env *dst, inline_irg_env *src, int loop_depth)
 941 {
 942         call_entry *entry, *nentry;
 943
 944         /* Note that the src list points to Call nodes in the inlined graph, but
 945            we need Call nodes in our graph. Luckily the inliner leaves this information
 946            in the link field. */
 947         list_for_each_entry(call_entry, entry, &src->calls, list) {
 948                 nentry = duplicate_call_entry(entry, (ir_node*)get_irn_link(entry->call), loop_depth);
 949                 list_add_tail(&nentry->list, &dst->calls);
 950         }
 951         dst->n_call_nodes += src->n_call_nodes;
 952         dst->n_nodes      += src->n_nodes;
 953 }
 954
 955 /*
 956  * Inlines small leaf methods at call sites where the called address comes
 957  * from a Const node that references the entity representing the called
 958  * method.
 959  * The size argument is a rough measure for the code size of the method:
 960  * Methods where the obstack containing the firm graph is smaller than
 961  * size are inlined.
 962  */
 963 void inline_leaf_functions(unsigned maxsize, unsigned leafsize,
 964                             unsigned size, int ignore_runtime)
 965 {
 966         inline_irg_env   *env;
 967         ir_graph         *irg;
 968         size_t           i, n_irgs;
 969         ir_graph         *rem;
 970         int              did_inline;
 971         wenv_t           wenv;
 972         call_entry       *entry, *next;
 973         const call_entry *centry;
 974         pmap             *copied_graphs;
 975         pmap_entry       *pm_entry;
 976
 977         rem = current_ir_graph;
 978         obstack_init(&temp_obst);
 979
 980         /* a map for the copied graphs, used to inline recursive calls */
 981         copied_graphs = pmap_create();
 982
 983         /* extend all irgs by a temporary data structure for inlining. */
 984         n_irgs = get_irp_n_irgs();
 985         for (i = 0; i < n_irgs; ++i)
 986                 set_irg_link(get_irp_irg(i), alloc_inline_irg_env());
 987
 988         /* Pre-compute information in temporary data structure. */
 989         wenv.ignore_runtime = ignore_runtime;
 990         wenv.ignore_callers = 0;
 991         for (i = 0; i < n_irgs; ++i) {
 992                 ir_graph *irg = get_irp_irg(i);
 993
 994                 assert(get_irg_phase_state(irg) != phase_building);
 995                 free_callee_info(irg);
 996
 997                 assure_loopinfo(irg);
 998                 wenv.x = (inline_irg_env*)get_irg_link(irg);
 999                 irg_walk_graph(irg, NULL, collect_calls2, &wenv);
1000         }
1001
1002         /* -- and now inline. -- */
1003
1004         /* Inline leafs recursively -- we might construct new leafs. */
1005         do {
1006                 did_inline = 0;
1007
1008                 for (i = 0; i < n_irgs; ++i) {
1009                         ir_node *call;
1010                         int phiproj_computed = 0;
1011
1012                         current_ir_graph = get_irp_irg(i);
1013                         env              = (inline_irg_env*)get_irg_link(current_ir_graph);
1014
1015                         ir_reserve_resources(current_ir_graph, IR_RESOURCE_IRN_LINK|IR_RESOURCE_PHI_LIST);
1016                         list_for_each_entry_safe(call_entry, entry, next, &env->calls, list) {
1017                                 ir_graph            *callee;
1018                                 irg_inline_property  prop;
1019
1020                                 if (env->n_nodes > maxsize)
1021                                         break;
1022
1023                                 call   = entry->call;
1024                                 callee = entry->callee;
1025
1026                                 prop = get_irg_inline_property(callee);
1027                                 if (prop == irg_inline_forbidden) {
1028                                         continue;
1029                                 }
1030
1031                                 if (is_leaf(callee) && (
1032                                     is_smaller(callee, leafsize) || prop >= irg_inline_forced)) {
1033                                         if (!phiproj_computed) {
1034                                                 phiproj_computed = 1;
1035                                                 collect_phiprojs(current_ir_graph);
1036                                         }
1037                                         did_inline = inline_method(call, callee);
1038
1039                                         if (did_inline) {
1040                                                 inline_irg_env *callee_env = (inline_irg_env*)get_irg_link(callee);
1041
1042                                                 /* call was inlined, Phi/Projs for current graph must be recomputed */
1043                                                 phiproj_computed = 0;
1044
1045                                                 /* Do some statistics */
1046                                                 env->got_inline = 1;
1047                                                 --env->n_call_nodes;
1048                                                 env->n_nodes += callee_env->n_nodes;
1049                                                 --callee_env->n_callers;
1050
1051                                                 /* remove this call from the list */
1052                                                 list_del(&entry->list);
1053                                                 continue;
1054                                         }
1055                                 }
1056                         }
1057                         ir_free_resources(current_ir_graph, IR_RESOURCE_IRN_LINK|IR_RESOURCE_PHI_LIST);
1058                 }
1059         } while (did_inline);
1060
1061         /* inline other small functions. */
1062         for (i = 0; i < n_irgs; ++i) {
1063                 ir_node *call;
1064                 int phiproj_computed = 0;
1065
1066                 current_ir_graph = get_irp_irg(i);
1067                 env              = (inline_irg_env*)get_irg_link(current_ir_graph);
1068
1069                 ir_reserve_resources(current_ir_graph, IR_RESOURCE_IRN_LINK|IR_RESOURCE_PHI_LIST);
1070
1071                 /* note that the list of possible calls is updated during the process */
1072                 list_for_each_entry_safe(call_entry, entry, next, &env->calls, list) {
1073                         irg_inline_property prop;
1074                         ir_graph            *callee;
1075                         ir_graph            *calleee;
1076
1077                         call   = entry->call;
1078                         callee = entry->callee;
1079
1080                         prop = get_irg_inline_property(callee);
1081                         if (prop == irg_inline_forbidden) {
1082                                 continue;
1083                         }
1084
1085                         calleee = (ir_graph*)pmap_get(copied_graphs, callee);
1086                         if (calleee != NULL) {
1087                                 /*
1088                                  * Remap callee if we have a copy.
1089                                  * FIXME: Should we do this only for recursive Calls ?
1090                                  */
1091                                 callee = calleee;
1092                         }
1093
1094                         if (prop >= irg_inline_forced ||
1095                             (is_smaller(callee, size) && env->n_nodes < maxsize) /* small function */) {
1096                                 if (current_ir_graph == callee) {
1097                                         /*
1098                                          * Recursive call: we cannot directly inline because we cannot walk
1099                                          * the graph and change it. So we have to make a copy of the graph
1100                                          * first.
1101                                          */
1102
1103                                         inline_irg_env *callee_env;
1104                                         ir_graph       *copy;
1105
1106                                         ir_free_resources(current_ir_graph, IR_RESOURCE_IRN_LINK|IR_RESOURCE_PHI_LIST);
1107
1108                                         /*
1109                                          * No copy yet, create one.
1110                                          * Note that recursive methods are never leafs, so it is sufficient
1111                                          * to test this condition here.
1112                                          */
1113                                         copy = create_irg_copy(callee);
1114
1115                                         /* create_irg_copy() destroys the Proj links, recompute them */
1116                                         phiproj_computed = 0;
1117
1118                                         ir_reserve_resources(current_ir_graph, IR_RESOURCE_IRN_LINK|IR_RESOURCE_PHI_LIST);
1119
1120                                         /* allocate new environment */
1121                                         callee_env = alloc_inline_irg_env();
1122                                         set_irg_link(copy, callee_env);
1123
1124                                         assure_loopinfo(copy);
1125                                         wenv.x              = callee_env;
1126                                         wenv.ignore_callers = 1;
1127                                         irg_walk_graph(copy, NULL, collect_calls2, &wenv);
1128
1129                                         /*
1130                                          * Enter the entity of the original graph. This is needed
1131                                          * for inline_method(). However, note that ent->irg still points
1132                                          * to callee, NOT to copy.
1133                                          */
1134                                         set_irg_entity(copy, get_irg_entity(callee));
1135
1136                                         pmap_insert(copied_graphs, callee, copy);
1137                                         callee = copy;
1138
1139                                         /* we have only one caller: the original graph */
1140                                         callee_env->n_callers      = 1;
1141                                         callee_env->n_callers_orig = 1;
1142                                 }
1143                                 if (! phiproj_computed) {
1144                                         phiproj_computed = 1;
1145                                         collect_phiprojs(current_ir_graph);
1146                                 }
1147                                 did_inline = inline_method(call, callee);
1148                                 if (did_inline) {
1149                                         inline_irg_env *callee_env = (inline_irg_env *)get_irg_link(callee);
1150
1151                                         /* call was inlined, Phi/Projs for current graph must be recomputed */
1152                                         phiproj_computed = 0;
1153
1154                                         /* callee was inline. Append its call list. */
1155                                         env->got_inline = 1;
1156                                         --env->n_call_nodes;
1157                                         append_call_list(env, callee_env, entry->loop_depth);
1158                                         --callee_env->n_callers;
1159
1160                                         /* after we have inlined callee, all called methods inside callee
1161                                            are now called once more */
1162                                         list_for_each_entry(call_entry, centry, &callee_env->calls, list) {
1163                                                 inline_irg_env *penv = (inline_irg_env*)get_irg_link(centry->callee);
1164                                                 ++penv->n_callers;
1165                                         }
1166
1167                                         /* remove this call from the list */
1168                                         list_del(&entry->list);
1169                                         continue;
1170                                 }
1171                         }
1172                 }
1173                 ir_free_resources(current_ir_graph, IR_RESOURCE_IRN_LINK|IR_RESOURCE_PHI_LIST);
1174         }
1175
1176         for (i = 0; i < n_irgs; ++i) {
1177                 irg = get_irp_irg(i);
1178                 env = (inline_irg_env*)get_irg_link(irg);
1179
1180                 if (env->got_inline) {
1181                         optimize_graph_df(irg);
1182                         optimize_cf(irg);
1183                 }
1184                 if (env->got_inline || (env->n_callers_orig != env->n_callers)) {
1185                         DB((dbg, LEVEL_1, "Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
1186                         env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
1187                         env->n_callers_orig, env->n_callers,
1188                         get_entity_name(get_irg_entity(irg))));
1189                 }
1190         }
1191
1192         /* kill the copied graphs: we don't need them anymore */
1193         foreach_pmap(copied_graphs, pm_entry) {
1194                 ir_graph *copy = (ir_graph*)pm_entry->value;
1195
1196                 /* reset the entity, otherwise it will be deleted in the next step ... */
1197                 set_irg_entity(copy, NULL);
1198                 free_ir_graph(copy);
1199         }
1200         pmap_destroy(copied_graphs);
1201
1202         obstack_free(&temp_obst, NULL);
1203         current_ir_graph = rem;
1204 }
1205
1206 typedef struct inline_leaf_functions_pass_t {
1207         ir_prog_pass_t pass;
1208         unsigned       maxsize;
1209         unsigned       leafsize;
1210         unsigned       size;
1211         int            ignore_runtime;
1212 } inline_leaf_functions_pass_t;
1213
1214 /**
1215  * Wrapper to run inline_leaf_functions() as a ir_prog pass.
1216  */
1217 static int inline_leaf_functions_wrapper(ir_prog *irp, void *context)
1218 {
1219         inline_leaf_functions_pass_t *pass = (inline_leaf_functions_pass_t*)context;
1220
1221         (void)irp;
1222         inline_leaf_functions(
1223                 pass->maxsize, pass->leafsize,
1224                 pass->size, pass->ignore_runtime);
1225         return 0;
1226 }
1227
1228 /* create a pass for inline_leaf_functions() */
1229 ir_prog_pass_t *inline_leaf_functions_pass(
1230         const char *name, unsigned maxsize, unsigned leafsize,
1231         unsigned size, int ignore_runtime)
1232 {
1233         inline_leaf_functions_pass_t *pass = XMALLOCZ(inline_leaf_functions_pass_t);
1234
1235         pass->maxsize        = maxsize;
1236         pass->leafsize       = leafsize;
1237         pass->size           = size;
1238         pass->ignore_runtime = ignore_runtime;
1239
1240         return def_prog_pass_constructor(
1241                 &pass->pass,
1242                 name ? name : "inline_leaf_functions",
1243                 inline_leaf_functions_wrapper);
1244 }
1245
1246 /**
1247  * Calculate the parameter weights for transmitting the address of a local variable.
1248  */
1249 static unsigned calc_method_local_weight(ir_node *arg)
1250 {
1251         int      i, j, k;
1252         unsigned v, weight = 0;
1253
1254         for (i = get_irn_n_outs(arg) - 1; i >= 0; --i) {
1255                 ir_node *succ = get_irn_out(arg, i);
1256
1257                 switch (get_irn_opcode(succ)) {
1258                 case iro_Load:
1259                 case iro_Store:
1260                         /* Loads and Store can be removed */
1261                         weight += 3;
1262                         break;
1263                 case iro_Sel:
1264                         /* check if all args are constant */
1265                         for (j = get_Sel_n_indexs(succ) - 1; j >= 0; --j) {
1266                                 ir_node *idx = get_Sel_index(succ, j);
1267                                 if (! is_Const(idx))
1268                                         return 0;
1269                         }
1270                         /* Check users on this Sel. Note: if a 0 is returned here, there was
1271                            some unsupported node. */
1272                         v = calc_method_local_weight(succ);
1273                         if (v == 0)
1274                                 return 0;
1275                         /* we can kill one Sel with constant indexes, this is cheap */
1276                         weight += v + 1;
1277                         break;
1278                 case iro_Id:
1279                         /* when looking backward we might find Id nodes */
1280                         weight += calc_method_local_weight(succ);
1281                         break;
1282                 case iro_Tuple:
1283                         /* unoptimized tuple */
1284                         for (j = get_Tuple_n_preds(succ) - 1; j >= 0; --j) {
1285                                 ir_node *pred = get_Tuple_pred(succ, j);
1286                                 if (pred == arg) {
1287                                         /* look for Proj(j) */
1288                                         for (k = get_irn_n_outs(succ) - 1; k >= 0; --k) {
1289                                                 ir_node *succ_succ = get_irn_out(succ, k);
1290                                                 if (is_Proj(succ_succ)) {
1291                                                         if (get_Proj_proj(succ_succ) == j) {
1292                                                                 /* found */
1293                                                                 weight += calc_method_local_weight(succ_succ);
1294                                                         }
1295                                                 } else {
1296                                                         /* this should NOT happen */
1297                                                         return 0;
1298                                                 }
1299                                         }
1300                                 }
1301                         }
1302                         break;
1303                 default:
1304                         /* any other node: unsupported yet or bad. */
1305                         return 0;
1306                 }
1307         }
1308         return weight;
1309 }
1310
1311 /**
1312  * Calculate the parameter weights for transmitting the address of a local variable.
1313  */
1314 static void analyze_irg_local_weights(inline_irg_env *env, ir_graph *irg)
1315 {
1316         ir_entity *ent = get_irg_entity(irg);
1317         ir_type  *mtp;
1318         size_t   nparams;
1319         int      i;
1320         long     proj_nr;
1321         ir_node  *irg_args, *arg;
1322
1323         mtp      = get_entity_type(ent);
1324         nparams  = get_method_n_params(mtp);
1325
1326         /* allocate a new array. currently used as 'analysed' flag */
1327         env->local_weights = NEW_ARR_D(unsigned, &temp_obst, nparams);
1328
1329         /* If the method haven't parameters we have nothing to do. */
1330         if (nparams <= 0)
1331                 return;
1332
1333         assure_irg_outs(irg);
1334         irg_args = get_irg_args(irg);
1335         for (i = get_irn_n_outs(irg_args) - 1; i >= 0; --i) {
1336                 arg     = get_irn_out(irg_args, i);
1337                 proj_nr = get_Proj_proj(arg);
1338                 env->local_weights[proj_nr] = calc_method_local_weight(arg);
1339         }
1340 }
1341
1342 /**
1343  * Calculate the benefice for transmitting an local variable address.
1344  * After inlining, the local variable might be transformed into a
1345  * SSA variable by scalar_replacement().
1346  */
1347 static unsigned get_method_local_adress_weight(ir_graph *callee, size_t pos)
1348 {
1349         inline_irg_env *env = (inline_irg_env*)get_irg_link(callee);
1350
1351         if (env->local_weights == NULL)
1352                 analyze_irg_local_weights(env, callee);
1353
1354         if (pos < ARR_LEN(env->local_weights))
1355                 return env->local_weights[pos];
1356         return 0;
1357 }
1358
1359 /**
1360  * Calculate a benefice value for inlining the given call.
1361  *
1362  * @param call       the call node we have to inspect
1363  * @param callee     the called graph
1364  */
1365 static int calc_inline_benefice(call_entry *entry, ir_graph *callee)
1366 {
1367         ir_node   *call = entry->call;
1368         ir_entity *ent  = get_irg_entity(callee);
1369         ir_type   *callee_frame;
1370         size_t    i, n_members, n_params;
1371         ir_node   *frame_ptr;
1372         ir_type   *mtp;
1373         int       weight = 0;
1374         int       all_const;
1375         unsigned  cc, v;
1376         irg_inline_property prop;
1377
1378         inline_irg_env *callee_env;
1379
1380         prop = get_irg_inline_property(callee);
1381         if (prop == irg_inline_forbidden) {
1382                 DB((dbg, LEVEL_2, "In %+F Call to %+F: inlining forbidden\n",
1383                     call, callee));
1384                 return entry->benefice = INT_MIN;
1385         }
1386
1387         callee_frame = get_irg_frame_type(callee);
1388         n_members = get_class_n_members(callee_frame);
1389         for (i = 0; i < n_members; ++i) {
1390                 ir_entity *frame_ent = get_class_member(callee_frame, i);
1391                 if (is_parameter_entity(frame_ent)) {
1392                         // TODO inliner should handle parameter entities by inserting Store operations
1393                         DB((dbg, LEVEL_2, "In %+F Call to %+F: inlining forbidden due to parameter entity\n", call, callee));
1394                         set_irg_inline_property(callee, irg_inline_forbidden);
1395                         return entry->benefice = INT_MIN;
1396                 }
1397         }
1398
1399         if (get_irg_additional_properties(callee) & mtp_property_noreturn) {
1400                 DB((dbg, LEVEL_2, "In %+F Call to %+F: not inlining noreturn or weak\n",
1401                     call, callee));
1402                 return entry->benefice = INT_MIN;
1403         }
1404
1405         /* costs for every passed parameter */
1406         n_params = get_Call_n_params(call);
1407         mtp      = get_entity_type(ent);
1408         cc       = get_method_calling_convention(mtp);
1409         if (cc & cc_reg_param) {
1410                 /* register parameter, smaller costs for register parameters */
1411                 size_t max_regs = cc & ~cc_bits;
1412
1413                 if (max_regs < n_params)
1414                         weight += max_regs * 2 + (n_params - max_regs) * 5;
1415                 else
1416                         weight += n_params * 2;
1417         } else {
1418                 /* parameters are passed an stack */
1419                 weight += 5 * n_params;
1420         }
1421
1422         /* constant parameters improve the benefice */
1423         frame_ptr = get_irg_frame(current_ir_graph);
1424         all_const = 1;
1425         for (i = 0; i < n_params; ++i) {
1426                 ir_node *param = get_Call_param(call, i);
1427
1428                 if (is_Const(param)) {
1429                         weight += get_method_param_weight(ent, i);
1430                 } else {
1431                         all_const = 0;
1432                         if (is_SymConst(param))
1433                                 weight += get_method_param_weight(ent, i);
1434                         else if (is_Sel(param) && get_Sel_ptr(param) == frame_ptr) {
1435                                 /*
1436                                  * An address of a local variable is transmitted. After
1437                                  * inlining, scalar_replacement might be able to remove the
1438                                  * local variable, so honor this.
1439                                  */
1440                                 v = get_method_local_adress_weight(callee, i);
1441                                 weight += v;
1442                                 if (v > 0)
1443                                         entry->local_adr = 1;
1444                         }
1445                 }
1446         }
1447         entry->all_const = all_const;
1448
1449         callee_env = (inline_irg_env*)get_irg_link(callee);
1450         if (callee_env->n_callers == 1 &&
1451             callee != current_ir_graph &&
1452             !entity_is_externally_visible(ent)) {
1453                 weight += 700;
1454         }
1455
1456         /* give a bonus for functions with one block */
1457         if (callee_env->n_blocks == 1)
1458                 weight = weight * 3 / 2;
1459
1460         /* and one for small non-recursive functions: we want them to be inlined in mostly every case */
1461         if (callee_env->n_nodes < 30 && !callee_env->recursive)
1462                 weight += 2000;
1463
1464         /* and finally for leafs: they do not increase the register pressure
1465            because of callee safe registers */
1466         if (callee_env->n_call_nodes == 0)
1467                 weight += 400;
1468
1469         /** it's important to inline inner loops first */
1470         if (entry->loop_depth > 30)
1471                 weight += 30 * 1024;
1472         else
1473                 weight += entry->loop_depth * 1024;
1474
1475         /*
1476          * All arguments constant is probably a good sign, give an extra bonus
1477          */
1478         if (all_const)
1479                 weight += 1024;
1480
1481         return entry->benefice = weight;
1482 }
1483
1484 typedef struct walk_env_t {
1485         ir_graph **irgs;
1486         size_t   last_irg;
1487 } walk_env_t;
1488
1489 /**
1490  * Callgraph walker, collect all visited graphs.
1491  */
1492 static void callgraph_walker(ir_graph *irg, void *data)
1493 {
1494         walk_env_t *env = (walk_env_t *)data;
1495         env->irgs[env->last_irg++] = irg;
1496 }
1497
1498 /**
1499  * Creates an inline order for all graphs.
1500  *
1501  * @return the list of graphs.
1502  */
1503 static ir_graph **create_irg_list(void)
1504 {
1505         ir_entity  **free_methods;
1506         size_t     n_irgs = get_irp_n_irgs();
1507         walk_env_t env;
1508
1509         cgana(&free_methods);
1510         xfree(free_methods);
1511
1512         compute_callgraph();
1513
1514         env.irgs     = XMALLOCNZ(ir_graph*, n_irgs);
1515         env.last_irg = 0;
1516
1517         callgraph_walk(NULL, callgraph_walker, &env);
1518         assert(n_irgs == env.last_irg);
1519
1520         free_callgraph();
1521
1522         return env.irgs;
1523 }
1524
1525 /**
1526  * Push a call onto the priority list if its benefice is big enough.
1527  *
1528  * @param pqueue   the priority queue of calls
1529  * @param call     the call entry
1530  * @param inlien_threshold
1531  *                 the threshold value
1532  */
1533 static void maybe_push_call(pqueue_t *pqueue, call_entry *call,
1534                             int inline_threshold)
1535 {
1536         ir_graph            *callee  = call->callee;
1537         irg_inline_property prop     = get_irg_inline_property(callee);
1538         int                 benefice = calc_inline_benefice(call, callee);
1539
1540         DB((dbg, LEVEL_2, "In %+F Call %+F to %+F has benefice %d\n",
1541             get_irn_irg(call->call), call->call, callee, benefice));
1542
1543         if (prop < irg_inline_forced && benefice < inline_threshold) {
1544                 return;
1545         }
1546
1547         pqueue_put(pqueue, call, benefice);
1548 }
1549
1550 /**
1551  * Try to inline calls into a graph.
1552  *
1553  * @param irg      the graph into which we inline
1554  * @param maxsize  do NOT inline if the size of irg gets
1555  *                 bigger than this amount
1556  * @param inline_threshold
1557  *                 threshold value for inline decision
1558  * @param copied_graphs
1559  *                 map containing copied of recursive graphs
1560  */
1561 static void inline_into(ir_graph *irg, unsigned maxsize,
1562                         int inline_threshold, pmap *copied_graphs)
1563 {
1564         int            phiproj_computed = 0;
1565         inline_irg_env *env = (inline_irg_env*)get_irg_link(irg);
1566         call_entry     *curr_call;
1567         wenv_t         wenv;
1568         pqueue_t       *pqueue;
1569
1570         if (env->n_call_nodes == 0)
1571                 return;
1572
1573         if (env->n_nodes > maxsize) {
1574                 DB((dbg, LEVEL_2, "%+F: too big (%d)\n", irg, env->n_nodes));
1575                 return;
1576         }
1577
1578         current_ir_graph = irg;
1579         ir_reserve_resources(irg, IR_RESOURCE_IRN_LINK|IR_RESOURCE_PHI_LIST);
1580
1581         /* put irgs into the pqueue */
1582         pqueue = new_pqueue();
1583
1584         list_for_each_entry(call_entry, curr_call, &env->calls, list) {
1585                 assert(is_Call(curr_call->call));
1586                 maybe_push_call(pqueue, curr_call, inline_threshold);
1587         }
1588
1589         /* note that the list of possible calls is updated during the process */
1590         while (!pqueue_empty(pqueue)) {
1591                 int                 did_inline;
1592                 call_entry          *curr_call  = (call_entry*)pqueue_pop_front(pqueue);
1593                 ir_graph            *callee     = curr_call->callee;
1594                 ir_node             *call_node  = curr_call->call;
1595                 inline_irg_env      *callee_env = (inline_irg_env*)get_irg_link(callee);
1596                 irg_inline_property prop        = get_irg_inline_property(callee);
1597                 ir_graph            *calleee;
1598                 int                 loop_depth;
1599                 const call_entry    *centry;
1600
1601                 if ((prop < irg_inline_forced) && env->n_nodes + callee_env->n_nodes > maxsize) {
1602                         DB((dbg, LEVEL_2, "%+F: too big (%d) + %+F (%d)\n", irg,
1603                                                 env->n_nodes, callee, callee_env->n_nodes));
1604                         continue;
1605                 }
1606
1607                 calleee = (ir_graph*)pmap_get(copied_graphs, callee);
1608                 if (calleee != NULL) {
1609                         int benefice = curr_call->benefice;
1610                         /*
1611                          * Reduce the weight for recursive function IFF not all arguments are const.
1612                          * inlining recursive functions is rarely good.
1613                          */
1614                         if (!curr_call->all_const)
1615                                 benefice -= 2000;
1616                         if (benefice < inline_threshold)
1617                                 continue;
1618
1619                         /*
1620                          * Remap callee if we have a copy.
1621                          */
1622                         callee     = calleee;
1623                         callee_env = (inline_irg_env*)get_irg_link(callee);
1624                 }
1625
1626                 if (current_ir_graph == callee) {
1627                         /*
1628                          * Recursive call: we cannot directly inline because we cannot
1629                          * walk the graph and change it. So we have to make a copy of
1630                          * the graph first.
1631                          */
1632                         int benefice = curr_call->benefice;
1633                         ir_graph *copy;
1634
1635                         /*
1636                          * Reduce the weight for recursive function IFF not all arguments are const.
1637                          * inlining recursive functions is rarely good.
1638                          */
1639                         if (!curr_call->all_const)
1640                                 benefice -= 2000;
1641                         if (benefice < inline_threshold)
1642                                 continue;
1643
1644                         ir_free_resources(irg, IR_RESOURCE_IRN_LINK|IR_RESOURCE_PHI_LIST);
1645
1646                         /*
1647                          * No copy yet, create one.
1648                          * Note that recursive methods are never leafs, so it is
1649                          * sufficient to test this condition here.
1650                          */
1651                         copy = create_irg_copy(callee);
1652
1653                         /* create_irg_copy() destroys the Proj links, recompute them */
1654                         phiproj_computed = 0;
1655
1656                         ir_reserve_resources(irg, IR_RESOURCE_IRN_LINK|IR_RESOURCE_PHI_LIST);
1657
1658                         /* allocate a new environment */
1659                         callee_env = alloc_inline_irg_env();
1660                         set_irg_link(copy, callee_env);
1661
1662                         assure_loopinfo(copy);
1663                         memset(&wenv, 0, sizeof(wenv));
1664                         wenv.x              = callee_env;
1665                         wenv.ignore_callers = 1;
1666                         irg_walk_graph(copy, NULL, collect_calls2, &wenv);
1667
1668                         /*
1669                          * Enter the entity of the original graph. This is needed
1670                          * for inline_method(). However, note that ent->irg still points
1671                          * to callee, NOT to copy.
1672                          */
1673                         set_irg_entity(copy, get_irg_entity(callee));
1674
1675                         pmap_insert(copied_graphs, callee, copy);
1676                         callee = copy;
1677
1678                         /* we have only one caller: the original graph */
1679                         callee_env->n_callers      = 1;
1680                         callee_env->n_callers_orig = 1;
1681                 }
1682                 if (! phiproj_computed) {
1683                         phiproj_computed = 1;
1684                         collect_phiprojs(current_ir_graph);
1685                 }
1686                 did_inline = inline_method(call_node, callee);
1687                 if (!did_inline)
1688                         continue;
1689
1690                 /* call was inlined, Phi/Projs for current graph must be recomputed */
1691                 phiproj_computed = 0;
1692
1693                 /* remove it from the caller list */
1694                 list_del(&curr_call->list);
1695
1696                 /* callee was inline. Append its call list. */
1697                 env->got_inline = 1;
1698                 --env->n_call_nodes;
1699
1700                 /* we just generate a bunch of new calls */
1701                 loop_depth = curr_call->loop_depth;
1702                 list_for_each_entry(call_entry, centry, &callee_env->calls, list) {
1703                         inline_irg_env *penv = (inline_irg_env*)get_irg_link(centry->callee);
1704                         ir_node        *new_call;
1705                         call_entry     *new_entry;
1706
1707                         /* after we have inlined callee, all called methods inside
1708                          * callee are now called once more */
1709                         ++penv->n_callers;
1710
1711                         /* Note that the src list points to Call nodes in the inlined graph,
1712                          * but we need Call nodes in our graph. Luckily the inliner leaves
1713                          * this information in the link field. */
1714                         new_call = (ir_node*)get_irn_link(centry->call);
1715                         if (get_irn_irg(new_call) != irg) {
1716                                 /* centry->call has not been copied, which means it is dead.
1717                                  * This might happen during inlining, if a const function,
1718                                  * which cannot be inlined is only used as an unused argument
1719                                  * of another function, which is inlined. */
1720                                 continue;
1721                         }
1722                         assert(is_Call(new_call));
1723
1724                         new_entry = duplicate_call_entry(centry, new_call, loop_depth);
1725                         list_add_tail(&new_entry->list, &env->calls);
1726                         maybe_push_call(pqueue, new_entry, inline_threshold);
1727                 }
1728
1729                 env->n_call_nodes += callee_env->n_call_nodes;
1730                 env->n_nodes += callee_env->n_nodes;
1731                 --callee_env->n_callers;
1732         }
1733         ir_free_resources(irg, IR_RESOURCE_IRN_LINK|IR_RESOURCE_PHI_LIST);
1734         del_pqueue(pqueue);
1735 }
1736
1737 /*
1738  * Heuristic inliner. Calculates a benefice value for every call and inlines
1739  * those calls with a value higher than the threshold.
1740  */
1741 void inline_functions(unsigned maxsize, int inline_threshold,
1742                       opt_ptr after_inline_opt)
1743 {
1744         inline_irg_env   *env;
1745         size_t           i, n_irgs;
1746         ir_graph         *rem;
1747         wenv_t           wenv;
1748         pmap             *copied_graphs;
1749         pmap_entry       *pm_entry;
1750         ir_graph         **irgs;
1751
1752         rem = current_ir_graph;
1753         obstack_init(&temp_obst);
1754
1755         irgs = create_irg_list();
1756
1757         /* a map for the copied graphs, used to inline recursive calls */
1758         copied_graphs = pmap_create();
1759
1760         /* extend all irgs by a temporary data structure for inlining. */
1761         n_irgs = get_irp_n_irgs();
1762         for (i = 0; i < n_irgs; ++i)
1763                 set_irg_link(irgs[i], alloc_inline_irg_env());
1764
1765         /* Pre-compute information in temporary data structure. */
1766         wenv.ignore_runtime = 0;
1767         wenv.ignore_callers = 0;
1768         for (i = 0; i < n_irgs; ++i) {
1769                 ir_graph *irg = irgs[i];
1770
1771                 free_callee_info(irg);
1772
1773                 wenv.x = (inline_irg_env*)get_irg_link(irg);
1774                 assure_loopinfo(irg);
1775                 irg_walk_graph(irg, NULL, collect_calls2, &wenv);
1776         }
1777
1778         /* -- and now inline. -- */
1779         for (i = 0; i < n_irgs; ++i) {
1780                 ir_graph *irg = irgs[i];
1781
1782                 inline_into(irg, maxsize, inline_threshold, copied_graphs);
1783         }
1784
1785         for (i = 0; i < n_irgs; ++i) {
1786                 ir_graph *irg = irgs[i];
1787
1788                 env = (inline_irg_env*)get_irg_link(irg);
1789                 if (env->got_inline && after_inline_opt != NULL) {
1790                         /* this irg got calls inlined: optimize it */
1791                         after_inline_opt(irg);
1792                 }
1793                 if (env->got_inline || (env->n_callers_orig != env->n_callers)) {
1794                         DB((dbg, LEVEL_1, "Nodes:%3d ->%3d, calls:%3d ->%3d, callers:%3d ->%3d, -- %s\n",
1795                         env->n_nodes_orig, env->n_nodes, env->n_call_nodes_orig, env->n_call_nodes,
1796                         env->n_callers_orig, env->n_callers,
1797                         get_entity_name(get_irg_entity(irg))));
1798                 }
1799         }
1800
1801         /* kill the copied graphs: we don't need them anymore */
1802         foreach_pmap(copied_graphs, pm_entry) {
1803                 ir_graph *copy = (ir_graph*)pm_entry->value;
1804
1805                 /* reset the entity, otherwise it will be deleted in the next step ... */
1806                 set_irg_entity(copy, NULL);
1807                 free_ir_graph(copy);
1808         }
1809         pmap_destroy(copied_graphs);
1810
1811         xfree(irgs);
1812
1813         obstack_free(&temp_obst, NULL);
1814         current_ir_graph = rem;
1815 }
1816
1817 typedef struct inline_functions_pass_t {
1818         ir_prog_pass_t pass;
1819         unsigned       maxsize;
1820         int            inline_threshold;
1821         opt_ptr        after_inline_opt;
1822 } inline_functions_pass_t;
1823
1824 /**
1825  * Wrapper to run inline_functions() as a ir_prog pass.
1826  */
1827 static int inline_functions_wrapper(ir_prog *irp, void *context)
1828 {
1829         inline_functions_pass_t *pass = (inline_functions_pass_t*)context;
1830
1831         (void)irp;
1832         inline_functions(pass->maxsize, pass->inline_threshold,
1833                          pass->after_inline_opt);
1834         return 0;
1835 }
1836
1837 /* create a ir_prog pass for inline_functions */
1838 ir_prog_pass_t *inline_functions_pass(
1839           const char *name, unsigned maxsize, int inline_threshold,
1840           opt_ptr after_inline_opt)
1841 {
1842         inline_functions_pass_t *pass = XMALLOCZ(inline_functions_pass_t);
1843
1844         pass->maxsize          = maxsize;
1845         pass->inline_threshold = inline_threshold;
1846         pass->after_inline_opt = after_inline_opt;
1847
1848         return def_prog_pass_constructor(
1849                 &pass->pass, name ? name : "inline_functions",
1850                 inline_functions_wrapper);
1851 }
1852
1853 void firm_init_inline(void)
1854 {
1855         FIRM_DBG_REGISTER(dbg, "firm.opt.inline");
1856 }