gcc/tree-tailcall.c

   1 /* Tail call optimization on trees.
   2    Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
   3    Free Software Foundation, Inc.
   4
   5 This file is part of GCC.
   6
   7 GCC is free software; you can redistribute it and/or modify
   8 it under the terms of the GNU General Public License as published by
   9 the Free Software Foundation; either version 3, or (at your option)
  10 any later version.
  11
  12 GCC is distributed in the hope that it will be useful,
  13 but WITHOUT ANY WARRANTY; without even the implied warranty of
  14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15 GNU General Public License for more details.
  16
  17 You should have received a copy of the GNU General Public License
  18 along with GCC; see the file COPYING3.  If not see
  19 <http://www.gnu.org/licenses/>.  */
  20
  21 #include "config.h"
  22 #include "system.h"
  23 #include "coretypes.h"
  24 #include "tm.h"
  25 #include "tree.h"
  26 #include "rtl.h"
  27 #include "tm_p.h"
  28 #include "hard-reg-set.h"
  29 #include "basic-block.h"
  30 #include "function.h"
  31 #include "tree-flow.h"
  32 #include "tree-dump.h"
  33 #include "diagnostic.h"
  34 #include "except.h"
  35 #include "tree-pass.h"
  36 #include "flags.h"
  37 #include "langhooks.h"
  38 #include "dbgcnt.h"
  39
  40 /* The file implements the tail recursion elimination.  It is also used to
  41    analyze the tail calls in general, passing the results to the rtl level
  42    where they are used for sibcall optimization.
  43
  44    In addition to the standard tail recursion elimination, we handle the most
  45    trivial cases of making the call tail recursive by creating accumulators.
  46    For example the following function
  47
  48    int sum (int n)
  49    {
  50      if (n > 0)
  51        return n + sum (n - 1);
  52      else
  53        return 0;
  54    }
  55
  56    is transformed into
  57
  58    int sum (int n)
  59    {
  60      int acc = 0;
  61
  62      while (n > 0)
  63        acc += n--;
  64
  65      return acc;
  66    }
  67
  68    To do this, we maintain two accumulators (a_acc and m_acc) that indicate
  69    when we reach the return x statement, we should return a_acc + x * m_acc
  70    instead.  They are initially initialized to 0 and 1, respectively,
  71    so the semantics of the function is obviously preserved.  If we are
  72    guaranteed that the value of the accumulator never change, we
  73    omit the accumulator.
  74
  75    There are three cases how the function may exit.  The first one is
  76    handled in adjust_return_value, the other two in adjust_accumulator_values
  77    (the second case is actually a special case of the third one and we
  78    present it separately just for clarity):
  79
  80    1) Just return x, where x is not in any of the remaining special shapes.
  81       We rewrite this to a gimple equivalent of return m_acc * x + a_acc.
  82
  83    2) return f (...), where f is the current function, is rewritten in a
  84       classical tail-recursion elimination way, into assignment of arguments
  85       and jump to the start of the function.  Values of the accumulators
  86       are unchanged.
  87
  88    3) return a + m * f(...), where a and m do not depend on call to f.
  89       To preserve the semantics described before we want this to be rewritten
  90       in such a way that we finally return
  91
  92       a_acc + (a + m * f(...)) * m_acc = (a_acc + a * m_acc) + (m * m_acc) * f(...).
  93
  94       I.e. we increase a_acc by a * m_acc, multiply m_acc by m and
  95       eliminate the tail call to f.  Special cases when the value is just
  96       added or just multiplied are obtained by setting a = 0 or m = 1.
  97
  98    TODO -- it is possible to do similar tricks for other operations.  */
  99
 100 /* A structure that describes the tailcall.  */
 101
 102 struct tailcall
 103 {
 104   /* The iterator pointing to the call statement.  */
 105   gimple_stmt_iterator call_gsi;
 106
 107   /* True if it is a call to the current function.  */
 108   bool tail_recursion;
 109
 110   /* The return value of the caller is mult * f + add, where f is the return
 111      value of the call.  */
 112   tree mult, add;
 113
 114   /* Next tailcall in the chain.  */
 115   struct tailcall *next;
 116 };
 117
 118 /* The variables holding the value of multiplicative and additive
 119    accumulator.  */
 120 static tree m_acc, a_acc;
 121
 122 static bool suitable_for_tail_opt_p (void);
 123 static bool optimize_tail_call (struct tailcall *, bool);
 124 static void eliminate_tail_call (struct tailcall *);
 125 static void find_tail_calls (basic_block, struct tailcall **);
 126
 127 /* Returns false when the function is not suitable for tail call optimization
 128    from some reason (e.g. if it takes variable number of arguments).  */
 129
 130 static bool
 131 suitable_for_tail_opt_p (void)
 132 {
 133   if (cfun->stdarg)
 134     return false;
 135
 136   return true;
 137 }
 138 /* Returns false when the function is not suitable for tail call optimization
 139    from some reason (e.g. if it takes variable number of arguments).
 140    This test must pass in addition to suitable_for_tail_opt_p in order to make
 141    tail call discovery happen.  */
 142
 143 static bool
 144 suitable_for_tail_call_opt_p (void)
 145 {
 146   tree param;
 147
 148   /* alloca (until we have stack slot life analysis) inhibits
 149      sibling call optimizations, but not tail recursion.  */
 150   if (cfun->calls_alloca)
 151     return false;
 152
 153   /* If we are using sjlj exceptions, we may need to add a call to
 154      _Unwind_SjLj_Unregister at exit of the function.  Which means
 155      that we cannot do any sibcall transformations.  */
 156   if (USING_SJLJ_EXCEPTIONS && current_function_has_exception_handlers ())
 157     return false;
 158
 159   /* Any function that calls setjmp might have longjmp called from
 160      any called function.  ??? We really should represent this
 161      properly in the CFG so that this needn't be special cased.  */
 162   if (cfun->calls_setjmp)
 163     return false;
 164
 165   /* ??? It is OK if the argument of a function is taken in some cases,
 166      but not in all cases.  See PR15387 and PR19616.  Revisit for 4.1.  */
 167   for (param = DECL_ARGUMENTS (current_function_decl);
 168        param;
 169        param = TREE_CHAIN (param))
 170     if (TREE_ADDRESSABLE (param))
 171       return false;
 172
 173   return true;
 174 }
 175
 176 /* Checks whether the expression EXPR in stmt AT is independent of the
 177    statement pointed to by GSI (in a sense that we already know EXPR's value
 178    at GSI).  We use the fact that we are only called from the chain of
 179    basic blocks that have only single successor.  Returns the expression
 180    containing the value of EXPR at GSI.  */
 181
 182 static tree
 183 independent_of_stmt_p (tree expr, gimple at, gimple_stmt_iterator gsi)
 184 {
 185   basic_block bb, call_bb, at_bb;
 186   edge e;
 187   edge_iterator ei;
 188
 189   if (is_gimple_min_invariant (expr))
 190     return expr;
 191
 192   if (TREE_CODE (expr) != SSA_NAME)
 193     return NULL_TREE;
 194
 195   /* Mark the blocks in the chain leading to the end.  */
 196   at_bb = gimple_bb (at);
 197   call_bb = gimple_bb (gsi_stmt (gsi));
 198   for (bb = call_bb; bb != at_bb; bb = single_succ (bb))
 199     bb->aux = &bb->aux;
 200   bb->aux = &bb->aux;
 201
 202   while (1)
 203     {
 204       at = SSA_NAME_DEF_STMT (expr);
 205       bb = gimple_bb (at);
 206
 207       /* The default definition or defined before the chain.  */
 208       if (!bb || !bb->aux)
 209         break;
 210
 211       if (bb == call_bb)
 212         {
 213           for (; !gsi_end_p (gsi); gsi_next (&gsi))
 214             if (gsi_stmt (gsi) == at)
 215               break;
 216
 217           if (!gsi_end_p (gsi))
 218             expr = NULL_TREE;
 219           break;
 220         }
 221
 222       if (gimple_code (at) != GIMPLE_PHI)
 223         {
 224           expr = NULL_TREE;
 225           break;
 226         }
 227
 228       FOR_EACH_EDGE (e, ei, bb->preds)
 229         if (e->src->aux)
 230           break;
 231       gcc_assert (e);
 232
 233       expr = PHI_ARG_DEF_FROM_EDGE (at, e);
 234       if (TREE_CODE (expr) != SSA_NAME)
 235         {
 236           /* The value is a constant.  */
 237           break;
 238         }
 239     }
 240
 241   /* Unmark the blocks.  */
 242   for (bb = call_bb; bb != at_bb; bb = single_succ (bb))
 243     bb->aux = NULL;
 244   bb->aux = NULL;
 245
 246   return expr;
 247 }
 248
 249 /* Simulates the effect of an assignment STMT on the return value of the tail
 250    recursive CALL passed in ASS_VAR.  M and A are the multiplicative and the
 251    additive factor for the real return value.  */
 252
 253 static bool
 254 process_assignment (gimple stmt, gimple_stmt_iterator call, tree *m,
 255                     tree *a, tree *ass_var)
 256 {
 257   tree op0, op1, non_ass_var;
 258   tree dest = gimple_assign_lhs (stmt);
 259   enum tree_code code = gimple_assign_rhs_code (stmt);
 260   enum gimple_rhs_class rhs_class = get_gimple_rhs_class (code);
 261   tree src_var = gimple_assign_rhs1 (stmt);
 262
 263   /* See if this is a simple copy operation of an SSA name to the function
 264      result.  In that case we may have a simple tail call.  Ignore type
 265      conversions that can never produce extra code between the function
 266      call and the function return.  */
 267   if ((rhs_class == GIMPLE_SINGLE_RHS || gimple_assign_cast_p (stmt))
 268       && (TREE_CODE (src_var) == SSA_NAME))
 269     {
 270       /* Reject a tailcall if the type conversion might need
 271          additional code.  */
 272       if (gimple_assign_cast_p (stmt)
 273           && TYPE_MODE (TREE_TYPE (dest)) != TYPE_MODE (TREE_TYPE (src_var)))
 274         return false;
 275
 276       if (src_var != *ass_var)
 277         return false;
 278
 279       *ass_var = dest;
 280       return true;
 281     }
 282
 283   if (rhs_class != GIMPLE_BINARY_RHS)
 284     return false;
 285
 286   /* Accumulator optimizations will reverse the order of operations.
 287      We can only do that for floating-point types if we're assuming
 288      that addition and multiplication are associative.  */
 289   if (!flag_associative_math)
 290     if (FLOAT_TYPE_P (TREE_TYPE (DECL_RESULT (current_function_decl))))
 291       return false;
 292
 293   /* We only handle the code like
 294
 295      x = call ();
 296      y = m * x;
 297      z = y + a;
 298      return z;
 299
 300      TODO -- Extend it for cases where the linear transformation of the output
 301      is expressed in a more complicated way.  */
 302
 303   op0 = gimple_assign_rhs1 (stmt);
 304   op1 = gimple_assign_rhs2 (stmt);
 305
 306   if (op0 == *ass_var
 307       && (non_ass_var = independent_of_stmt_p (op1, stmt, call)))
 308     ;
 309   else if (op1 == *ass_var
 310            && (non_ass_var = independent_of_stmt_p (op0, stmt, call)))
 311     ;
 312   else
 313     return false;
 314
 315   switch (code)
 316     {
 317     case PLUS_EXPR:
 318       *a = non_ass_var;
 319       *ass_var = dest;
 320       return true;
 321
 322     case MULT_EXPR:
 323       *m = non_ass_var;
 324       *ass_var = dest;
 325       return true;
 326
 327       /* TODO -- Handle other codes (NEGATE_EXPR, MINUS_EXPR,
 328          POINTER_PLUS_EXPR).  */
 329
 330     default:
 331       return false;
 332     }
 333 }
 334
 335 /* Propagate VAR through phis on edge E.  */
 336
 337 static tree
 338 propagate_through_phis (tree var, edge e)
 339 {
 340   basic_block dest = e->dest;
 341   gimple_stmt_iterator gsi;
 342
 343   for (gsi = gsi_start_phis (dest); !gsi_end_p (gsi); gsi_next (&gsi))
 344     {
 345       gimple phi = gsi_stmt (gsi);
 346       if (PHI_ARG_DEF_FROM_EDGE (phi, e) == var)
 347         return PHI_RESULT (phi);
 348     }
 349   return var;
 350 }
 351
 352 /* Finds tailcalls falling into basic block BB. The list of found tailcalls is
 353    added to the start of RET.  */
 354
 355 static void
 356 find_tail_calls (basic_block bb, struct tailcall **ret)
 357 {
 358   tree ass_var = NULL_TREE, ret_var, func, param;
 359   gimple stmt, call = NULL;
 360   gimple_stmt_iterator gsi, agsi;
 361   bool tail_recursion;
 362   struct tailcall *nw;
 363   edge e;
 364   tree m, a;
 365   basic_block abb;
 366   size_t idx;
 367   tree var;
 368   referenced_var_iterator rvi;
 369
 370   if (!single_succ_p (bb))
 371     return;
 372
 373   for (gsi = gsi_last_bb (bb); !gsi_end_p (gsi); gsi_prev (&gsi))
 374     {
 375       stmt = gsi_stmt (gsi);
 376
 377       /* Ignore labels.  */
 378       if (gimple_code (stmt) == GIMPLE_LABEL || is_gimple_debug (stmt))
 379         continue;
 380
 381       /* Check for a call.  */
 382       if (is_gimple_call (stmt))
 383         {
 384           call = stmt;
 385           ass_var = gimple_call_lhs (stmt);
 386           break;
 387         }
 388
 389       /* If the statement references memory or volatile operands, fail.  */
 390       if (gimple_references_memory_p (stmt)
 391           || gimple_has_volatile_ops (stmt))
 392         return;
 393     }
 394
 395   if (gsi_end_p (gsi))
 396     {
 397       edge_iterator ei;
 398       /* Recurse to the predecessors.  */
 399       FOR_EACH_EDGE (e, ei, bb->preds)
 400         find_tail_calls (e->src, ret);
 401
 402       return;
 403     }
 404
 405   /* If the LHS of our call is not just a simple register, we can't
 406      transform this into a tail or sibling call.  This situation happens,
 407      in (e.g.) "*p = foo()" where foo returns a struct.  In this case
 408      we won't have a temporary here, but we need to carry out the side
 409      effect anyway, so tailcall is impossible.
 410
 411      ??? In some situations (when the struct is returned in memory via
 412      invisible argument) we could deal with this, e.g. by passing 'p'
 413      itself as that argument to foo, but it's too early to do this here,
 414      and expand_call() will not handle it anyway.  If it ever can, then
 415      we need to revisit this here, to allow that situation.  */
 416   if (ass_var && !is_gimple_reg (ass_var))
 417     return;
 418
 419   /* We found the call, check whether it is suitable.  */
 420   tail_recursion = false;
 421   func = gimple_call_fndecl (call);
 422   if (func == current_function_decl)
 423     {
 424       tree arg;
 425
 426       for (param = DECL_ARGUMENTS (func), idx = 0;
 427            param && idx < gimple_call_num_args (call);
 428            param = TREE_CHAIN (param), idx ++)
 429         {
 430           arg = gimple_call_arg (call, idx);
 431           if (param != arg)
 432             {
 433               /* Make sure there are no problems with copying.  The parameter
 434                  have a copyable type and the two arguments must have reasonably
 435                  equivalent types.  The latter requirement could be relaxed if
 436                  we emitted a suitable type conversion statement.  */
 437               if (!is_gimple_reg_type (TREE_TYPE (param))
 438                   || !useless_type_conversion_p (TREE_TYPE (param),
 439                                                  TREE_TYPE (arg)))
 440                 break;
 441
 442               /* The parameter should be a real operand, so that phi node
 443                  created for it at the start of the function has the meaning
 444                  of copying the value.  This test implies is_gimple_reg_type
 445                  from the previous condition, however this one could be
 446                  relaxed by being more careful with copying the new value
 447                  of the parameter (emitting appropriate GIMPLE_ASSIGN and
 448                  updating the virtual operands).  */
 449               if (!is_gimple_reg (param))
 450                 break;
 451             }
 452         }
 453       if (idx == gimple_call_num_args (call) && !param)
 454         tail_recursion = true;
 455     }
 456
 457   /* Make sure the tail invocation of this function does not refer
 458      to local variables.  */
 459   FOR_EACH_REFERENCED_VAR (var, rvi)
 460     {
 461       if (TREE_CODE (var) != PARM_DECL
 462           && auto_var_in_fn_p (var, cfun->decl)
 463           && ref_maybe_used_by_stmt_p (call, var))
 464         return;
 465     }
 466
 467   /* Now check the statements after the call.  None of them has virtual
 468      operands, so they may only depend on the call through its return
 469      value.  The return value should also be dependent on each of them,
 470      since we are running after dce.  */
 471   m = NULL_TREE;
 472   a = NULL_TREE;
 473
 474   abb = bb;
 475   agsi = gsi;
 476   while (1)
 477     {
 478       tree tmp_a = NULL_TREE;
 479       tree tmp_m = NULL_TREE;
 480       gsi_next (&agsi);
 481
 482       while (gsi_end_p (agsi))
 483         {
 484           ass_var = propagate_through_phis (ass_var, single_succ_edge (abb));
 485           abb = single_succ (abb);
 486           agsi = gsi_start_bb (abb);
 487         }
 488
 489       stmt = gsi_stmt (agsi);
 490
 491       if (gimple_code (stmt) == GIMPLE_LABEL)
 492         continue;
 493
 494       if (gimple_code (stmt) == GIMPLE_RETURN)
 495         break;
 496
 497       if (is_gimple_debug (stmt))
 498         continue;
 499
 500       if (gimple_code (stmt) != GIMPLE_ASSIGN)
 501         return;
 502
 503       /* This is a gimple assign. */
 504       if (! process_assignment (stmt, gsi, &tmp_m, &tmp_a, &ass_var))
 505         return;
 506
 507       if (tmp_a)
 508         {
 509           if (a)
 510             a = fold_build2 (PLUS_EXPR, TREE_TYPE (tmp_a), a, tmp_a);
 511           else
 512             a = tmp_a;
 513         }
 514       if (tmp_m)
 515         {
 516           if (m)
 517             m = fold_build2 (MULT_EXPR, TREE_TYPE (tmp_m), m, tmp_m);
 518           else
 519             m = tmp_m;
 520
 521           if (a)
 522             a = fold_build2 (MULT_EXPR, TREE_TYPE (tmp_m), a, tmp_m);
 523         }
 524     }
 525
 526   /* See if this is a tail call we can handle.  */
 527   ret_var = gimple_return_retval (stmt);
 528
 529   /* We may proceed if there either is no return value, or the return value
 530      is identical to the call's return.  */
 531   if (ret_var
 532       && (ret_var != ass_var))
 533     return;
 534
 535   /* If this is not a tail recursive call, we cannot handle addends or
 536      multiplicands.  */
 537   if (!tail_recursion && (m || a))
 538     return;
 539
 540   nw = XNEW (struct tailcall);
 541
 542   nw->call_gsi = gsi;
 543
 544   nw->tail_recursion = tail_recursion;
 545
 546   nw->mult = m;
 547   nw->add = a;
 548
 549   nw->next = *ret;
 550   *ret = nw;
 551 }
 552
 553 /* Helper to insert PHI_ARGH to the phi of VAR in the destination of edge E.  */
 554
 555 static void
 556 add_successor_phi_arg (edge e, tree var, tree phi_arg)
 557 {
 558   gimple_stmt_iterator gsi;
 559
 560   for (gsi = gsi_start_phis (e->dest); !gsi_end_p (gsi); gsi_next (&gsi))
 561     if (PHI_RESULT (gsi_stmt (gsi)) == var)
 562       break;
 563
 564   gcc_assert (!gsi_end_p (gsi));
 565   add_phi_arg (gsi_stmt (gsi), phi_arg, e, UNKNOWN_LOCATION);
 566 }
 567
 568 /* Creates a GIMPLE statement which computes the operation specified by
 569    CODE, OP0 and OP1 to a new variable with name LABEL and inserts the
 570    statement in the position specified by GSI and UPDATE.  Returns the
 571    tree node of the statement's result.  */
 572
 573 static tree
 574 adjust_return_value_with_ops (enum tree_code code, const char *label,
 575                               tree acc, tree op1, gimple_stmt_iterator gsi)
 576 {
 577
 578   tree ret_type = TREE_TYPE (DECL_RESULT (current_function_decl));
 579   tree tmp = create_tmp_reg (ret_type, label);
 580   gimple stmt;
 581   tree result;
 582
 583   add_referenced_var (tmp);
 584
 585   if (types_compatible_p (TREE_TYPE (acc), TREE_TYPE (op1)))
 586     stmt = gimple_build_assign_with_ops (code, tmp, acc, op1);
 587   else
 588     {
 589       tree rhs = fold_convert (TREE_TYPE (acc),
 590                                fold_build2 (code,
 591                                             TREE_TYPE (op1),
 592                                             fold_convert (TREE_TYPE (op1), acc),
 593                                             op1));
 594       rhs = force_gimple_operand_gsi (&gsi, rhs,
 595                                       false, NULL, true, GSI_CONTINUE_LINKING);
 596       stmt = gimple_build_assign (NULL_TREE, rhs);
 597     }
 598
 599   result = make_ssa_name (tmp, stmt);
 600   gimple_assign_set_lhs (stmt, result);
 601   update_stmt (stmt);
 602   gsi_insert_before (&gsi, stmt, GSI_NEW_STMT);
 603   return result;
 604 }
 605
 606 /* Creates a new GIMPLE statement that adjusts the value of accumulator ACC by
 607    the computation specified by CODE and OP1 and insert the statement
 608    at the position specified by GSI as a new statement.  Returns new SSA name
 609    of updated accumulator.  */
 610
 611 static tree
 612 update_accumulator_with_ops (enum tree_code code, tree acc, tree op1,
 613                              gimple_stmt_iterator gsi)
 614 {
 615   gimple stmt;
 616   tree var;
 617   if (types_compatible_p (TREE_TYPE (acc), TREE_TYPE (op1)))
 618     stmt = gimple_build_assign_with_ops (code, SSA_NAME_VAR (acc), acc, op1);
 619   else
 620     {
 621       tree rhs = fold_convert (TREE_TYPE (acc),
 622                                fold_build2 (code,
 623                                             TREE_TYPE (op1),
 624                                             fold_convert (TREE_TYPE (op1), acc),
 625                                             op1));
 626       rhs = force_gimple_operand_gsi (&gsi, rhs,
 627                                       false, NULL, false, GSI_CONTINUE_LINKING);
 628       stmt = gimple_build_assign (NULL_TREE, rhs);
 629     }
 630   var = make_ssa_name (SSA_NAME_VAR (acc), stmt);
 631   gimple_assign_set_lhs (stmt, var);
 632   update_stmt (stmt);
 633   gsi_insert_after (&gsi, stmt, GSI_NEW_STMT);
 634   return var;
 635 }
 636
 637 /* Adjust the accumulator values according to A and M after GSI, and update
 638    the phi nodes on edge BACK.  */
 639
 640 static void
 641 adjust_accumulator_values (gimple_stmt_iterator gsi, tree m, tree a, edge back)
 642 {
 643   tree var, a_acc_arg, m_acc_arg;
 644
 645   if (m)
 646     m = force_gimple_operand_gsi (&gsi, m, true, NULL, true, GSI_SAME_STMT);
 647   if (a)
 648     a = force_gimple_operand_gsi (&gsi, a, true, NULL, true, GSI_SAME_STMT);
 649
 650   a_acc_arg = a_acc;
 651   m_acc_arg = m_acc;
 652   if (a)
 653     {
 654       if (m_acc)
 655         {
 656           if (integer_onep (a))
 657             var = m_acc;
 658           else
 659             var = adjust_return_value_with_ops (MULT_EXPR, "acc_tmp", m_acc,
 660                                                 a, gsi);
 661         }
 662       else
 663         var = a;
 664
 665       a_acc_arg = update_accumulator_with_ops (PLUS_EXPR, a_acc, var, gsi);
 666     }
 667
 668   if (m)
 669     m_acc_arg = update_accumulator_with_ops (MULT_EXPR, m_acc, m, gsi);
 670
 671   if (a_acc)
 672     add_successor_phi_arg (back, a_acc, a_acc_arg);
 673
 674   if (m_acc)
 675     add_successor_phi_arg (back, m_acc, m_acc_arg);
 676 }
 677
 678 /* Adjust value of the return at the end of BB according to M and A
 679    accumulators.  */
 680
 681 static void
 682 adjust_return_value (basic_block bb, tree m, tree a)
 683 {
 684   tree retval;
 685   gimple ret_stmt = gimple_seq_last_stmt (bb_seq (bb));
 686   gimple_stmt_iterator gsi = gsi_last_bb (bb);
 687
 688   gcc_assert (gimple_code (ret_stmt) == GIMPLE_RETURN);
 689
 690   retval = gimple_return_retval (ret_stmt);
 691   if (!retval || retval == error_mark_node)
 692     return;
 693
 694   if (m)
 695     retval = adjust_return_value_with_ops (MULT_EXPR, "mul_tmp", m_acc, retval,
 696                                            gsi);
 697   if (a)
 698     retval = adjust_return_value_with_ops (PLUS_EXPR, "acc_tmp", a_acc, retval,
 699                                            gsi);
 700   gimple_return_set_retval (ret_stmt, retval);
 701   update_stmt (ret_stmt);
 702 }
 703
 704 /* Subtract COUNT and FREQUENCY from the basic block and it's
 705    outgoing edge.  */
 706 static void
 707 decrease_profile (basic_block bb, gcov_type count, int frequency)
 708 {
 709   edge e;
 710   bb->count -= count;
 711   if (bb->count < 0)
 712     bb->count = 0;
 713   bb->frequency -= frequency;
 714   if (bb->frequency < 0)
 715     bb->frequency = 0;
 716   if (!single_succ_p (bb))
 717     {
 718       gcc_assert (!EDGE_COUNT (bb->succs));
 719       return;
 720     }
 721   e = single_succ_edge (bb);
 722   e->count -= count;
 723   if (e->count < 0)
 724     e->count = 0;
 725 }
 726
 727 /* Returns true if argument PARAM of the tail recursive call needs to be copied
 728    when the call is eliminated.  */
 729
 730 static bool
 731 arg_needs_copy_p (tree param)
 732 {
 733   tree def;
 734
 735   if (!is_gimple_reg (param) || !var_ann (param))
 736     return false;
 737
 738   /* Parameters that are only defined but never used need not be copied.  */
 739   def = gimple_default_def (cfun, param);
 740   if (!def)
 741     return false;
 742
 743   return true;
 744 }
 745
 746 /* Eliminates tail call described by T.  TMP_VARS is a list of
 747    temporary variables used to copy the function arguments.  */
 748
 749 static void
 750 eliminate_tail_call (struct tailcall *t)
 751 {
 752   tree param, rslt;
 753   gimple stmt, call;
 754   tree arg;
 755   size_t idx;
 756   basic_block bb, first;
 757   edge e;
 758   gimple phi;
 759   gimple_stmt_iterator gsi;
 760   gimple orig_stmt;
 761
 762   stmt = orig_stmt = gsi_stmt (t->call_gsi);
 763   bb = gsi_bb (t->call_gsi);
 764
 765   if (dump_file && (dump_flags & TDF_DETAILS))
 766     {
 767       fprintf (dump_file, "Eliminated tail recursion in bb %d : ",
 768                bb->index);
 769       print_gimple_stmt (dump_file, stmt, 0, TDF_SLIM);
 770       fprintf (dump_file, "\n");
 771     }
 772
 773   gcc_assert (is_gimple_call (stmt));
 774
 775   first = single_succ (ENTRY_BLOCK_PTR);
 776
 777   /* Remove the code after call_gsi that will become unreachable.  The
 778      possibly unreachable code in other blocks is removed later in
 779      cfg cleanup.  */
 780   gsi = t->call_gsi;
 781   gsi_next (&gsi);
 782   while (!gsi_end_p (gsi))
 783     {
 784       gimple t = gsi_stmt (gsi);
 785       /* Do not remove the return statement, so that redirect_edge_and_branch
 786          sees how the block ends.  */
 787       if (gimple_code (t) == GIMPLE_RETURN)
 788         break;
 789
 790       gsi_remove (&gsi, true);
 791       release_defs (t);
 792     }
 793
 794   /* Number of executions of function has reduced by the tailcall.  */
 795   e = single_succ_edge (gsi_bb (t->call_gsi));
 796   decrease_profile (EXIT_BLOCK_PTR, e->count, EDGE_FREQUENCY (e));
 797   decrease_profile (ENTRY_BLOCK_PTR, e->count, EDGE_FREQUENCY (e));
 798   if (e->dest != EXIT_BLOCK_PTR)
 799     decrease_profile (e->dest, e->count, EDGE_FREQUENCY (e));
 800
 801   /* Replace the call by a jump to the start of function.  */
 802   e = redirect_edge_and_branch (single_succ_edge (gsi_bb (t->call_gsi)),
 803                                 first);
 804   gcc_assert (e);
 805   PENDING_STMT (e) = NULL;
 806
 807   /* Add phi node entries for arguments.  The ordering of the phi nodes should
 808      be the same as the ordering of the arguments.  */
 809   for (param = DECL_ARGUMENTS (current_function_decl),
 810          idx = 0, gsi = gsi_start_phis (first);
 811        param;
 812        param = TREE_CHAIN (param), idx++)
 813     {
 814       if (!arg_needs_copy_p (param))
 815         continue;
 816
 817       arg = gimple_call_arg (stmt, idx);
 818       phi = gsi_stmt (gsi);
 819       gcc_assert (param == SSA_NAME_VAR (PHI_RESULT (phi)));
 820
 821       add_phi_arg (phi, arg, e, gimple_location (stmt));
 822       gsi_next (&gsi);
 823     }
 824
 825   /* Update the values of accumulators.  */
 826   adjust_accumulator_values (t->call_gsi, t->mult, t->add, e);
 827
 828   call = gsi_stmt (t->call_gsi);
 829   rslt = gimple_call_lhs (call);
 830   if (rslt != NULL_TREE)
 831     {
 832       /* Result of the call will no longer be defined.  So adjust the
 833          SSA_NAME_DEF_STMT accordingly.  */
 834       SSA_NAME_DEF_STMT (rslt) = gimple_build_nop ();
 835     }
 836
 837   gsi_remove (&t->call_gsi, true);
 838   release_defs (call);
 839 }
 840
 841 /* Add phi nodes for the virtual operands defined in the function to the
 842    header of the loop created by tail recursion elimination.
 843
 844    Originally, we used to add phi nodes only for call clobbered variables,
 845    as the value of the non-call clobbered ones obviously cannot be used
 846    or changed within the recursive call.  However, the local variables
 847    from multiple calls now share the same location, so the virtual ssa form
 848    requires us to say that the location dies on further iterations of the loop,
 849    which requires adding phi nodes.
 850 */
 851 static void
 852 add_virtual_phis (void)
 853 {
 854   referenced_var_iterator rvi;
 855   tree var;
 856
 857   /* The problematic part is that there is no way how to know what
 858      to put into phi nodes (there in fact does not have to be such
 859      ssa name available).  A solution would be to have an artificial
 860      use/kill for all virtual operands in EXIT node.  Unless we have
 861      this, we cannot do much better than to rebuild the ssa form for
 862      possibly affected virtual ssa names from scratch.  */
 863
 864   FOR_EACH_REFERENCED_VAR (var, rvi)
 865     {
 866       if (!is_gimple_reg (var) && gimple_default_def (cfun, var) != NULL_TREE)
 867         mark_sym_for_renaming (var);
 868     }
 869 }
 870
 871 /* Optimizes the tailcall described by T.  If OPT_TAILCALLS is true, also
 872    mark the tailcalls for the sibcall optimization.  */
 873
 874 static bool
 875 optimize_tail_call (struct tailcall *t, bool opt_tailcalls)
 876 {
 877   if (t->tail_recursion)
 878     {
 879       eliminate_tail_call (t);
 880       return true;
 881     }
 882
 883   if (opt_tailcalls)
 884     {
 885       gimple stmt = gsi_stmt (t->call_gsi);
 886
 887       gimple_call_set_tail (stmt, true);
 888       if (dump_file && (dump_flags & TDF_DETAILS))
 889         {
 890           fprintf (dump_file, "Found tail call ");
 891           print_gimple_stmt (dump_file, stmt, 0, dump_flags);
 892           fprintf (dump_file, " in bb %i\n", (gsi_bb (t->call_gsi))->index);
 893         }
 894     }
 895
 896   return false;
 897 }
 898
 899 /* Creates a tail-call accumulator of the same type as the return type of the
 900    current function.  LABEL is the name used to creating the temporary
 901    variable for the accumulator.  The accumulator will be inserted in the
 902    phis of a basic block BB with single predecessor with an initial value
 903    INIT converted to the current function return type.  */
 904
 905 static tree
 906 create_tailcall_accumulator (const char *label, basic_block bb, tree init)
 907 {
 908   tree ret_type = TREE_TYPE (DECL_RESULT (current_function_decl));
 909   tree tmp = create_tmp_reg (ret_type, label);
 910   gimple phi;
 911
 912   add_referenced_var (tmp);
 913   phi = create_phi_node (tmp, bb);
 914   /* RET_TYPE can be a float when -ffast-maths is enabled.  */
 915   add_phi_arg (phi, fold_convert (ret_type, init), single_pred_edge (bb),
 916                UNKNOWN_LOCATION);
 917   return PHI_RESULT (phi);
 918 }
 919
 920 /* Optimizes tail calls in the function, turning the tail recursion
 921    into iteration.  */
 922
 923 static unsigned int
 924 tree_optimize_tail_calls_1 (bool opt_tailcalls)
 925 {
 926   edge e;
 927   bool phis_constructed = false;
 928   struct tailcall *tailcalls = NULL, *act, *next;
 929   bool changed = false;
 930   basic_block first = single_succ (ENTRY_BLOCK_PTR);
 931   tree param;
 932   gimple stmt;
 933   edge_iterator ei;
 934
 935   if (!suitable_for_tail_opt_p ())
 936     return 0;
 937   if (opt_tailcalls)
 938     opt_tailcalls = suitable_for_tail_call_opt_p ();
 939
 940   FOR_EACH_EDGE (e, ei, EXIT_BLOCK_PTR->preds)
 941     {
 942       /* Only traverse the normal exits, i.e. those that end with return
 943          statement.  */
 944       stmt = last_stmt (e->src);
 945
 946       if (stmt
 947           && gimple_code (stmt) == GIMPLE_RETURN)
 948         find_tail_calls (e->src, &tailcalls);
 949     }
 950
 951   /* Construct the phi nodes and accumulators if necessary.  */
 952   a_acc = m_acc = NULL_TREE;
 953   for (act = tailcalls; act; act = act->next)
 954     {
 955       if (!act->tail_recursion)
 956         continue;
 957
 958       if (!phis_constructed)
 959         {
 960           /* Ensure that there is only one predecessor of the block
 961              or if there are existing degenerate PHI nodes.  */
 962           if (!single_pred_p (first)
 963               || !gimple_seq_empty_p (phi_nodes (first)))
 964             first = split_edge (single_succ_edge (ENTRY_BLOCK_PTR));
 965
 966           /* Copy the args if needed.  */
 967           for (param = DECL_ARGUMENTS (current_function_decl);
 968                param;
 969                param = TREE_CHAIN (param))
 970             if (arg_needs_copy_p (param))
 971               {
 972                 tree name = gimple_default_def (cfun, param);
 973                 tree new_name = make_ssa_name (param, SSA_NAME_DEF_STMT (name));
 974                 gimple phi;
 975
 976                 set_default_def (param, new_name);
 977                 phi = create_phi_node (name, first);
 978                 SSA_NAME_DEF_STMT (name) = phi;
 979                 add_phi_arg (phi, new_name, single_pred_edge (first),
 980                              EXPR_LOCATION (param));
 981               }
 982           phis_constructed = true;
 983         }
 984
 985       if (act->add && !a_acc)
 986         a_acc = create_tailcall_accumulator ("add_acc", first,
 987                                              integer_zero_node);
 988
 989       if (act->mult && !m_acc)
 990         m_acc = create_tailcall_accumulator ("mult_acc", first,
 991                                              integer_one_node);
 992     }
 993
 994   for (; tailcalls; tailcalls = next)
 995     {
 996       next = tailcalls->next;
 997       changed |= optimize_tail_call (tailcalls, opt_tailcalls);
 998       free (tailcalls);
 999     }
1000
1001   if (a_acc || m_acc)
1002     {
1003       /* Modify the remaining return statements.  */
1004       FOR_EACH_EDGE (e, ei, EXIT_BLOCK_PTR->preds)
1005         {
1006           stmt = last_stmt (e->src);
1007
1008           if (stmt
1009               && gimple_code (stmt) == GIMPLE_RETURN)
1010             adjust_return_value (e->src, m_acc, a_acc);
1011         }
1012     }
1013
1014   if (changed)
1015     free_dominance_info (CDI_DOMINATORS);
1016
1017   if (phis_constructed)
1018     add_virtual_phis ();
1019   if (changed)
1020     return TODO_cleanup_cfg | TODO_update_ssa_only_virtuals;
1021   return 0;
1022 }
1023
1024 static unsigned int
1025 execute_tail_recursion (void)
1026 {
1027   return tree_optimize_tail_calls_1 (false);
1028 }
1029
1030 static bool
1031 gate_tail_calls (void)
1032 {
1033   return flag_optimize_sibling_calls != 0 && dbg_cnt (tail_call);
1034 }
1035
1036 static unsigned int
1037 execute_tail_calls (void)
1038 {
1039   return tree_optimize_tail_calls_1 (true);
1040 }
1041
1042 struct gimple_opt_pass pass_tail_recursion =
1043 {
1044  {
1045   GIMPLE_PASS,
1046   "tailr",                              /* name */
1047   gate_tail_calls,                      /* gate */
1048   execute_tail_recursion,               /* execute */
1049   NULL,                                 /* sub */
1050   NULL,                                 /* next */
1051   0,                                    /* static_pass_number */
1052   TV_NONE,                              /* tv_id */
1053   PROP_cfg | PROP_ssa,                  /* properties_required */
1054   0,                                    /* properties_provided */
1055   0,                                    /* properties_destroyed */
1056   0,                                    /* todo_flags_start */
1057   TODO_dump_func | TODO_verify_ssa      /* todo_flags_finish */
1058  }
1059 };
1060
1061 struct gimple_opt_pass pass_tail_calls =
1062 {
1063  {
1064   GIMPLE_PASS,
1065   "tailc",                              /* name */
1066   gate_tail_calls,                      /* gate */
1067   execute_tail_calls,                   /* execute */
1068   NULL,                                 /* sub */
1069   NULL,                                 /* next */
1070   0,                                    /* static_pass_number */
1071   TV_NONE,                              /* tv_id */
1072   PROP_cfg | PROP_ssa,                  /* properties_required */
1073   0,                                    /* properties_provided */
1074   0,                                    /* properties_destroyed */
1075   0,                                    /* todo_flags_start */
1076   TODO_dump_func | TODO_verify_ssa      /* todo_flags_finish */
1077  }
1078 };