re PR target/79027 (fold-const.c:11104:1: internal compiler error: Floating point...
[official-gcc.git] / gcc / tree-tailcall.c
blobf586edcd730ffb702b5f7692f2939d25708cb2b6
1 /* Tail call optimization on trees.
2 Copyright (C) 2003-2017 Free Software Foundation, Inc.
4 This file is part of GCC.
6 GCC is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 3, or (at your option)
9 any later version.
11 GCC is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with GCC; see the file COPYING3. If not see
18 <http://www.gnu.org/licenses/>. */
20 #include "config.h"
21 #include "system.h"
22 #include "coretypes.h"
23 #include "backend.h"
24 #include "rtl.h"
25 #include "tree.h"
26 #include "gimple.h"
27 #include "cfghooks.h"
28 #include "tree-pass.h"
29 #include "ssa.h"
30 #include "cgraph.h"
31 #include "gimple-pretty-print.h"
32 #include "fold-const.h"
33 #include "stor-layout.h"
34 #include "gimple-iterator.h"
35 #include "gimplify-me.h"
36 #include "tree-cfg.h"
37 #include "tree-into-ssa.h"
38 #include "tree-dfa.h"
39 #include "except.h"
40 #include "dbgcnt.h"
41 #include "cfgloop.h"
42 #include "common/common-target.h"
43 #include "ipa-utils.h"
45 /* The file implements the tail recursion elimination. It is also used to
46 analyze the tail calls in general, passing the results to the rtl level
47 where they are used for sibcall optimization.
49 In addition to the standard tail recursion elimination, we handle the most
50 trivial cases of making the call tail recursive by creating accumulators.
51 For example the following function
53 int sum (int n)
55 if (n > 0)
56 return n + sum (n - 1);
57 else
58 return 0;
61 is transformed into
63 int sum (int n)
65 int acc = 0;
67 while (n > 0)
68 acc += n--;
70 return acc;
73 To do this, we maintain two accumulators (a_acc and m_acc) that indicate
74 when we reach the return x statement, we should return a_acc + x * m_acc
75 instead. They are initially initialized to 0 and 1, respectively,
76 so the semantics of the function is obviously preserved. If we are
77 guaranteed that the value of the accumulator never change, we
78 omit the accumulator.
80 There are three cases how the function may exit. The first one is
81 handled in adjust_return_value, the other two in adjust_accumulator_values
82 (the second case is actually a special case of the third one and we
83 present it separately just for clarity):
85 1) Just return x, where x is not in any of the remaining special shapes.
86 We rewrite this to a gimple equivalent of return m_acc * x + a_acc.
88 2) return f (...), where f is the current function, is rewritten in a
89 classical tail-recursion elimination way, into assignment of arguments
90 and jump to the start of the function. Values of the accumulators
91 are unchanged.
93 3) return a + m * f(...), where a and m do not depend on call to f.
94 To preserve the semantics described before we want this to be rewritten
95 in such a way that we finally return
97 a_acc + (a + m * f(...)) * m_acc = (a_acc + a * m_acc) + (m * m_acc) * f(...).
99 I.e. we increase a_acc by a * m_acc, multiply m_acc by m and
100 eliminate the tail call to f. Special cases when the value is just
101 added or just multiplied are obtained by setting a = 0 or m = 1.
103 TODO -- it is possible to do similar tricks for other operations. */
105 /* A structure that describes the tailcall. */
107 struct tailcall
109 /* The iterator pointing to the call statement. */
110 gimple_stmt_iterator call_gsi;
112 /* True if it is a call to the current function. */
113 bool tail_recursion;
115 /* The return value of the caller is mult * f + add, where f is the return
116 value of the call. */
117 tree mult, add;
119 /* Next tailcall in the chain. */
120 struct tailcall *next;
123 /* The variables holding the value of multiplicative and additive
124 accumulator. */
125 static tree m_acc, a_acc;
127 static bool optimize_tail_call (struct tailcall *, bool);
128 static void eliminate_tail_call (struct tailcall *);
130 /* Returns false when the function is not suitable for tail call optimization
131 from some reason (e.g. if it takes variable number of arguments). */
133 static bool
134 suitable_for_tail_opt_p (void)
136 if (cfun->stdarg)
137 return false;
139 return true;
141 /* Returns false when the function is not suitable for tail call optimization
142 for some reason (e.g. if it takes variable number of arguments).
143 This test must pass in addition to suitable_for_tail_opt_p in order to make
144 tail call discovery happen. */
146 static bool
147 suitable_for_tail_call_opt_p (void)
149 tree param;
151 /* alloca (until we have stack slot life analysis) inhibits
152 sibling call optimizations, but not tail recursion. */
153 if (cfun->calls_alloca)
154 return false;
156 /* If we are using sjlj exceptions, we may need to add a call to
157 _Unwind_SjLj_Unregister at exit of the function. Which means
158 that we cannot do any sibcall transformations. */
159 if (targetm_common.except_unwind_info (&global_options) == UI_SJLJ
160 && current_function_has_exception_handlers ())
161 return false;
163 /* Any function that calls setjmp might have longjmp called from
164 any called function. ??? We really should represent this
165 properly in the CFG so that this needn't be special cased. */
166 if (cfun->calls_setjmp)
167 return false;
169 /* ??? It is OK if the argument of a function is taken in some cases,
170 but not in all cases. See PR15387 and PR19616. Revisit for 4.1. */
171 for (param = DECL_ARGUMENTS (current_function_decl);
172 param;
173 param = DECL_CHAIN (param))
174 if (TREE_ADDRESSABLE (param))
175 return false;
177 return true;
180 /* Checks whether the expression EXPR in stmt AT is independent of the
181 statement pointed to by GSI (in a sense that we already know EXPR's value
182 at GSI). We use the fact that we are only called from the chain of
183 basic blocks that have only single successor. Returns the expression
184 containing the value of EXPR at GSI. */
186 static tree
187 independent_of_stmt_p (tree expr, gimple *at, gimple_stmt_iterator gsi)
189 basic_block bb, call_bb, at_bb;
190 edge e;
191 edge_iterator ei;
193 if (is_gimple_min_invariant (expr))
194 return expr;
196 if (TREE_CODE (expr) != SSA_NAME)
197 return NULL_TREE;
199 /* Mark the blocks in the chain leading to the end. */
200 at_bb = gimple_bb (at);
201 call_bb = gimple_bb (gsi_stmt (gsi));
202 for (bb = call_bb; bb != at_bb; bb = single_succ (bb))
203 bb->aux = &bb->aux;
204 bb->aux = &bb->aux;
206 while (1)
208 at = SSA_NAME_DEF_STMT (expr);
209 bb = gimple_bb (at);
211 /* The default definition or defined before the chain. */
212 if (!bb || !bb->aux)
213 break;
215 if (bb == call_bb)
217 for (; !gsi_end_p (gsi); gsi_next (&gsi))
218 if (gsi_stmt (gsi) == at)
219 break;
221 if (!gsi_end_p (gsi))
222 expr = NULL_TREE;
223 break;
226 if (gimple_code (at) != GIMPLE_PHI)
228 expr = NULL_TREE;
229 break;
232 FOR_EACH_EDGE (e, ei, bb->preds)
233 if (e->src->aux)
234 break;
235 gcc_assert (e);
237 expr = PHI_ARG_DEF_FROM_EDGE (at, e);
238 if (TREE_CODE (expr) != SSA_NAME)
240 /* The value is a constant. */
241 break;
245 /* Unmark the blocks. */
246 for (bb = call_bb; bb != at_bb; bb = single_succ (bb))
247 bb->aux = NULL;
248 bb->aux = NULL;
250 return expr;
253 /* Simulates the effect of an assignment STMT on the return value of the tail
254 recursive CALL passed in ASS_VAR. M and A are the multiplicative and the
255 additive factor for the real return value. */
257 static bool
258 process_assignment (gassign *stmt, gimple_stmt_iterator call, tree *m,
259 tree *a, tree *ass_var)
261 tree op0, op1 = NULL_TREE, non_ass_var = NULL_TREE;
262 tree dest = gimple_assign_lhs (stmt);
263 enum tree_code code = gimple_assign_rhs_code (stmt);
264 enum gimple_rhs_class rhs_class = get_gimple_rhs_class (code);
265 tree src_var = gimple_assign_rhs1 (stmt);
267 /* See if this is a simple copy operation of an SSA name to the function
268 result. In that case we may have a simple tail call. Ignore type
269 conversions that can never produce extra code between the function
270 call and the function return. */
271 if ((rhs_class == GIMPLE_SINGLE_RHS || gimple_assign_cast_p (stmt))
272 && src_var == *ass_var)
274 /* Reject a tailcall if the type conversion might need
275 additional code. */
276 if (gimple_assign_cast_p (stmt))
278 if (TYPE_MODE (TREE_TYPE (dest)) != TYPE_MODE (TREE_TYPE (src_var)))
279 return false;
281 /* Even if the type modes are the same, if the precision of the
282 type is smaller than mode's precision,
283 reduce_to_bit_field_precision would generate additional code. */
284 if (INTEGRAL_TYPE_P (TREE_TYPE (dest))
285 && (GET_MODE_PRECISION (TYPE_MODE (TREE_TYPE (dest)))
286 > TYPE_PRECISION (TREE_TYPE (dest))))
287 return false;
290 *ass_var = dest;
291 return true;
294 switch (rhs_class)
296 case GIMPLE_BINARY_RHS:
297 op1 = gimple_assign_rhs2 (stmt);
299 /* Fall through. */
301 case GIMPLE_UNARY_RHS:
302 op0 = gimple_assign_rhs1 (stmt);
303 break;
305 default:
306 return false;
309 /* Accumulator optimizations will reverse the order of operations.
310 We can only do that for floating-point types if we're assuming
311 that addition and multiplication are associative. */
312 if (!flag_associative_math)
313 if (FLOAT_TYPE_P (TREE_TYPE (DECL_RESULT (current_function_decl))))
314 return false;
316 if (rhs_class == GIMPLE_UNARY_RHS)
318 else if (op0 == *ass_var
319 && (non_ass_var = independent_of_stmt_p (op1, stmt, call)))
321 else if (op1 == *ass_var
322 && (non_ass_var = independent_of_stmt_p (op0, stmt, call)))
324 else
325 return false;
327 switch (code)
329 case PLUS_EXPR:
330 *a = non_ass_var;
331 *ass_var = dest;
332 return true;
334 case POINTER_PLUS_EXPR:
335 if (op0 != *ass_var)
336 return false;
337 *a = non_ass_var;
338 *ass_var = dest;
339 return true;
341 case MULT_EXPR:
342 *m = non_ass_var;
343 *ass_var = dest;
344 return true;
346 case NEGATE_EXPR:
347 *m = build_minus_one_cst (TREE_TYPE (op0));
348 *ass_var = dest;
349 return true;
351 case MINUS_EXPR:
352 if (*ass_var == op0)
353 *a = fold_build1 (NEGATE_EXPR, TREE_TYPE (non_ass_var), non_ass_var);
354 else
356 *m = build_minus_one_cst (TREE_TYPE (non_ass_var));
357 *a = fold_build1 (NEGATE_EXPR, TREE_TYPE (non_ass_var), non_ass_var);
360 *ass_var = dest;
361 return true;
363 /* TODO -- Handle POINTER_PLUS_EXPR. */
365 default:
366 return false;
370 /* Propagate VAR through phis on edge E. */
372 static tree
373 propagate_through_phis (tree var, edge e)
375 basic_block dest = e->dest;
376 gphi_iterator gsi;
378 for (gsi = gsi_start_phis (dest); !gsi_end_p (gsi); gsi_next (&gsi))
380 gphi *phi = gsi.phi ();
381 if (PHI_ARG_DEF_FROM_EDGE (phi, e) == var)
382 return PHI_RESULT (phi);
384 return var;
387 /* Finds tailcalls falling into basic block BB. The list of found tailcalls is
388 added to the start of RET. */
390 static void
391 find_tail_calls (basic_block bb, struct tailcall **ret)
393 tree ass_var = NULL_TREE, ret_var, func, param;
394 gimple *stmt;
395 gcall *call = NULL;
396 gimple_stmt_iterator gsi, agsi;
397 bool tail_recursion;
398 struct tailcall *nw;
399 edge e;
400 tree m, a;
401 basic_block abb;
402 size_t idx;
403 tree var;
405 if (!single_succ_p (bb))
406 return;
408 for (gsi = gsi_last_bb (bb); !gsi_end_p (gsi); gsi_prev (&gsi))
410 stmt = gsi_stmt (gsi);
412 /* Ignore labels, returns, nops, clobbers and debug stmts. */
413 if (gimple_code (stmt) == GIMPLE_LABEL
414 || gimple_code (stmt) == GIMPLE_RETURN
415 || gimple_code (stmt) == GIMPLE_NOP
416 || gimple_clobber_p (stmt)
417 || is_gimple_debug (stmt))
418 continue;
420 /* Check for a call. */
421 if (is_gimple_call (stmt))
423 call = as_a <gcall *> (stmt);
424 ass_var = gimple_call_lhs (call);
425 break;
428 /* Allow simple copies between local variables, even if they're
429 aggregates. */
430 if (is_gimple_assign (stmt)
431 && auto_var_in_fn_p (gimple_assign_lhs (stmt), cfun->decl)
432 && auto_var_in_fn_p (gimple_assign_rhs1 (stmt), cfun->decl))
433 continue;
435 /* If the statement references memory or volatile operands, fail. */
436 if (gimple_references_memory_p (stmt)
437 || gimple_has_volatile_ops (stmt))
438 return;
441 if (gsi_end_p (gsi))
443 edge_iterator ei;
444 /* Recurse to the predecessors. */
445 FOR_EACH_EDGE (e, ei, bb->preds)
446 find_tail_calls (e->src, ret);
448 return;
451 /* If the LHS of our call is not just a simple register or local
452 variable, we can't transform this into a tail or sibling call.
453 This situation happens, in (e.g.) "*p = foo()" where foo returns a
454 struct. In this case we won't have a temporary here, but we need
455 to carry out the side effect anyway, so tailcall is impossible.
457 ??? In some situations (when the struct is returned in memory via
458 invisible argument) we could deal with this, e.g. by passing 'p'
459 itself as that argument to foo, but it's too early to do this here,
460 and expand_call() will not handle it anyway. If it ever can, then
461 we need to revisit this here, to allow that situation. */
462 if (ass_var
463 && !is_gimple_reg (ass_var)
464 && !auto_var_in_fn_p (ass_var, cfun->decl))
465 return;
467 /* We found the call, check whether it is suitable. */
468 tail_recursion = false;
469 func = gimple_call_fndecl (call);
470 if (func
471 && !DECL_BUILT_IN (func)
472 && recursive_call_p (current_function_decl, func))
474 tree arg;
476 for (param = DECL_ARGUMENTS (func), idx = 0;
477 param && idx < gimple_call_num_args (call);
478 param = DECL_CHAIN (param), idx ++)
480 arg = gimple_call_arg (call, idx);
481 if (param != arg)
483 /* Make sure there are no problems with copying. The parameter
484 have a copyable type and the two arguments must have reasonably
485 equivalent types. The latter requirement could be relaxed if
486 we emitted a suitable type conversion statement. */
487 if (!is_gimple_reg_type (TREE_TYPE (param))
488 || !useless_type_conversion_p (TREE_TYPE (param),
489 TREE_TYPE (arg)))
490 break;
492 /* The parameter should be a real operand, so that phi node
493 created for it at the start of the function has the meaning
494 of copying the value. This test implies is_gimple_reg_type
495 from the previous condition, however this one could be
496 relaxed by being more careful with copying the new value
497 of the parameter (emitting appropriate GIMPLE_ASSIGN and
498 updating the virtual operands). */
499 if (!is_gimple_reg (param))
500 break;
503 if (idx == gimple_call_num_args (call) && !param)
504 tail_recursion = true;
507 /* Make sure the tail invocation of this function does not indirectly
508 refer to local variables. (Passing variables directly by value
509 is OK.) */
510 FOR_EACH_LOCAL_DECL (cfun, idx, var)
512 if (TREE_CODE (var) != PARM_DECL
513 && auto_var_in_fn_p (var, cfun->decl)
514 && may_be_aliased (var)
515 && (ref_maybe_used_by_stmt_p (call, var)
516 || call_may_clobber_ref_p (call, var)))
517 return;
520 /* Now check the statements after the call. None of them has virtual
521 operands, so they may only depend on the call through its return
522 value. The return value should also be dependent on each of them,
523 since we are running after dce. */
524 m = NULL_TREE;
525 a = NULL_TREE;
527 abb = bb;
528 agsi = gsi;
529 while (1)
531 tree tmp_a = NULL_TREE;
532 tree tmp_m = NULL_TREE;
533 gsi_next (&agsi);
535 while (gsi_end_p (agsi))
537 ass_var = propagate_through_phis (ass_var, single_succ_edge (abb));
538 abb = single_succ (abb);
539 agsi = gsi_start_bb (abb);
542 stmt = gsi_stmt (agsi);
544 if (gimple_code (stmt) == GIMPLE_LABEL
545 || gimple_code (stmt) == GIMPLE_NOP)
546 continue;
548 if (gimple_code (stmt) == GIMPLE_RETURN)
549 break;
551 if (gimple_clobber_p (stmt))
552 continue;
554 if (is_gimple_debug (stmt))
555 continue;
557 if (gimple_code (stmt) != GIMPLE_ASSIGN)
558 return;
560 /* This is a gimple assign. */
561 if (! process_assignment (as_a <gassign *> (stmt), gsi, &tmp_m,
562 &tmp_a, &ass_var))
563 return;
565 if (tmp_a)
567 tree type = TREE_TYPE (tmp_a);
568 if (a)
569 a = fold_build2 (PLUS_EXPR, type, fold_convert (type, a), tmp_a);
570 else
571 a = tmp_a;
573 if (tmp_m)
575 tree type = TREE_TYPE (tmp_m);
576 if (m)
577 m = fold_build2 (MULT_EXPR, type, fold_convert (type, m), tmp_m);
578 else
579 m = tmp_m;
581 if (a)
582 a = fold_build2 (MULT_EXPR, type, fold_convert (type, a), tmp_m);
586 /* See if this is a tail call we can handle. */
587 ret_var = gimple_return_retval (as_a <greturn *> (stmt));
589 /* We may proceed if there either is no return value, or the return value
590 is identical to the call's return. */
591 if (ret_var
592 && (ret_var != ass_var))
593 return;
595 /* If this is not a tail recursive call, we cannot handle addends or
596 multiplicands. */
597 if (!tail_recursion && (m || a))
598 return;
600 /* For pointers only allow additions. */
601 if (m && POINTER_TYPE_P (TREE_TYPE (DECL_RESULT (current_function_decl))))
602 return;
604 nw = XNEW (struct tailcall);
606 nw->call_gsi = gsi;
608 nw->tail_recursion = tail_recursion;
610 nw->mult = m;
611 nw->add = a;
613 nw->next = *ret;
614 *ret = nw;
617 /* Helper to insert PHI_ARGH to the phi of VAR in the destination of edge E. */
619 static void
620 add_successor_phi_arg (edge e, tree var, tree phi_arg)
622 gphi_iterator gsi;
624 for (gsi = gsi_start_phis (e->dest); !gsi_end_p (gsi); gsi_next (&gsi))
625 if (PHI_RESULT (gsi.phi ()) == var)
626 break;
628 gcc_assert (!gsi_end_p (gsi));
629 add_phi_arg (gsi.phi (), phi_arg, e, UNKNOWN_LOCATION);
632 /* Creates a GIMPLE statement which computes the operation specified by
633 CODE, ACC and OP1 to a new variable with name LABEL and inserts the
634 statement in the position specified by GSI. Returns the
635 tree node of the statement's result. */
637 static tree
638 adjust_return_value_with_ops (enum tree_code code, const char *label,
639 tree acc, tree op1, gimple_stmt_iterator gsi)
642 tree ret_type = TREE_TYPE (DECL_RESULT (current_function_decl));
643 tree result = make_temp_ssa_name (ret_type, NULL, label);
644 gassign *stmt;
646 if (POINTER_TYPE_P (ret_type))
648 gcc_assert (code == PLUS_EXPR && TREE_TYPE (acc) == sizetype);
649 code = POINTER_PLUS_EXPR;
651 if (types_compatible_p (TREE_TYPE (acc), TREE_TYPE (op1))
652 && code != POINTER_PLUS_EXPR)
653 stmt = gimple_build_assign (result, code, acc, op1);
654 else
656 tree tem;
657 if (code == POINTER_PLUS_EXPR)
658 tem = fold_build2 (code, TREE_TYPE (op1), op1, acc);
659 else
660 tem = fold_build2 (code, TREE_TYPE (op1),
661 fold_convert (TREE_TYPE (op1), acc), op1);
662 tree rhs = fold_convert (ret_type, tem);
663 rhs = force_gimple_operand_gsi (&gsi, rhs,
664 false, NULL, true, GSI_SAME_STMT);
665 stmt = gimple_build_assign (result, rhs);
668 gsi_insert_before (&gsi, stmt, GSI_NEW_STMT);
669 return result;
672 /* Creates a new GIMPLE statement that adjusts the value of accumulator ACC by
673 the computation specified by CODE and OP1 and insert the statement
674 at the position specified by GSI as a new statement. Returns new SSA name
675 of updated accumulator. */
677 static tree
678 update_accumulator_with_ops (enum tree_code code, tree acc, tree op1,
679 gimple_stmt_iterator gsi)
681 gassign *stmt;
682 tree var = copy_ssa_name (acc);
683 if (types_compatible_p (TREE_TYPE (acc), TREE_TYPE (op1)))
684 stmt = gimple_build_assign (var, code, acc, op1);
685 else
687 tree rhs = fold_convert (TREE_TYPE (acc),
688 fold_build2 (code,
689 TREE_TYPE (op1),
690 fold_convert (TREE_TYPE (op1), acc),
691 op1));
692 rhs = force_gimple_operand_gsi (&gsi, rhs,
693 false, NULL, false, GSI_CONTINUE_LINKING);
694 stmt = gimple_build_assign (var, rhs);
696 gsi_insert_after (&gsi, stmt, GSI_NEW_STMT);
697 return var;
700 /* Adjust the accumulator values according to A and M after GSI, and update
701 the phi nodes on edge BACK. */
703 static void
704 adjust_accumulator_values (gimple_stmt_iterator gsi, tree m, tree a, edge back)
706 tree var, a_acc_arg, m_acc_arg;
708 if (m)
709 m = force_gimple_operand_gsi (&gsi, m, true, NULL, true, GSI_SAME_STMT);
710 if (a)
711 a = force_gimple_operand_gsi (&gsi, a, true, NULL, true, GSI_SAME_STMT);
713 a_acc_arg = a_acc;
714 m_acc_arg = m_acc;
715 if (a)
717 if (m_acc)
719 if (integer_onep (a))
720 var = m_acc;
721 else
722 var = adjust_return_value_with_ops (MULT_EXPR, "acc_tmp", m_acc,
723 a, gsi);
725 else
726 var = a;
728 a_acc_arg = update_accumulator_with_ops (PLUS_EXPR, a_acc, var, gsi);
731 if (m)
732 m_acc_arg = update_accumulator_with_ops (MULT_EXPR, m_acc, m, gsi);
734 if (a_acc)
735 add_successor_phi_arg (back, a_acc, a_acc_arg);
737 if (m_acc)
738 add_successor_phi_arg (back, m_acc, m_acc_arg);
741 /* Adjust value of the return at the end of BB according to M and A
742 accumulators. */
744 static void
745 adjust_return_value (basic_block bb, tree m, tree a)
747 tree retval;
748 greturn *ret_stmt = as_a <greturn *> (gimple_seq_last_stmt (bb_seq (bb)));
749 gimple_stmt_iterator gsi = gsi_last_bb (bb);
751 gcc_assert (gimple_code (ret_stmt) == GIMPLE_RETURN);
753 retval = gimple_return_retval (ret_stmt);
754 if (!retval || retval == error_mark_node)
755 return;
757 if (m)
758 retval = adjust_return_value_with_ops (MULT_EXPR, "mul_tmp", m_acc, retval,
759 gsi);
760 if (a)
761 retval = adjust_return_value_with_ops (PLUS_EXPR, "acc_tmp", a_acc, retval,
762 gsi);
763 gimple_return_set_retval (ret_stmt, retval);
764 update_stmt (ret_stmt);
767 /* Subtract COUNT and FREQUENCY from the basic block and it's
768 outgoing edge. */
769 static void
770 decrease_profile (basic_block bb, gcov_type count, int frequency)
772 edge e;
773 bb->count -= count;
774 if (bb->count < 0)
775 bb->count = 0;
776 bb->frequency -= frequency;
777 if (bb->frequency < 0)
778 bb->frequency = 0;
779 if (!single_succ_p (bb))
781 gcc_assert (!EDGE_COUNT (bb->succs));
782 return;
784 e = single_succ_edge (bb);
785 e->count -= count;
786 if (e->count < 0)
787 e->count = 0;
790 /* Returns true if argument PARAM of the tail recursive call needs to be copied
791 when the call is eliminated. */
793 static bool
794 arg_needs_copy_p (tree param)
796 tree def;
798 if (!is_gimple_reg (param))
799 return false;
801 /* Parameters that are only defined but never used need not be copied. */
802 def = ssa_default_def (cfun, param);
803 if (!def)
804 return false;
806 return true;
809 /* Eliminates tail call described by T. TMP_VARS is a list of
810 temporary variables used to copy the function arguments. */
812 static void
813 eliminate_tail_call (struct tailcall *t)
815 tree param, rslt;
816 gimple *stmt, *call;
817 tree arg;
818 size_t idx;
819 basic_block bb, first;
820 edge e;
821 gphi *phi;
822 gphi_iterator gpi;
823 gimple_stmt_iterator gsi;
824 gimple *orig_stmt;
826 stmt = orig_stmt = gsi_stmt (t->call_gsi);
827 bb = gsi_bb (t->call_gsi);
829 if (dump_file && (dump_flags & TDF_DETAILS))
831 fprintf (dump_file, "Eliminated tail recursion in bb %d : ",
832 bb->index);
833 print_gimple_stmt (dump_file, stmt, 0, TDF_SLIM);
834 fprintf (dump_file, "\n");
837 gcc_assert (is_gimple_call (stmt));
839 first = single_succ (ENTRY_BLOCK_PTR_FOR_FN (cfun));
841 /* Remove the code after call_gsi that will become unreachable. The
842 possibly unreachable code in other blocks is removed later in
843 cfg cleanup. */
844 gsi = t->call_gsi;
845 gimple_stmt_iterator gsi2 = gsi_last_bb (gimple_bb (gsi_stmt (gsi)));
846 while (gsi_stmt (gsi2) != gsi_stmt (gsi))
848 gimple *t = gsi_stmt (gsi2);
849 /* Do not remove the return statement, so that redirect_edge_and_branch
850 sees how the block ends. */
851 if (gimple_code (t) != GIMPLE_RETURN)
853 gimple_stmt_iterator gsi3 = gsi2;
854 gsi_prev (&gsi2);
855 gsi_remove (&gsi3, true);
856 release_defs (t);
858 else
859 gsi_prev (&gsi2);
862 /* Number of executions of function has reduced by the tailcall. */
863 e = single_succ_edge (gsi_bb (t->call_gsi));
864 decrease_profile (EXIT_BLOCK_PTR_FOR_FN (cfun), e->count, EDGE_FREQUENCY (e));
865 decrease_profile (ENTRY_BLOCK_PTR_FOR_FN (cfun), e->count,
866 EDGE_FREQUENCY (e));
867 if (e->dest != EXIT_BLOCK_PTR_FOR_FN (cfun))
868 decrease_profile (e->dest, e->count, EDGE_FREQUENCY (e));
870 /* Replace the call by a jump to the start of function. */
871 e = redirect_edge_and_branch (single_succ_edge (gsi_bb (t->call_gsi)),
872 first);
873 gcc_assert (e);
874 PENDING_STMT (e) = NULL;
876 /* Add phi node entries for arguments. The ordering of the phi nodes should
877 be the same as the ordering of the arguments. */
878 for (param = DECL_ARGUMENTS (current_function_decl),
879 idx = 0, gpi = gsi_start_phis (first);
880 param;
881 param = DECL_CHAIN (param), idx++)
883 if (!arg_needs_copy_p (param))
884 continue;
886 arg = gimple_call_arg (stmt, idx);
887 phi = gpi.phi ();
888 gcc_assert (param == SSA_NAME_VAR (PHI_RESULT (phi)));
890 add_phi_arg (phi, arg, e, gimple_location (stmt));
891 gsi_next (&gpi);
894 /* Update the values of accumulators. */
895 adjust_accumulator_values (t->call_gsi, t->mult, t->add, e);
897 call = gsi_stmt (t->call_gsi);
898 rslt = gimple_call_lhs (call);
899 if (rslt != NULL_TREE && TREE_CODE (rslt) == SSA_NAME)
901 /* Result of the call will no longer be defined. So adjust the
902 SSA_NAME_DEF_STMT accordingly. */
903 SSA_NAME_DEF_STMT (rslt) = gimple_build_nop ();
906 gsi_remove (&t->call_gsi, true);
907 release_defs (call);
910 /* Optimizes the tailcall described by T. If OPT_TAILCALLS is true, also
911 mark the tailcalls for the sibcall optimization. */
913 static bool
914 optimize_tail_call (struct tailcall *t, bool opt_tailcalls)
916 if (t->tail_recursion)
918 eliminate_tail_call (t);
919 return true;
922 if (opt_tailcalls)
924 gcall *stmt = as_a <gcall *> (gsi_stmt (t->call_gsi));
926 gimple_call_set_tail (stmt, true);
927 cfun->tail_call_marked = true;
928 if (dump_file && (dump_flags & TDF_DETAILS))
930 fprintf (dump_file, "Found tail call ");
931 print_gimple_stmt (dump_file, stmt, 0, dump_flags);
932 fprintf (dump_file, " in bb %i\n", (gsi_bb (t->call_gsi))->index);
936 return false;
939 /* Creates a tail-call accumulator of the same type as the return type of the
940 current function. LABEL is the name used to creating the temporary
941 variable for the accumulator. The accumulator will be inserted in the
942 phis of a basic block BB with single predecessor with an initial value
943 INIT converted to the current function return type. */
945 static tree
946 create_tailcall_accumulator (const char *label, basic_block bb, tree init)
948 tree ret_type = TREE_TYPE (DECL_RESULT (current_function_decl));
949 if (POINTER_TYPE_P (ret_type))
950 ret_type = sizetype;
952 tree tmp = make_temp_ssa_name (ret_type, NULL, label);
953 gphi *phi;
955 phi = create_phi_node (tmp, bb);
956 /* RET_TYPE can be a float when -ffast-maths is enabled. */
957 add_phi_arg (phi, fold_convert (ret_type, init), single_pred_edge (bb),
958 UNKNOWN_LOCATION);
959 return PHI_RESULT (phi);
962 /* Optimizes tail calls in the function, turning the tail recursion
963 into iteration. */
965 static unsigned int
966 tree_optimize_tail_calls_1 (bool opt_tailcalls)
968 edge e;
969 bool phis_constructed = false;
970 struct tailcall *tailcalls = NULL, *act, *next;
971 bool changed = false;
972 basic_block first = single_succ (ENTRY_BLOCK_PTR_FOR_FN (cfun));
973 tree param;
974 gimple *stmt;
975 edge_iterator ei;
977 if (!suitable_for_tail_opt_p ())
978 return 0;
979 if (opt_tailcalls)
980 opt_tailcalls = suitable_for_tail_call_opt_p ();
982 FOR_EACH_EDGE (e, ei, EXIT_BLOCK_PTR_FOR_FN (cfun)->preds)
984 /* Only traverse the normal exits, i.e. those that end with return
985 statement. */
986 stmt = last_stmt (e->src);
988 if (stmt
989 && gimple_code (stmt) == GIMPLE_RETURN)
990 find_tail_calls (e->src, &tailcalls);
993 /* Construct the phi nodes and accumulators if necessary. */
994 a_acc = m_acc = NULL_TREE;
995 for (act = tailcalls; act; act = act->next)
997 if (!act->tail_recursion)
998 continue;
1000 if (!phis_constructed)
1002 /* Ensure that there is only one predecessor of the block
1003 or if there are existing degenerate PHI nodes. */
1004 if (!single_pred_p (first)
1005 || !gimple_seq_empty_p (phi_nodes (first)))
1006 first =
1007 split_edge (single_succ_edge (ENTRY_BLOCK_PTR_FOR_FN (cfun)));
1009 /* Copy the args if needed. */
1010 for (param = DECL_ARGUMENTS (current_function_decl);
1011 param;
1012 param = DECL_CHAIN (param))
1013 if (arg_needs_copy_p (param))
1015 tree name = ssa_default_def (cfun, param);
1016 tree new_name = make_ssa_name (param, SSA_NAME_DEF_STMT (name));
1017 gphi *phi;
1019 set_ssa_default_def (cfun, param, new_name);
1020 phi = create_phi_node (name, first);
1021 add_phi_arg (phi, new_name, single_pred_edge (first),
1022 EXPR_LOCATION (param));
1024 phis_constructed = true;
1027 if (act->add && !a_acc)
1028 a_acc = create_tailcall_accumulator ("add_acc", first,
1029 integer_zero_node);
1031 if (act->mult && !m_acc)
1032 m_acc = create_tailcall_accumulator ("mult_acc", first,
1033 integer_one_node);
1036 if (a_acc || m_acc)
1038 /* When the tail call elimination using accumulators is performed,
1039 statements adding the accumulated value are inserted at all exits.
1040 This turns all other tail calls to non-tail ones. */
1041 opt_tailcalls = false;
1044 for (; tailcalls; tailcalls = next)
1046 next = tailcalls->next;
1047 changed |= optimize_tail_call (tailcalls, opt_tailcalls);
1048 free (tailcalls);
1051 if (a_acc || m_acc)
1053 /* Modify the remaining return statements. */
1054 FOR_EACH_EDGE (e, ei, EXIT_BLOCK_PTR_FOR_FN (cfun)->preds)
1056 stmt = last_stmt (e->src);
1058 if (stmt
1059 && gimple_code (stmt) == GIMPLE_RETURN)
1060 adjust_return_value (e->src, m_acc, a_acc);
1064 if (changed)
1066 /* We may have created new loops. Make them magically appear. */
1067 loops_state_set (LOOPS_NEED_FIXUP);
1068 free_dominance_info (CDI_DOMINATORS);
1071 /* Add phi nodes for the virtual operands defined in the function to the
1072 header of the loop created by tail recursion elimination. Do so
1073 by triggering the SSA renamer. */
1074 if (phis_constructed)
1075 mark_virtual_operands_for_renaming (cfun);
1077 if (changed)
1078 return TODO_cleanup_cfg | TODO_update_ssa_only_virtuals;
1079 return 0;
1082 static bool
1083 gate_tail_calls (void)
1085 return flag_optimize_sibling_calls != 0 && dbg_cnt (tail_call);
1088 static unsigned int
1089 execute_tail_calls (void)
1091 return tree_optimize_tail_calls_1 (true);
1094 namespace {
1096 const pass_data pass_data_tail_recursion =
1098 GIMPLE_PASS, /* type */
1099 "tailr", /* name */
1100 OPTGROUP_NONE, /* optinfo_flags */
1101 TV_NONE, /* tv_id */
1102 ( PROP_cfg | PROP_ssa ), /* properties_required */
1103 0, /* properties_provided */
1104 0, /* properties_destroyed */
1105 0, /* todo_flags_start */
1106 0, /* todo_flags_finish */
1109 class pass_tail_recursion : public gimple_opt_pass
1111 public:
1112 pass_tail_recursion (gcc::context *ctxt)
1113 : gimple_opt_pass (pass_data_tail_recursion, ctxt)
1116 /* opt_pass methods: */
1117 opt_pass * clone () { return new pass_tail_recursion (m_ctxt); }
1118 virtual bool gate (function *) { return gate_tail_calls (); }
1119 virtual unsigned int execute (function *)
1121 return tree_optimize_tail_calls_1 (false);
1124 }; // class pass_tail_recursion
1126 } // anon namespace
1128 gimple_opt_pass *
1129 make_pass_tail_recursion (gcc::context *ctxt)
1131 return new pass_tail_recursion (ctxt);
1134 namespace {
1136 const pass_data pass_data_tail_calls =
1138 GIMPLE_PASS, /* type */
1139 "tailc", /* name */
1140 OPTGROUP_NONE, /* optinfo_flags */
1141 TV_NONE, /* tv_id */
1142 ( PROP_cfg | PROP_ssa ), /* properties_required */
1143 0, /* properties_provided */
1144 0, /* properties_destroyed */
1145 0, /* todo_flags_start */
1146 0, /* todo_flags_finish */
1149 class pass_tail_calls : public gimple_opt_pass
1151 public:
1152 pass_tail_calls (gcc::context *ctxt)
1153 : gimple_opt_pass (pass_data_tail_calls, ctxt)
1156 /* opt_pass methods: */
1157 virtual bool gate (function *) { return gate_tail_calls (); }
1158 virtual unsigned int execute (function *) { return execute_tail_calls (); }
1160 }; // class pass_tail_calls
1162 } // anon namespace
1164 gimple_opt_pass *
1165 make_pass_tail_calls (gcc::context *ctxt)
1167 return new pass_tail_calls (ctxt);