3 * File name: ir/ir/iropt.c
4 * Purpose: iropt --- optimizations intertwined with IR construction.
5 * Author: Christian Schaefer
6 * Modified by: Goetz Lindenmaier
9 * Copyright: (c) 1998-2003 Universität Karlsruhe
10 * Licence: This file protected by GPL - GNU GENERAL PUBLIC LICENSE.
17 # include "irnode_t.h"
18 # include "irgraph_t.h"
19 # include "irmode_t.h"
21 # include "ircons_t.h"
25 # include "dbginfo_t.h"
26 # include "iropt_dbg.h"
27 # include "irflag_t.h"
28 # include "firmstat.h"
30 /* Make types visible to allow most efficient access */
31 # include "entity_t.h"
34 * Trivial INLINEable routine for copy propagation.
35 * Does follow Ids, needed to optimize INLINEd code.
37 static INLINE ir_node *
38 follow_Id (ir_node *n)
40 while (get_irn_op (n) == op_Id) n = get_Id_pred (n);
45 * Returns the tarval of a Const node or tarval_bad for all other nodes.
47 static INLINE tarval *
50 if ((n != NULL) && (get_irn_op(n) == op_Const))
51 return get_Const_tarval(n); /* might return tarval_bad */
57 * return the value of a Constant
59 static tarval *computed_value_Const(ir_node *n)
61 return get_Const_tarval(n);
65 * return the value of a 'sizeof' SymConst
67 static tarval *computed_value_SymConst(ir_node *n)
69 if ((get_SymConst_kind(n) == symconst_size) &&
70 (get_type_state(get_SymConst_type(n))) == layout_fixed)
71 return new_tarval_from_long(get_type_size_bytes(get_SymConst_type(n)), get_irn_mode(n));
75 static tarval *computed_value_Add(ir_node *n)
77 ir_node *a = get_Add_left(n);
78 ir_node *b = get_Add_right(n);
80 tarval *ta = value_of(a);
81 tarval *tb = value_of(b);
83 if ((ta != tarval_bad) && (tb != tarval_bad)
84 && (get_irn_mode(a) == get_irn_mode(b))
85 && !(get_mode_sort(get_irn_mode(a)) == irms_reference)) {
86 return tarval_add(ta, tb);
91 static tarval *computed_value_Sub(ir_node *n)
93 ir_node *a = get_Sub_left(n);
94 ir_node *b = get_Sub_right(n);
96 tarval *ta = value_of(a);
97 tarval *tb = value_of(b);
99 if ((ta != tarval_bad) && (tb != tarval_bad)
100 && (get_irn_mode(a) == get_irn_mode(b))
101 && !(get_mode_sort(get_irn_mode(a)) == irms_reference)) {
102 return tarval_sub(ta, tb);
107 static tarval *computed_value_Minus(ir_node *n)
109 ir_node *a = get_Minus_op(n);
110 tarval *ta = value_of(a);
112 if ((ta != tarval_bad) && mode_is_signed(get_irn_mode(a)))
113 return tarval_neg(ta);
118 static tarval *computed_value_Mul(ir_node *n)
120 ir_node *a = get_Mul_left(n);
121 ir_node *b = get_Mul_right(n);
123 tarval *ta = value_of(a);
124 tarval *tb = value_of(b);
126 if ((ta != tarval_bad) && (tb != tarval_bad) && (get_irn_mode(a) == get_irn_mode(b))) {
127 return tarval_mul(ta, tb);
129 /* a*0 = 0 or 0*b = 0:
130 calls computed_value recursive and returns the 0 with proper
134 if ( ( ((v = ta) != tarval_bad)
135 && (v == get_mode_null(get_tarval_mode(v))) )
136 || ( ((v = tb) != tarval_bad)
137 && (v == get_mode_null(get_tarval_mode(v))) )) {
144 static tarval *computed_value_Quot(ir_node *n)
146 ir_node *a = get_Quot_left(n);
147 ir_node *b = get_Quot_right(n);
149 tarval *ta = value_of(a);
150 tarval *tb = value_of(b);
152 /* This was missing in original implementation. Why? */
153 if ((ta != tarval_bad) && (tb != tarval_bad) && (get_irn_mode(a) == get_irn_mode(b))) {
154 if (tb != get_mode_null(get_tarval_mode(tb))) /* div by zero: return tarval_bad */
155 return tarval_quo(ta, tb);
160 static tarval *computed_value_Div(ir_node *n)
162 ir_node *a = get_Div_left(n);
163 ir_node *b = get_Div_right(n);
165 tarval *ta = value_of(a);
166 tarval *tb = value_of(b);
168 /* This was missing in original implementation. Why? */
169 if ((ta != tarval_bad) && (tb != tarval_bad) && (get_irn_mode(a) == get_irn_mode(b))) {
170 if (tb != get_mode_null(get_tarval_mode(tb))) /* div by zero: return tarval_bad */
171 return tarval_div(ta, tb);
176 static tarval *computed_value_Mod(ir_node *n)
178 ir_node *a = get_Mod_left(n);
179 ir_node *b = get_Mod_right(n);
181 tarval *ta = value_of(a);
182 tarval *tb = value_of(b);
184 /* This was missing in original implementation. Why? */
185 if ((ta != tarval_bad) && (tb != tarval_bad) && (get_irn_mode(a) == get_irn_mode(b))) {
186 if (tb != get_mode_null(get_tarval_mode(tb))) /* div by zero: return tarval_bad */
187 return tarval_mod(ta, tb);
192 static tarval *computed_value_Abs(ir_node *n)
194 ir_node *a = get_Abs_op(n);
195 tarval *ta = value_of(a);
197 if (ta != tarval_bad)
198 return tarval_abs(ta);
203 static tarval *computed_value_And(ir_node *n)
205 ir_node *a = get_And_left(n);
206 ir_node *b = get_And_right(n);
208 tarval *ta = value_of(a);
209 tarval *tb = value_of(b);
211 if ((ta != tarval_bad) && (tb != tarval_bad)) {
212 return tarval_and (ta, tb);
216 if ( (classify_tarval ((v = computed_value (a))) == TV_CLASSIFY_NULL)
217 || (classify_tarval ((v = computed_value (b))) == TV_CLASSIFY_NULL)) {
224 static tarval *computed_value_Or(ir_node *n)
226 ir_node *a = get_Or_left(n);
227 ir_node *b = get_Or_right(n);
229 tarval *ta = value_of(a);
230 tarval *tb = value_of(b);
232 if ((ta != tarval_bad) && (tb != tarval_bad)) {
233 return tarval_or (ta, tb);
236 if ( (classify_tarval ((v = computed_value (a))) == TV_CLASSIFY_ALL_ONE)
237 || (classify_tarval ((v = computed_value (b))) == TV_CLASSIFY_ALL_ONE)) {
244 static tarval *computed_value_Eor(ir_node *n)
246 ir_node *a = get_Eor_left(n);
247 ir_node *b = get_Eor_right(n);
249 tarval *ta = value_of(a);
250 tarval *tb = value_of(b);
252 if ((ta != tarval_bad) && (tb != tarval_bad)) {
253 return tarval_eor (ta, tb);
258 static tarval *computed_value_Not(ir_node *n)
260 ir_node *a = get_Not_op(n);
261 tarval *ta = value_of(a);
263 if (ta != tarval_bad)
264 return tarval_not(ta);
269 static tarval *computed_value_Shl(ir_node *n)
271 ir_node *a = get_Shl_left(n);
272 ir_node *b = get_Shl_right(n);
274 tarval *ta = value_of(a);
275 tarval *tb = value_of(b);
277 if ((ta != tarval_bad) && (tb != tarval_bad)) {
278 return tarval_shl (ta, tb);
283 static tarval *computed_value_Shr(ir_node *n)
285 ir_node *a = get_Shr_left(n);
286 ir_node *b = get_Shr_right(n);
288 tarval *ta = value_of(a);
289 tarval *tb = value_of(b);
291 if ((ta != tarval_bad) && (tb != tarval_bad)) {
292 return tarval_shr (ta, tb);
297 static tarval *computed_value_Shrs(ir_node *n)
299 ir_node *a = get_Shrs_left(n);
300 ir_node *b = get_Shrs_right(n);
302 tarval *ta = value_of(a);
303 tarval *tb = value_of(b);
305 if ((ta != tarval_bad) && (tb != tarval_bad)) {
306 return tarval_shrs (ta, tb);
311 static tarval *computed_value_Rot(ir_node *n)
313 ir_node *a = get_Rot_left(n);
314 ir_node *b = get_Rot_right(n);
316 tarval *ta = value_of(a);
317 tarval *tb = value_of(b);
319 if ((ta != tarval_bad) && (tb != tarval_bad)) {
320 return tarval_rot (ta, tb);
325 static tarval *computed_value_Conv(ir_node *n)
327 ir_node *a = get_Conv_op(n);
328 tarval *ta = value_of(a);
330 if (ta != tarval_bad)
331 return tarval_convert_to(ta, get_irn_mode(n));
336 static tarval *computed_value_Proj(ir_node *n)
338 ir_node *a = get_Proj_pred(n), *b;
341 /* Optimize Cmp nodes.
342 This performs a first step of unreachable code elimination.
343 Proj can not be computed, but folding a Cmp above the Proj here is
344 not as wasteful as folding a Cmp into a Tuple of 16 Consts of which
346 There are several case where we can evaluate a Cmp node:
347 1. The nodes compared are both the same. If we compare for
348 equal, greater equal, ... this will return true, else it
349 will return false. This step relies on cse.
350 2. The predecessors of Cmp are target values. We can evaluate
352 3. The predecessors are Allocs or void* constants. Allocs never
353 return NULL, they raise an exception. Therefore we can predict
355 switch (get_irn_opcode(a)) {
357 aa = get_Cmp_left(a);
358 ab = get_Cmp_right(a);
360 if (aa == ab) { /* 1.: */
361 /* This is a trick with the bits used for encoding the Cmp
362 Proj numbers, the following statement is not the same:
363 return new_tarval_from_long ((get_Proj_proj(n) == Eq), mode_b) */
364 return new_tarval_from_long ((get_Proj_proj(n) & Eq), mode_b);
366 tarval *taa = computed_value (aa);
367 tarval *tab = computed_value (ab);
369 if ((taa != tarval_bad) && (tab != tarval_bad)) { /* 2.: */
370 /* strange checks... */
371 pnc_number flags = tarval_cmp (taa, tab);
372 if (flags != False) {
373 return new_tarval_from_long (get_Proj_proj(n) & flags, mode_b);
375 } else { /* check for 3.: */
376 ir_node *aaa = skip_Id(skip_Proj(aa));
377 ir_node *aba = skip_Id(skip_Proj(ab));
379 if ( ( (/* aa is ProjP and aaa is Alloc */
380 (get_irn_op(aa) == op_Proj)
381 && (mode_is_reference(get_irn_mode(aa)))
382 && (get_irn_op(aaa) == op_Alloc))
383 && ( (/* ab is constant void */
384 (get_irn_op(ab) == op_Const)
385 && (mode_is_reference(get_irn_mode(ab)))
386 && (get_Const_tarval(ab) == get_mode_null(get_irn_mode(ab))))
387 || (/* ab is other Alloc */
388 (get_irn_op(ab) == op_Proj)
389 && (mode_is_reference(get_irn_mode(ab)))
390 && (get_irn_op(aba) == op_Alloc)
392 || (/* aa is void and aba is Alloc */
393 (get_irn_op(aa) == op_Const)
394 && (mode_is_reference(get_irn_mode(aa)))
395 && (get_Const_tarval(aa) == get_mode_null(get_irn_mode(aa)))
396 && (get_irn_op(ab) == op_Proj)
397 && (mode_is_reference(get_irn_mode(ab)))
398 && (get_irn_op(aba) == op_Alloc)))
400 return new_tarval_from_long (get_Proj_proj(n) & Ne, mode_b);
407 tarval *tb = value_of(b = get_DivMod_right(a));
408 tarval *ta = value_of(a = get_DivMod_left(a));
410 if ((ta != tarval_bad) && (tb != tarval_bad) && (get_irn_mode(a) == get_irn_mode(b))) {
411 if (tb == get_mode_null(get_tarval_mode(tb))) /* div by zero: return tarval_bad */
413 if (get_Proj_proj(n)== pn_DivMod_res_div)
414 return tarval_div(ta, tb);
415 else if (get_Proj_proj(n)== pn_DivMod_res_mod)
416 return tarval_mod(ta, tb);
423 tarval *tb = value_of(b = get_Div_right(a));
424 tarval *ta = value_of(a = get_Div_left(a));
426 if ((ta != tarval_bad) && (tb != tarval_bad) && (get_irn_mode(a) == get_irn_mode(b))) {
427 if (tb == get_mode_null(get_tarval_mode(tb))) /* div by zero: return tarval_bad */
429 if (get_Proj_proj(n)== pn_Div_res)
430 return tarval_div(ta, tb);
437 tarval *tb = value_of(b = get_Mod_right(a));
438 tarval *ta = value_of(a = get_Mod_left(a));
440 if ((ta != tarval_bad) && (tb != tarval_bad) && (get_irn_mode(a) == get_irn_mode(b))) {
441 if (tb == get_mode_null(get_tarval_mode(tb))) /* div by zero: return tarval_bad */
443 if (get_Proj_proj(n)== pn_Mod_res)
444 return tarval_mod(ta, tb);
456 * If the parameter n can be computed, return its value, else tarval_bad.
457 * Performs constant folding.
459 * @param n The node this should be evaluated
461 tarval *computed_value(ir_node *n)
463 if (n->op->computed_value)
464 return n->op->computed_value(n);
469 * set the default computed_value evaluator
471 static ir_op *firm_set_default_computed_value(ir_op *op)
475 op->computed_value = computed_value_##a; \
500 op->computed_value = NULL;
508 /* returns 1 if the a and b are pointers to different locations. */
510 different_identity (ir_node *a, ir_node *b)
512 assert (mode_is_reference(get_irn_mode (a))
513 && mode_is_reference(get_irn_mode (b)));
515 if (get_irn_op (a) == op_Proj && get_irn_op(b) == op_Proj) {
516 ir_node *a1 = get_Proj_pred (a);
517 ir_node *b1 = get_Proj_pred (b);
518 if (a1 != b1 && get_irn_op (a1) == op_Alloc
519 && get_irn_op (b1) == op_Alloc)
526 static ir_node *equivalent_node_Block(ir_node *n)
530 /* The Block constructor does not call optimize, but mature_immBlock
531 calls the optimization. */
532 assert(get_Block_matured(n));
534 /* Straightening: a single entry Block following a single exit Block
535 can be merged, if it is not the Start block. */
536 /* !!! Beware, all Phi-nodes of n must have been optimized away.
537 This should be true, as the block is matured before optimize is called.
538 But what about Phi-cycles with the Phi0/Id that could not be resolved?
539 Remaining Phi nodes are just Ids. */
540 if ((get_Block_n_cfgpreds(n) == 1) &&
541 (get_irn_op(get_Block_cfgpred(n, 0)) == op_Jmp)) {
542 ir_node *predblock = get_nodes_block(get_Block_cfgpred(n, 0));
543 if (predblock == oldn) {
544 /* Jmp jumps into the block it is in -- deal self cycle. */
545 n = new_Bad(); DBG_OPT_DEAD;
546 } else if (get_opt_control_flow_straightening()) {
547 n = predblock; DBG_OPT_STG;
550 else if ((get_Block_n_cfgpreds(n) == 1) &&
551 (get_irn_op(skip_Proj(get_Block_cfgpred(n, 0))) == op_Cond)) {
552 ir_node *predblock = get_nodes_block(get_Block_cfgpred(n, 0));
553 if (predblock == oldn) {
554 /* Jmp jumps into the block it is in -- deal self cycle. */
555 n = new_Bad(); DBG_OPT_DEAD;
558 else if ((get_Block_n_cfgpreds(n) == 2) &&
559 (get_opt_control_flow_weak_simplification())) {
560 /* Test whether Cond jumps twice to this block
561 @@@ we could do this also with two loops finding two preds from several ones. */
562 ir_node *a = get_Block_cfgpred(n, 0);
563 ir_node *b = get_Block_cfgpred(n, 1);
565 if ((get_irn_op(a) == op_Proj) &&
566 (get_irn_op(b) == op_Proj) &&
567 (get_Proj_pred(a) == get_Proj_pred(b)) &&
568 (get_irn_op(get_Proj_pred(a)) == op_Cond) &&
569 (get_irn_mode(get_Cond_selector(get_Proj_pred(a))) == mode_b)) {
570 /* Also a single entry Block following a single exit Block. Phis have
571 twice the same operand and will be optimized away. */
572 n = get_nodes_block(a); DBG_OPT_IFSIM;
574 } else if (get_opt_unreachable_code() &&
575 (n != current_ir_graph->start_block) &&
576 (n != current_ir_graph->end_block) ) {
578 /* If all inputs are dead, this block is dead too, except if it is
579 the start or end block. This is a step of unreachable code
581 for (i = 0; i < get_Block_n_cfgpreds(n); i++) {
582 if (!is_Bad(get_Block_cfgpred(n, i))) break;
584 if (i == get_Block_n_cfgpreds(n))
592 * Returns a equivalent node for a Jmp, a Bad :-)
593 * Of course this only happens if the Block of the Jmp is Bad.
595 static ir_node *equivalent_node_Jmp(ir_node *n)
597 /* GL: Why not same for op_Raise?? */
598 /* unreachable code elimination */
599 if (is_Bad(get_nodes_block(n)))
605 static ir_node *equivalent_node_Cond(ir_node *n)
607 /* We do not evaluate Cond here as we replace it by a new node, a Jmp.
608 See cases for iro_Cond and iro_Proj in transform_node. */
613 * Use algebraic simplification a v a = a.
615 static ir_node *equivalent_node_Or(ir_node *n)
619 ir_node *a = get_Or_left(n);
620 ir_node *b = get_Or_right(n);
624 n = a; DBG_OPT_ALGSIM1;
631 * optimize operations that are commutative and have neutral 0,
632 * so a op 0 = 0 op a = a.
634 static ir_node *equivalent_node_neutral_zero(ir_node *n)
638 ir_node *a = get_binop_left(n);
639 ir_node *b = get_binop_right(n);
644 /* After running compute_node there is only one constant predecessor.
645 Find this predecessors value and remember the other node: */
646 if ((tv = computed_value(a)) != tarval_bad) {
648 } else if ((tv = computed_value(b)) != tarval_bad) {
653 /* If this predecessors constant value is zero, the operation is
654 unnecessary. Remove it: */
655 if (classify_tarval (tv) == TV_CLASSIFY_NULL) {
656 n = on; DBG_OPT_ALGSIM1;
662 #define equivalent_node_Add equivalent_node_neutral_zero
663 #define equivalent_node_Eor equivalent_node_neutral_zero
666 * optimize operations that are not commutative but have neutral 0 on left,
669 static ir_node *equivalent_node_left_zero(ir_node *n)
673 ir_node *a = get_binop_left(n);
674 ir_node *b = get_binop_right(n);
676 if (classify_tarval(computed_value(b)) == TV_CLASSIFY_NULL) {
677 n = a; DBG_OPT_ALGSIM1;
683 #define equivalent_node_Sub equivalent_node_left_zero
684 #define equivalent_node_Shl equivalent_node_left_zero
685 #define equivalent_node_Shr equivalent_node_left_zero
686 #define equivalent_node_Shrs equivalent_node_left_zero
687 #define equivalent_node_Rot equivalent_node_left_zero
690 * Er, a "symmetic unop", ie op(op(n)) = n.
692 static ir_node *equivalent_node_symmetric_unop(ir_node *n)
695 ir_node *pred = get_unop_op(n);
697 /* optimize symmetric unop */
698 if (get_irn_op(pred) == get_irn_op(n)) {
699 n = get_unop_op(pred); DBG_OPT_ALGSIM2;
705 #define equivalent_node_Not equivalent_node_symmetric_unop
707 /* --x == x */ /* ??? Is this possible or can --x raise an
708 out of bounds exception if min =! max? */
709 #define equivalent_node_Minus equivalent_node_symmetric_unop
712 * Optimize a * 1 = 1 * a = a.
714 static ir_node *equivalent_node_Mul(ir_node *n)
718 ir_node *a = get_Mul_left(n);
719 ir_node *b = get_Mul_right(n);
721 /* Mul is commutative and has again an other neutral element. */
722 if (classify_tarval (computed_value (a)) == TV_CLASSIFY_ONE) {
723 n = b; DBG_OPT_ALGSIM1;
724 } else if (classify_tarval (computed_value (b)) == TV_CLASSIFY_ONE) {
725 n = a; DBG_OPT_ALGSIM1;
731 * Optimize a / 1 = a.
733 static ir_node *equivalent_node_Div(ir_node *n)
735 ir_node *a = get_Div_left(n);
736 ir_node *b = get_Div_right(n);
738 /* Div is not commutative. */
739 if (classify_tarval(computed_value(b)) == TV_CLASSIFY_ONE) { /* div(x, 1) == x */
740 /* Turn Div into a tuple (mem, bad, a) */
741 ir_node *mem = get_Div_mem(n);
742 turn_into_tuple(n, 3);
743 set_Tuple_pred(n, pn_Div_M, mem);
744 set_Tuple_pred(n, pn_Div_X_except, new_Bad()); /* no exception */
745 set_Tuple_pred(n, pn_Div_res, a);
751 * Optimize a & 0b1...1 = 0b1...1 & a = a & a = a.
753 static ir_node *equivalent_node_And(ir_node *n)
757 ir_node *a = get_And_left(n);
758 ir_node *b = get_And_right(n);
761 n = a; /* And has it's own neutral element */
762 } else if (classify_tarval(computed_value(a)) == TV_CLASSIFY_ALL_ONE) {
764 } else if (classify_tarval(computed_value(b)) == TV_CLASSIFY_ALL_ONE) {
767 if (n != oldn) DBG_OPT_ALGSIM1;
772 * Try to remove useless conv's:
774 static ir_node *equivalent_node_Conv(ir_node *n)
777 ir_node *a = get_Conv_op(n);
780 ir_mode *n_mode = get_irn_mode(n);
781 ir_mode *a_mode = get_irn_mode(a);
783 if (n_mode == a_mode) { /* No Conv necessary */
784 n = a; DBG_OPT_ALGSIM3;
785 } else if (get_irn_op(a) == op_Conv) { /* Conv(Conv(b)) */
789 n_mode = get_irn_mode(n);
790 b_mode = get_irn_mode(b);
792 if (n_mode == b_mode) {
793 if (n_mode == mode_b) {
794 n = b; /* Convb(Conv*(xxxb(...))) == xxxb(...) */ DBG_OPT_ALGSIM1;
796 else if (mode_is_int(n_mode) || mode_is_character(n_mode)) {
797 if (smaller_mode(b_mode, a_mode)){
798 n = b; /* ConvS(ConvL(xxxS(...))) == xxxS(...) */ DBG_OPT_ALGSIM1;
806 static ir_node *equivalent_node_Phi(ir_node *n)
808 /* Several optimizations:
809 - no Phi in start block.
810 - remove Id operators that are inputs to Phi
811 - fold Phi-nodes, iff they have only one predecessor except
817 ir_node *block = NULL; /* to shutup gcc */
818 ir_node *first_val = NULL; /* to shutup gcc */
819 ir_node *scnd_val = NULL; /* to shutup gcc */
821 if (!get_opt_normalize()) return n;
823 n_preds = get_Phi_n_preds(n);
825 block = get_nodes_block(n);
826 /* @@@ fliegt 'raus, sollte aber doch immer wahr sein!!!
827 assert(get_irn_arity(block) == n_preds && "phi in wrong block!"); */
828 if ((is_Bad(block)) || /* Control dead */
829 (block == current_ir_graph->start_block)) /* There should be no Phi nodes */
830 return new_Bad(); /* in the Start Block. */
832 if (n_preds == 0) return n; /* Phi of dead Region without predecessors. */
835 /* first we test for a special case: */
836 /* Confirm is a special node fixing additional information for a
837 value that is known at a certain point. This is useful for
838 dataflow analysis. */
840 ir_node *a = get_Phi_pred(n, 0);
841 ir_node *b = get_Phi_pred(n, 1);
842 if ( (get_irn_op(a) == op_Confirm)
843 && (get_irn_op(b) == op_Confirm)
844 && follow_Id (get_irn_n(a, 0) == get_irn_n(b, 0))
845 && (get_irn_n(a, 1) == get_irn_n (b, 1))
846 && (a->data.num == (~b->data.num & irpn_True) )) {
847 return get_irn_n(a, 0);
852 /* If the Block has a Bad pred, we also have one. */
853 for (i = 0; i < n_preds; ++i)
854 if (is_Bad (get_Block_cfgpred(block, i)))
855 set_Phi_pred(n, i, new_Bad());
857 /* Find first non-self-referencing input */
858 for (i = 0; i < n_preds; ++i) {
859 first_val = get_Phi_pred(n, i);
860 if ( (first_val != n) /* not self pointer */
862 && (get_irn_op(first_val) != op_Bad)
864 ) { /* value not dead */
865 break; /* then found first value. */
869 /* A totally Bad or self-referencing Phi (we didn't break the above loop) */
870 if (i >= n_preds) { return new_Bad(); }
874 /* follow_Id () for rest of inputs, determine if any of these
875 are non-self-referencing */
876 while (++i < n_preds) {
877 scnd_val = get_Phi_pred(n, i);
879 && (scnd_val != first_val)
881 && (get_irn_op(scnd_val) != op_Bad)
888 /* Fold, if no multiple distinct non-self-referencing inputs */
890 n = first_val; DBG_OPT_PHI;
892 /* skip the remaining Ids (done in get_Phi_pred). */
893 /* superfluous, since we walk all to propagate Block's Bads.
894 while (++i < n_preds) get_Phi_pred(n, i); */
900 * optimize Proj(Tuple) and gigo for ProjX in Bad block
902 static ir_node *equivalent_node_Proj(ir_node *n)
906 ir_node *a = get_Proj_pred(n);
908 if ( get_irn_op(a) == op_Tuple) {
909 /* Remove the Tuple/Proj combination. */
910 if ( get_Proj_proj(n) <= get_Tuple_n_preds(a) ) {
911 n = get_Tuple_pred(a, get_Proj_proj(n)); DBG_OPT_TUPLE;
913 assert(0); /* This should not happen! */
916 } else if (get_irn_mode(n) == mode_X &&
917 is_Bad(get_nodes_block(n))) {
918 /* Remove dead control flow -- early gigo. */
927 static ir_node *equivalent_node_Id(ir_node *n)
931 n = follow_Id(n); DBG_OPT_ID;
936 * equivalent_node() returns a node equivalent to input n. It skips all nodes that
937 * perform no actual computation, as, e.g., the Id nodes. It does not create
938 * new nodes. It is therefore safe to free n if the node returned is not n.
939 * If a node returns a Tuple we can not just skip it. If the size of the
940 * in array fits, we transform n into a tuple (e.g., Div).
943 equivalent_node(ir_node *n)
945 if (n->op->equivalent_node)
946 return n->op->equivalent_node(n);
951 * set the default equivalent node operation
953 static ir_op *firm_set_default_equivalent_node(ir_op *op)
957 op->equivalent_node = equivalent_node_##a; \
982 op->equivalent_node = NULL;
990 * Do node specific optimizations of nodes predecessors.
993 optimize_preds(ir_node *n) {
994 ir_node *a = NULL, *b = NULL;
996 /* get the operands we will work on for simple cases. */
998 a = get_binop_left(n);
999 b = get_binop_right(n);
1000 } else if (is_unop(n)) {
1004 switch (get_irn_opcode(n)) {
1007 /* We don't want Cast as input to Cmp. */
1008 if (get_irn_op(a) == op_Cast) {
1012 if (get_irn_op(b) == op_Cast) {
1014 set_Cmp_right(n, b);
1022 static ir_node *transform_node_Div(ir_node *n)
1024 tarval *tv = computed_value(n);
1026 /* BEWARE: it is NOT possible to optimize a/a to 1, as this may cause a exception */
1028 if (tv != tarval_bad) {
1029 /* Turn Div into a tuple (mem, bad, value) */
1030 ir_node *mem = get_Div_mem(n);
1032 turn_into_tuple(n, 3);
1033 set_Tuple_pred(n, pn_Div_M, mem);
1034 set_Tuple_pred(n, pn_Div_X_except, new_Bad());
1035 set_Tuple_pred(n, pn_Div_res, new_Const(get_tarval_mode(tv), tv));
1040 static ir_node *transform_node_Mod(ir_node *n)
1042 tarval *tv = computed_value(n);
1044 /* BEWARE: it is NOT possible to optimize a%a to 0, as this may cause a exception */
1046 if (tv != tarval_bad) {
1047 /* Turn Mod into a tuple (mem, bad, value) */
1048 ir_node *mem = get_Mod_mem(n);
1049 turn_into_tuple(n, 3);
1050 set_Tuple_pred(n, pn_Mod_M, mem);
1051 set_Tuple_pred(n, pn_Mod_X_except, new_Bad());
1052 set_Tuple_pred(n, pn_Mod_res, new_Const(get_tarval_mode(tv), tv));
1057 static ir_node *transform_node_DivMod(ir_node *n)
1061 ir_node *a = get_DivMod_left(n);
1062 ir_node *b = get_DivMod_right(n);
1063 ir_mode *mode = get_irn_mode(a);
1064 tarval *ta = value_of(a);
1065 tarval *tb = value_of(b);
1067 if (!(mode_is_int(mode) && mode_is_int(get_irn_mode(b))))
1070 /* BEWARE: it is NOT possible to optimize a/a to 1, as this may cause a exception */
1072 if (tb != tarval_bad) {
1073 if (tb == get_mode_one(get_tarval_mode(tb))) {
1074 b = new_Const (mode, get_mode_null(mode));
1076 } else if (ta != tarval_bad) {
1077 tarval *resa, *resb;
1078 resa = tarval_div (ta, tb);
1079 if (resa == tarval_bad) return n; /* Causes exception!!! Model by replacing through
1080 Jmp for X result!? */
1081 resb = tarval_mod (ta, tb);
1082 if (resb == tarval_bad) return n; /* Causes exception! */
1083 a = new_Const (mode, resa);
1084 b = new_Const (mode, resb);
1087 } else if (ta == get_mode_null(mode)) {
1091 if (evaluated) { /* replace by tuple */
1092 ir_node *mem = get_DivMod_mem(n);
1093 turn_into_tuple(n, 4);
1094 set_Tuple_pred(n, pn_DivMod_M, mem);
1095 set_Tuple_pred(n, pn_DivMod_X_except, new_Bad()); /* no exception */
1096 set_Tuple_pred(n, pn_DivMod_res_div, a);
1097 set_Tuple_pred(n, pn_DivMod_res_mod, b);
1098 assert(get_nodes_block(n));
1104 static ir_node *transform_node_Cond(ir_node *n)
1106 /* Replace the Cond by a Jmp if it branches on a constant
1109 ir_node *a = get_Cond_selector(n);
1110 tarval *ta = value_of(a);
1112 if ((ta != tarval_bad) &&
1113 (get_irn_mode(a) == mode_b) &&
1114 (get_opt_unreachable_code())) {
1115 /* It's a boolean Cond, branching on a boolean constant.
1116 Replace it by a tuple (Bad, Jmp) or (Jmp, Bad) */
1117 jmp = new_r_Jmp(current_ir_graph, get_nodes_block(n));
1118 turn_into_tuple(n, 2);
1119 if (ta == tarval_b_true) {
1120 set_Tuple_pred(n, pn_Cond_false, new_Bad());
1121 set_Tuple_pred(n, pn_Cond_true, jmp);
1123 set_Tuple_pred(n, pn_Cond_false, jmp);
1124 set_Tuple_pred(n, pn_Cond_true, new_Bad());
1126 /* We might generate an endless loop, so keep it alive. */
1127 add_End_keepalive(get_irg_end(current_ir_graph), get_nodes_block(n));
1128 } else if ((ta != tarval_bad) &&
1129 (get_irn_mode(a) == mode_Iu) &&
1130 (get_Cond_kind(n) == dense) &&
1131 (get_opt_unreachable_code())) {
1132 /* I don't want to allow Tuples smaller than the biggest Proj.
1133 Also this tuple might get really big...
1134 I generate the Jmp here, and remember it in link. Link is used
1135 when optimizing Proj. */
1136 set_irn_link(n, new_r_Jmp(current_ir_graph, get_nodes_block(n)));
1137 /* We might generate an endless loop, so keep it alive. */
1138 add_End_keepalive(get_irg_end(current_ir_graph), get_nodes_block(n));
1139 } else if ((get_irn_op(a) == op_Eor)
1140 && (get_irn_mode(a) == mode_b)
1141 && (classify_tarval(computed_value(get_Eor_right(a))) == TV_CLASSIFY_ONE)) {
1142 /* The Eor is a negate. Generate a new Cond without the negate,
1143 simulate the negate by exchanging the results. */
1144 set_irn_link(n, new_r_Cond(current_ir_graph, get_nodes_block(n),
1146 } else if ((get_irn_op(a) == op_Not)
1147 && (get_irn_mode(a) == mode_b)) {
1148 /* A Not before the Cond. Generate a new Cond without the Not,
1149 simulate the Not by exchanging the results. */
1150 set_irn_link(n, new_r_Cond(current_ir_graph, get_nodes_block(n),
1156 static ir_node *transform_node_Eor(ir_node *n)
1158 ir_node *a = get_Eor_left(n);
1159 ir_node *b = get_Eor_right(n);
1161 if ((get_irn_mode(n) == mode_b)
1162 && (get_irn_op(a) == op_Proj)
1163 && (get_irn_mode(a) == mode_b)
1164 && (classify_tarval (computed_value (b)) == TV_CLASSIFY_ONE)
1165 && (get_irn_op(get_Proj_pred(a)) == op_Cmp))
1166 /* The Eor negates a Cmp. The Cmp has the negated result anyways! */
1167 n = new_r_Proj(current_ir_graph, get_nodes_block(n), get_Proj_pred(a),
1168 mode_b, get_negated_pnc(get_Proj_proj(a)));
1169 else if ((get_irn_mode(n) == mode_b)
1170 && (classify_tarval (computed_value (b)) == TV_CLASSIFY_ONE))
1171 /* The Eor is a Not. Replace it by a Not. */
1172 /* ????!!!Extend to bitfield 1111111. */
1173 n = new_r_Not(current_ir_graph, get_nodes_block(n), a, mode_b);
1179 * Transfor a boolean Not.
1181 static ir_node *transform_node_Not(ir_node *n)
1183 ir_node *a = get_Not_op(n);
1185 if ( (get_irn_mode(n) == mode_b)
1186 && (get_irn_op(a) == op_Proj)
1187 && (get_irn_mode(a) == mode_b)
1188 && (get_irn_op(get_Proj_pred(a)) == op_Cmp))
1189 /* We negate a Cmp. The Cmp has the negated result anyways! */
1190 n = new_r_Proj(current_ir_graph, get_nodes_block(n), get_Proj_pred(a),
1191 mode_b, get_negated_pnc(get_Proj_proj(a)));
1197 * Transform a Div/Mod/DivMod with a non-zero constant. Must be
1198 * done here instead of equivalent node because in creates new
1201 static ir_node *transform_node_Proj(ir_node *proj)
1203 ir_node *n = get_Proj_pred(proj);
1207 switch (get_irn_opcode(n)) {
1209 b = get_Div_right(n);
1210 tb = computed_value(b);
1212 if (tb != tarval_bad && classify_tarval(tb) != TV_CLASSIFY_NULL) { /* div(x, c) && c != 0 */
1213 ir_node *div, *proj;
1214 ir_node *a = get_Div_left(n);
1215 ir_node *mem = get_Div_mem(n);
1216 int rem = get_optimize();
1220 div = new_rd_Div(get_irn_dbg_info(n), current_ir_graph,
1221 get_nodes_block(n), get_irg_initial_mem(current_ir_graph), a, b);
1223 proj = new_r_Proj(current_ir_graph, get_nodes_block(n), div, get_irn_mode(a), pn_Div_res);
1227 turn_into_tuple(n, 3);
1228 set_Tuple_pred(n, pn_Mod_M, mem);
1229 set_Tuple_pred(n, pn_Mod_X_except, new_Bad());
1230 set_Tuple_pred(n, pn_Mod_res, proj);
1234 b = get_Mod_right(n);
1235 tb = computed_value(b);
1237 if (tb != tarval_bad && classify_tarval(tb) != TV_CLASSIFY_NULL) { /* mod(x, c) && c != 0 */
1238 ir_node *mod, *proj;
1239 ir_node *a = get_Mod_left(n);
1240 ir_node *mem = get_Mod_mem(n);
1241 int rem = get_optimize();
1245 mod = new_rd_Mod(get_irn_dbg_info(n), current_ir_graph,
1246 get_nodes_block(n), get_irg_initial_mem(current_ir_graph), a, b);
1248 proj = new_r_Proj(current_ir_graph, get_nodes_block(n), mod, get_irn_mode(a), pn_Mod_res);
1252 turn_into_tuple(n, 3);
1253 set_Tuple_pred(n, pn_Mod_M, mem);
1254 set_Tuple_pred(n, pn_Mod_X_except, new_Bad());
1255 set_Tuple_pred(n, pn_Mod_res, proj);
1259 b = get_DivMod_right(n);
1260 tb = computed_value(b);
1262 if (tb != tarval_bad && classify_tarval(tb) != TV_CLASSIFY_NULL) { /* DivMod(x, c) && c != 0 */
1263 ir_node *div_mod, *proj_div, *proj_mod;
1264 ir_node *a = get_Mod_left(n);
1265 ir_node *mem = get_Mod_mem(n);
1266 int rem = get_optimize();
1270 div_mod = new_rd_DivMod(get_irn_dbg_info(n), current_ir_graph,
1271 get_nodes_block(n), get_irg_initial_mem(current_ir_graph), a, b);
1273 proj_div = new_r_Proj(current_ir_graph, get_nodes_block(n), div_mod, get_irn_mode(a), pn_DivMod_res_div);
1274 proj_mod = new_r_Proj(current_ir_graph, get_nodes_block(n), div_mod, get_irn_mode(a), pn_DivMod_res_mod);
1278 turn_into_tuple(n, 4);
1279 set_Tuple_pred(n, pn_DivMod_M, mem);
1280 set_Tuple_pred(n, pn_DivMod_X_except, new_Bad());
1281 set_Tuple_pred(n, pn_DivMod_res_div, proj_div);
1282 set_Tuple_pred(n, pn_DivMod_res_mod, proj_mod);
1287 if (get_opt_unreachable_code()) {
1288 b = get_Cond_selector(n);
1289 tb = computed_value(b);
1291 if (tb != tarval_bad && mode_is_int(get_tarval_mode(tb))) {
1292 /* we have a constant switch */
1293 long num = get_Proj_proj(proj);
1295 if (num != get_Cond_defaultProj(n)) { /* we cannot optimize default Proj's yet */
1296 if (get_tarval_long(tb) == num) {
1297 /* Do NOT create a jump here, or we will have 2 control flow ops
1298 * in a block. This case is optimized away in optimize_cf(). */
1309 /* should not happen, but if it doest will optimize */
1317 /* we have added a Tuple, optimize it for the current Proj away */
1318 return equivalent_node_Proj(proj);
1322 * returns the operands of a commutative bin-op, if one operand is
1323 * a const, it is returned as the second one.
1325 static void get_comm_Binop_Ops(ir_node *binop, ir_node **a, ir_node **c)
1327 ir_node *op_a = get_binop_left(binop);
1328 ir_node *op_b = get_binop_right(binop);
1330 assert(is_op_commutative(get_irn_op(binop)));
1332 if (get_irn_op(op_a) == op_Const) {
1343 * Optimize a Or(And(Or(And(v,c4),c3),c2),c1) pattern if possible.
1344 * Such pattern may arise in bitfield stores.
1346 * value c4 value c4 & c2
1347 * AND c3 AND c1 | c3
1352 static ir_node *transform_node_Or(ir_node *or)
1356 ir_node *and_l, *c3;
1357 ir_node *value, *c4;
1358 ir_node *new_and, *new_const, *block;
1359 ir_mode *mode = get_irn_mode(or);
1361 tarval *tv1, *tv2, *tv3, *tv4, *tv, *n_tv4, *n_tv2;
1363 get_comm_Binop_Ops(or, &and, &c1);
1364 if ((get_irn_op(c1) != op_Const) || (get_irn_op(and) != op_And))
1367 get_comm_Binop_Ops(and, &or_l, &c2);
1368 if ((get_irn_op(c2) != op_Const) || (get_irn_op(or_l) != op_Or))
1371 get_comm_Binop_Ops(or_l, &and_l, &c3);
1372 if ((get_irn_op(c3) != op_Const) || (get_irn_op(and_l) != op_And))
1375 get_comm_Binop_Ops(and_l, &value, &c4);
1376 if (get_irn_op(c4) != op_Const)
1379 /* ok, found the pattern, check for conditions */
1380 assert(mode == get_irn_mode(and));
1381 assert(mode == get_irn_mode(or_l));
1382 assert(mode == get_irn_mode(and_l));
1384 tv1 = get_Const_tarval(c1);
1385 tv2 = get_Const_tarval(c2);
1386 tv3 = get_Const_tarval(c3);
1387 tv4 = get_Const_tarval(c4);
1389 tv = tarval_or(tv4, tv2);
1390 if (classify_tarval(tv) != TV_CLASSIFY_ALL_ONE) {
1391 /* have at least one 0 at the same bit position */
1395 n_tv4 = tarval_not(tv4);
1396 if (tv3 != tarval_and(tv3, n_tv4)) {
1397 /* bit in the or_mask is outside the and_mask */
1401 n_tv2 = tarval_not(tv2);
1402 if (tv1 != tarval_and(tv1, n_tv2)) {
1403 /* bit in the or_mask is outside the and_mask */
1407 /* ok, all conditions met */
1408 block = get_nodes_block(or);
1410 new_and = new_r_And(current_ir_graph, block,
1411 value, new_r_Const(current_ir_graph, block, mode, tarval_and(tv4, tv2)), mode);
1413 new_const = new_r_Const(current_ir_graph, block, mode, tarval_or(tv3, tv1));
1415 set_Or_left(or, new_and);
1416 set_Or_right(or, new_const);
1418 /* check for more */
1419 return transform_node_Or(or);
1423 * Tries several [inplace] [optimizing] transformations and returns an
1424 * equivalent node. The difference to equivalent_node() is that these
1425 * transformations _do_ generate new nodes, and thus the old node must
1426 * not be freed even if the equivalent node isn't the old one.
1428 static ir_node *transform_node(ir_node *n)
1430 if (n->op->transform_node)
1431 n = n->op->transform_node(n);
1436 * set the default transform node operation
1438 static ir_op *firm_set_default_transform_node(ir_op *op)
1442 op->transform_node = transform_node_##a; \
1455 op->transform_node = NULL;
1463 /* **************** Common Subexpression Elimination **************** */
1465 /** The size of the hash table used, should estimate the number of nodes
1467 #define N_IR_NODES 512
1469 static int node_cmp_attr_Const(ir_node *a, ir_node *b)
1471 return (get_Const_tarval(a) != get_Const_tarval(b))
1472 || (get_Const_type(a) != get_Const_type(b));
1475 static int node_cmp_attr_Proj(ir_node *a, ir_node *b)
1477 return get_irn_proj_attr (a) != get_irn_proj_attr (b);
1480 static int node_cmp_attr_Filter(ir_node *a, ir_node *b)
1482 return get_Filter_proj(a) != get_Filter_proj(b);
1485 static int node_cmp_attr_Alloc(ir_node *a, ir_node *b)
1487 return (get_irn_alloc_attr(a).where != get_irn_alloc_attr(b).where)
1488 || (get_irn_alloc_attr(a).type != get_irn_alloc_attr(b).type);
1491 static int node_cmp_attr_Free(ir_node *a, ir_node *b)
1493 return (get_irn_free_attr(a) != get_irn_free_attr(b));
1496 static int node_cmp_attr_SymConst(ir_node *a, ir_node *b)
1498 return (get_irn_symconst_attr(a).num != get_irn_symconst_attr(b).num)
1499 || (get_irn_symconst_attr(a).sym.type_p != get_irn_symconst_attr(b).sym.type_p);
1502 static int node_cmp_attr_Call(ir_node *a, ir_node *b)
1504 return (get_irn_call_attr(a) != get_irn_call_attr(b));
1507 static int node_cmp_attr_FuncCall(ir_node *a, ir_node *b)
1509 return (get_irn_funccall_attr(a) != get_irn_funccall_attr(b));
1512 static int node_cmp_attr_Sel(ir_node *a, ir_node *b)
1514 return (get_irn_sel_attr(a).ent->kind != get_irn_sel_attr(b).ent->kind)
1515 || (get_irn_sel_attr(a).ent->name != get_irn_sel_attr(b).ent->name)
1516 || (get_irn_sel_attr(a).ent->owner != get_irn_sel_attr(b).ent->owner)
1517 || (get_irn_sel_attr(a).ent->ld_name != get_irn_sel_attr(b).ent->ld_name)
1518 || (get_irn_sel_attr(a).ent->type != get_irn_sel_attr(b).ent->type);
1521 static int node_cmp_attr_Phi(ir_node *a, ir_node *b)
1523 return get_irn_phi_attr (a) != get_irn_phi_attr (b);
1526 static int node_cmp_attr_Cast(ir_node *a, ir_node *b)
1528 return get_Cast_type(a) != get_Cast_type(b);
1531 static int node_cmp_attr_Load(ir_node *a, ir_node *b)
1533 if (get_Load_volatility(a) == volatility_is_volatile ||
1534 get_Load_volatility(b) == volatility_is_volatile)
1535 /* NEVER do CSE on volatile Loads */
1538 return get_Load_mode(a) != get_Load_mode(b);
1541 static int node_cmp_attr_Store(ir_node *a, ir_node *b)
1543 /* NEVER do CSE on volatile Stores */
1544 return (get_Store_volatility(a) == volatility_is_volatile ||
1545 get_Load_volatility(b) == volatility_is_volatile);
1549 * set the default node attribute compare operation
1551 static ir_op *firm_set_default_node_cmp_attr(ir_op *op)
1555 op->node_cmp_attr = node_cmp_attr_##a; \
1573 op->node_cmp_attr = NULL;
1581 * Compare function for two nodes in the hash table. Gets two
1582 * nodes as parameters. Returns 0 if the nodes are a cse.
1585 vt_cmp (const void *elt, const void *key)
1593 if (a == b) return 0;
1595 if ((get_irn_op(a) != get_irn_op(b)) ||
1596 (get_irn_mode(a) != get_irn_mode(b))) return 1;
1598 /* compare if a's in and b's in are equal */
1599 irn_arity_a = get_irn_arity (a);
1600 if (irn_arity_a != get_irn_arity(b))
1603 /* for block-local cse and op_pin_state_pinned nodes: */
1604 if (!get_opt_global_cse() || (get_op_pinned(get_irn_op(a)) == op_pin_state_pinned)) {
1605 if (get_irn_n(a, -1) != get_irn_n(b, -1))
1609 /* compare a->in[0..ins] with b->in[0..ins] */
1610 for (i = 0; i < irn_arity_a; i++)
1611 if (get_irn_n(a, i) != get_irn_n(b, i))
1615 * here, we already now that the nodes are identical except their
1618 if (a->op->node_cmp_attr)
1619 return a->op->node_cmp_attr(a, b);
1625 * Calculate a hash value of a node.
1628 ir_node_hash (ir_node *node)
1633 if (node->op == op_Const) {
1634 /* special value for const, as they only differ in their tarval. */
1635 h = ((unsigned) node->attr.con.tv)>>3 ;
1636 h = 9*h + (unsigned)get_irn_mode(node);
1637 } else if (node->op == op_SymConst) {
1638 /* special value for const, as they only differ in their symbol. */
1639 h = ((unsigned) node->attr.i.sym.type_p)>>3 ;
1640 h = 9*h + (unsigned)get_irn_mode(node);
1643 /* hash table value = 9*(9*(9*(9*(9*arity+in[0])+in[1])+ ...)+mode)+code */
1644 h = irn_arity = get_irn_arity(node);
1646 /* consider all in nodes... except the block. */
1647 for (i = 0; i < irn_arity; i++) {
1648 h = 9*h + (unsigned)get_irn_n(node, i);
1652 h = 9*h + (unsigned) get_irn_mode (node);
1654 h = 9*h + (unsigned) get_irn_op (node);
1661 new_identities (void)
1663 return new_pset (vt_cmp, N_IR_NODES);
1667 del_identities (pset *value_table)
1669 del_pset (value_table);
1673 * Return the canonical node computing the same value as n.
1674 * Looks up the node in a hash table.
1676 * For Const nodes this is performed in the constructor, too. Const
1677 * nodes are extremely time critical because of their frequent use in
1678 * constant string arrays.
1680 static INLINE ir_node *
1681 identify (pset *value_table, ir_node *n)
1685 if (!value_table) return n;
1687 /* TODO: use a generic commutative attribute */
1688 if (get_opt_reassociation()) {
1689 if (is_op_commutative(get_irn_op(n))) {
1690 /* for commutative operators perform a OP b == b OP a */
1691 if (get_binop_left(n) > get_binop_right(n)) {
1692 ir_node *h = get_binop_left(n);
1693 set_binop_left(n, get_binop_right(n));
1694 set_binop_right(n, h);
1699 o = pset_find (value_table, n, ir_node_hash (n));
1706 * During construction we set the op_pin_state_pinned flag in the graph right when the
1707 * optimization is performed. The flag turning on procedure global cse could
1708 * be changed between two allocations. This way we are safe.
1710 static INLINE ir_node *
1711 identify_cons (pset *value_table, ir_node *n) {
1713 n = identify(value_table, n);
1714 if (get_irn_n(old, -1) != get_irn_n(n, -1))
1715 set_irg_pinned(current_ir_graph, op_pin_state_floats);
1720 * Return the canonical node computing the same value as n.
1721 * Looks up the node in a hash table, enters it in the table
1722 * if it isn't there yet.
1725 identify_remember (pset *value_table, ir_node *node)
1729 if (!value_table) return node;
1731 /* lookup or insert in hash table with given hash key. */
1732 o = pset_insert (value_table, node, ir_node_hash (node));
1734 if (o == node) return node;
1740 add_identities (pset *value_table, ir_node *node) {
1741 identify_remember (value_table, node);
1745 * garbage in, garbage out. If a node has a dead input, i.e., the
1746 * Bad node is input to the node, return the Bad node.
1748 static INLINE ir_node *
1749 gigo (ir_node *node)
1752 ir_op* op = get_irn_op(node);
1754 /* remove garbage blocks by looking at control flow that leaves the block
1755 and replacing the control flow by Bad. */
1756 if (get_irn_mode(node) == mode_X) {
1757 ir_node *block = get_nodes_block(node);
1758 if (op == op_End) return node; /* Don't optimize End, may have Bads. */
1759 if (get_irn_op(block) == op_Block && get_Block_matured(block)) {
1760 irn_arity = get_irn_arity(block);
1761 for (i = 0; i < irn_arity; i++) {
1762 if (!is_Bad(get_irn_n(block, i))) break;
1764 if (i == irn_arity) return new_Bad();
1768 /* Blocks, Phis and Tuples may have dead inputs, e.g., if one of the
1769 blocks predecessors is dead. */
1770 if ( op != op_Block && op != op_Phi && op != op_Tuple) {
1771 irn_arity = get_irn_arity(node);
1772 for (i = -1; i < irn_arity; i++) {
1773 if (is_Bad(get_irn_n(node, i))) {
1779 /* With this code we violate the agreement that local_optimize
1780 only leaves Bads in Block, Phi and Tuple nodes. */
1781 /* If Block has only Bads as predecessors it's garbage. */
1782 /* If Phi has only Bads as predecessors it's garbage. */
1783 if ((op == op_Block && get_Block_matured(node)) || op == op_Phi) {
1784 irn_arity = get_irn_arity(node);
1785 for (i = 0; i < irn_arity; i++) {
1786 if (!is_Bad(get_irn_n(node, i))) break;
1788 if (i == irn_arity) node = new_Bad();
1796 * These optimizations deallocate nodes from the obstack.
1797 * It can only be called if it is guaranteed that no other nodes
1798 * reference this one, i.e., right after construction of a node.
1801 optimize_node (ir_node *n)
1805 opcode iro = get_irn_opcode(n);
1807 /* Allways optimize Phi nodes: part of the construction. */
1808 if ((!get_opt_optimize()) && (iro != iro_Phi)) return n;
1810 /* constant expression evaluation / constant folding */
1811 if (get_opt_constant_folding()) {
1812 /* constants can not be evaluated */
1813 if (iro != iro_Const) {
1814 /* try to evaluate */
1815 tv = computed_value (n);
1816 if ((get_irn_mode(n) != mode_T) && (tv != tarval_bad)) {
1818 * we MUST copy the node here temporary, because it's still needed
1819 * for DBG_OPT_ALGSIM0
1821 int node_size = offsetof(ir_node, attr) + n->op->attr_size;
1822 ir_node *x = alloca(node_size);
1824 memcpy(x, n, node_size);
1827 /* evaluation was successful -- replace the node. */
1828 obstack_free (current_ir_graph->obst, n);
1829 n = new_Const (get_tarval_mode (tv), tv);
1836 /* remove unnecessary nodes */
1837 if (get_opt_constant_folding() ||
1838 (iro == iro_Phi) || /* always optimize these nodes. */
1840 (iro == iro_Proj) ||
1841 (iro == iro_Block) ) /* Flags tested local. */
1842 n = equivalent_node (n);
1844 optimize_preds(n); /* do node specific optimizations of nodes predecessors. */
1846 /** common subexpression elimination **/
1847 /* Checks whether n is already available. */
1848 /* The block input is used to distinguish different subexpressions. Right
1849 now all nodes are op_pin_state_pinned to blocks, i.e., the cse only finds common
1850 subexpressions within a block. */
1852 n = identify_cons (current_ir_graph->value_table, n);
1855 /* We found an existing, better node, so we can deallocate the old node. */
1856 obstack_free (current_ir_graph->obst, oldn);
1861 /* Some more constant expression evaluation that does not allow to
1863 iro = get_irn_opcode(n);
1864 if (get_opt_constant_folding() ||
1865 (iro == iro_Cond) ||
1866 (iro == iro_Proj)) /* Flags tested local. */
1867 n = transform_node (n);
1869 /* Remove nodes with dead (Bad) input.
1870 Run always for transformation induced Bads. */
1873 /* Now we have a legal, useful node. Enter it in hash table for cse */
1874 if (get_opt_cse() && (get_irn_opcode(n) != iro_Block)) {
1875 n = identify_remember (current_ir_graph->value_table, n);
1883 * These optimizations never deallocate nodes. This can cause dead
1884 * nodes lying on the obstack. Remove these by a dead node elimination,
1885 * i.e., a copying garbage collection.
1888 optimize_in_place_2 (ir_node *n)
1892 opcode iro = get_irn_opcode(n);
1894 if (!get_opt_optimize() && (get_irn_op(n) != op_Phi)) return n;
1896 /* if not optimize return n */
1899 /* Here this is possible. Why? */
1904 /* constant expression evaluation / constant folding */
1905 if (get_opt_constant_folding()) {
1906 /* constants can not be evaluated */
1907 if (iro != iro_Const) {
1908 /* try to evaluate */
1909 tv = computed_value (n);
1910 if ((get_irn_mode(n) != mode_T) && (tv != tarval_bad)) {
1911 /* evaluation was successful -- replace the node. */
1912 n = new_Const (get_tarval_mode (tv), tv);
1919 /* remove unnecessary nodes */
1920 if (get_opt_constant_folding() ||
1921 (iro == iro_Phi) || /* always optimize these nodes. */
1922 (iro == iro_Id) || /* ... */
1923 (iro == iro_Proj) || /* ... */
1924 (iro == iro_Block) ) /* Flags tested local. */
1925 n = equivalent_node (n);
1927 optimize_preds(n); /* do node specific optimizations of nodes predecessors. */
1929 /** common subexpression elimination **/
1930 /* Checks whether n is already available. */
1931 /* The block input is used to distinguish different subexpressions. Right
1932 now all nodes are op_pin_state_pinned to blocks, i.e., the cse only finds common
1933 subexpressions within a block. */
1934 if (get_opt_cse()) {
1935 n = identify (current_ir_graph->value_table, n);
1938 /* Some more constant expression evaluation. */
1939 iro = get_irn_opcode(n);
1940 if (get_opt_constant_folding() ||
1941 (iro == iro_Cond) ||
1942 (iro == iro_Proj)) /* Flags tested local. */
1943 n = transform_node (n);
1945 /* Remove nodes with dead (Bad) input.
1946 Run always for transformation induced Bads. */
1949 /* Now we can verify the node, as it has no dead inputs any more. */
1952 /* Now we have a legal, useful node. Enter it in hash table for cse.
1953 Blocks should be unique anyways. (Except the successor of start:
1954 is cse with the start block!) */
1955 if (get_opt_cse() && (get_irn_opcode(n) != iro_Block))
1956 n = identify_remember (current_ir_graph->value_table, n);
1962 * Wrapper for external use, set proper status bits after optimization.
1965 optimize_in_place (ir_node *n)
1967 /* Handle graph state */
1968 assert(get_irg_phase_state(current_ir_graph) != phase_building);
1970 if (get_opt_global_cse())
1971 set_irg_pinned(current_ir_graph, op_pin_state_floats);
1972 if (get_irg_outs_state(current_ir_graph) == outs_consistent)
1973 set_irg_outs_inconsistent(current_ir_graph);
1975 /* Maybe we could also test whether optimizing the node can
1976 change the control graph. */
1977 if (get_irg_dom_state(current_ir_graph) == dom_consistent)
1978 set_irg_dom_inconsistent(current_ir_graph);
1979 return optimize_in_place_2 (n);
1983 * set the default ir op operations
1985 ir_op *firm_set_default_operations(ir_op *op)
1987 op = firm_set_default_computed_value(op);
1988 op = firm_set_default_equivalent_node(op);
1989 op = firm_set_default_transform_node(op);
1990 op = firm_set_default_node_cmp_attr(op);