1 /* Copyright (C) 1998 - 2000 by Universitaet Karlsruhe
2 ** All rights reserved.
4 ** Authors: Christian Schaefer, Goetz Lindenmaier
6 ** iropt --- optimizations intertwined with IR construction.
10 # include "irgraph_t.h"
19 /* Make types visible to allow most efficient access */
20 # include "entity_t.h"
22 /* Trivial inlineable routine for copy propagation.
23 Does follow Ids, needed to optimize inlined code. */
24 static inline ir_node *
25 follow_Id (ir_node *n)
27 while (get_irn_op (n) == op_Id) n = get_Id_pred (n);
31 static inline tarval *
34 if ((n != NULL) && (get_irn_op(n) == op_Const))
35 return get_Const_tarval(n);
40 /* if n can be computed, return the value, else NULL. Performs
41 constant folding. GL: Only if n is arithmetic operator? */
43 computed_value (ir_node *n)
47 ir_node *a = NULL, *b = NULL; /* initialized to shut up gcc */
48 tarval *ta = NULL, *tb = NULL; /* initialized to shut up gcc */
52 /* get the operands we will work on for simple cases. */
54 a = get_binop_left(n);
55 b = get_binop_right(n);
56 } else if (is_unop(n)) {
60 /* if the operands are constants, get the target value, else set it NULL.
61 (a and b may be NULL if we treat a node that is no computation.) */
65 /* Perform the constant evaluation / computation. */
66 switch (get_irn_opcode(n)) {
68 res = get_Const_tarval(n);
70 if (ta && tb && (get_irn_mode(a) == get_irn_mode(b))
71 && (get_irn_mode(a) != mode_p)) {
72 res = tarval_add (ta, tb);
76 if (ta && tb && (get_irn_mode(a) == get_irn_mode(b))
77 && (get_irn_mode(a) != mode_p)) {
78 res = tarval_sub (ta, tb);
80 res = tarval_mode_null [get_irn_modecode (n)];
84 if (ta && mode_is_float(get_irn_mode(a)))
85 res = /*tarval_minus (ta);*/ res;
88 if (ta && tb) /* tarval_mul tests for equivalent modes itself */ {
89 res = tarval_mul (ta, tb);
91 /* calls computed_value recursive and returns the 0 with proper
92 mode. Why is this an extra case? */
94 if ( (tarval_classify ((v = computed_value (a))) == 0)
95 || (tarval_classify ((v = computed_value (b))) == 0)) {
102 res = /*tarval_abs (ta);*/ res;
103 /* allowed or problems with max/min ?? */
107 res = tarval_and (ta, tb);
110 if ( (tarval_classify ((v = computed_value (a))) == 0)
111 || (tarval_classify ((v = computed_value (b))) == 0)) {
118 res = tarval_or (ta, tb);
121 if ( (tarval_classify ((v = computed_value (a))) == -1)
122 || (tarval_classify ((v = computed_value (b))) == -1)) {
127 case iro_Eor: if (ta && tb) { res = tarval_eor (ta, tb); } break;
128 case iro_Not: if (ta) { /*res = tarval_not (ta)*/; } break;
129 case iro_Shl: if (ta && tb) { res = tarval_shl (ta, tb); } break;
130 case iro_Shr: if (ta && tb) { res = tarval_shr (ta, tb); } break;
131 case iro_Shrs: if(ta && tb) { /*res = tarval_shrs (ta, tb)*/; } break;
132 case iro_Rot: if (ta && tb) { /*res = tarval_rot (ta, tb)*/; } break;
133 case iro_Conv: if(ta) { res = tarval_convert_to (ta, get_irn_mode (n)); }
139 a = get_Proj_pred(n);
140 /* Optimize Cmp nodes.
141 This performs a first step of unreachable code elimination.
142 Proj can not be computed, but folding a Cmp above the Proj here is
143 not as wasteful as folding a Cmp into a Tuple of 16 Consts of which
145 There are several case where we can evaluate a Cmp node:
146 1. The nodes compared are both the same. If we compare for
147 equal, this will return true, else it will return false.
148 This step relies on cse.
149 2. The predecessors of Cmp are target values. We can evaluate
151 3. The predecessors are Allocs or void* constants. Allocs never
152 return NULL, they raise an exception. Therefore we can predict
154 if (get_irn_op(a) == op_Cmp) {
155 aa = get_Cmp_left(a);
156 ab = get_Cmp_right(a);
157 if (aa == ab) { /* 1.: */
158 /* This is a tric with the bits used for encoding the Cmp
159 Proj numbers, the following statement is not the same:
160 res = tarval_from_long (mode_b, (get_Proj_proj(n) == Eq)): */
161 res = tarval_from_long (mode_b, (get_Proj_proj(n) & irpn_Eq));
163 tarval *taa = computed_value (aa);
164 tarval *tab = computed_value (ab);
165 if (taa && tab) { /* 2.: */
166 /* strange checks... */
167 ir_pncmp flags = tarval_comp (taa, tab);
168 if (flags != irpn_False) {
169 res = tarval_from_long (mode_b, get_Proj_proj(n) & flags);
171 } else { /* check for 3.: */
172 ir_node *aaa = skip_nop(skip_Proj(aa));
173 ir_node *aba = skip_nop(skip_Proj(ab));
174 if ( ( (/* aa is ProjP and aaa is Alloc */
175 (get_irn_op(aa) == op_Proj)
176 && (get_irn_mode(aa) == mode_p)
177 && (get_irn_op(aaa) == op_Alloc))
178 && ( (/* ab is constant void */
179 (get_irn_op(ab) == op_Const)
180 && (get_irn_mode(ab) == mode_p)
181 && (get_Const_tarval(ab) == tarval_p_void))
182 || (/* ab is other Alloc */
183 (get_irn_op(ab) == op_Proj)
184 && (get_irn_mode(ab) == mode_p)
185 && (get_irn_op(aba) == op_Alloc)
187 || (/* aa is void and aba is Alloc */
188 (get_irn_op(aa) == op_Const)
189 && (get_irn_mode(aa) == mode_p)
190 && (get_Const_tarval(aa) == tarval_p_void)
191 && (get_irn_op(ab) == op_Proj)
192 && (get_irn_mode(ab) == mode_p)
193 && (get_irn_op(aba) == op_Alloc)))
195 res = tarval_from_long (mode_b, get_Proj_proj(n) & irpn_Ne);
199 /* printf(" # comp_val: Proj node, not optimized\n"); */
211 /* returns 1 if the a and b are pointers to different locations. */
213 different_identity (ir_node *a, ir_node *b)
215 assert (get_irn_mode (a) == mode_p
216 && get_irn_mode (b) == mode_p);
218 if (get_irn_op (a) == op_Proj && get_irn_op(b) == op_Proj) {
219 ir_node *a1 = get_Proj_pred (a);
220 ir_node *b1 = get_Proj_pred (b);
221 if (a1 != b1 && get_irn_op (a1) == op_Alloc
222 && get_irn_op (b1) == op_Alloc)
229 /* equivalent_node returns a node equivalent to N. It skips all nodes that
230 perform no actual computation, as, e.g., the Id nodes. It does not create
231 new nodes. It is therefore safe to free N if the node returned is not N.
232 If a node returns a Tuple we can not just skip it. If the size of the
233 in array fits, we transform n into a tuple (e.g., Div). */
235 equivalent_node (ir_node *n)
238 ir_node *a = NULL; /* to shutup gcc */
239 ir_node *b = NULL; /* to shutup gcc */
240 ir_node *c = NULL; /* to shutup gcc */
242 ins = get_irn_arity (n);
244 /* get the operands we will work on */
246 a = get_binop_left(n);
247 b = get_binop_right(n);
248 } else if (is_unop(n)) {
252 /* skip unnecessary nodes. */
253 switch (get_irn_opcode (n)) {
256 /* The Block constructor does not call optimize, but mature_block
257 calls the optimization. */
258 assert(get_Block_matured(n));
260 /* A single entry Block following a single exit Block can be merged,
261 if it is not the Start block. */
262 /* !!! Beware, all Phi-nodes of n must have been optimized away.
263 This is true, as the block is matured before optimize is called. */
264 if (get_Block_n_cfgpreds(n) == 1
265 && get_irn_op(get_Block_cfgpred(n, 0)) == op_Jmp) {
266 n = get_nodes_Block(get_Block_cfgpred(n, 0));
268 } else if (n != current_ir_graph->start_block) {
270 /* If all inputs are dead, this block is dead too, except if it is
271 the start block. This is a step of unreachable code elimination */
272 for (i = 0; i < get_Block_n_cfgpreds(n); i++) {
273 if (!is_Bad(get_Block_cfgpred(n, i))) break;
275 if (i == get_Block_n_cfgpreds(n))
281 case iro_Jmp: /* GL: Why not same for op_Raise?? */
282 /* unreachable code elimination */
283 if (is_Bad(get_nodes_Block(n))) n = new_Bad();
285 /* We do not evaluate Cond here as we replace it by a new node, a Jmp.
286 See cases for iro_Cond and iro_Proj in transform_node. */
287 /** remove stuff as x+0, x*1 x&true ... constant expression evaluation **/
288 case iro_Or: if (a == b) {n = a; break;}
293 /* After running compute_node there is only one constant predecessor.
294 Find this predecessors value and remember the other node: */
295 if ((tv = computed_value (a))) {
297 } else if ((tv = computed_value (b))) {
301 /* If this predecessors constant value is zero, the operation is
302 unnecessary. Remove it: */
303 if (tarval_classify (tv) == 0) {
313 /* these operations are not commutative. Test only one predecessor. */
314 if (tarval_classify (computed_value (b)) == 0) {
316 /* Test if b > #bits of a ==> return 0 / divide b by #bits
317 --> transform node? */
320 case iro_Not: /* NotNot x == x */
321 case iro_Minus: /* --x == x */ /* ??? Is this possible or can --x raise an
322 out of bounds exception if min =! max? */
323 if (get_irn_op(get_unop_op(n)) == get_irn_op(n))
324 n = get_unop_op(get_unop_op(n));
327 /* Mul is commutative and has again an other neutral element. */
328 if (tarval_classify (computed_value (a)) == 1) {
330 } else if (tarval_classify (computed_value (b)) == 1) {
335 /* Div is not commutative. */
336 if (tarval_classify (computed_value (b)) == 1) { /* div(x, 1) == x */
337 /* Turn Div into a tuple (mem, bad, a) */
338 ir_node *mem = get_Div_mem(n);
339 turn_into_tuple(n, 3);
340 set_Tuple_pred(n, 0, mem);
341 set_Tuple_pred(n, 1, new_Bad());
342 set_Tuple_pred(n, 2, a);
345 /* GL: Why are they skipped? DivMod allocates new nodes --> it's
346 teated in transform node.
347 case iro_Mod, Quot, DivMod
351 /* And has it's own neutral element */
352 else if (tarval_classify (computed_value (a)) == -1) {
354 } else if (tarval_classify (computed_value (b)) == -1) {
359 if (get_irn_mode(n) == get_irn_mode(a)) { /* No Conv necessary */
361 } else if (get_irn_mode(n) == mode_b) {
362 if (get_irn_op(a) == op_Conv &&
363 get_irn_mode (get_Conv_op(a)) == mode_b) {
364 n = get_Conv_op(a); /* Convb(Conv*(xxxb(...))) == xxxb(...) */
371 /* Several optimizations:
372 - no Phi in start block.
373 - remove Id operators that are inputs to Phi
374 - fold Phi-nodes, iff they have only one predecessor except
378 ir_node *block = NULL; /* to shutup gcc */
379 ir_node *first_val = NULL; /* to shutup gcc */
380 ir_node *scnd_val = NULL; /* to shutup gcc */
382 n_preds = get_Phi_n_preds(n);
384 block = get_nodes_Block(n);
385 assert(get_irn_op (block) == op_Block);
387 /* there should be no Phi nodes in the Start region. */
388 if (block == current_ir_graph->start_block) {
393 if (n_preds == 0) { /* Phi of dead Region without predecessors. */
394 /* GL: why not return new_Bad? */
399 /* first we test for a special case: */
400 /* Confirm is a special node fixing additional information for a
401 value that is known at a certain point. This is useful for
402 dataflow analysis. */
404 ir_node *a = follow_Id (get_Phi_pred(n, 0));
405 ir_node *b = follow_Id (get_Phi_pred(n, 1));
406 if ( (get_irn_op(a) == op_Confirm)
407 && (get_irn_op(b) == op_Confirm)
408 && (follow_Id (get_irn_n(a, 0)) == follow_Id(get_irn_n(b, 0)))
409 && (get_irn_n(a, 1) == get_irn_n (b, 1))
410 && (a->data.num == (~b->data.num & irpn_True) )) {
411 n = follow_Id (get_irn_n(a, 0));
416 /* Find first non-self-referencing input */
417 for (i = 0; i < n_preds; ++i) {
418 first_val = follow_Id(get_Phi_pred(n, i));
420 set_Phi_pred(n, i, first_val);
421 if ( (first_val != n) /* not self pointer */
422 && (get_irn_op(first_val) != op_Bad) /* value not dead */
423 && !(is_Bad (get_Block_cfgpred(block, i))) ) { /* not dead control flow */
424 break; /* then found first value. */
428 /* A totally Bad or self-referencing Phi (we didn't break the above loop) */
429 if (i > n_preds) { n = new_Bad(); break; }
433 /* follow_Id () for rest of inputs, determine if any of these
434 are non-self-referencing */
435 while (++i < n_preds) {
436 scnd_val = follow_Id(get_Phi_pred(n, i));
438 set_Phi_pred(n, i, scnd_val);
440 && (scnd_val != first_val)
441 && (get_irn_op(scnd_val) != op_Bad)
442 && !(is_Bad (get_Block_cfgpred(block, i))) ) {
447 /* Fold, if no multiple distinct non-self-referencing inputs */
451 /* skip the remaining Ids. */
452 while (++i < n_preds) {
453 set_Phi_pred(n, i, follow_Id(get_Phi_pred(n, i)));
461 a = skip_Proj(get_Load_mem(n));
462 b = skip_Proj(get_Load_ptr(n));
464 if (get_irn_op(a) == op_Store) {
465 if ( different_identity (b, get_Store_ptr(a))) {
466 /* load and store use different pointers, therefore load
467 needs not take store's memory but the state before. */
468 set_Load_mem (n, get_Store_mem(a));
469 } else if (( 0 /* ???didn't get cryptic test that returns 0 */ )) {
475 /* remove unnecessary store. */
477 a = skip_Proj(get_Store_mem(n));
478 b = get_Store_ptr(n);
479 c = skip_Proj(get_Store_value(n));
481 if (get_irn_op(a) == op_Store
482 && get_Store_ptr(a) == b
483 && skip_Proj(get_Store_value(a)) == c) {
484 /* We have twice exactly the same store -- a write after write. */
486 } else if (get_irn_op(c) == op_Load
487 && (a == c || skip_Proj(get_Load_mem(c)) == a)
488 && get_Load_ptr(c) == b )
489 /* !!!??? and a cryptic test */ {
490 /* We just loaded the value from the same memory, i.e., the store
491 doesn't change the memory -- a write after read. */
492 turn_into_tuple(n, 2);
493 set_Tuple_pred(n, 0, a);
494 set_Tuple_pred(n, 1, new_Bad());
501 a = get_Proj_pred(n);
503 if ( get_irn_op(a) == op_Tuple) {
504 /* Remove the Tuple/Proj combination. */
505 if ( get_Proj_proj(n) <= get_Tuple_n_preds(a) ) {
506 n = get_Tuple_pred(a, get_Proj_proj(n));
508 assert(0); /* This should not happen! (GL added this assert) */
511 } else if (get_irn_mode(n) == mode_X &&
512 is_Bad(get_nodes_Block(n))) {
513 /* Remove dead control flow. */
527 } /* end equivalent_node() */
530 /* tries several [inplace] [optimizing] transformations and returns a
531 equivalent node. The difference to equivalent_node is that these
532 transformations _do_ generate new nodes, and thus the old node must
533 not be freed even if the equivalent node isn't the old one. */
535 transform_node (ir_node *n)
538 ir_node *a = NULL, *b;
541 switch (get_irn_opcode(n)) {
547 a = get_DivMod_left(n);
548 b = get_DivMod_right(n);
549 mode = get_irn_mode(a);
551 if (!( mode_is_int(get_irn_mode(a))
552 && mode_is_int(get_irn_mode(b))))
556 a = new_Const (mode, tarval_from_long (mode, 1));
557 b = new_Const (mode, tarval_from_long (mode, 0));
564 if (tarval_classify(tb) == 1) {
565 b = new_Const (mode, tarval_from_long (mode, 0));
569 resa = tarval_div (ta, tb);
570 if (!resa) break; /* Causes exception!!! Model by replacing through
571 Jmp for X result!? */
572 resb = tarval_mod (ta, tb);
573 if (!resb) break; /* Causes exception! */
574 a = new_Const (mode, resa);
575 b = new_Const (mode, resb);
578 } else if (tarval_classify (ta) == 0) {
583 if (evaluated) { /* replace by tuple */
584 ir_node *mem = get_DivMod_mem(n);
585 turn_into_tuple(n, 4);
586 set_Tuple_pred(n, 0, mem);
587 set_Tuple_pred(n, 1, new_Bad()); /* no exception */
588 set_Tuple_pred(n, 2, a);
589 set_Tuple_pred(n, 3, b);
590 assert(get_nodes_Block(n));
596 /* Replace the Cond by a Jmp if it branches on a constant
599 a = get_Cond_selector(n);
602 if (ta && (get_irn_mode(a) == mode_b)) {
603 /* It's a boolean Cond, branching on a boolean constant.
604 Replace it by a tuple (Bad, Jmp) or (Jmp, Bad) */
605 jmp = new_r_Jmp(current_ir_graph, get_nodes_Block(n));
606 turn_into_tuple(n, 2);
607 if (tv_val_b(ta) == 1) /* GL: I hope this returns 1 if true */ {
608 set_Tuple_pred(n, 0, new_Bad());
609 set_Tuple_pred(n, 1, jmp);
611 set_Tuple_pred(n, 0, jmp);
612 set_Tuple_pred(n, 1, new_Bad());
614 } else if (ta && (get_irn_mode(a) == mode_I)) {
615 /* I don't want to allow Tuples smaller than the biggest Proj.
616 Also this tuple might get really big...
617 I generate the Jmp here, and remember it in link. Link is used
618 when optimizing Proj. */
619 set_irn_link(n, new_r_Jmp(current_ir_graph, get_nodes_Block(n)));
620 } else if ( (get_irn_op(get_Cond_selector(n)) == op_Eor)
621 && (get_irn_mode(get_Cond_selector(n)) == mode_b)
622 && (tarval_classify(computed_value(get_Eor_right(a))) == 1)) {
623 /* The Eor is a negate. Generate a new Cond without the negate,
624 simulate the negate by exchanging the results. */
625 set_irn_link(n, new_r_Cond(current_ir_graph, get_nodes_Block(n),
627 } else if ( (get_irn_op(get_Cond_selector(n)) == op_Not)
628 && (get_irn_mode(get_Cond_selector(n)) == mode_b)) {
629 /* A Not before the Cond. Generate a new Cond without the Not,
630 simulate the Not by exchanging the results. */
631 set_irn_link(n, new_r_Cond(current_ir_graph, get_nodes_Block(n),
638 a = get_Proj_pred(n);
640 if ( (get_irn_op(a) == op_Cond)
642 && get_irn_op(get_irn_link(a)) == op_Cond) {
643 /* Use the better Cond if the Proj projs from a Cond which get's
644 its result from an Eor/Not. */
645 assert ( ( (get_irn_op(get_Cond_selector(a)) == op_Eor)
646 || (get_irn_op(get_Cond_selector(a)) == op_Not))
647 && (get_irn_mode(get_Cond_selector(a)) == mode_b)
648 && (get_irn_op(get_irn_link(a)) == op_Cond)
649 && (get_Cond_selector(get_irn_link(a)) ==
650 get_Eor_left(get_Cond_selector(a))));
651 set_Proj_pred(n, get_irn_link(a));
652 if (get_Proj_proj(n) == 0)
656 } else if ( (get_irn_op(a) == op_Cond)
657 && (get_irn_mode(get_Cond_selector(a)) == mode_I)
659 /* The Cond is a Switch on a Constant */
660 if (get_Proj_proj(n) == tv_val_CHIL(value_of(a))) {
661 /* The always taken branch, reuse the existing Jmp. */
662 if (!get_irn_link(a)) /* well, if it exists ;-> */
663 set_irn_link(a, new_r_Jmp(current_ir_graph, get_nodes_Block(n)));
664 assert(get_irn_op(get_irn_link(a)) == op_Jmp);
667 /* a never taken branch */
673 case iro_Eor: { /* @@@ not tested as boolean Eor not allowed any more. */
675 b = get_Eor_right(n);
677 if ( (get_irn_mode(n) == mode_b)
678 && (get_irn_op(a) == op_Proj)
679 && (get_irn_mode(a) == mode_b)
680 && (tarval_classify (computed_value (b)) == 1)
681 && (get_irn_op(get_Proj_pred(a)) == op_Cmp))
682 /* The Eor negates a Cmp. The Cmp has the negated result anyways! */
683 n = new_r_Proj(current_ir_graph, get_nodes_Block(n), get_Proj_pred(a),
684 mode_b, get_negated_pnc(get_Proj_proj(a)));
685 else if ( (get_irn_mode(n) == mode_b)
686 && (tarval_classify (computed_value (b)) == 1))
687 /* The Eor is a Not. Replace it by a Not. */
688 /* ????!!!Extend to bitfield 1111111. */
689 n = new_r_Not(current_ir_graph, get_nodes_Block(n), a, mode_b);
695 if ( (get_irn_mode(n) == mode_b)
696 && (get_irn_op(a) == op_Proj)
697 && (get_irn_mode(a) == mode_b)
698 && (get_irn_op(get_Proj_pred(a)) == op_Cmp))
699 /* We negate a Cmp. The Cmp has the negated result anyways! */
700 n = new_r_Proj(current_ir_graph, get_nodes_Block(n), get_Proj_pred(a),
701 mode_b, get_negated_pnc(get_Proj_proj(a)));
709 /***************** Common Subexpression Elimination *****************/
711 /* Compare function for two nodes in the hash table. Gets two */
712 /* nodes as parameters. */
713 /* @@@ a+b != b+a ? */
715 vt_cmp (const void *elt, const void *key)
723 if (a == b) return 0;
725 if ((get_irn_op(a) != get_irn_op(b)) ||
726 (get_irn_mode(a) != get_irn_mode(b))) return 1;
728 /* compare if a's in and b's in are equal */
729 /* GL: we optimize only nodes with in arrays of fixed sizes.
730 if (get_irn_arity (a) != -2) {
731 ins = get_irn_arity (a);
732 if (ins != get_irn_arity (b)) return 1;
733 ain = get_irn_in (a);
734 bin = get_irn_in (b);
737 if (get_irn_arity (a) != get_irn_arity(b))
740 /* compare a->in[0..ins] with b->in[0..ins], i.e., include the block. */
741 /* do if (*ain++ != *bin++) return 1; while (ins--); */
742 for (i = -1; i < get_irn_arity(a); i++)
743 if (get_irn_n(a, i) != get_irn_n(b, i))
747 switch (get_irn_opcode(a)) {
749 return get_irn_const_attr (a) != get_irn_const_attr (b);
751 return get_irn_proj_attr (a) != get_irn_proj_attr (b);
753 return (get_irn_alloc_attr(a).where != get_irn_alloc_attr(b).where)
754 || (get_irn_alloc_attr(a).type != get_irn_alloc_attr(b).type);
756 return (get_irn_free_attr(a) != get_irn_free_attr(b));
758 return (get_irn_symconst_attr(a).num != get_irn_symconst_attr(b).num)
759 || (get_irn_symconst_attr(a).tori.typ != get_irn_symconst_attr(b).tori.typ);
761 return (get_irn_call_attr(a)->kind != get_irn_call_attr(b)->kind)
762 || (get_irn_call_attr(a)->arity != get_irn_call_attr(b)->arity);
764 return (get_irn_sel_attr(a).ent->kind != get_irn_sel_attr(b).ent->kind)
765 || (get_irn_sel_attr(a).ent->name != get_irn_sel_attr(b).ent->name)
766 || (get_irn_sel_attr(a).ent->owner != get_irn_sel_attr(b).ent->owner)
767 || (get_irn_sel_attr(a).ent->ld_name != get_irn_sel_attr(b).ent->ld_name)
768 || (get_irn_sel_attr(a).ent->type != get_irn_sel_attr(b).ent->type)
769 || (get_irn_sel_attr(a).ltyp != get_irn_sel_attr(b).ltyp);
771 return get_irn_phi_attr (a) != get_irn_phi_attr (b);
779 ir_node_hash (ir_node *node)
784 /* hash table value = 9*(9*(9*(9*(9*arity+in[0])+in[1])+ ...)+mode)+code */
785 h = get_irn_arity(node);
787 /* consider all in nodes... except the block. */
788 for (i = 0; i < get_irn_arity(node); i++) {
789 h = 9*h + (unsigned long)get_irn_n(node, i);
793 h = 9*h + (unsigned long) get_irn_mode (node);
795 h = 9*h + (unsigned long) get_irn_op (node);
801 new_identities (void)
803 return new_pset (vt_cmp, TUNE_NIR_NODES);
807 del_identities (pset *value_table)
809 del_pset (value_table);
812 /* Return the canonical node computing the same value as n.
813 Looks up the node in a hash table. */
814 static inline ir_node *
815 identify (pset *value_table, ir_node *n)
819 if (!value_table) return n;
821 switch (get_irn_opcode (n)) {
828 /* for commutative operators perform a OP b == b OP a */
829 if (get_binop_left(n) > get_binop_right(n)) {
830 ir_node *h = get_binop_left(n);
831 set_binop_left(n, get_binop_right(n));
832 set_binop_right(n, h);
838 o = pset_find (value_table, n, ir_node_hash (n));
844 /* Return the canonical node computing the same value as n.
845 Looks up the node in a hash table, enters it in the table
846 if it isn't there yet. */
848 identify_remember (pset *value_table, ir_node *node)
852 if (!value_table) return node;
854 /* lookup or insert in hash table with given hash key. */
855 o = pset_insert (value_table, node, ir_node_hash (node));
857 if (o == node) return node;
862 /* garbage in, garbage out. If a node has a dead input, i.e., the
863 Bad node is input to the node, return the Bad node. */
864 static inline ir_node *
868 ir_op* op = get_irn_op(node);
870 /* Blocks, Phis and Tuples may have dead inputs, e.g., if one of the
871 blocks predecessors is dead. */
872 if ( op != op_Block && op != op_Phi && op != op_Tuple) {
873 for (i = -1; i < get_irn_arity(node); i++) {
874 if (is_Bad(get_irn_n(node, i))) {
884 /* These optimizations deallocate nodes from the obstack.
885 It can only be called if it is guaranteed that no other nodes
886 reference this one, i.e., right after construction of a node. */
888 optimize (ir_node *n)
893 /* Allways optimize Phi nodes: part of the construction. */
894 if ((!get_optimize()) && (get_irn_op(n) != op_Phi)) return n;
896 /* if not optimize return n */
898 printf(" attention: empty node!!! \n");
902 /* constant expression evaluation / constant folding */
903 if (get_opt_constant_folding()) {
904 /* constants can not be evaluated */
905 if (get_irn_op(n) != op_Const) {
906 /* try to evaluate */
907 tv = computed_value (n);
909 /* evaluation was succesful -- replace the node. */
910 obstack_free (current_ir_graph->obst, n);
911 return new_Const (get_tv_mode (tv), tv);
916 /* remove unnecessary nodes */
917 if (get_opt_constant_folding() || get_irn_op(n) == op_Phi)
918 n = equivalent_node (n);
920 /** common subexpression elimination **/
921 /* Checks whether n is already available. */
922 /* The block input is used to distinguish different subexpressions. Right
923 now all nodes are pinned to blocks, i.e., the cse only finds common
924 subexpressions within a block. */
926 n = identify (current_ir_graph->value_table, n);
927 /* identify found a cse, so deallocate the old node. */
929 obstack_free (current_ir_graph->obst, old_n);
930 /* The AmRoq fiasco returns n here. Martin's version doesn't. */
933 /* Some more constant expression evaluation that does not allow to
935 if (get_opt_constant_folding())
936 n = transform_node (n);
938 /* Remove nodes with dead (Bad) input. */
940 /* Now we can verify the node, as it has no dead inputs any more. */
943 /* Now we have a legal, useful node. Enter it in hash table for cse */
945 n = identify_remember (current_ir_graph->value_table, n);
948 #if 0 /* GL: what's the use of this?? */
949 if ((current_ir_graph->state & irgs_building) && IR_KEEP_ALIVE (n)) {
950 assert (~current_ir_graph->state & irgs_keep_alives_in_arr);
951 pdeq_putr (current_ir_graph->keep.living, n);
958 /* These optimizations never deallocate nodes. This can cause dead
959 nodes lying on the obstack. Remove these by a dead node elimination,
960 i.e., a copying garbage collection. */
962 optimize_in_place (ir_node *n)
968 /* if not optimize return n */
970 /* Here this is possible. Why? */
974 /* constant expression evaluation / constant folding */
975 if (get_opt_constant_folding()) {
976 /* constants can not be evaluated */
977 if (get_irn_op(n) != op_Const) {
978 /* try to evaluate */
979 tv = computed_value (n);
981 /* evaluation was succesful -- replace the node. */
982 return new_Const (get_tv_mode (tv), tv);
983 /* xprintf("* optimize: computed node %I\n", n->op->name);*/
988 /* remove unnecessary nodes */
989 if (get_opt_constant_folding())
990 // if (get_opt_constant_folding() || get_irn_op(n) == op_Phi)
991 n = equivalent_node (n);
993 /** common subexpression elimination **/
994 /* Checks whether n is already available. */
995 /* The block input is used to distinguish different subexpressions. Right
996 now all nodes are pinned to blocks, i.e., the cse only finds common
997 subexpressions within a block. */
999 n = identify (current_ir_graph->value_table, n);
1001 /* identify found a cse, so deallocate the old node. */
1003 /* The AmRoq fiasco returns n here. Martin's version doesn't. */
1006 /* Some more constant expression evaluation. */
1007 if (get_opt_constant_folding())
1008 n = transform_node (n);
1010 /* Remove nodes with dead (Bad) input. */
1012 /* Now we can verify the node, as it has no dead inputs any more. */
1015 /* Now we have a legal, useful node. Enter it in hash table for cse */
1016 if (get_opt_cse()) {
1017 /* aborts ??! set/pset can not handle several hash tables??!
1018 No, suddenly it works. */
1019 n = identify_remember (current_ir_graph->value_table, n);