/* SSA-PRE for trees. Copyright (C) 2001, 2002, 2003, 2004 Free Software Foundation, Inc. Contributed by Daniel Berlin and Steven Bosscher This file is part of GCC. GCC is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2, or (at your option) any later version. GCC is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with GCC; see the file COPYING. If not, write to the Free Software Foundation, 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */ #include "config.h" #include "system.h" #include "coretypes.h" #include "tm.h" #include "errors.h" #include "ggc.h" #include "tree.h" #include "basic-block.h" #include "diagnostic.h" #include "tree-inline.h" #include "tree-flow.h" #include "tree-gimple.h" #include "tree-dump.h" #include "timevar.h" #include "fibheap.h" #include "hashtab.h" #include "tree-iterator.h" #include "real.h" #include "alloc-pool.h" #include "tree-pass.h" #include "flags.h" #include "splay-tree.h" #include "bitmap.h" #include "langhooks.h" /* TODO: 1. Avail sets can be shared by making an avail_find_leader that walks up the dominator tree and looks in those avail sets. This might affect code optimality, it's unclear right now. 2. Load motion can be performed by value numbering the loads the same as we do other expressions. This requires iterative hashing the vuses into the values. Right now we simply assign a new value every time we see a statement with a vuse. 3. Strength reduction can be performed by anticipating expressions we can repair later on. 4. Our canonicalization of expressions during lookups don't take constants into account very well. In particular, we don't fold anywhere, so we can get situations where we stupidly think something is a new value (a + 1 + 1 vs a + 2). This is somewhat expensive to fix, but it does expose a lot more eliminations. It may or not be worth it, depending on how critical you consider PRE vs just plain GRE. */ /* For ease of terminology, "expression node" in the below refers to every expression node but MODIFY_EXPR, because MODIFY_EXPR's represent the actual statement containing the expressions we care about, and we cache the value number by putting it in the expression. */ /* Basic algorithm First we walk the statements to generate the AVAIL sets, the EXP_GEN sets, and the tmp_gen sets. EXP_GEN sets represent the generation of values/expressions by a given block. We use them when computing the ANTIC sets. The AVAIL sets consist of SSA_NAME's that represent values, so we know what values are available in what blocks. AVAIL is a forward dataflow problem. In SSA, values are never killed, so we don't need a kill set, or a fixpoint iteration, in order to calculate the AVAIL sets. In traditional parlance, AVAIL sets tell us the downsafety of the expressions/values. Next, we generate the ANTIC sets. These sets represent the anticipatable expressions. ANTIC is a backwards dataflow problem.An expression is anticipatable in a given block if it could be generated in that block. This means that if we had to perform an insertion in that block, of the value of that expression, we could. Calculating the ANTIC sets requires phi translation of expressions, because the flow goes backwards through phis. We must iterate to a fixpoint of the ANTIC sets, because we have a kill set. Even in SSA form, values are not live over the entire function, only from their definition point onwards. So we have to remove values from the ANTIC set once we go past the definition point of the leaders that make them up. compute_antic/compute_antic_aux performs this computation. Third, we perform insertions to make partially redundant expressions fully redundant. An expression is partially redundant (excluding partial anticipation) if: 1. It is AVAIL in some, but not all, of the predecessors of a given block. 2. It is ANTIC in all the predecessors. In order to make it fully redundant, we insert the expression into the predecessors where it is not available, but is ANTIC. insert/insert_aux performs this insertion. Fourth, we eliminate fully redundant expressions. This is a simple statement walk that replaces redundant calculations with the now available values. */ /* Representations of value numbers: Value numbers are represented using the "value handle" approach. This means that each SSA_NAME (and for other reasons to be disclosed in a moment, expression nodes) has a value handle that can be retrieved through get_value_handle. This value handle, *is* the value number of the SSA_NAME. You can pointer compare the value handles for equivalence purposes. For debugging reasons, the value handle is internally more than just a number, it is a VAR_DECL named "value.x", where x is a unique number for each value number in use. This allows expressions with SSA_NAMES replaced by value handles to still be pretty printed in a sane way. They simply print as "value.3 * value.5", etc. Expression nodes have value handles associated with them as a cache. Otherwise, we'd have to look them up again in the hash table This makes significant difference (factor of two or more) on some test cases. They can be thrown away after the pass is finished. */ /* Representation of expressions on value numbers: In some portions of this code, you will notice we allocate "fake" analogues to the expression we are value numbering, and replace the operands with the values of the expression. Since we work on values, and not just names, we canonicalize expressions to value expressions for use in the ANTIC sets, the EXP_GEN set, etc. This is theoretically unnecessary, it just saves a bunch of repeated get_value_handle and find_leader calls in the remainder of the code, trading off temporary memory usage for speed. The tree nodes aren't actually creating more garbage, since they are allocated in a special pools which are thrown away at the end of this pass. All of this also means that if you print the EXP_GEN or ANTIC sets, you will see "value.5 + value.7" in the set, instead of "a_55 + b_66" or something. The only thing that actually cares about seeing the value leaders is phi translation, and it needs to be able to find the leader for a value in an arbitrary block, so this "value expression" form is perfect for it (otherwise you'd do get_value_handle->find_leader->translate->get_value_handle->find_leader).*/ /* Representation of sets: There are currently two types of sets used, hopefully to be unified soon. The AVAIL sets do not need to be sorted in any particular order, and thus, are simply represented as two bitmaps, one that keeps track of values present in the set, and one that keeps track of expressions present in the set. The other sets are represented as doubly linked lists kept in topological order, with an optional supporting bitmap of values present in the set. The sets represent values, and the elements can be values or expressions. The elements can appear in different sets, but each element can only appear once in each set. Since each node in the set represents a value, we also want to be able to map expression, set pairs to something that tells us whether the value is present is a set. We use a per-set bitmap for that. The value handles also point to a linked list of the expressions they represent via a tree annotation. This is mainly useful only for debugging, since we don't do identity lookups. */ /* A value set element. Basically a single linked list of expressions/values. */ typedef struct value_set_node { /* An expression. */ tree expr; /* A pointer to the next element of the value set. */ struct value_set_node *next; } *value_set_node_t; /* A value set. This is a singly linked list of value_set_node elements with a possible bitmap that tells us what values exist in the set. This set must be kept in topologically sorted order. */ typedef struct value_set { /* The head of the list. Used for iterating over the list in order. */ value_set_node_t head; /* The tail of the list. Used for tail insertions, which are necessary to keep the set in topologically sorted order because of how the set is built. */ value_set_node_t tail; /* The length of the list. */ size_t length; /* True if the set is indexed, which means it contains a backing bitmap for quick determination of whether certain values exist in the set. */ bool indexed; /* The bitmap of values that exist in the set. May be NULL in an empty or non-indexed set. */ bitmap values; } *value_set_t; /* An unordered bitmap set. One bitmap tracks values, the other, expressions. */ typedef struct bitmap_set { bitmap expressions; bitmap values; } *bitmap_set_t; /* Sets that we need to keep track of. */ typedef struct bb_value_sets { /* The EXP_GEN set, which represents expressions/values generated in a basic block. */ value_set_t exp_gen; /* The PHI_GEN set, which represents PHI results generated in a basic block. */ bitmap_set_t phi_gen; /* The TMP_GEN set, which represents results/temporaries generated in a basic block. IE the LHS of an expression. */ bitmap_set_t tmp_gen; /* The AVAIL_OUT set, which represents which values are available in a given basic block. */ bitmap_set_t avail_out; /* The ANTIC_IN set, which represents which values are anticiptable in a given basic block. */ value_set_t antic_in; /* The NEW_SETS set, which is used during insertion to augment the AVAIL_OUT set of blocks with the new insertions performed during the current iteration. */ bitmap_set_t new_sets; } *bb_value_sets_t; #define EXP_GEN(BB) ((bb_value_sets_t) ((BB)->aux))->exp_gen #define PHI_GEN(BB) ((bb_value_sets_t) ((BB)->aux))->phi_gen #define TMP_GEN(BB) ((bb_value_sets_t) ((BB)->aux))->tmp_gen #define AVAIL_OUT(BB) ((bb_value_sets_t) ((BB)->aux))->avail_out #define ANTIC_IN(BB) ((bb_value_sets_t) ((BB)->aux))->antic_in #define NEW_SETS(BB) ((bb_value_sets_t) ((BB)->aux))->new_sets /* This structure is used to keep track of statistics on what optimization PRE was able to perform. */ static struct { /* The number of RHS computations eliminated by PRE. */ int eliminations; /* The number of new expressions/temporaries generated by PRE. */ int insertions; /* The number of new PHI nodes added by PRE. */ int phis; } pre_stats; static tree bitmap_find_leader (bitmap_set_t, tree); static tree find_leader (value_set_t, tree); static void value_insert_into_set (value_set_t, tree); static void bitmap_value_insert_into_set (bitmap_set_t, tree); static void bitmap_value_replace_in_set (bitmap_set_t, tree); static void insert_into_set (value_set_t, tree); static void bitmap_set_copy (bitmap_set_t, bitmap_set_t); static bool bitmap_set_contains_value (bitmap_set_t, tree); static bitmap_set_t bitmap_set_new (void); static value_set_t set_new (bool); static bool is_undefined_value (tree); static tree create_expression_by_pieces (basic_block, tree, tree); /* We can add and remove elements and entries to and from sets and hash tables, so we use alloc pools for them. */ static alloc_pool value_set_pool; static alloc_pool bitmap_set_pool; static alloc_pool value_set_node_pool; static alloc_pool binary_node_pool; static alloc_pool unary_node_pool; static alloc_pool reference_node_pool; /* The phi_translate_table caches phi translations for a given expression and predecessor. */ static htab_t phi_translate_table; /* A three tuple {e, pred, v} used to cache phi translations in the phi_translate_table. */ typedef struct expr_pred_trans_d { /* The expression. */ tree e; /* The predecessor block along which we translated the expression. */ basic_block pred; /* The value that resulted from the translation. */ tree v; /* The hashcode for the expression, pred pair. This is cached for speed reasons. */ hashval_t hashcode; } *expr_pred_trans_t; /* Return the hash value for a phi translation table entry. */ static hashval_t expr_pred_trans_hash (const void *p) { const expr_pred_trans_t ve = (expr_pred_trans_t) p; return ve->hashcode; } /* Return true if two phi translation table entries are the same. P1 and P2 should point to the expr_pred_trans_t's to be compared.*/ static int expr_pred_trans_eq (const void *p1, const void *p2) { const expr_pred_trans_t ve1 = (expr_pred_trans_t) p1; const expr_pred_trans_t ve2 = (expr_pred_trans_t) p2; basic_block b1 = ve1->pred; basic_block b2 = ve2->pred; /* If they are not translations for the same basic block, they can't be equal. */ if (b1 != b2) return false; /* If they are for the same basic block, determine if the expressions are equal. */ if (expressions_equal_p (ve1->e, ve2->e)) return true; return false; } /* Search in the phi translation table for the translation of expression E in basic block PRED. Return the translated value, if found, NULL otherwise. */ static inline tree phi_trans_lookup (tree e, basic_block pred) { void **slot; struct expr_pred_trans_d ept; ept.e = e; ept.pred = pred; ept.hashcode = vn_compute (e, (unsigned long) pred, NULL); slot = htab_find_slot_with_hash (phi_translate_table, &ept, ept.hashcode, NO_INSERT); if (!slot) return NULL; else return ((expr_pred_trans_t) *slot)->v; } /* Add the tuple mapping from {expression E, basic block PRED} to value V, to the phi translation table. */ static inline void phi_trans_add (tree e, tree v, basic_block pred) { void **slot; expr_pred_trans_t new_pair = xmalloc (sizeof (*new_pair)); new_pair->e = e; new_pair->pred = pred; new_pair->v = v; new_pair->hashcode = vn_compute (e, (unsigned long) pred, NULL); slot = htab_find_slot_with_hash (phi_translate_table, new_pair, new_pair->hashcode, INSERT); if (*slot) free (*slot); *slot = (void *) new_pair; } /* Add expression E to the expression set of value V. */ void add_to_value (tree v, tree e) { /* Constants have no expression sets. */ if (is_gimple_min_invariant (v)) return; if (VALUE_HANDLE_EXPR_SET (v) == NULL) VALUE_HANDLE_EXPR_SET (v) = set_new (false); insert_into_set (VALUE_HANDLE_EXPR_SET (v), e); } /* Return true if value V exists in the bitmap for SET. */ static inline bool value_exists_in_set_bitmap (value_set_t set, tree v) { if (!set->values) return false; return bitmap_bit_p (set->values, VALUE_HANDLE_ID (v)); } /* Remove value V from the bitmap for SET. */ static void value_remove_from_set_bitmap (value_set_t set, tree v) { #ifdef ENABLE_CHECKING if (!set->indexed) abort (); #endif if (!set->values) return; bitmap_clear_bit (set->values, VALUE_HANDLE_ID (v)); } /* Insert the value number V into the bitmap of values existing in SET. */ static inline void value_insert_into_set_bitmap (value_set_t set, tree v) { #ifdef ENABLE_CHECKING if (!set->indexed) abort (); #endif if (set->values == NULL) { set->values = BITMAP_GGC_ALLOC (); bitmap_clear (set->values); } bitmap_set_bit (set->values, VALUE_HANDLE_ID (v)); } /* Create a new bitmap set and return it. */ static bitmap_set_t bitmap_set_new (void) { bitmap_set_t ret = pool_alloc (bitmap_set_pool); ret->expressions = BITMAP_GGC_ALLOC (); ret->values = BITMAP_GGC_ALLOC (); bitmap_clear (ret->expressions); bitmap_clear (ret->values); return ret; } /* Create a new set. */ static value_set_t set_new (bool indexed) { value_set_t ret; ret = pool_alloc (value_set_pool); ret->head = ret->tail = NULL; ret->length = 0; ret->indexed = indexed; ret->values = NULL; return ret; } /* Insert an expression EXPR into a bitmapped set. */ static void bitmap_insert_into_set (bitmap_set_t set, tree expr) { tree val; /* XXX: For now, we only let SSA_NAMES into the bitmap sets. */ if (TREE_CODE (expr) != SSA_NAME) abort (); val = get_value_handle (expr); if (val == NULL) abort (); if (!is_gimple_min_invariant (val)) bitmap_set_bit (set->values, VALUE_HANDLE_ID (val)); bitmap_set_bit (set->expressions, SSA_NAME_VERSION (expr)); } /* Insert EXPR into SET. */ static void insert_into_set (value_set_t set, tree expr) { value_set_node_t newnode = pool_alloc (value_set_node_pool); tree val = get_value_handle (expr); if (val == NULL) abort (); /* For indexed sets, insert the value into the set value bitmap. For all sets, add it to the linked list and increment the list length. */ if (set->indexed) value_insert_into_set_bitmap (set, val); newnode->next = NULL; newnode->expr = expr; set->length ++; if (set->head == NULL) { set->head = set->tail = newnode; } else { set->tail->next = newnode; set->tail = newnode; } } /* Copy a bitmapped set ORIG, into bitmapped set DEST. */ static void bitmap_set_copy (bitmap_set_t dest, bitmap_set_t orig) { bitmap_copy (dest->expressions, orig->expressions); bitmap_copy (dest->values, orig->values); } /* Copy the set ORIG to the set DEST. */ static void set_copy (value_set_t dest, value_set_t orig) { value_set_node_t node; if (!orig || !orig->head) return; for (node = orig->head; node; node = node->next) { insert_into_set (dest, node->expr); } } /* Remove EXPR from SET. */ static void set_remove (value_set_t set, tree expr) { value_set_node_t node, prev; /* Remove the value of EXPR from the bitmap, decrement the set length, and remove it from the actual double linked list. */ value_remove_from_set_bitmap (set, get_value_handle (expr)); set->length--; prev = NULL; for (node = set->head; node != NULL; prev = node, node = node->next) { if (node->expr == expr) { if (prev == NULL) set->head = node->next; else prev->next= node->next; if (node == set->tail) set->tail = prev; pool_free (value_set_node_pool, node); return; } } } /* Return true if SET contains the value VAL. */ static bool set_contains_value (value_set_t set, tree val) { /* All constants are in every set. */ if (is_gimple_min_invariant (val)) return true; if (set->length == 0) return false; return value_exists_in_set_bitmap (set, val); } /* Return true if bitmapped set SET contains the expression EXPR. */ static bool bitmap_set_contains (bitmap_set_t set, tree expr) { /* XXX: Bitmapped sets only contain SSA_NAME's for now. */ if (TREE_CODE (expr) != SSA_NAME) return false; return bitmap_bit_p (set->expressions, SSA_NAME_VERSION (expr)); } /* Return true if bitmapped set SET contains the value VAL. */ static bool bitmap_set_contains_value (bitmap_set_t set, tree val) { if (is_gimple_min_invariant (val)) return true; return bitmap_bit_p (set->values, VALUE_HANDLE_ID (val)); } /* Replace an instance of value LOOKFOR with expression EXPR in SET. */ static void bitmap_set_replace_value (bitmap_set_t set, tree lookfor, tree expr) { value_set_t exprset; value_set_node_t node; if (is_gimple_min_invariant (lookfor)) return; if (!bitmap_set_contains_value (set, lookfor)) return; /* The number of expressions having a given value is usually significantly less than the total number of expressions in SET. Thus, rather than check, for each expression in SET, whether it has the value LOOKFOR, we walk the reverse mapping that tells us what expressions have a given value, and see if any of those expressions are in our set. For large testcases, this is about 5-10x faster than walking the bitmap. If this is somehow a significant lose for some cases, we can choose which set to walk based on the set size. */ exprset = VALUE_HANDLE_EXPR_SET (lookfor); for (node = exprset->head; node; node = node->next) { if (TREE_CODE (node->expr) == SSA_NAME) { if (bitmap_bit_p (set->expressions, SSA_NAME_VERSION (node->expr))) { bitmap_clear_bit (set->expressions, SSA_NAME_VERSION (node->expr)); bitmap_set_bit (set->expressions, SSA_NAME_VERSION (expr)); return; } } } } /* Subtract bitmapped set B from value set A, and return the new set. */ static value_set_t bitmap_set_subtract_from_value_set (value_set_t a, bitmap_set_t b, bool indexed) { value_set_t ret = set_new (indexed); value_set_node_t node; for (node = a->head; node; node = node->next) { if (!bitmap_set_contains (b, node->expr)) insert_into_set (ret, node->expr); } return ret; } /* Return true if two sets are equal. */ static bool set_equal (value_set_t a, value_set_t b) { value_set_node_t node; if (a->length != b->length) return false; for (node = a->head; node; node = node->next) { if (!set_contains_value (b, get_value_handle (node->expr))) return false; } return true; } /* Replace an instance of EXPR's VALUE with EXPR in SET. */ static void bitmap_value_replace_in_set (bitmap_set_t set, tree expr) { tree val = get_value_handle (expr); bitmap_set_replace_value (set, val, expr); } /* Insert EXPR into SET if EXPR's value is not already present in SET. */ static void bitmap_value_insert_into_set (bitmap_set_t set, tree expr) { tree val = get_value_handle (expr); if (is_gimple_min_invariant (val)) return; if (!bitmap_set_contains_value (set, val)) bitmap_insert_into_set (set, expr); } /* Insert the value for EXPR into SET, if it doesn't exist already. */ static void value_insert_into_set (value_set_t set, tree expr) { tree val = get_value_handle (expr); /* Constant and invariant values exist everywhere, and thus, actually keeping them in the sets is pointless. */ if (is_gimple_min_invariant (val)) return; if (!set_contains_value (set, val)) insert_into_set (set, expr); } /* Print out SET to OUTFILE. */ static void bitmap_print_value_set (FILE *outfile, bitmap_set_t set, const char *setname, int blockindex) { fprintf (outfile, "%s[%d] := { ", setname, blockindex); if (set) { int i; EXECUTE_IF_SET_IN_BITMAP (set->expressions, 0, i, { print_generic_expr (outfile, ssa_name (i), 0); fprintf (outfile, " ("); print_generic_expr (outfile, get_value_handle (ssa_name (i)), 0); fprintf (outfile, ") "); if (bitmap_last_set_bit (set->expressions) != i) fprintf (outfile, ", "); }); } fprintf (outfile, " }\n"); } /* Print out the value_set SET to OUTFILE. */ static void print_value_set (FILE *outfile, value_set_t set, const char *setname, int blockindex) { value_set_node_t node; fprintf (outfile, "%s[%d] := { ", setname, blockindex); if (set) { for (node = set->head; node; node = node->next) { print_generic_expr (outfile, node->expr, 0); fprintf (outfile, " ("); print_generic_expr (outfile, get_value_handle (node->expr), 0); fprintf (outfile, ") "); if (node->next) fprintf (outfile, ", "); } } fprintf (outfile, " }\n"); } /* Print out the expressions that have VAL to OUTFILE. */ void print_value_expressions (FILE *outfile, tree val) { if (VALUE_HANDLE_EXPR_SET (val)) { char s[10]; sprintf (s, "VH.%04d", VALUE_HANDLE_ID (val)); print_value_set (outfile, VALUE_HANDLE_EXPR_SET (val), s, 0); } } void debug_value_expressions (tree val) { print_value_expressions (stderr, val); } void debug_value_set (value_set_t, const char *, int); void debug_value_set (value_set_t set, const char *setname, int blockindex) { print_value_set (stderr, set, setname, blockindex); } /* Translate EXPR using phis in PHIBLOCK, so that it has the values of the phis in PRED. Return NULL if we can't find a leader for each part of the translated expression. */ static tree phi_translate (tree expr, value_set_t set, basic_block pred, basic_block phiblock) { tree phitrans = NULL; tree oldexpr = expr; if (expr == NULL) return NULL; /* Phi translations of a given expression don't change, */ phitrans = phi_trans_lookup (expr, pred); if (phitrans) return phitrans; switch (TREE_CODE_CLASS (TREE_CODE (expr))) { case '2': { tree oldop1 = TREE_OPERAND (expr, 0); tree oldop2 = TREE_OPERAND (expr, 1); tree newop1; tree newop2; tree newexpr; newop1 = phi_translate (find_leader (set, oldop1), set, pred, phiblock); if (newop1 == NULL) return NULL; newop2 = phi_translate (find_leader (set, oldop2), set, pred, phiblock); if (newop2 == NULL) return NULL; if (newop1 != oldop1 || newop2 != oldop2) { newexpr = pool_alloc (binary_node_pool); memcpy (newexpr, expr, tree_size (expr)); create_tree_ann (newexpr); TREE_OPERAND (newexpr, 0) = newop1 == oldop1 ? oldop1 : get_value_handle (newop1); TREE_OPERAND (newexpr, 1) = newop2 == oldop2 ? oldop2 : get_value_handle (newop2); vn_lookup_or_add (newexpr, NULL); expr = newexpr; phi_trans_add (oldexpr, newexpr, pred); } } break; /* XXX: Until we have PRE of loads working, none will be ANTIC. */ case 'r': return NULL; break; case '1': { tree oldop1 = TREE_OPERAND (expr, 0); tree newop1; tree newexpr; newop1 = phi_translate (find_leader (set, oldop1), set, pred, phiblock); if (newop1 == NULL) return NULL; if (newop1 != oldop1) { newexpr = pool_alloc (unary_node_pool); memcpy (newexpr, expr, tree_size (expr)); create_tree_ann (newexpr); TREE_OPERAND (newexpr, 0) = get_value_handle (newop1); vn_lookup_or_add (newexpr, NULL); expr = newexpr; phi_trans_add (oldexpr, newexpr, pred); } } break; case 'd': abort (); case 'x': { tree phi = NULL; int i; if (TREE_CODE (expr) != SSA_NAME) abort (); if (TREE_CODE (SSA_NAME_DEF_STMT (expr)) == PHI_NODE) phi = SSA_NAME_DEF_STMT (expr); else return expr; for (i = 0; i < PHI_NUM_ARGS (phi); i++) if (PHI_ARG_EDGE (phi, i)->src == pred) { tree val; if (is_undefined_value (PHI_ARG_DEF (phi, i))) return NULL; val = vn_lookup_or_add (PHI_ARG_DEF (phi, i), NULL); return PHI_ARG_DEF (phi, i); } } break; } return expr; } static void phi_translate_set (value_set_t dest, value_set_t set, basic_block pred, basic_block phiblock) { value_set_node_t node; for (node = set->head; node; node = node->next) { tree translated; translated = phi_translate (node->expr, set, pred, phiblock); phi_trans_add (node->expr, translated, pred); if (translated != NULL) value_insert_into_set (dest, translated); } } /* Find the leader for a value (i.e., the name representing that value) in a given set, and return it. Return NULL if no leader is found. */ static tree bitmap_find_leader (bitmap_set_t set, tree val) { if (val == NULL) return NULL; if (is_gimple_min_invariant (val)) return val; if (bitmap_set_contains_value (set, val)) { /* Rather than walk the entire bitmap of expressions, and see whether any of them has the value we are looking for, we look at the reverse mapping, which tells us the set of expressions that have a given value (IE value->expressions with that value) and see if any of those expressions are in our set. The number of expressions per value is usually significantly less than the number of expressions in the set. In fact, for large testcases, doing it this way is roughly 5-10x faster than walking the bitmap. If this is somehow a significant lose for some cases, we can choose which set to walk based on which set is smaller. */ value_set_t exprset; value_set_node_t node; exprset = VALUE_HANDLE_EXPR_SET (val); for (node = exprset->head; node; node = node->next) { if (TREE_CODE (node->expr) == SSA_NAME) { if (bitmap_bit_p (set->expressions, SSA_NAME_VERSION (node->expr))) return node->expr; } } } return NULL; } /* Find the leader for a value (i.e., the name representing that value) in a given set, and return it. Return NULL if no leader is found. */ static tree find_leader (value_set_t set, tree val) { value_set_node_t node; if (val == NULL) return NULL; /* Constants represent themselves. */ if (is_gimple_min_invariant (val)) return val; if (set->length == 0) return NULL; if (value_exists_in_set_bitmap (set, val)) { for (node = set->head; node; node = node->next) { if (get_value_handle (node->expr) == val) return node->expr; } } return NULL; } /* Determine if the expression EXPR is valid in SET. This means that we have a leader for each part of the expression (if it consists of values), or the expression is an SSA_NAME. NB: We never should run into a case where we have SSA_NAME + SSA_NAME or SSA_NAME + value. The sets valid_in_set is called on, the ANTIC sets, will only ever have SSA_NAME's or binary value expression (IE VALUE1 + VALUE2) */ static bool valid_in_set (value_set_t set, tree expr) { switch (TREE_CODE_CLASS (TREE_CODE (expr))) { case '2': { tree op1 = TREE_OPERAND (expr, 0); tree op2 = TREE_OPERAND (expr, 1); return set_contains_value (set, op1) && set_contains_value (set, op2); } break; case '1': { tree op1 = TREE_OPERAND (expr, 0); return set_contains_value (set, op1); } break; /* XXX: Until PRE of loads works, no reference nodes are ANTIC. */ case 'r': { return false; } case 'x': { if (TREE_CODE (expr) == SSA_NAME) return true; abort (); } case 'c': abort (); } return false; } /* Clean the set of expressions that are no longer valid in SET. This means expressions that are made up of values we have no leaders for in SET. */ static void clean (value_set_t set) { value_set_node_t node; value_set_node_t next; node = set->head; while (node) { next = node->next; if (!valid_in_set (set, node->expr)) set_remove (set, node->expr); node = next; } } /* Compute the ANTIC set for BLOCK. ANTIC_OUT[BLOCK] = intersection of ANTIC_IN[b] for all succ(BLOCK), if succs(BLOCK) > 1 ANTIC_OUT[BLOCK] = phi_translate (ANTIC_IN[succ(BLOCK)]) if succs(BLOCK) == 1 ANTIC_IN[BLOCK] = clean(ANTIC_OUT[BLOCK] U EXP_GEN[BLOCK] - TMP_GEN[BLOCK]) Iterate until fixpointed. XXX: It would be nice to either write a set_clear, and use it for antic_out, or to mark the antic_out set as deleted at the end of this routine, so that the pool can hand the same memory back out again for the next antic_out. */ static bool compute_antic_aux (basic_block block) { basic_block son; edge e; bool changed = false; value_set_t S, old, ANTIC_OUT; value_set_node_t node; ANTIC_OUT = S = NULL; /* If any edges from predecessors are abnormal, antic_in is empty, so punt. Remember that the block has an incoming abnormal edge by setting the BB_VISITED flag. */ if (! (block->flags & BB_VISITED)) { for (e = block->pred; e; e = e->pred_next) if (e->flags & EDGE_ABNORMAL) { block->flags |= BB_VISITED; break; } } if (block->flags & BB_VISITED) { S = NULL; goto visit_sons; } old = set_new (false); set_copy (old, ANTIC_IN (block)); ANTIC_OUT = set_new (true); /* If the block has no successors, ANTIC_OUT is empty, because it is the exit block. */ if (block->succ == NULL); /* If we have one successor, we could have some phi nodes to translate through. */ else if (block->succ->succ_next == NULL) { phi_translate_set (ANTIC_OUT, ANTIC_IN(block->succ->dest), block, block->succ->dest); } /* If we have multiple successors, we take the intersection of all of them. */ else { varray_type worklist; edge e; size_t i; basic_block bprime, first; VARRAY_BB_INIT (worklist, 1, "succ"); e = block->succ; while (e) { VARRAY_PUSH_BB (worklist, e->dest); e = e->succ_next; } first = VARRAY_BB (worklist, 0); set_copy (ANTIC_OUT, ANTIC_IN (first)); for (i = 1; i < VARRAY_ACTIVE_SIZE (worklist); i++) { bprime = VARRAY_BB (worklist, i); node = ANTIC_OUT->head; while (node) { tree val; value_set_node_t next = node->next; val = get_value_handle (node->expr); if (!set_contains_value (ANTIC_IN (bprime), val)) set_remove (ANTIC_OUT, node->expr); node = next; } } VARRAY_CLEAR (worklist); } /* Generate ANTIC_OUT - TMP_GEN */ S = bitmap_set_subtract_from_value_set (ANTIC_OUT, TMP_GEN (block), false); /* Start ANTIC_IN with EXP_GEN - TMP_GEN */ ANTIC_IN (block) = bitmap_set_subtract_from_value_set (EXP_GEN (block), TMP_GEN (block), true); /* Then union in the ANTIC_OUT - TMP_GEN values, to get ANTIC_OUT U EXP_GEN - TMP_GEN */ for (node = S->head; node; node = node->next) { value_insert_into_set (ANTIC_IN (block), node->expr); } clean (ANTIC_IN (block)); if (!set_equal (old, ANTIC_IN (block))) changed = true; visit_sons: if (dump_file && (dump_flags & TDF_DETAILS)) { if (ANTIC_OUT) print_value_set (dump_file, ANTIC_OUT, "ANTIC_OUT", block->index); print_value_set (dump_file, ANTIC_IN (block), "ANTIC_IN", block->index); if (S) print_value_set (dump_file, S, "S", block->index); } for (son = first_dom_son (CDI_POST_DOMINATORS, block); son; son = next_dom_son (CDI_POST_DOMINATORS, son)) { changed |= compute_antic_aux (son); } return changed; } /* Compute ANTIC sets. */ static void compute_antic (void) { bool changed = true; basic_block bb; int num_iterations = 0; FOR_ALL_BB (bb) { ANTIC_IN (bb) = set_new (true); if (bb->flags & BB_VISITED) abort (); } while (changed) { num_iterations++; changed = false; changed = compute_antic_aux (EXIT_BLOCK_PTR); } FOR_ALL_BB (bb) { bb->flags &= ~BB_VISITED; } if (num_iterations > 2 && dump_file && (dump_flags & TDF_STATS)) fprintf (dump_file, "compute_antic required %d iterations\n", num_iterations); } /* Find a leader for an expression, or generate one using create_expression_by_pieces if it's ANTIC but complex. BLOCK is the basic_block we are looking for leaders in. EXPR is the expression to find a leader or generate for. STMTS is the statement list to put the inserted expressions on. Returns the SSA_NAME of the LHS of the generated expression or the leader. */ static tree find_or_generate_expression (basic_block block, tree expr, tree stmts) { tree genop; genop = bitmap_find_leader (AVAIL_OUT (block), expr); /* Depending on the order we process DOM branches in, the value may not have propagated to all the dom children yet during this iteration. In this case, the value will always be in the NEW_SETS for us already, having been propagated from our dominator. */ if (genop == NULL) genop = bitmap_find_leader (NEW_SETS (block), expr); /* If it's still NULL, see if it is a complex expression, and if so, generate it recursively, otherwise, abort, because it's not really . */ if (genop == NULL) { genop = VALUE_HANDLE_EXPR_SET (expr)->head->expr; if (TREE_CODE_CLASS (TREE_CODE (genop)) != '1' && TREE_CODE_CLASS (TREE_CODE (genop)) != '2' && TREE_CODE_CLASS (TREE_CODE (genop)) != 'r') abort (); genop = create_expression_by_pieces (block, genop, stmts); } return genop; } /* Create an expression in pieces, so that we can handle very complex expressions that may be ANTIC, but not necessary GIMPLE. BLOCK is the basic block the expression will be inserted into, EXPR is the expression to insert (in value form) STMTS is a statement list to append the necessary insertions into. This function will abort if we hit some value that shouldn't be ANTIC but is (IE there is no leader for it, or its components). This function may also generate expressions that are themselves partially or fully redundant. Those that are will be either made fully redundant during the next iteration of insert (for partially redundant ones), or eliminated by eliminate (for fully redundant ones). */ static tree create_expression_by_pieces (basic_block block, tree expr, tree stmts) { tree name = NULL_TREE; tree newexpr = NULL_TREE; tree v; switch (TREE_CODE_CLASS (TREE_CODE (expr))) { case '2': { tree_stmt_iterator tsi; tree genop1, genop2; tree temp; tree op1 = TREE_OPERAND (expr, 0); tree op2 = TREE_OPERAND (expr, 1); genop1 = find_or_generate_expression (block, op1, stmts); genop2 = find_or_generate_expression (block, op2, stmts); temp = create_tmp_var (TREE_TYPE (expr), "pretmp"); add_referenced_tmp_var (temp); newexpr = build (TREE_CODE (expr), TREE_TYPE (expr), genop1, genop2); newexpr = build (MODIFY_EXPR, TREE_TYPE (expr), temp, newexpr); name = make_ssa_name (temp, newexpr); TREE_OPERAND (newexpr, 0) = name; tsi = tsi_last (stmts); tsi_link_after (&tsi, newexpr, TSI_CONTINUE_LINKING); pre_stats.insertions++; break; } case '1': { tree_stmt_iterator tsi; tree genop1; tree temp; tree op1 = TREE_OPERAND (expr, 0); genop1 = find_or_generate_expression (block, op1, stmts); temp = create_tmp_var (TREE_TYPE (expr), "pretmp"); add_referenced_tmp_var (temp); newexpr = build (TREE_CODE (expr), TREE_TYPE (expr), genop1); newexpr = build (MODIFY_EXPR, TREE_TYPE (expr), temp, newexpr); name = make_ssa_name (temp, newexpr); TREE_OPERAND (newexpr, 0) = name; tsi = tsi_last (stmts); tsi_link_after (&tsi, newexpr, TSI_CONTINUE_LINKING); pre_stats.insertions++; break; } default: abort (); } v = get_value_handle (expr); vn_add (name, v, NULL); bitmap_insert_into_set (NEW_SETS (block), name); bitmap_value_insert_into_set (AVAIL_OUT (block), name); if (dump_file && (dump_flags & TDF_DETAILS)) { fprintf (dump_file, "Inserted "); print_generic_expr (dump_file, newexpr, 0); fprintf (dump_file, " in predecessor %d\n", block->index); } return name; } /* Perform insertion of partially redundant values. For BLOCK, do the following: 1. Propagate the NEW_SETS of the dominator into the current block. If the block has multiple predecessors, 2a. Iterate over the ANTIC expressions for the block to see if any of them are partially redundant. 2b. If so, insert them into the necessary predecessors to make the expression fully redundant. 2c. Insert a new PHI merging the values of the predecessors. 2d. Insert the new PHI, and the new expressions, into the NEW_SETS set. 3. Recursively call ourselves on the dominator children of BLOCK. */ static bool insert_aux (basic_block block) { basic_block son; bool new_stuff = false; if (block) { basic_block dom; dom = get_immediate_dominator (CDI_DOMINATORS, block); if (dom) { int i; bitmap_set_t newset = NEW_SETS (dom); EXECUTE_IF_SET_IN_BITMAP (newset->expressions, 0, i, { bitmap_insert_into_set (NEW_SETS (block), ssa_name (i)); bitmap_value_replace_in_set (AVAIL_OUT (block), ssa_name (i)); }); if (block->pred->pred_next) { value_set_node_t node; for (node = ANTIC_IN (block)->head; node; node = node->next) { if (TREE_CODE_CLASS (TREE_CODE (node->expr)) == '2' || TREE_CODE_CLASS (TREE_CODE (node->expr)) == '1') { tree *avail; tree val; bool by_some = false; bool cant_insert = false; bool all_same = true; tree first_s = NULL; edge pred; basic_block bprime; tree eprime; val = get_value_handle (node->expr); if (bitmap_set_contains_value (PHI_GEN (block), val)) continue; if (bitmap_set_contains_value (AVAIL_OUT (dom), val)) { if (dump_file && (dump_flags & TDF_DETAILS)) fprintf (dump_file, "Found fully redundant value\n"); continue; } avail = xcalloc (last_basic_block, sizeof (tree)); for (pred = block->pred; pred; pred = pred->pred_next) { tree vprime; tree edoubleprime; bprime = pred->src; eprime = phi_translate (node->expr, ANTIC_IN (block), bprime, block); /* eprime will generally only be NULL if the value of the expression, translated through the PHI for this predecessor, is undefined. If that is the case, we can't make the expression fully redundant, because its value is undefined along a predecessor path. We can thus break out early because it doesn't matter what the rest of the results are. */ if (eprime == NULL) { cant_insert = true; break; } vprime = get_value_handle (eprime); if (!vprime) abort (); edoubleprime = bitmap_find_leader (AVAIL_OUT (bprime), vprime); if (edoubleprime == NULL) { avail[bprime->index] = eprime; all_same = false; } else { avail[bprime->index] = edoubleprime; by_some = true; if (first_s == NULL) first_s = edoubleprime; else if (first_s != edoubleprime) all_same = false; if (first_s != edoubleprime && operand_equal_p (first_s, edoubleprime, 0)) abort (); } } /* If we can insert it, it's not the same value already existing along every predecessor, and it's defined by some predecessor, it is partially redundant. */ if (!cant_insert && !all_same && by_some) { tree type = TREE_TYPE (avail[block->pred->src->index]); tree temp; if (dump_file && (dump_flags & TDF_DETAILS)) { fprintf (dump_file, "Found partial redundancy for expression "); print_generic_expr (dump_file, node->expr, 0); fprintf (dump_file, "\n"); } /* Make the necessary insertions. */ for (pred = block->pred; pred; pred = pred->pred_next) { tree stmts = alloc_stmt_list (); tree builtexpr; bprime = pred->src; eprime = avail[bprime->index]; if (TREE_CODE_CLASS (TREE_CODE (eprime)) == '2' || TREE_CODE_CLASS (TREE_CODE (eprime)) == '1') { builtexpr = create_expression_by_pieces (bprime, eprime, stmts); bsi_insert_on_edge (pred, stmts); bsi_commit_edge_inserts (NULL); avail[bprime->index] = builtexpr; } } /* Now build a phi for the new variable. */ temp = create_tmp_var (type, "prephitmp"); add_referenced_tmp_var (temp); temp = create_phi_node (temp, block); vn_add (PHI_RESULT (temp), val, NULL); #if 0 if (!set_contains_value (AVAIL_OUT (block), val)) insert_into_set (AVAIL_OUT (block), PHI_RESULT (temp)); else #endif bitmap_value_replace_in_set (AVAIL_OUT (block), PHI_RESULT (temp)); for (pred = block->pred; pred; pred = pred->pred_next) { add_phi_arg (&temp, avail[pred->src->index], pred); } if (dump_file && (dump_flags & TDF_DETAILS)) { fprintf (dump_file, "Created phi "); print_generic_expr (dump_file, temp, 0); fprintf (dump_file, " in block %d\n", block->index); } pre_stats.phis++; new_stuff = true; bitmap_insert_into_set (NEW_SETS (block), PHI_RESULT (temp)); bitmap_insert_into_set (PHI_GEN (block), PHI_RESULT (temp)); } free (avail); } } } } } for (son = first_dom_son (CDI_DOMINATORS, block); son; son = next_dom_son (CDI_DOMINATORS, son)) { new_stuff |= insert_aux (son); } return new_stuff; } /* Perform insertion of partially redundant values. */ static void insert (void) { bool new_stuff = true; basic_block bb; int num_iterations = 0; FOR_ALL_BB (bb) NEW_SETS (bb) = bitmap_set_new (); while (new_stuff) { num_iterations++; new_stuff = false; new_stuff = insert_aux (ENTRY_BLOCK_PTR); } if (num_iterations > 2 && dump_file && (dump_flags & TDF_STATS)) fprintf (dump_file, "insert required %d iterations\n", num_iterations); } /* Return true if VAR is an SSA variable with no defining statement in this procedure, *AND* isn't a live-on-entry parameter. */ static bool is_undefined_value (tree expr) { return (TREE_CODE (expr) == SSA_NAME && IS_EMPTY_STMT (SSA_NAME_DEF_STMT (expr)) /* PARM_DECLs and hard registers are always defined. */ && TREE_CODE (SSA_NAME_VAR (expr)) != PARM_DECL && !DECL_HARD_REGISTER (SSA_NAME_VAR (expr))); } /* Given an SSA variable VAR and an expression EXPR, compute the value number for EXPR and create a value handle (VAL) for it. If VAR and EXPR are not the same, associate VAL with VAR. Finally, add VAR to S1 and its value handle to S2. VUSES represent the virtual use operands associated with EXPR (if any). They are used when computing the hash value for EXPR. */ static inline void add_to_sets (tree var, tree expr, vuse_optype vuses, bitmap_set_t s1, bitmap_set_t s2) { tree val = vn_lookup_or_add (expr, vuses); /* VAR and EXPR may be the same when processing statements for which we are not computing value numbers (e.g., non-assignments, or statements that make aliased stores). In those cases, we are only interested in making VAR available as its own value. */ if (var != expr) vn_add (var, val, NULL); bitmap_insert_into_set (s1, var); bitmap_value_insert_into_set (s2, var); } /* Given a unary or binary expression EXPR, create and return a new expression with the same structure as EXPR but with its operands replaced with the value handles of each of the operands of EXPR. Insert EXPR's operands into the EXP_GEN set for BLOCK. VUSES represent the virtual use operands associated with EXPR (if any). They are used when computing the hash value for EXPR. */ static inline tree create_value_expr_from (tree expr, basic_block block, vuse_optype vuses) { int i; enum tree_code code = TREE_CODE (expr); tree vexpr; #if defined ENABLE_CHECKING if (TREE_CODE_CLASS (code) != '1' && TREE_CODE_CLASS (code) != '2' && TREE_CODE_CLASS (code) != 'r') abort (); #endif if (TREE_CODE_CLASS (code) == '1') vexpr = pool_alloc (unary_node_pool); else if (TREE_CODE_CLASS (code) == 'r') vexpr = pool_alloc (reference_node_pool); else vexpr = pool_alloc (binary_node_pool); memcpy (vexpr, expr, tree_size (expr)); for (i = 0; i < TREE_CODE_LENGTH (code); i++) { tree op = TREE_OPERAND (expr, i); if (op != NULL) { tree val = vn_lookup_or_add (op, vuses); if (!is_undefined_value (op)) value_insert_into_set (EXP_GEN (block), op); TREE_TYPE (val) = TREE_TYPE (TREE_OPERAND (vexpr, i)); TREE_OPERAND (vexpr, i) = val; } } return vexpr; } /* Compute the AVAIL set for BLOCK. This function performs value numbering of the statements in BLOCK. The AVAIL sets are built from information we glean while doing this value numbering, since the AVAIL sets contain only one entry per value. AVAIL_IN[BLOCK] = AVAIL_OUT[dom(BLOCK)]. AVAIL_OUT[BLOCK] = AVAIL_IN[BLOCK] U PHI_GEN[BLOCK] U TMP_GEN[BLOCK]. */ static void compute_avail (basic_block block) { basic_block son; /* For arguments with default definitions, we pretend they are defined in the entry block. */ if (block == ENTRY_BLOCK_PTR) { tree param; for (param = DECL_ARGUMENTS (current_function_decl); param; param = TREE_CHAIN (param)) { if (default_def (param) != NULL) { tree val; tree def = default_def (param); val = vn_lookup_or_add (def, NULL); bitmap_insert_into_set (TMP_GEN (block), def); bitmap_value_insert_into_set (AVAIL_OUT (block), def); } } } else if (block) { block_stmt_iterator bsi; tree stmt, phi; basic_block dom; /* Initially, the set of available values in BLOCK is that of its immediate dominator. */ dom = get_immediate_dominator (CDI_DOMINATORS, block); if (dom) bitmap_set_copy (AVAIL_OUT (block), AVAIL_OUT (dom)); /* Generate values for PHI nodes. */ for (phi = phi_nodes (block); phi; phi = PHI_CHAIN (phi)) /* We have no need for virtual phis, as they don't represent actual computations. */ if (is_gimple_reg (PHI_RESULT (phi))) add_to_sets (PHI_RESULT (phi), PHI_RESULT (phi), NULL, PHI_GEN (block), AVAIL_OUT (block)); /* Now compute value numbers and populate value sets with all the expressions computed in BLOCK. */ for (bsi = bsi_start (block); !bsi_end_p (bsi); bsi_next (&bsi)) { stmt_ann_t ann; size_t j; stmt = bsi_stmt (bsi); ann = stmt_ann (stmt); get_stmt_operands (stmt); /* We are only interested in assignments of the form X_i = EXPR, where EXPR represents an "interesting" computation, it has no volatile operands and X_i doesn't flow through an abnormal edge. */ if (TREE_CODE (stmt) == MODIFY_EXPR && !ann->has_volatile_ops && TREE_CODE (TREE_OPERAND (stmt, 0)) == SSA_NAME && !SSA_NAME_OCCURS_IN_ABNORMAL_PHI (TREE_OPERAND (stmt, 0))) { tree lhs = TREE_OPERAND (stmt, 0); tree rhs = TREE_OPERAND (stmt, 1); vuse_optype vuses = STMT_VUSE_OPS (stmt); STRIP_USELESS_TYPE_CONVERSION (rhs); if (TREE_CODE (rhs) == SSA_NAME || is_gimple_min_invariant (rhs)) { /* Compute a value number for the RHS of the statement and add its value to the AVAIL_OUT set for the block. Add the LHS to TMP_GEN. */ add_to_sets (lhs, rhs, vuses, TMP_GEN (block), AVAIL_OUT (block)); if (TREE_CODE (rhs) == SSA_NAME && !is_undefined_value (rhs)) value_insert_into_set (EXP_GEN (block), rhs); continue; } else if (TREE_CODE_CLASS (TREE_CODE (rhs)) == '1' || TREE_CODE_CLASS (TREE_CODE (rhs)) == '2' || TREE_CODE (rhs) == INDIRECT_REF) { /* For binary, unary, and reference expressions, create a duplicate expression with the operands replaced with the value handles of the original RHS. */ tree newt = create_value_expr_from (rhs, block, vuses); add_to_sets (lhs, newt, vuses, TMP_GEN (block), AVAIL_OUT (block)); value_insert_into_set (EXP_GEN (block), newt); continue; } } /* For any other statement that we don't recognize, simply make the names generated by the statement available in AVAIL_OUT and TMP_GEN. */ for (j = 0; j < NUM_DEFS (STMT_DEF_OPS (stmt)); j++) { tree def = DEF_OP (STMT_DEF_OPS (stmt), j); add_to_sets (def, def, NULL, TMP_GEN (block), AVAIL_OUT (block)); } for (j = 0; j < NUM_USES (STMT_USE_OPS (stmt)); j++) { tree use = USE_OP (STMT_USE_OPS (stmt), j); add_to_sets (use, use, NULL, TMP_GEN (block), AVAIL_OUT (block)); } } } /* Compute available sets for the dominator children of BLOCK. */ for (son = first_dom_son (CDI_DOMINATORS, block); son; son = next_dom_son (CDI_DOMINATORS, son)) compute_avail (son); } /* Eliminate fully redundant computations. */ static void eliminate (void) { basic_block b; FOR_EACH_BB (b) { block_stmt_iterator i; for (i = bsi_start (b); !bsi_end_p (i); bsi_next (&i)) { tree stmt = bsi_stmt (i); /* Lookup the RHS of the expression, see if we have an available computation for it. If so, replace the RHS with the available computation. */ if (TREE_CODE (stmt) == MODIFY_EXPR && TREE_CODE (TREE_OPERAND (stmt, 0)) == SSA_NAME && TREE_CODE (TREE_OPERAND (stmt ,1)) != SSA_NAME && !is_gimple_min_invariant (TREE_OPERAND (stmt, 1)) && !stmt_ann (stmt)->has_volatile_ops) { tree lhs = TREE_OPERAND (stmt, 0); tree *rhs_p = &TREE_OPERAND (stmt, 1); tree sprime; sprime = bitmap_find_leader (AVAIL_OUT (b), vn_lookup (lhs, NULL)); if (sprime && sprime != lhs && (TREE_CODE (*rhs_p) != SSA_NAME || may_propagate_copy (*rhs_p, sprime))) { if (sprime == *rhs_p) abort (); if (dump_file && (dump_flags & TDF_DETAILS)) { fprintf (dump_file, "Replaced "); print_generic_expr (dump_file, *rhs_p, 0); fprintf (dump_file, " with "); print_generic_expr (dump_file, sprime, 0); fprintf (dump_file, " in "); print_generic_stmt (dump_file, stmt, 0); } pre_stats.eliminations++; propagate_tree_value (rhs_p, sprime); modify_stmt (stmt); } } } } } /* Initialize data structures used by PRE. */ static void init_pre (void) { size_t tsize; basic_block bb; vn_init (); memset (&pre_stats, 0, sizeof (pre_stats)); FOR_ALL_BB (bb) bb->aux = xcalloc (1, sizeof (struct bb_value_sets)); phi_translate_table = htab_create (511, expr_pred_trans_hash, expr_pred_trans_eq, free); value_set_pool = create_alloc_pool ("Value sets", sizeof (struct value_set), 30); bitmap_set_pool = create_alloc_pool ("Bitmap sets", sizeof (struct bitmap_set), 30); value_set_node_pool = create_alloc_pool ("Value set nodes", sizeof (struct value_set_node), 30); calculate_dominance_info (CDI_POST_DOMINATORS); calculate_dominance_info (CDI_DOMINATORS); tsize = tree_size (build (PLUS_EXPR, void_type_node, NULL_TREE, NULL_TREE)); binary_node_pool = create_alloc_pool ("Binary tree nodes", tsize, 30); tsize = tree_size (build1 (NEGATE_EXPR, void_type_node, NULL_TREE)); unary_node_pool = create_alloc_pool ("Unary tree nodes", tsize, 30); tsize = tree_size (build (COMPONENT_REF, void_type_node, NULL_TREE, NULL_TREE, NULL_TREE)); reference_node_pool = create_alloc_pool ("Reference tree nodes", tsize, 30); FOR_ALL_BB (bb) { EXP_GEN (bb) = set_new (true); PHI_GEN (bb) = bitmap_set_new (); TMP_GEN (bb) = bitmap_set_new (); AVAIL_OUT (bb) = bitmap_set_new (); } } /* Deallocate data structures used by PRE. */ static void fini_pre (void) { basic_block bb; free_alloc_pool (value_set_pool); free_alloc_pool (bitmap_set_pool); free_alloc_pool (value_set_node_pool); free_alloc_pool (binary_node_pool); free_alloc_pool (reference_node_pool); free_alloc_pool (unary_node_pool); htab_delete (phi_translate_table); FOR_ALL_BB (bb) { free (bb->aux); bb->aux = NULL; } free_dominance_info (CDI_POST_DOMINATORS); vn_delete (); } /* Main entry point to the SSA-PRE pass. DO_FRE is true if the caller only wants to do full redundancy elimination. */ static void execute_pre (bool do_fre) { init_pre (); /* Collect and value number expressions computed in each basic block. */ compute_avail (ENTRY_BLOCK_PTR); if (dump_file && (dump_flags & TDF_DETAILS)) { basic_block bb; FOR_ALL_BB (bb) { print_value_set (dump_file, EXP_GEN (bb), "exp_gen", bb->index); bitmap_print_value_set (dump_file, TMP_GEN (bb), "tmp_gen", bb->index); bitmap_print_value_set (dump_file, AVAIL_OUT (bb), "avail_out", bb->index); } } /* Insert can get quite slow on an incredibly large number of basic blocks due to some quadratic behavior. Until this behavior is fixed, don't run it when he have an incredibly large number of bb's. If we aren't going to run insert, there is no point in computing ANTIC, either, even though it's plenty fast. */ if (!do_fre && n_basic_blocks < 4000) { compute_antic (); insert (); } /* Remove all the redundant expressions. */ eliminate (); if (dump_file && (dump_flags & TDF_STATS)) { fprintf (dump_file, "Insertions:%d\n", pre_stats.insertions); fprintf (dump_file, "New PHIs:%d\n", pre_stats.phis); fprintf (dump_file, "Eliminated:%d\n", pre_stats.eliminations); } fini_pre (); } /* Gate and execute functions for PRE. */ static void do_pre (void) { execute_pre (false); } static bool gate_pre (void) { return flag_tree_pre != 0; } struct tree_opt_pass pass_pre = { "pre", /* name */ gate_pre, /* gate */ do_pre, /* execute */ NULL, /* sub */ NULL, /* next */ 0, /* static_pass_number */ TV_TREE_PRE, /* tv_id */ PROP_no_crit_edges | PROP_cfg | PROP_ssa,/* properties_required */ 0, /* properties_provided */ 0, /* properties_destroyed */ 0, /* todo_flags_start */ TODO_dump_func | TODO_ggc_collect | TODO_verify_ssa /* todo_flags_finish */ }; /* Gate and execute functions for FRE. */ static void do_fre (void) { execute_pre (true); } static bool gate_fre (void) { return flag_tree_fre != 0; } struct tree_opt_pass pass_fre = { "fre", /* name */ gate_fre, /* gate */ do_fre, /* execute */ NULL, /* sub */ NULL, /* next */ 0, /* static_pass_number */ TV_TREE_FRE, /* tv_id */ PROP_no_crit_edges | PROP_cfg | PROP_ssa,/* properties_required */ 0, /* properties_provided */ 0, /* properties_destroyed */ 0, /* todo_flags_start */ TODO_dump_func | TODO_ggc_collect | TODO_verify_ssa /* todo_flags_finish */ };