/* Support routines for Value Range Propagation (VRP). Copyright (C) 2005, 2006 Free Software Foundation, Inc. Contributed by Diego Novillo . 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, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. */ #include "config.h" #include "system.h" #include "coretypes.h" #include "tm.h" #include "ggc.h" #include "flags.h" #include "tree.h" #include "basic-block.h" #include "tree-flow.h" #include "tree-pass.h" #include "tree-dump.h" #include "timevar.h" #include "diagnostic.h" #include "cfgloop.h" #include "tree-scalar-evolution.h" #include "tree-ssa-propagate.h" #include "tree-chrec.h" /* Set of SSA names found during the dominator traversal of a sub-graph in find_assert_locations. */ static sbitmap found_in_subgraph; /* Local functions. */ static int compare_values (tree val1, tree val2); /* Location information for ASSERT_EXPRs. Each instance of this structure describes an ASSERT_EXPR for an SSA name. Since a single SSA name may have more than one assertion associated with it, these locations are kept in a linked list attached to the corresponding SSA name. */ struct assert_locus_d { /* Basic block where the assertion would be inserted. */ basic_block bb; /* Some assertions need to be inserted on an edge (e.g., assertions generated by COND_EXPRs). In those cases, BB will be NULL. */ edge e; /* Pointer to the statement that generated this assertion. */ block_stmt_iterator si; /* Predicate code for the ASSERT_EXPR. Must be COMPARISON_CLASS_P. */ enum tree_code comp_code; /* Value being compared against. */ tree val; /* Next node in the linked list. */ struct assert_locus_d *next; }; typedef struct assert_locus_d *assert_locus_t; /* If bit I is present, it means that SSA name N_i has a list of assertions that should be inserted in the IL. */ static bitmap need_assert_for; /* Array of locations lists where to insert assertions. ASSERTS_FOR[I] holds a list of ASSERT_LOCUS_T nodes that describe where ASSERT_EXPRs for SSA name N_I should be inserted. */ static assert_locus_t *asserts_for; /* Set of blocks visited in find_assert_locations. Used to avoid visiting the same block more than once. */ static sbitmap blocks_visited; /* Value range array. After propagation, VR_VALUE[I] holds the range of values that SSA name N_I may take. */ static value_range_t **vr_value; /* Return true if ARG is marked with the nonnull attribute in the current function signature. */ static bool nonnull_arg_p (tree arg) { tree t, attrs, fntype; unsigned HOST_WIDE_INT arg_num; gcc_assert (TREE_CODE (arg) == PARM_DECL && POINTER_TYPE_P (TREE_TYPE (arg))); fntype = TREE_TYPE (current_function_decl); attrs = lookup_attribute ("nonnull", TYPE_ATTRIBUTES (fntype)); /* If "nonnull" wasn't specified, we know nothing about the argument. */ if (attrs == NULL_TREE) return false; /* If "nonnull" applies to all the arguments, then ARG is non-null. */ if (TREE_VALUE (attrs) == NULL_TREE) return true; /* Get the position number for ARG in the function signature. */ for (arg_num = 1, t = DECL_ARGUMENTS (current_function_decl); t; t = TREE_CHAIN (t), arg_num++) { if (t == arg) break; } gcc_assert (t == arg); /* Now see if ARG_NUM is mentioned in the nonnull list. */ for (t = TREE_VALUE (attrs); t; t = TREE_CHAIN (t)) { if (compare_tree_int (TREE_VALUE (t), arg_num) == 0) return true; } return false; } /* Set value range VR to {T, MIN, MAX, EQUIV}. */ static void set_value_range (value_range_t *vr, enum value_range_type t, tree min, tree max, bitmap equiv) { #if defined ENABLE_CHECKING /* Check the validity of the range. */ if (t == VR_RANGE || t == VR_ANTI_RANGE) { int cmp; gcc_assert (min && max); if (INTEGRAL_TYPE_P (TREE_TYPE (min)) && t == VR_ANTI_RANGE) gcc_assert (min != TYPE_MIN_VALUE (TREE_TYPE (min)) || max != TYPE_MAX_VALUE (TREE_TYPE (max))); cmp = compare_values (min, max); gcc_assert (cmp == 0 || cmp == -1 || cmp == -2); } if (t == VR_UNDEFINED || t == VR_VARYING) gcc_assert (min == NULL_TREE && max == NULL_TREE); if (t == VR_UNDEFINED || t == VR_VARYING) gcc_assert (equiv == NULL || bitmap_empty_p (equiv)); #endif vr->type = t; vr->min = min; vr->max = max; /* Since updating the equivalence set involves deep copying the bitmaps, only do it if absolutely necessary. */ if (vr->equiv == NULL) vr->equiv = BITMAP_ALLOC (NULL); if (equiv != vr->equiv) { if (equiv && !bitmap_empty_p (equiv)) bitmap_copy (vr->equiv, equiv); else bitmap_clear (vr->equiv); } } /* Copy value range FROM into value range TO. */ static inline void copy_value_range (value_range_t *to, value_range_t *from) { set_value_range (to, from->type, from->min, from->max, from->equiv); } /* Set value range VR to a non-NULL range of type TYPE. */ static inline void set_value_range_to_nonnull (value_range_t *vr, tree type) { tree zero = build_int_cst (type, 0); set_value_range (vr, VR_ANTI_RANGE, zero, zero, vr->equiv); } /* Set value range VR to a NULL range of type TYPE. */ static inline void set_value_range_to_null (value_range_t *vr, tree type) { tree zero = build_int_cst (type, 0); set_value_range (vr, VR_RANGE, zero, zero, vr->equiv); } /* Set value range VR to VR_VARYING. */ static inline void set_value_range_to_varying (value_range_t *vr) { vr->type = VR_VARYING; vr->min = vr->max = NULL_TREE; if (vr->equiv) bitmap_clear (vr->equiv); } /* Set value range VR to VR_UNDEFINED. */ static inline void set_value_range_to_undefined (value_range_t *vr) { vr->type = VR_UNDEFINED; vr->min = vr->max = NULL_TREE; if (vr->equiv) bitmap_clear (vr->equiv); } /* Return value range information for VAR. Create an empty range if none existed. */ static value_range_t * get_value_range (tree var) { value_range_t *vr; tree sym; unsigned ver = SSA_NAME_VERSION (var); vr = vr_value[ver]; if (vr) return vr; /* Create a default value range. */ vr_value[ver] = vr = XNEW (value_range_t); memset (vr, 0, sizeof (*vr)); /* Allocate an equivalence set. */ vr->equiv = BITMAP_ALLOC (NULL); /* If VAR is a default definition, the variable can take any value in VAR's type. */ sym = SSA_NAME_VAR (var); if (var == default_def (sym)) { /* Try to use the "nonnull" attribute to create ~[0, 0] anti-ranges for pointers. Note that this is only valid with default definitions of PARM_DECLs. */ if (TREE_CODE (sym) == PARM_DECL && POINTER_TYPE_P (TREE_TYPE (sym)) && nonnull_arg_p (sym)) set_value_range_to_nonnull (vr, TREE_TYPE (sym)); else set_value_range_to_varying (vr); } return vr; } /* Update the value range and equivalence set for variable VAR to NEW_VR. Return true if NEW_VR is different from VAR's previous value. NOTE: This function assumes that NEW_VR is a temporary value range object created for the sole purpose of updating VAR's range. The storage used by the equivalence set from NEW_VR will be freed by this function. Do not call update_value_range when NEW_VR is the range object associated with another SSA name. */ static inline bool update_value_range (tree var, value_range_t *new_vr) { value_range_t *old_vr; bool is_new; /* Update the value range, if necessary. */ old_vr = get_value_range (var); is_new = old_vr->type != new_vr->type || old_vr->min != new_vr->min || old_vr->max != new_vr->max || (old_vr->equiv == NULL && new_vr->equiv) || (old_vr->equiv && new_vr->equiv == NULL) || (!bitmap_equal_p (old_vr->equiv, new_vr->equiv)); if (is_new) set_value_range (old_vr, new_vr->type, new_vr->min, new_vr->max, new_vr->equiv); BITMAP_FREE (new_vr->equiv); new_vr->equiv = NULL; return is_new; } /* Add VAR and VAR's equivalence set to EQUIV. */ static void add_equivalence (bitmap equiv, tree var) { unsigned ver = SSA_NAME_VERSION (var); value_range_t *vr = vr_value[ver]; bitmap_set_bit (equiv, ver); if (vr && vr->equiv) bitmap_ior_into (equiv, vr->equiv); } /* Return true if VR is ~[0, 0]. */ static inline bool range_is_nonnull (value_range_t *vr) { return vr->type == VR_ANTI_RANGE && integer_zerop (vr->min) && integer_zerop (vr->max); } /* Return true if VR is [0, 0]. */ static inline bool range_is_null (value_range_t *vr) { return vr->type == VR_RANGE && integer_zerop (vr->min) && integer_zerop (vr->max); } /* Return true if value range VR involves at least one symbol. */ static inline bool symbolic_range_p (value_range_t *vr) { return (!is_gimple_min_invariant (vr->min) || !is_gimple_min_invariant (vr->max)); } /* Like tree_expr_nonzero_p, but this function uses value ranges obtained so far. */ static bool vrp_expr_computes_nonzero (tree expr) { if (tree_expr_nonzero_p (expr)) return true; /* If we have an expression of the form &X->a, then the expression is nonnull if X is nonnull. */ if (TREE_CODE (expr) == ADDR_EXPR) { tree base = get_base_address (TREE_OPERAND (expr, 0)); if (base != NULL_TREE && TREE_CODE (base) == INDIRECT_REF && TREE_CODE (TREE_OPERAND (base, 0)) == SSA_NAME) { value_range_t *vr = get_value_range (TREE_OPERAND (base, 0)); if (range_is_nonnull (vr)) return true; } } return false; } /* Compare two values VAL1 and VAL2. Return -2 if VAL1 and VAL2 cannot be compared at compile-time, -1 if VAL1 < VAL2, 0 if VAL1 == VAL2, +1 if VAL1 > VAL2, and +2 if VAL1 != VAL2 This is similar to tree_int_cst_compare but supports pointer values and values that cannot be compared at compile time. */ static int compare_values (tree val1, tree val2) { if (val1 == val2) return 0; /* Below we rely on the fact that VAL1 and VAL2 are both pointers or both integers. */ gcc_assert (POINTER_TYPE_P (TREE_TYPE (val1)) == POINTER_TYPE_P (TREE_TYPE (val2))); /* Do some limited symbolic comparisons. */ if (!POINTER_TYPE_P (TREE_TYPE (val1))) { /* We can determine some comparisons against +INF and -INF even if the other value is an expression. */ if (val1 == TYPE_MAX_VALUE (TREE_TYPE (val1)) && TREE_CODE (val2) == MINUS_EXPR) { /* +INF > NAME - CST. */ return 1; } else if (val1 == TYPE_MIN_VALUE (TREE_TYPE (val1)) && TREE_CODE (val2) == PLUS_EXPR) { /* -INF < NAME + CST. */ return -1; } else if (TREE_CODE (val1) == MINUS_EXPR && val2 == TYPE_MAX_VALUE (TREE_TYPE (val2))) { /* NAME - CST < +INF. */ return -1; } else if (TREE_CODE (val1) == PLUS_EXPR && val2 == TYPE_MIN_VALUE (TREE_TYPE (val2))) { /* NAME + CST > -INF. */ return 1; } } if ((TREE_CODE (val1) == SSA_NAME || TREE_CODE (val1) == PLUS_EXPR || TREE_CODE (val1) == MINUS_EXPR) && (TREE_CODE (val2) == SSA_NAME || TREE_CODE (val2) == PLUS_EXPR || TREE_CODE (val2) == MINUS_EXPR)) { tree n1, c1, n2, c2; /* If VAL1 and VAL2 are of the form 'NAME [+-] CST' or 'NAME', return -1 or +1 accordingly. If VAL1 and VAL2 don't use the same name, return -2. */ if (TREE_CODE (val1) == SSA_NAME) { n1 = val1; c1 = NULL_TREE; } else { n1 = TREE_OPERAND (val1, 0); c1 = TREE_OPERAND (val1, 1); } if (TREE_CODE (val2) == SSA_NAME) { n2 = val2; c2 = NULL_TREE; } else { n2 = TREE_OPERAND (val2, 0); c2 = TREE_OPERAND (val2, 1); } /* Both values must use the same name. */ if (n1 != n2) return -2; if (TREE_CODE (val1) == SSA_NAME) { if (TREE_CODE (val2) == SSA_NAME) /* NAME == NAME */ return 0; else if (TREE_CODE (val2) == PLUS_EXPR) /* NAME < NAME + CST */ return -1; else if (TREE_CODE (val2) == MINUS_EXPR) /* NAME > NAME - CST */ return 1; } else if (TREE_CODE (val1) == PLUS_EXPR) { if (TREE_CODE (val2) == SSA_NAME) /* NAME + CST > NAME */ return 1; else if (TREE_CODE (val2) == PLUS_EXPR) /* NAME + CST1 > NAME + CST2, if CST1 > CST2 */ return compare_values (c1, c2); else if (TREE_CODE (val2) == MINUS_EXPR) /* NAME + CST1 > NAME - CST2 */ return 1; } else if (TREE_CODE (val1) == MINUS_EXPR) { if (TREE_CODE (val2) == SSA_NAME) /* NAME - CST < NAME */ return -1; else if (TREE_CODE (val2) == PLUS_EXPR) /* NAME - CST1 < NAME + CST2 */ return -1; else if (TREE_CODE (val2) == MINUS_EXPR) /* NAME - CST1 > NAME - CST2, if CST1 < CST2. Notice that C1 and C2 are swapped in the call to compare_values. */ return compare_values (c2, c1); } gcc_unreachable (); } /* We cannot compare non-constants. */ if (!is_gimple_min_invariant (val1) || !is_gimple_min_invariant (val2)) return -2; if (!POINTER_TYPE_P (TREE_TYPE (val1))) { /* We cannot compare overflowed values. */ if (TREE_OVERFLOW (val1) || TREE_OVERFLOW (val2)) return -2; return tree_int_cst_compare (val1, val2); } else { tree t; /* First see if VAL1 and VAL2 are not the same. */ if (val1 == val2 || operand_equal_p (val1, val2, 0)) return 0; /* If VAL1 is a lower address than VAL2, return -1. */ t = fold_binary (LT_EXPR, boolean_type_node, val1, val2); if (t == boolean_true_node) return -1; /* If VAL1 is a higher address than VAL2, return +1. */ t = fold_binary (GT_EXPR, boolean_type_node, val1, val2); if (t == boolean_true_node) return 1; /* If VAL1 is different than VAL2, return +2. */ t = fold_binary (NE_EXPR, boolean_type_node, val1, val2); if (t == boolean_true_node) return 2; return -2; } } /* Return 1 if VAL is inside value range VR (VR->MIN <= VAL <= VR->MAX), 0 if VAL is not inside VR, -2 if we cannot tell either way. FIXME, the current semantics of this functions are a bit quirky when taken in the context of VRP. In here we do not care about VR's type. If VR is the anti-range ~[3, 5] the call value_inside_range (4, VR) will return 1. This is counter-intuitive in a strict sense, but the callers currently expect this. They are calling the function merely to determine whether VR->MIN <= VAL <= VR->MAX. The callers are applying the VR_RANGE/VR_ANTI_RANGE semantics themselves. This also applies to value_ranges_intersect_p and range_includes_zero_p. The semantics of VR_RANGE and VR_ANTI_RANGE should be encoded here, but that also means adapting the users of these functions to the new semantics. */ static inline int value_inside_range (tree val, value_range_t *vr) { int cmp1, cmp2; cmp1 = compare_values (val, vr->min); if (cmp1 == -2 || cmp1 == 2) return -2; cmp2 = compare_values (val, vr->max); if (cmp2 == -2 || cmp2 == 2) return -2; return (cmp1 == 0 || cmp1 == 1) && (cmp2 == -1 || cmp2 == 0); } /* Return true if value ranges VR0 and VR1 have a non-empty intersection. */ static inline bool value_ranges_intersect_p (value_range_t *vr0, value_range_t *vr1) { return (value_inside_range (vr1->min, vr0) == 1 || value_inside_range (vr1->max, vr0) == 1 || value_inside_range (vr0->min, vr1) == 1 || value_inside_range (vr0->max, vr1) == 1); } /* Return true if VR includes the value zero, false otherwise. FIXME, currently this will return false for an anti-range like ~[-4, 3]. This will be wrong when the semantics of value_inside_range are modified (currently the users of this function expect these semantics). */ static inline bool range_includes_zero_p (value_range_t *vr) { tree zero; gcc_assert (vr->type != VR_UNDEFINED && vr->type != VR_VARYING && !symbolic_range_p (vr)); zero = build_int_cst (TREE_TYPE (vr->min), 0); return (value_inside_range (zero, vr) == 1); } /* When extracting ranges from X_i = ASSERT_EXPR , we will initially consider X_i and Y_j equivalent, so the equivalence set of Y_j is added to the equivalence set of X_i. However, it is possible to have a chain of ASSERT_EXPRs whose predicates are actually incompatible. This is usually the result of nesting of contradictory if-then-else statements. For instance, in PR 24670: count_4 has range [-INF, 63] if (count_4 != 0) { count_19 = ASSERT_EXPR if (count_19 > 63) { count_18 = ASSERT_EXPR 63> if (count_18 <= 63) ... } } Notice that 'if (count_19 > 63)' is trivially false and will be folded out at the end. However, during propagation, the flowgraph is not cleaned up and so, VRP will evaluate predicates more predicates than necessary, so it must support these inconsistencies. The problem here is that because of the chaining of ASSERT_EXPRs, the equivalency set for count_18 includes count_4. Since count_4 has an incompatible range, we ICE when evaluating the ranges in the equivalency set. So, we need to remove count_4 from it. */ static void fix_equivalence_set (value_range_t *vr_p) { bitmap_iterator bi; unsigned i; bitmap e = vr_p->equiv; bitmap to_remove = BITMAP_ALLOC (NULL); /* Only detect inconsistencies on numeric ranges. */ if (vr_p->type == VR_VARYING || vr_p->type == VR_UNDEFINED || symbolic_range_p (vr_p)) return; EXECUTE_IF_SET_IN_BITMAP (e, 0, i, bi) { value_range_t *equiv_vr = vr_value[i]; if (equiv_vr->type == VR_VARYING || equiv_vr->type == VR_UNDEFINED || symbolic_range_p (equiv_vr)) continue; if (equiv_vr->type == VR_RANGE && vr_p->type == VR_RANGE && !value_ranges_intersect_p (vr_p, equiv_vr)) bitmap_set_bit (to_remove, i); else if ((equiv_vr->type == VR_RANGE && vr_p->type == VR_ANTI_RANGE) || (equiv_vr->type == VR_ANTI_RANGE && vr_p->type == VR_RANGE)) { /* A range and an anti-range have an empty intersection if their end points are the same. FIXME, value_ranges_intersect_p should handle this automatically. */ if (compare_values (equiv_vr->min, vr_p->min) == 0 && compare_values (equiv_vr->max, vr_p->max) == 0) bitmap_set_bit (to_remove, i); } } bitmap_and_compl_into (vr_p->equiv, to_remove); BITMAP_FREE (to_remove); } /* Extract value range information from an ASSERT_EXPR EXPR and store it in *VR_P. */ static void extract_range_from_assert (value_range_t *vr_p, tree expr) { tree var, cond, limit, min, max, type; value_range_t *var_vr, *limit_vr; enum tree_code cond_code; var = ASSERT_EXPR_VAR (expr); cond = ASSERT_EXPR_COND (expr); gcc_assert (COMPARISON_CLASS_P (cond)); /* Find VAR in the ASSERT_EXPR conditional. */ if (var == TREE_OPERAND (cond, 0)) { /* If the predicate is of the form VAR COMP LIMIT, then we just take LIMIT from the RHS and use the same comparison code. */ limit = TREE_OPERAND (cond, 1); cond_code = TREE_CODE (cond); } else { /* If the predicate is of the form LIMIT COMP VAR, then we need to flip around the comparison code to create the proper range for VAR. */ limit = TREE_OPERAND (cond, 0); cond_code = swap_tree_comparison (TREE_CODE (cond)); } type = TREE_TYPE (limit); gcc_assert (limit != var); /* For pointer arithmetic, we only keep track of pointer equality and inequality. */ if (POINTER_TYPE_P (type) && cond_code != NE_EXPR && cond_code != EQ_EXPR) { set_value_range_to_varying (vr_p); return; } /* If LIMIT is another SSA name and LIMIT has a range of its own, try to use LIMIT's range to avoid creating symbolic ranges unnecessarily. */ limit_vr = (TREE_CODE (limit) == SSA_NAME) ? get_value_range (limit) : NULL; /* LIMIT's range is only interesting if it has any useful information. */ if (limit_vr && (limit_vr->type == VR_UNDEFINED || limit_vr->type == VR_VARYING || symbolic_range_p (limit_vr))) limit_vr = NULL; /* Special handling for integral types with super-types. Some FEs construct integral types derived from other types and restrict the range of values these new types may take. It may happen that LIMIT is actually smaller than TYPE's minimum value. For instance, the Ada FE is generating code like this during bootstrap: D.1480_32 = nam_30 - 300000361; if (D.1480_32 <= 1) goto ; else goto ; :; D.1480_94 = ASSERT_EXPR ; All the names are of type types__name_id___XDLU_300000000__399999999 which has min == 300000000 and max == 399999999. This means that the ASSERT_EXPR would try to create the range [3000000, 1] which is invalid. The fact that the type specifies MIN and MAX values does not automatically mean that every variable of that type will always be within that range, so the predicate may well be true at run time. If we had symbolic -INF and +INF values, we could represent this range, but we currently represent -INF and +INF using the type's min and max values. So, the only sensible thing we can do for now is set the resulting range to VR_VARYING. TODO, would having symbolic -INF and +INF values be worth the trouble? */ if (TREE_CODE (limit) != SSA_NAME && INTEGRAL_TYPE_P (type) && TREE_TYPE (type)) { if (cond_code == LE_EXPR || cond_code == LT_EXPR) { tree type_min = TYPE_MIN_VALUE (type); int cmp = compare_values (limit, type_min); /* For < or <= comparisons, if LIMIT is smaller than TYPE_MIN, set the range to VR_VARYING. */ if (cmp == -1 || cmp == 0) { set_value_range_to_varying (vr_p); return; } } else if (cond_code == GE_EXPR || cond_code == GT_EXPR) { tree type_max = TYPE_MIN_VALUE (type); int cmp = compare_values (limit, type_max); /* For > or >= comparisons, if LIMIT is bigger than TYPE_MAX, set the range to VR_VARYING. */ if (cmp == 1 || cmp == 0) { set_value_range_to_varying (vr_p); return; } } } /* Initially, the new range has the same set of equivalences of VAR's range. This will be revised before returning the final value. Since assertions may be chained via mutually exclusive predicates, we will need to trim the set of equivalences before we are done. */ gcc_assert (vr_p->equiv == NULL); vr_p->equiv = BITMAP_ALLOC (NULL); add_equivalence (vr_p->equiv, var); /* Extract a new range based on the asserted comparison for VAR and LIMIT's value range. Notice that if LIMIT has an anti-range, we will only use it for equality comparisons (EQ_EXPR). For any other kind of assertion, we cannot derive a range from LIMIT's anti-range that can be used to describe the new range. For instance, ASSERT_EXPR . If b_4 is ~[2, 10], then b_4 takes on the ranges [-INF, 1] and [11, +INF]. There is no single range for x_2 that could describe LE_EXPR, so we might as well build the range [b_4, +INF] for it. */ if (cond_code == EQ_EXPR) { enum value_range_type range_type; if (limit_vr) { range_type = limit_vr->type; min = limit_vr->min; max = limit_vr->max; } else { range_type = VR_RANGE; min = limit; max = limit; } set_value_range (vr_p, range_type, min, max, vr_p->equiv); /* When asserting the equality VAR == LIMIT and LIMIT is another SSA name, the new range will also inherit the equivalence set from LIMIT. */ if (TREE_CODE (limit) == SSA_NAME) add_equivalence (vr_p->equiv, limit); } else if (cond_code == NE_EXPR) { /* As described above, when LIMIT's range is an anti-range and this assertion is an inequality (NE_EXPR), then we cannot derive anything from the anti-range. For instance, if LIMIT's range was ~[0, 0], the assertion 'VAR != LIMIT' does not imply that VAR's range is [0, 0]. So, in the case of anti-ranges, we just assert the inequality using LIMIT and not its anti-range. If LIMIT_VR is a range, we can only use it to build a new anti-range if LIMIT_VR is a single-valued range. For instance, if LIMIT_VR is [0, 1], the predicate VAR != [0, 1] does not mean that VAR's range is ~[0, 1]. Rather, it means that for value 0 VAR should be ~[0, 0] and for value 1, VAR should be ~[1, 1]. We cannot represent these ranges. The only situation in which we can build a valid anti-range is when LIMIT_VR is a single-valued range (i.e., LIMIT_VR->MIN == LIMIT_VR->MAX). In that case, build the anti-range ~[LIMIT_VR->MIN, LIMIT_VR->MAX]. */ if (limit_vr && limit_vr->type == VR_RANGE && compare_values (limit_vr->min, limit_vr->max) == 0) { min = limit_vr->min; max = limit_vr->max; } else { /* In any other case, we cannot use LIMIT's range to build a valid anti-range. */ min = max = limit; } /* If MIN and MAX cover the whole range for their type, then just use the original LIMIT. */ if (INTEGRAL_TYPE_P (type) && min == TYPE_MIN_VALUE (type) && max == TYPE_MAX_VALUE (type)) min = max = limit; set_value_range (vr_p, VR_ANTI_RANGE, min, max, vr_p->equiv); } else if (cond_code == LE_EXPR || cond_code == LT_EXPR) { min = TYPE_MIN_VALUE (type); if (limit_vr == NULL || limit_vr->type == VR_ANTI_RANGE) max = limit; else { /* If LIMIT_VR is of the form [N1, N2], we need to build the range [MIN, N2] for LE_EXPR and [MIN, N2 - 1] for LT_EXPR. */ max = limit_vr->max; } /* For LT_EXPR, we create the range [MIN, MAX - 1]. */ if (cond_code == LT_EXPR) { tree one = build_int_cst (type, 1); max = fold_build2 (MINUS_EXPR, type, max, one); } set_value_range (vr_p, VR_RANGE, min, max, vr_p->equiv); } else if (cond_code == GE_EXPR || cond_code == GT_EXPR) { max = TYPE_MAX_VALUE (type); if (limit_vr == NULL || limit_vr->type == VR_ANTI_RANGE) min = limit; else { /* If LIMIT_VR is of the form [N1, N2], we need to build the range [N1, MAX] for GE_EXPR and [N1 + 1, MAX] for GT_EXPR. */ min = limit_vr->min; } /* For GT_EXPR, we create the range [MIN + 1, MAX]. */ if (cond_code == GT_EXPR) { tree one = build_int_cst (type, 1); min = fold_build2 (PLUS_EXPR, type, min, one); } set_value_range (vr_p, VR_RANGE, min, max, vr_p->equiv); } else gcc_unreachable (); /* If VAR already had a known range, it may happen that the new range we have computed and VAR's range are not compatible. For instance, if (p_5 == NULL) p_6 = ASSERT_EXPR ; x_7 = p_6->fld; p_8 = ASSERT_EXPR ; While the above comes from a faulty program, it will cause an ICE later because p_8 and p_6 will have incompatible ranges and at the same time will be considered equivalent. A similar situation would arise from if (i_5 > 10) i_6 = ASSERT_EXPR 10>; if (i_5 < 5) i_7 = ASSERT_EXPR ; Again i_6 and i_7 will have incompatible ranges. It would be pointless to try and do anything with i_7's range because anything dominated by 'if (i_5 < 5)' will be optimized away. Note, due to the wa in which simulation proceeds, the statement i_7 = ASSERT_EXPR <...> we would never be visited because the conditional 'if (i_5 < 5)' always evaluates to false. However, this extra check does not hurt and may protect against future changes to VRP that may get into a situation similar to the NULL pointer dereference example. Note that these compatibility tests are only needed when dealing with ranges or a mix of range and anti-range. If VAR_VR and VR_P are both anti-ranges, they will always be compatible, because two anti-ranges will always have a non-empty intersection. */ var_vr = get_value_range (var); /* We may need to make adjustments when VR_P and VAR_VR are numeric ranges or anti-ranges. */ if (vr_p->type == VR_VARYING || vr_p->type == VR_UNDEFINED || var_vr->type == VR_VARYING || var_vr->type == VR_UNDEFINED || symbolic_range_p (vr_p) || symbolic_range_p (var_vr)) goto done; if (var_vr->type == VR_RANGE && vr_p->type == VR_RANGE) { /* If the two ranges have a non-empty intersection, we can refine the resulting range. Since the assert expression creates an equivalency and at the same time it asserts a predicate, we can take the intersection of the two ranges to get better precision. */ if (value_ranges_intersect_p (var_vr, vr_p)) { /* Use the larger of the two minimums. */ if (compare_values (vr_p->min, var_vr->min) == -1) min = var_vr->min; else min = vr_p->min; /* Use the smaller of the two maximums. */ if (compare_values (vr_p->max, var_vr->max) == 1) max = var_vr->max; else max = vr_p->max; set_value_range (vr_p, vr_p->type, min, max, vr_p->equiv); } else { /* The two ranges do not intersect, set the new range to VARYING, because we will not be able to do anything meaningful with it. */ set_value_range_to_varying (vr_p); } } else if ((var_vr->type == VR_RANGE && vr_p->type == VR_ANTI_RANGE) || (var_vr->type == VR_ANTI_RANGE && vr_p->type == VR_RANGE)) { /* A range and an anti-range will cancel each other only if their ends are the same. For instance, in the example above, p_8's range ~[0, 0] and p_6's range [0, 0] are incompatible, so VR_P should be set to VR_VARYING. */ if (compare_values (var_vr->min, vr_p->min) == 0 && compare_values (var_vr->max, vr_p->max) == 0) set_value_range_to_varying (vr_p); else { tree min, max, anti_min, anti_max, real_min, real_max; /* We want to compute the logical AND of the two ranges; there are three cases to consider. 1. The VR_ANTI_RANGE range is competely within the VR_RANGE and the endpoints of the ranges are different. In that case the resulting range should be whichever range is more precise. Typically that will be the VR_RANGE. 2. The VR_ANTI_RANGE is completely disjoint from the VR_RANGE. In this case the resulting range should be the VR_RANGE. 3. There is some overlap between the VR_ANTI_RANGE and the VR_RANGE. 3a. If the high limit of the VR_ANTI_RANGE resides within the VR_RANGE, then the result is a new VR_RANGE starting at the high limit of the the VR_ANTI_RANGE + 1 and extending to the high limit of the original VR_RANGE. 3b. If the low limit of the VR_ANTI_RANGE resides within the VR_RANGE, then the result is a new VR_RANGE starting at the low limit of the original VR_RANGE and extending to the low limit of the VR_ANTI_RANGE - 1. */ if (vr_p->type == VR_ANTI_RANGE) { anti_min = vr_p->min; anti_max = vr_p->max; real_min = var_vr->min; real_max = var_vr->max; } else { anti_min = var_vr->min; anti_max = var_vr->max; real_min = vr_p->min; real_max = vr_p->max; } /* Case 1, VR_ANTI_RANGE completely within VR_RANGE, not including any endpoints. */ if (compare_values (anti_max, real_max) == -1 && compare_values (anti_min, real_min) == 1) { set_value_range (vr_p, VR_RANGE, real_min, real_max, vr_p->equiv); } /* Case 2, VR_ANTI_RANGE completely disjoint from VR_RANGE. */ else if (compare_values (anti_min, real_max) == 1 || compare_values (anti_max, real_min) == -1) { set_value_range (vr_p, VR_RANGE, real_min, real_max, vr_p->equiv); } /* Case 3a, the anti-range extends into the low part of the real range. Thus creating a new low for the real reange. */ else if ((compare_values (anti_max, real_min) == 1 || compare_values (anti_max, real_min) == 0) && compare_values (anti_max, real_max) == -1) { min = fold_build2 (PLUS_EXPR, TREE_TYPE (var_vr->min), anti_max, build_int_cst (TREE_TYPE (var_vr->min), 1)); max = real_max; set_value_range (vr_p, VR_RANGE, min, max, vr_p->equiv); } /* Case 3b, the anti-range extends into the high part of the real range. Thus creating a new higher for the real reange. */ else if (compare_values (anti_min, real_min) == 1 && (compare_values (anti_min, real_max) == -1 || compare_values (anti_min, real_max) == 0)) { max = fold_build2 (MINUS_EXPR, TREE_TYPE (var_vr->min), anti_min, build_int_cst (TREE_TYPE (var_vr->min), 1)); min = real_min; set_value_range (vr_p, VR_RANGE, min, max, vr_p->equiv); } } } /* Remove names from the equivalence set that have ranges incompatible with VR_P. */ done: fix_equivalence_set (vr_p); } /* Extract range information from SSA name VAR and store it in VR. If VAR has an interesting range, use it. Otherwise, create the range [VAR, VAR] and return it. This is useful in situations where we may have conditionals testing values of VARYING names. For instance, x_3 = y_5; if (x_3 > y_5) ... Even if y_5 is deemed VARYING, we can determine that x_3 > y_5 is always false. */ static void extract_range_from_ssa_name (value_range_t *vr, tree var) { value_range_t *var_vr = get_value_range (var); if (var_vr->type != VR_UNDEFINED && var_vr->type != VR_VARYING) copy_value_range (vr, var_vr); else set_value_range (vr, VR_RANGE, var, var, NULL); add_equivalence (vr->equiv, var); } /* Wrapper around int_const_binop. If the operation overflows and we are not using wrapping arithmetic, then adjust the result to be -INF or +INF depending on CODE, VAL1 and VAL2. */ static inline tree vrp_int_const_binop (enum tree_code code, tree val1, tree val2) { tree res; if (flag_wrapv) return int_const_binop (code, val1, val2, 0); /* If we are not using wrapping arithmetic, operate symbolically on -INF and +INF. */ res = int_const_binop (code, val1, val2, 0); if (TYPE_UNSIGNED (TREE_TYPE (val1))) { int checkz = compare_values (res, val1); /* Ensure that res = val1 [+*] val2 >= val1 or that res = val1 - val2 <= val1. */ if (((code == PLUS_EXPR || code == MULT_EXPR) && !(checkz == 1 || checkz == 0)) || (code == MINUS_EXPR && !(checkz == 0 || checkz == -1))) { res = copy_node (res); TREE_OVERFLOW (res) = 1; } } else if (TREE_OVERFLOW (res) && !TREE_OVERFLOW (val1) && !TREE_OVERFLOW (val2)) { /* If the operation overflowed but neither VAL1 nor VAL2 are overflown, return -INF or +INF depending on the operation and the combination of signs of the operands. */ int sgn1 = tree_int_cst_sgn (val1); int sgn2 = tree_int_cst_sgn (val2); /* Notice that we only need to handle the restricted set of operations handled by extract_range_from_binary_expr. Among them, only multiplication, addition and subtraction can yield overflow without overflown operands because we are working with integral types only... except in the case VAL1 = -INF and VAL2 = -1 which overflows to +INF for division too. */ /* For multiplication, the sign of the overflow is given by the comparison of the signs of the operands. */ if ((code == MULT_EXPR && sgn1 == sgn2) /* For addition, the operands must be of the same sign to yield an overflow. Its sign is therefore that of one of the operands, for example the first. */ || (code == PLUS_EXPR && sgn1 > 0) /* For subtraction, the operands must be of different signs to yield an overflow. Its sign is therefore that of the first operand or the opposite of that of the second operand. A first operand of 0 counts as positive here, for the corner case 0 - (-INF), which overflows, but must yield +INF. */ || (code == MINUS_EXPR && sgn1 >= 0) /* For division, the only case is -INF / -1 = +INF. */ || code == TRUNC_DIV_EXPR || code == FLOOR_DIV_EXPR || code == CEIL_DIV_EXPR || code == EXACT_DIV_EXPR || code == ROUND_DIV_EXPR) return TYPE_MAX_VALUE (TREE_TYPE (res)); else return TYPE_MIN_VALUE (TREE_TYPE (res)); } return res; } /* Extract range information from a binary expression EXPR based on the ranges of each of its operands and the expression code. */ static void extract_range_from_binary_expr (value_range_t *vr, tree expr) { enum tree_code code = TREE_CODE (expr); enum value_range_type type; tree op0, op1, min, max; int cmp; value_range_t vr0 = { VR_UNDEFINED, NULL_TREE, NULL_TREE, NULL }; value_range_t vr1 = { VR_UNDEFINED, NULL_TREE, NULL_TREE, NULL }; /* Not all binary expressions can be applied to ranges in a meaningful way. Handle only arithmetic operations. */ if (code != PLUS_EXPR && code != MINUS_EXPR && code != MULT_EXPR && code != TRUNC_DIV_EXPR && code != FLOOR_DIV_EXPR && code != CEIL_DIV_EXPR && code != EXACT_DIV_EXPR && code != ROUND_DIV_EXPR && code != MIN_EXPR && code != MAX_EXPR && code != BIT_AND_EXPR && code != TRUTH_ANDIF_EXPR && code != TRUTH_ORIF_EXPR && code != TRUTH_AND_EXPR && code != TRUTH_OR_EXPR && code != TRUTH_XOR_EXPR) { set_value_range_to_varying (vr); return; } /* Get value ranges for each operand. For constant operands, create a new value range with the operand to simplify processing. */ op0 = TREE_OPERAND (expr, 0); if (TREE_CODE (op0) == SSA_NAME) vr0 = *(get_value_range (op0)); else if (is_gimple_min_invariant (op0)) set_value_range (&vr0, VR_RANGE, op0, op0, NULL); else set_value_range_to_varying (&vr0); op1 = TREE_OPERAND (expr, 1); if (TREE_CODE (op1) == SSA_NAME) vr1 = *(get_value_range (op1)); else if (is_gimple_min_invariant (op1)) set_value_range (&vr1, VR_RANGE, op1, op1, NULL); else set_value_range_to_varying (&vr1); /* If either range is UNDEFINED, so is the result. */ if (vr0.type == VR_UNDEFINED || vr1.type == VR_UNDEFINED) { set_value_range_to_undefined (vr); return; } /* The type of the resulting value range defaults to VR0.TYPE. */ type = vr0.type; /* Refuse to operate on VARYING ranges, ranges of different kinds and symbolic ranges. As an exception, we allow BIT_AND_EXPR because we may be able to derive a useful range even if one of the operands is VR_VARYING or symbolic range. TODO, we may be able to derive anti-ranges in some cases. */ if (code != BIT_AND_EXPR && code != TRUTH_AND_EXPR && code != TRUTH_OR_EXPR && (vr0.type == VR_VARYING || vr1.type == VR_VARYING || vr0.type != vr1.type || symbolic_range_p (&vr0) || symbolic_range_p (&vr1))) { set_value_range_to_varying (vr); return; } /* Now evaluate the expression to determine the new range. */ if (POINTER_TYPE_P (TREE_TYPE (expr)) || POINTER_TYPE_P (TREE_TYPE (op0)) || POINTER_TYPE_P (TREE_TYPE (op1))) { /* For pointer types, we are really only interested in asserting whether the expression evaluates to non-NULL. FIXME, we used to gcc_assert (code == PLUS_EXPR || code == MINUS_EXPR), but ivopts is generating expressions with pointer multiplication in them. */ if (code == PLUS_EXPR) { if (range_is_nonnull (&vr0) || range_is_nonnull (&vr1)) set_value_range_to_nonnull (vr, TREE_TYPE (expr)); else if (range_is_null (&vr0) && range_is_null (&vr1)) set_value_range_to_null (vr, TREE_TYPE (expr)); else set_value_range_to_varying (vr); } else { /* Subtracting from a pointer, may yield 0, so just drop the resulting range to varying. */ set_value_range_to_varying (vr); } return; } /* For integer ranges, apply the operation to each end of the range and see what we end up with. */ if (code == TRUTH_ANDIF_EXPR || code == TRUTH_ORIF_EXPR || code == TRUTH_AND_EXPR || code == TRUTH_OR_EXPR || code == TRUTH_XOR_EXPR) { /* If one of the operands is zero, we know that the whole expression evaluates zero. */ if (code == TRUTH_AND_EXPR && ((vr0.type == VR_RANGE && integer_zerop (vr0.min) && integer_zerop (vr0.max)) || (vr1.type == VR_RANGE && integer_zerop (vr1.min) && integer_zerop (vr1.max)))) { type = VR_RANGE; min = max = build_int_cst (TREE_TYPE (expr), 0); } /* If one of the operands is one, we know that the whole expression evaluates one. */ else if (code == TRUTH_OR_EXPR && ((vr0.type == VR_RANGE && integer_onep (vr0.min) && integer_onep (vr0.max)) || (vr1.type == VR_RANGE && integer_onep (vr1.min) && integer_onep (vr1.max)))) { type = VR_RANGE; min = max = build_int_cst (TREE_TYPE (expr), 1); } else if (vr0.type != VR_VARYING && vr1.type != VR_VARYING && vr0.type == vr1.type && !symbolic_range_p (&vr0) && !symbolic_range_p (&vr1)) { /* Boolean expressions cannot be folded with int_const_binop. */ min = fold_binary (code, TREE_TYPE (expr), vr0.min, vr1.min); max = fold_binary (code, TREE_TYPE (expr), vr0.max, vr1.max); } else { set_value_range_to_varying (vr); return; } } else if (code == PLUS_EXPR || code == MIN_EXPR || code == MAX_EXPR) { /* If we have a PLUS_EXPR with two VR_ANTI_RANGEs, drop to VR_VARYING. It would take more effort to compute a precise range for such a case. For example, if we have op0 == 1 and op1 == -1 with their ranges both being ~[0,0], we would have op0 + op1 == 0, so we cannot claim that the sum is in ~[0,0]. Note that we are guaranteed to have vr0.type == vr1.type at this point. */ if (code == PLUS_EXPR && vr0.type == VR_ANTI_RANGE) { set_value_range_to_varying (vr); return; } /* For operations that make the resulting range directly proportional to the original ranges, apply the operation to the same end of each range. */ min = vrp_int_const_binop (code, vr0.min, vr1.min); max = vrp_int_const_binop (code, vr0.max, vr1.max); } else if (code == MULT_EXPR || code == TRUNC_DIV_EXPR || code == FLOOR_DIV_EXPR || code == CEIL_DIV_EXPR || code == EXACT_DIV_EXPR || code == ROUND_DIV_EXPR) { tree val[4]; size_t i; /* If we have an unsigned MULT_EXPR with two VR_ANTI_RANGEs, drop to VR_VARYING. It would take more effort to compute a precise range for such a case. For example, if we have op0 == 65536 and op1 == 65536 with their ranges both being ~[0,0] on a 32-bit machine, we would have op0 * op1 == 0, so we cannot claim that the product is in ~[0,0]. Note that we are guaranteed to have vr0.type == vr1.type at this point. */ if (code == MULT_EXPR && vr0.type == VR_ANTI_RANGE && (flag_wrapv || TYPE_UNSIGNED (TREE_TYPE (op0)))) { set_value_range_to_varying (vr); return; } /* Multiplications and divisions are a bit tricky to handle, depending on the mix of signs we have in the two ranges, we need to operate on different values to get the minimum and maximum values for the new range. One approach is to figure out all the variations of range combinations and do the operations. However, this involves several calls to compare_values and it is pretty convoluted. It's simpler to do the 4 operations (MIN0 OP MIN1, MIN0 OP MAX1, MAX0 OP MIN1 and MAX0 OP MAX0 OP MAX1) and then figure the smallest and largest values to form the new range. */ /* Divisions by zero result in a VARYING value. */ if (code != MULT_EXPR && (vr0.type == VR_ANTI_RANGE || range_includes_zero_p (&vr1))) { set_value_range_to_varying (vr); return; } /* Compute the 4 cross operations. */ val[0] = vrp_int_const_binop (code, vr0.min, vr1.min); val[1] = (vr1.max != vr1.min) ? vrp_int_const_binop (code, vr0.min, vr1.max) : NULL_TREE; val[2] = (vr0.max != vr0.min) ? vrp_int_const_binop (code, vr0.max, vr1.min) : NULL_TREE; val[3] = (vr0.min != vr0.max && vr1.min != vr1.max) ? vrp_int_const_binop (code, vr0.max, vr1.max) : NULL_TREE; /* Set MIN to the minimum of VAL[i] and MAX to the maximum of VAL[i]. */ min = val[0]; max = val[0]; for (i = 1; i < 4; i++) { if (!is_gimple_min_invariant (min) || TREE_OVERFLOW (min) || !is_gimple_min_invariant (max) || TREE_OVERFLOW (max)) break; if (val[i]) { if (!is_gimple_min_invariant (val[i]) || TREE_OVERFLOW (val[i])) { /* If we found an overflowed value, set MIN and MAX to it so that we set the resulting range to VARYING. */ min = max = val[i]; break; } if (compare_values (val[i], min) == -1) min = val[i]; if (compare_values (val[i], max) == 1) max = val[i]; } } } else if (code == MINUS_EXPR) { /* If we have a MINUS_EXPR with two VR_ANTI_RANGEs, drop to VR_VARYING. It would take more effort to compute a precise range for such a case. For example, if we have op0 == 1 and op1 == 1 with their ranges both being ~[0,0], we would have op0 - op1 == 0, so we cannot claim that the difference is in ~[0,0]. Note that we are guaranteed to have vr0.type == vr1.type at this point. */ if (vr0.type == VR_ANTI_RANGE) { set_value_range_to_varying (vr); return; } /* For MINUS_EXPR, apply the operation to the opposite ends of each range. */ min = vrp_int_const_binop (code, vr0.min, vr1.max); max = vrp_int_const_binop (code, vr0.max, vr1.min); } else if (code == BIT_AND_EXPR) { if (vr0.type == VR_RANGE && vr0.min == vr0.max && tree_expr_nonnegative_p (vr0.max) && TREE_CODE (vr0.max) == INTEGER_CST) { min = build_int_cst (TREE_TYPE (expr), 0); max = vr0.max; } else if (vr1.type == VR_RANGE && vr1.min == vr1.max && tree_expr_nonnegative_p (vr1.max) && TREE_CODE (vr1.max) == INTEGER_CST) { type = VR_RANGE; min = build_int_cst (TREE_TYPE (expr), 0); max = vr1.max; } else { set_value_range_to_varying (vr); return; } } else gcc_unreachable (); /* If either MIN or MAX overflowed, then set the resulting range to VARYING. */ if (!is_gimple_min_invariant (min) || TREE_OVERFLOW (min) || !is_gimple_min_invariant (max) || TREE_OVERFLOW (max)) { set_value_range_to_varying (vr); return; } cmp = compare_values (min, max); if (cmp == -2 || cmp == 1) { /* If the new range has its limits swapped around (MIN > MAX), then the operation caused one of them to wrap around, mark the new range VARYING. */ set_value_range_to_varying (vr); } else set_value_range (vr, type, min, max, NULL); } /* Extract range information from a unary expression EXPR based on the range of its operand and the expression code. */ static void extract_range_from_unary_expr (value_range_t *vr, tree expr) { enum tree_code code = TREE_CODE (expr); tree min, max, op0; int cmp; value_range_t vr0 = { VR_UNDEFINED, NULL_TREE, NULL_TREE, NULL }; /* Refuse to operate on certain unary expressions for which we cannot easily determine a resulting range. */ if (code == FIX_TRUNC_EXPR || code == FIX_CEIL_EXPR || code == FIX_FLOOR_EXPR || code == FIX_ROUND_EXPR || code == FLOAT_EXPR || code == BIT_NOT_EXPR || code == NON_LVALUE_EXPR || code == CONJ_EXPR) { set_value_range_to_varying (vr); return; } /* Get value ranges for the operand. For constant operands, create a new value range with the operand to simplify processing. */ op0 = TREE_OPERAND (expr, 0); if (TREE_CODE (op0) == SSA_NAME) vr0 = *(get_value_range (op0)); else if (is_gimple_min_invariant (op0)) set_value_range (&vr0, VR_RANGE, op0, op0, NULL); else set_value_range_to_varying (&vr0); /* If VR0 is UNDEFINED, so is the result. */ if (vr0.type == VR_UNDEFINED) { set_value_range_to_undefined (vr); return; } /* Refuse to operate on varying and symbolic ranges. Also, if the operand is neither a pointer nor an integral type, set the resulting range to VARYING. TODO, in some cases we may be able to derive anti-ranges (like nonzero values). */ if (vr0.type == VR_VARYING || (!INTEGRAL_TYPE_P (TREE_TYPE (op0)) && !POINTER_TYPE_P (TREE_TYPE (op0))) || symbolic_range_p (&vr0)) { set_value_range_to_varying (vr); return; } /* If the expression involves pointers, we are only interested in determining if it evaluates to NULL [0, 0] or non-NULL (~[0, 0]). */ if (POINTER_TYPE_P (TREE_TYPE (expr)) || POINTER_TYPE_P (TREE_TYPE (op0))) { if (range_is_nonnull (&vr0) || tree_expr_nonzero_p (expr)) set_value_range_to_nonnull (vr, TREE_TYPE (expr)); else if (range_is_null (&vr0)) set_value_range_to_null (vr, TREE_TYPE (expr)); else set_value_range_to_varying (vr); return; } /* Handle unary expressions on integer ranges. */ if (code == NOP_EXPR || code == CONVERT_EXPR) { tree inner_type = TREE_TYPE (op0); tree outer_type = TREE_TYPE (expr); /* If VR0 represents a simple range, then try to convert the min and max values for the range to the same type as OUTER_TYPE. If the results compare equal to VR0's min and max values and the new min is still less than or equal to the new max, then we can safely use the newly computed range for EXPR. This allows us to compute accurate ranges through many casts. */ if (vr0.type == VR_RANGE) { tree new_min, new_max; /* Convert VR0's min/max to OUTER_TYPE. */ new_min = fold_convert (outer_type, vr0.min); new_max = fold_convert (outer_type, vr0.max); /* Verify the new min/max values are gimple values and that they compare equal to VR0's min/max values. */ if (is_gimple_val (new_min) && is_gimple_val (new_max) && tree_int_cst_equal (new_min, vr0.min) && tree_int_cst_equal (new_max, vr0.max) && compare_values (new_min, new_max) <= 0 && compare_values (new_min, new_max) >= -1) { set_value_range (vr, VR_RANGE, new_min, new_max, vr->equiv); return; } } /* When converting types of different sizes, set the result to VARYING. Things like sign extensions and precision loss may change the range. For instance, if x_3 is of type 'long long int' and 'y_5 = (unsigned short) x_3', if x_3 is ~[0, 0], it is impossible to know at compile time whether y_5 will be ~[0, 0]. */ if (TYPE_SIZE (inner_type) != TYPE_SIZE (outer_type) || TYPE_PRECISION (inner_type) != TYPE_PRECISION (outer_type)) { set_value_range_to_varying (vr); return; } } /* Apply the operation to each end of the range and see what we end up with. */ if (code == NEGATE_EXPR && !TYPE_UNSIGNED (TREE_TYPE (expr))) { /* NEGATE_EXPR flips the range around. */ min = (vr0.max == TYPE_MAX_VALUE (TREE_TYPE (expr)) && !flag_wrapv) ? TYPE_MIN_VALUE (TREE_TYPE (expr)) : fold_unary_to_constant (code, TREE_TYPE (expr), vr0.max); max = (vr0.min == TYPE_MIN_VALUE (TREE_TYPE (expr)) && !flag_wrapv) ? TYPE_MAX_VALUE (TREE_TYPE (expr)) : fold_unary_to_constant (code, TREE_TYPE (expr), vr0.min); } else if (code == ABS_EXPR && !TYPE_UNSIGNED (TREE_TYPE (expr))) { /* -TYPE_MIN_VALUE = TYPE_MIN_VALUE with flag_wrapv so we can't get a useful range. */ if (flag_wrapv && ((vr0.type == VR_RANGE && vr0.min == TYPE_MIN_VALUE (TREE_TYPE (expr))) || (vr0.type == VR_ANTI_RANGE && vr0.min != TYPE_MIN_VALUE (TREE_TYPE (expr)) && !range_includes_zero_p (&vr0)))) { set_value_range_to_varying (vr); return; } /* ABS_EXPR may flip the range around, if the original range included negative values. */ min = (vr0.min == TYPE_MIN_VALUE (TREE_TYPE (expr))) ? TYPE_MAX_VALUE (TREE_TYPE (expr)) : fold_unary_to_constant (code, TREE_TYPE (expr), vr0.min); max = fold_unary_to_constant (code, TREE_TYPE (expr), vr0.max); cmp = compare_values (min, max); /* If a VR_ANTI_RANGEs contains zero, then we have ~[-INF, min(MIN, MAX)]. */ if (vr0.type == VR_ANTI_RANGE) { if (range_includes_zero_p (&vr0)) { tree type_min_value = TYPE_MIN_VALUE (TREE_TYPE (expr)); /* Take the lower of the two values. */ if (cmp != 1) max = min; /* Create ~[-INF, min (abs(MIN), abs(MAX))] or ~[-INF + 1, min (abs(MIN), abs(MAX))] when flag_wrapv is set and the original anti-range doesn't include TYPE_MIN_VALUE, remember -TYPE_MIN_VALUE = TYPE_MIN_VALUE. */ min = (flag_wrapv && vr0.min != type_min_value ? int_const_binop (PLUS_EXPR, type_min_value, integer_one_node, 0) : type_min_value); } else { /* All else has failed, so create the range [0, INF], even for flag_wrapv since TYPE_MIN_VALUE is in the original anti-range. */ vr0.type = VR_RANGE; min = build_int_cst (TREE_TYPE (expr), 0); max = TYPE_MAX_VALUE (TREE_TYPE (expr)); } } /* If the range contains zero then we know that the minimum value in the range will be zero. */ else if (range_includes_zero_p (&vr0)) { if (cmp == 1) max = min; min = build_int_cst (TREE_TYPE (expr), 0); } else { /* If the range was reversed, swap MIN and MAX. */ if (cmp == 1) { tree t = min; min = max; max = t; } } } else { /* Otherwise, operate on each end of the range. */ min = fold_unary_to_constant (code, TREE_TYPE (expr), vr0.min); max = fold_unary_to_constant (code, TREE_TYPE (expr), vr0.max); } cmp = compare_values (min, max); if (cmp == -2 || cmp == 1) { /* If the new range has its limits swapped around (MIN > MAX), then the operation caused one of them to wrap around, mark the new range VARYING. */ set_value_range_to_varying (vr); } else set_value_range (vr, vr0.type, min, max, NULL); } /* Extract range information from a comparison expression EXPR based on the range of its operand and the expression code. */ static void extract_range_from_comparison (value_range_t *vr, tree expr) { tree val = vrp_evaluate_conditional (expr, false); if (val) { /* Since this expression was found on the RHS of an assignment, its type may be different from _Bool. Convert VAL to EXPR's type. */ val = fold_convert (TREE_TYPE (expr), val); set_value_range (vr, VR_RANGE, val, val, vr->equiv); } else set_value_range_to_varying (vr); } /* Try to compute a useful range out of expression EXPR and store it in *VR. */ static void extract_range_from_expr (value_range_t *vr, tree expr) { enum tree_code code = TREE_CODE (expr); if (code == ASSERT_EXPR) extract_range_from_assert (vr, expr); else if (code == SSA_NAME) extract_range_from_ssa_name (vr, expr); else if (TREE_CODE_CLASS (code) == tcc_binary || code == TRUTH_ANDIF_EXPR || code == TRUTH_ORIF_EXPR || code == TRUTH_AND_EXPR || code == TRUTH_OR_EXPR || code == TRUTH_XOR_EXPR) extract_range_from_binary_expr (vr, expr); else if (TREE_CODE_CLASS (code) == tcc_unary) extract_range_from_unary_expr (vr, expr); else if (TREE_CODE_CLASS (code) == tcc_comparison) extract_range_from_comparison (vr, expr); else if (is_gimple_min_invariant (expr)) set_value_range (vr, VR_RANGE, expr, expr, NULL); else if (vrp_expr_computes_nonzero (expr)) set_value_range_to_nonnull (vr, TREE_TYPE (expr)); else set_value_range_to_varying (vr); } /* Given a range VR, a LOOP and a variable VAR, determine whether it would be profitable to adjust VR using scalar evolution information for VAR. If so, update VR with the new limits. */ static void adjust_range_with_scev (value_range_t *vr, struct loop *loop, tree stmt, tree var) { tree init, step, chrec; bool init_is_max, unknown_max; /* TODO. Don't adjust anti-ranges. An anti-range may provide better opportunities than a regular range, but I'm not sure. */ if (vr->type == VR_ANTI_RANGE) return; chrec = instantiate_parameters (loop, analyze_scalar_evolution (loop, var)); if (TREE_CODE (chrec) != POLYNOMIAL_CHREC) return; init = initial_condition_in_loop_num (chrec, loop->num); step = evolution_part_in_loop_num (chrec, loop->num); /* If STEP is symbolic, we can't know whether INIT will be the minimum or maximum value in the range. */ if (step == NULL_TREE || !is_gimple_min_invariant (step)) return; /* Do not adjust ranges when chrec may wrap. */ if (scev_probably_wraps_p (chrec_type (chrec), init, step, stmt, current_loops->parray[CHREC_VARIABLE (chrec)], &init_is_max, &unknown_max) || unknown_max) return; if (!POINTER_TYPE_P (TREE_TYPE (init)) && (vr->type == VR_VARYING || vr->type == VR_UNDEFINED)) { /* For VARYING or UNDEFINED ranges, just about anything we get from scalar evolutions should be better. */ tree min = TYPE_MIN_VALUE (TREE_TYPE (init)); tree max = TYPE_MAX_VALUE (TREE_TYPE (init)); if (init_is_max) max = init; else min = init; /* If we would create an invalid range, then just assume we know absolutely nothing. This may be over-conservative, but it's clearly safe. */ if (compare_values (min, max) == 1) return; set_value_range (vr, VR_RANGE, min, max, vr->equiv); } else if (vr->type == VR_RANGE) { tree min = vr->min; tree max = vr->max; if (init_is_max) { /* INIT is the maximum value. If INIT is lower than VR->MAX but no smaller than VR->MIN, set VR->MAX to INIT. */ if (compare_values (init, max) == -1) { max = init; /* If we just created an invalid range with the minimum greater than the maximum, take the minimum all the way to -INF. */ if (compare_values (min, max) == 1) min = TYPE_MIN_VALUE (TREE_TYPE (min)); } } else { /* If INIT is bigger than VR->MIN, set VR->MIN to INIT. */ if (compare_values (init, min) == 1) { min = init; /* If we just created an invalid range with the minimum greater than the maximum, take the maximum all the way to +INF. */ if (compare_values (min, max) == 1) max = TYPE_MAX_VALUE (TREE_TYPE (max)); } } set_value_range (vr, VR_RANGE, min, max, vr->equiv); } } /* Given two numeric value ranges VR0, VR1 and a comparison code COMP: - Return BOOLEAN_TRUE_NODE if VR0 COMP VR1 always returns true for all the values in the ranges. - Return BOOLEAN_FALSE_NODE if the comparison always returns false. - Return NULL_TREE if it is not always possible to determine the value of the comparison. */ static tree compare_ranges (enum tree_code comp, value_range_t *vr0, value_range_t *vr1) { /* VARYING or UNDEFINED ranges cannot be compared. */ if (vr0->type == VR_VARYING || vr0->type == VR_UNDEFINED || vr1->type == VR_VARYING || vr1->type == VR_UNDEFINED) return NULL_TREE; /* Anti-ranges need to be handled separately. */ if (vr0->type == VR_ANTI_RANGE || vr1->type == VR_ANTI_RANGE) { /* If both are anti-ranges, then we cannot compute any comparison. */ if (vr0->type == VR_ANTI_RANGE && vr1->type == VR_ANTI_RANGE) return NULL_TREE; /* These comparisons are never statically computable. */ if (comp == GT_EXPR || comp == GE_EXPR || comp == LT_EXPR || comp == LE_EXPR) return NULL_TREE; /* Equality can be computed only between a range and an anti-range. ~[VAL1, VAL2] == [VAL1, VAL2] is always false. */ if (vr0->type == VR_RANGE) { /* To simplify processing, make VR0 the anti-range. */ value_range_t *tmp = vr0; vr0 = vr1; vr1 = tmp; } gcc_assert (comp == NE_EXPR || comp == EQ_EXPR); if (compare_values (vr0->min, vr1->min) == 0 && compare_values (vr0->max, vr1->max) == 0) return (comp == NE_EXPR) ? boolean_true_node : boolean_false_node; return NULL_TREE; } /* Simplify processing. If COMP is GT_EXPR or GE_EXPR, switch the operands around and change the comparison code. */ if (comp == GT_EXPR || comp == GE_EXPR) { value_range_t *tmp; comp = (comp == GT_EXPR) ? LT_EXPR : LE_EXPR; tmp = vr0; vr0 = vr1; vr1 = tmp; } if (comp == EQ_EXPR) { /* Equality may only be computed if both ranges represent exactly one value. */ if (compare_values (vr0->min, vr0->max) == 0 && compare_values (vr1->min, vr1->max) == 0) { int cmp_min = compare_values (vr0->min, vr1->min); int cmp_max = compare_values (vr0->max, vr1->max); if (cmp_min == 0 && cmp_max == 0) return boolean_true_node; else if (cmp_min != -2 && cmp_max != -2) return boolean_false_node; } /* If [V0_MIN, V1_MAX] < [V1_MIN, V1_MAX] then V0 != V1. */ else if (compare_values (vr0->min, vr1->max) == 1 || compare_values (vr1->min, vr0->max) == 1) return boolean_false_node; return NULL_TREE; } else if (comp == NE_EXPR) { int cmp1, cmp2; /* If VR0 is completely to the left or completely to the right of VR1, they are always different. Notice that we need to make sure that both comparisons yield similar results to avoid comparing values that cannot be compared at compile-time. */ cmp1 = compare_values (vr0->max, vr1->min); cmp2 = compare_values (vr0->min, vr1->max); if ((cmp1 == -1 && cmp2 == -1) || (cmp1 == 1 && cmp2 == 1)) return boolean_true_node; /* If VR0 and VR1 represent a single value and are identical, return false. */ else if (compare_values (vr0->min, vr0->max) == 0 && compare_values (vr1->min, vr1->max) == 0 && compare_values (vr0->min, vr1->min) == 0 && compare_values (vr0->max, vr1->max) == 0) return boolean_false_node; /* Otherwise, they may or may not be different. */ else return NULL_TREE; } else if (comp == LT_EXPR || comp == LE_EXPR) { int tst; /* If VR0 is to the left of VR1, return true. */ tst = compare_values (vr0->max, vr1->min); if ((comp == LT_EXPR && tst == -1) || (comp == LE_EXPR && (tst == -1 || tst == 0))) return boolean_true_node; /* If VR0 is to the right of VR1, return false. */ tst = compare_values (vr0->min, vr1->max); if ((comp == LT_EXPR && (tst == 0 || tst == 1)) || (comp == LE_EXPR && tst == 1)) return boolean_false_node; /* Otherwise, we don't know. */ return NULL_TREE; } gcc_unreachable (); } /* Given a value range VR, a value VAL and a comparison code COMP, return BOOLEAN_TRUE_NODE if VR COMP VAL always returns true for all the values in VR. Return BOOLEAN_FALSE_NODE if the comparison always returns false. Return NULL_TREE if it is not always possible to determine the value of the comparison. */ static tree compare_range_with_value (enum tree_code comp, value_range_t *vr, tree val) { if (vr->type == VR_VARYING || vr->type == VR_UNDEFINED) return NULL_TREE; /* Anti-ranges need to be handled separately. */ if (vr->type == VR_ANTI_RANGE) { /* For anti-ranges, the only predicates that we can compute at compile time are equality and inequality. */ if (comp == GT_EXPR || comp == GE_EXPR || comp == LT_EXPR || comp == LE_EXPR) return NULL_TREE; /* ~[VAL_1, VAL_2] OP VAL is known if VAL_1 <= VAL <= VAL_2. */ if (value_inside_range (val, vr) == 1) return (comp == NE_EXPR) ? boolean_true_node : boolean_false_node; return NULL_TREE; } if (comp == EQ_EXPR) { /* EQ_EXPR may only be computed if VR represents exactly one value. */ if (compare_values (vr->min, vr->max) == 0) { int cmp = compare_values (vr->min, val); if (cmp == 0) return boolean_true_node; else if (cmp == -1 || cmp == 1 || cmp == 2) return boolean_false_node; } else if (compare_values (val, vr->min) == -1 || compare_values (vr->max, val) == -1) return boolean_false_node; return NULL_TREE; } else if (comp == NE_EXPR) { /* If VAL is not inside VR, then they are always different. */ if (compare_values (vr->max, val) == -1 || compare_values (vr->min, val) == 1) return boolean_true_node; /* If VR represents exactly one value equal to VAL, then return false. */ if (compare_values (vr->min, vr->max) == 0 && compare_values (vr->min, val) == 0) return boolean_false_node; /* Otherwise, they may or may not be different. */ return NULL_TREE; } else if (comp == LT_EXPR || comp == LE_EXPR) { int tst; /* If VR is to the left of VAL, return true. */ tst = compare_values (vr->max, val); if ((comp == LT_EXPR && tst == -1) || (comp == LE_EXPR && (tst == -1 || tst == 0))) return boolean_true_node; /* If VR is to the right of VAL, return false. */ tst = compare_values (vr->min, val); if ((comp == LT_EXPR && (tst == 0 || tst == 1)) || (comp == LE_EXPR && tst == 1)) return boolean_false_node; /* Otherwise, we don't know. */ return NULL_TREE; } else if (comp == GT_EXPR || comp == GE_EXPR) { int tst; /* If VR is to the right of VAL, return true. */ tst = compare_values (vr->min, val); if ((comp == GT_EXPR && tst == 1) || (comp == GE_EXPR && (tst == 0 || tst == 1))) return boolean_true_node; /* If VR is to the left of VAL, return false. */ tst = compare_values (vr->max, val); if ((comp == GT_EXPR && (tst == -1 || tst == 0)) || (comp == GE_EXPR && tst == -1)) return boolean_false_node; /* Otherwise, we don't know. */ return NULL_TREE; } gcc_unreachable (); } /* Debugging dumps. */ void dump_value_range (FILE *, value_range_t *); void debug_value_range (value_range_t *); void dump_all_value_ranges (FILE *); void debug_all_value_ranges (void); void dump_vr_equiv (FILE *, bitmap); void debug_vr_equiv (bitmap); /* Dump value range VR to FILE. */ void dump_value_range (FILE *file, value_range_t *vr) { if (vr == NULL) fprintf (file, "[]"); else if (vr->type == VR_UNDEFINED) fprintf (file, "UNDEFINED"); else if (vr->type == VR_RANGE || vr->type == VR_ANTI_RANGE) { tree type = TREE_TYPE (vr->min); fprintf (file, "%s[", (vr->type == VR_ANTI_RANGE) ? "~" : ""); if (INTEGRAL_TYPE_P (type) && !TYPE_UNSIGNED (type) && vr->min == TYPE_MIN_VALUE (type)) fprintf (file, "-INF"); else print_generic_expr (file, vr->min, 0); fprintf (file, ", "); if (INTEGRAL_TYPE_P (type) && vr->max == TYPE_MAX_VALUE (type)) fprintf (file, "+INF"); else print_generic_expr (file, vr->max, 0); fprintf (file, "]"); if (vr->equiv) { bitmap_iterator bi; unsigned i, c = 0; fprintf (file, " EQUIVALENCES: { "); EXECUTE_IF_SET_IN_BITMAP (vr->equiv, 0, i, bi) { print_generic_expr (file, ssa_name (i), 0); fprintf (file, " "); c++; } fprintf (file, "} (%u elements)", c); } } else if (vr->type == VR_VARYING) fprintf (file, "VARYING"); else fprintf (file, "INVALID RANGE"); } /* Dump value range VR to stderr. */ void debug_value_range (value_range_t *vr) { dump_value_range (stderr, vr); } /* Dump value ranges of all SSA_NAMEs to FILE. */ void dump_all_value_ranges (FILE *file) { size_t i; for (i = 0; i < num_ssa_names; i++) { if (vr_value[i]) { print_generic_expr (file, ssa_name (i), 0); fprintf (file, ": "); dump_value_range (file, vr_value[i]); fprintf (file, "\n"); } } fprintf (file, "\n"); } /* Dump all value ranges to stderr. */ void debug_all_value_ranges (void) { dump_all_value_ranges (stderr); } /* Given a COND_EXPR COND of the form 'V OP W', and an SSA name V, create a new SSA name N and return the assertion assignment 'V = ASSERT_EXPR '. */ static tree build_assert_expr_for (tree cond, tree v) { tree n, assertion; gcc_assert (TREE_CODE (v) == SSA_NAME); n = duplicate_ssa_name (v, NULL_TREE); if (COMPARISON_CLASS_P (cond)) { tree a = build2 (ASSERT_EXPR, TREE_TYPE (v), v, cond); assertion = build2 (MODIFY_EXPR, TREE_TYPE (v), n, a); } else if (TREE_CODE (cond) == TRUTH_NOT_EXPR) { /* Given !V, build the assignment N = false. */ tree op0 = TREE_OPERAND (cond, 0); gcc_assert (op0 == v); assertion = build2 (MODIFY_EXPR, TREE_TYPE (v), n, boolean_false_node); } else if (TREE_CODE (cond) == SSA_NAME) { /* Given V, build the assignment N = true. */ gcc_assert (v == cond); assertion = build2 (MODIFY_EXPR, TREE_TYPE (v), n, boolean_true_node); } else gcc_unreachable (); SSA_NAME_DEF_STMT (n) = assertion; /* The new ASSERT_EXPR, creates a new SSA name that replaces the operand of the ASSERT_EXPR. Register the new name and the old one in the replacement table so that we can fix the SSA web after adding all the ASSERT_EXPRs. */ register_new_name_mapping (n, v); return assertion; } /* Return false if EXPR is a predicate expression involving floating point values. */ static inline bool fp_predicate (tree expr) { return (COMPARISON_CLASS_P (expr) && FLOAT_TYPE_P (TREE_TYPE (TREE_OPERAND (expr, 0)))); } /* If the range of values taken by OP can be inferred after STMT executes, return the comparison code (COMP_CODE_P) and value (VAL_P) that describes the inferred range. Return true if a range could be inferred. */ static bool infer_value_range (tree stmt, tree op, enum tree_code *comp_code_p, tree *val_p) { *val_p = NULL_TREE; *comp_code_p = ERROR_MARK; /* Do not attempt to infer anything in names that flow through abnormal edges. */ if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (op)) return false; /* Similarly, don't infer anything from statements that may throw exceptions. */ if (tree_could_throw_p (stmt)) return false; /* If STMT is the last statement of a basic block with no successors, there is no point inferring anything about any of its operands. We would not be able to find a proper insertion point for the assertion, anyway. */ if (stmt_ends_bb_p (stmt) && EDGE_COUNT (bb_for_stmt (stmt)->succs) == 0) return false; if (POINTER_TYPE_P (TREE_TYPE (op))) { bool is_store; unsigned num_uses, num_derefs; count_uses_and_derefs (op, stmt, &num_uses, &num_derefs, &is_store); if (num_derefs > 0 && flag_delete_null_pointer_checks) { /* We can only assume that a pointer dereference will yield non-NULL if -fdelete-null-pointer-checks is enabled. */ *val_p = build_int_cst (TREE_TYPE (op), 0); *comp_code_p = NE_EXPR; return true; } } return false; } void dump_asserts_for (FILE *, tree); void debug_asserts_for (tree); void dump_all_asserts (FILE *); void debug_all_asserts (void); /* Dump all the registered assertions for NAME to FILE. */ void dump_asserts_for (FILE *file, tree name) { assert_locus_t loc; fprintf (file, "Assertions to be inserted for "); print_generic_expr (file, name, 0); fprintf (file, "\n"); loc = asserts_for[SSA_NAME_VERSION (name)]; while (loc) { fprintf (file, "\t"); print_generic_expr (file, bsi_stmt (loc->si), 0); fprintf (file, "\n\tBB #%d", loc->bb->index); if (loc->e) { fprintf (file, "\n\tEDGE %d->%d", loc->e->src->index, loc->e->dest->index); dump_edge_info (file, loc->e, 0); } fprintf (file, "\n\tPREDICATE: "); print_generic_expr (file, name, 0); fprintf (file, " %s ", tree_code_name[(int)loc->comp_code]); print_generic_expr (file, loc->val, 0); fprintf (file, "\n\n"); loc = loc->next; } fprintf (file, "\n"); } /* Dump all the registered assertions for NAME to stderr. */ void debug_asserts_for (tree name) { dump_asserts_for (stderr, name); } /* Dump all the registered assertions for all the names to FILE. */ void dump_all_asserts (FILE *file) { unsigned i; bitmap_iterator bi; fprintf (file, "\nASSERT_EXPRs to be inserted\n\n"); EXECUTE_IF_SET_IN_BITMAP (need_assert_for, 0, i, bi) dump_asserts_for (file, ssa_name (i)); fprintf (file, "\n"); } /* Dump all the registered assertions for all the names to stderr. */ void debug_all_asserts (void) { dump_all_asserts (stderr); } /* If NAME doesn't have an ASSERT_EXPR registered for asserting 'NAME COMP_CODE VAL' at a location that dominates block BB or E->DEST, then register this location as a possible insertion point for ASSERT_EXPR . BB, E and SI provide the exact insertion point for the new ASSERT_EXPR. If BB is NULL, then the ASSERT_EXPR is to be inserted on edge E. Otherwise, if E is NULL, the ASSERT_EXPR is inserted on BB. If SI points to a COND_EXPR or a SWITCH_EXPR statement, then E must not be NULL. */ static void register_new_assert_for (tree name, enum tree_code comp_code, tree val, basic_block bb, edge e, block_stmt_iterator si) { assert_locus_t n, loc, last_loc; bool found; basic_block dest_bb; #if defined ENABLE_CHECKING gcc_assert (bb == NULL || e == NULL); if (e == NULL) gcc_assert (TREE_CODE (bsi_stmt (si)) != COND_EXPR && TREE_CODE (bsi_stmt (si)) != SWITCH_EXPR); #endif /* The new assertion A will be inserted at BB or E. We need to determine if the new location is dominated by a previously registered location for A. If we are doing an edge insertion, assume that A will be inserted at E->DEST. Note that this is not necessarily true. If E is a critical edge, it will be split. But even if E is split, the new block will dominate the same set of blocks that E->DEST dominates. The reverse, however, is not true, blocks dominated by E->DEST will not be dominated by the new block created to split E. So, if the insertion location is on a critical edge, we will not use the new location to move another assertion previously registered at a block dominated by E->DEST. */ dest_bb = (bb) ? bb : e->dest; /* If NAME already has an ASSERT_EXPR registered for COMP_CODE and VAL at a block dominating DEST_BB, then we don't need to insert a new one. Similarly, if the same assertion already exists at a block dominated by DEST_BB and the new location is not on a critical edge, then update the existing location for the assertion (i.e., move the assertion up in the dominance tree). Note, this is implemented as a simple linked list because there should not be more than a handful of assertions registered per name. If this becomes a performance problem, a table hashed by COMP_CODE and VAL could be implemented. */ loc = asserts_for[SSA_NAME_VERSION (name)]; last_loc = loc; found = false; while (loc) { if (loc->comp_code == comp_code && (loc->val == val || operand_equal_p (loc->val, val, 0))) { /* If the assertion NAME COMP_CODE VAL has already been registered at a basic block that dominates DEST_BB, then we don't need to insert the same assertion again. Note that we don't check strict dominance here to avoid replicating the same assertion inside the same basic block more than once (e.g., when a pointer is dereferenced several times inside a block). An exception to this rule are edge insertions. If the new assertion is to be inserted on edge E, then it will dominate all the other insertions that we may want to insert in DEST_BB. So, if we are doing an edge insertion, don't do this dominance check. */ if (e == NULL && dominated_by_p (CDI_DOMINATORS, dest_bb, loc->bb)) return; /* Otherwise, if E is not a critical edge and DEST_BB dominates the existing location for the assertion, move the assertion up in the dominance tree by updating its location information. */ if ((e == NULL || !EDGE_CRITICAL_P (e)) && dominated_by_p (CDI_DOMINATORS, loc->bb, dest_bb)) { loc->bb = dest_bb; loc->e = e; loc->si = si; return; } } /* Update the last node of the list and move to the next one. */ last_loc = loc; loc = loc->next; } /* If we didn't find an assertion already registered for NAME COMP_CODE VAL, add a new one at the end of the list of assertions associated with NAME. */ n = XNEW (struct assert_locus_d); n->bb = dest_bb; n->e = e; n->si = si; n->comp_code = comp_code; n->val = val; n->next = NULL; if (last_loc) last_loc->next = n; else asserts_for[SSA_NAME_VERSION (name)] = n; bitmap_set_bit (need_assert_for, SSA_NAME_VERSION (name)); } /* Try to register an edge assertion for SSA name NAME on edge E for the conditional jump pointed to by SI. Return true if an assertion for NAME could be registered. */ static bool register_edge_assert_for (tree name, edge e, block_stmt_iterator si) { tree val, stmt; enum tree_code comp_code; stmt = bsi_stmt (si); /* Do not attempt to infer anything in names that flow through abnormal edges. */ if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (name)) return false; /* If NAME was not found in the sub-graph reachable from E, then there's nothing to do. */ if (!TEST_BIT (found_in_subgraph, SSA_NAME_VERSION (name))) return false; /* We found a use of NAME in the sub-graph rooted at E->DEST. Register an assertion for NAME according to the value that NAME takes on edge E. */ if (TREE_CODE (stmt) == COND_EXPR) { /* If BB ends in a COND_EXPR then NAME then we should insert the original predicate on EDGE_TRUE_VALUE and the opposite predicate on EDGE_FALSE_VALUE. */ tree cond = COND_EXPR_COND (stmt); bool is_else_edge = (e->flags & EDGE_FALSE_VALUE) != 0; /* Predicates may be a single SSA name or NAME OP VAL. */ if (cond == name) { /* If the predicate is a name, it must be NAME, in which case we create the predicate NAME == true or NAME == false accordingly. */ comp_code = EQ_EXPR; val = (is_else_edge) ? boolean_false_node : boolean_true_node; } else { /* Otherwise, we have a comparison of the form NAME COMP VAL or VAL COMP NAME. */ if (name == TREE_OPERAND (cond, 1)) { /* If the predicate is of the form VAL COMP NAME, flip COMP around because we need to register NAME as the first operand in the predicate. */ comp_code = swap_tree_comparison (TREE_CODE (cond)); val = TREE_OPERAND (cond, 0); } else { /* The comparison is of the form NAME COMP VAL, so the comparison code remains unchanged. */ comp_code = TREE_CODE (cond); val = TREE_OPERAND (cond, 1); } /* If we are inserting the assertion on the ELSE edge, we need to invert the sign comparison. */ if (is_else_edge) comp_code = invert_tree_comparison (comp_code, 0); /* Do not register always-false predicates. FIXME, this works around a limitation in fold() when dealing with enumerations. Given 'enum { N1, N2 } x;', fold will not fold 'if (x > N2)' to 'if (0)'. */ if ((comp_code == GT_EXPR || comp_code == LT_EXPR) && (INTEGRAL_TYPE_P (TREE_TYPE (val)) || SCALAR_FLOAT_TYPE_P (TREE_TYPE (val)))) { tree min = TYPE_MIN_VALUE (TREE_TYPE (val)); tree max = TYPE_MAX_VALUE (TREE_TYPE (val)); if (comp_code == GT_EXPR && compare_values (val, max) == 0) return false; if (comp_code == LT_EXPR && compare_values (val, min) == 0) return false; } } } else { /* FIXME. Handle SWITCH_EXPR. */ gcc_unreachable (); } register_new_assert_for (name, comp_code, val, NULL, e, si); return true; } static bool find_assert_locations (basic_block bb); /* Determine whether the outgoing edges of BB should receive an ASSERT_EXPR for each of the operands of BB's last statement. The last statement of BB must be a COND_EXPR or a SWITCH_EXPR. If any of the sub-graphs rooted at BB have an interesting use of the predicate operands, an assert location node is added to the list of assertions for the corresponding operands. */ static bool find_conditional_asserts (basic_block bb) { bool need_assert; block_stmt_iterator last_si; tree op, last; edge_iterator ei; edge e; ssa_op_iter iter; need_assert = false; last_si = bsi_last (bb); last = bsi_stmt (last_si); /* Look for uses of the operands in each of the sub-graphs rooted at BB. We need to check each of the outgoing edges separately, so that we know what kind of ASSERT_EXPR to insert. */ FOR_EACH_EDGE (e, ei, bb->succs) { if (e->dest == bb) continue; /* Remove the COND_EXPR operands from the FOUND_IN_SUBGRAPH bitmap. Otherwise, when we finish traversing each of the sub-graphs, we won't know whether the variables were found in the sub-graphs or if they had been found in a block upstream from BB. This is actually a bad idea is some cases, particularly jump threading. Consider a CFG like the following: 0 /| 1 | \| 2 / \ 3 4 Assume that one or more operands in the conditional at the end of block 0 are used in a conditional in block 2, but not anywhere in block 1. In this case we will not insert any assert statements in block 1, which may cause us to miss opportunities to optimize, particularly for jump threading. */ FOR_EACH_SSA_TREE_OPERAND (op, last, iter, SSA_OP_USE) RESET_BIT (found_in_subgraph, SSA_NAME_VERSION (op)); /* Traverse the strictly dominated sub-graph rooted at E->DEST to determine if any of the operands in the conditional predicate are used. */ if (e->dest != bb) need_assert |= find_assert_locations (e->dest); /* Register the necessary assertions for each operand in the conditional predicate. */ FOR_EACH_SSA_TREE_OPERAND (op, last, iter, SSA_OP_USE) need_assert |= register_edge_assert_for (op, e, last_si); } /* Finally, indicate that we have found the operands in the conditional. */ FOR_EACH_SSA_TREE_OPERAND (op, last, iter, SSA_OP_USE) SET_BIT (found_in_subgraph, SSA_NAME_VERSION (op)); return need_assert; } /* Traverse all the statements in block BB looking for statements that may generate useful assertions for the SSA names in their operand. If a statement produces a useful assertion A for name N_i, then the list of assertions already generated for N_i is scanned to determine if A is actually needed. If N_i already had the assertion A at a location dominating the current location, then nothing needs to be done. Otherwise, the new location for A is recorded instead. 1- For every statement S in BB, all the variables used by S are added to bitmap FOUND_IN_SUBGRAPH. 2- If statement S uses an operand N in a way that exposes a known value range for N, then if N was not already generated by an ASSERT_EXPR, create a new assert location for N. For instance, if N is a pointer and the statement dereferences it, we can assume that N is not NULL. 3- COND_EXPRs are a special case of #2. We can derive range information from the predicate but need to insert different ASSERT_EXPRs for each of the sub-graphs rooted at the conditional block. If the last statement of BB is a conditional expression of the form 'X op Y', then a) Remove X and Y from the set FOUND_IN_SUBGRAPH. b) If the conditional is the only entry point to the sub-graph corresponding to the THEN_CLAUSE, recurse into it. On return, if X and/or Y are marked in FOUND_IN_SUBGRAPH, then an ASSERT_EXPR is added for the corresponding variable. c) Repeat step (b) on the ELSE_CLAUSE. d) Mark X and Y in FOUND_IN_SUBGRAPH. For instance, if (a == 9) b = a; else b = c + 1; In this case, an assertion on the THEN clause is useful to determine that 'a' is always 9 on that edge. However, an assertion on the ELSE clause would be unnecessary. 4- If BB does not end in a conditional expression, then we recurse into BB's dominator children. At the end of the recursive traversal, every SSA name will have a list of locations where ASSERT_EXPRs should be added. When a new location for name N is found, it is registered by calling register_new_assert_for. That function keeps track of all the registered assertions to prevent adding unnecessary assertions. For instance, if a pointer P_4 is dereferenced more than once in a dominator tree, only the location dominating all the dereference of P_4 will receive an ASSERT_EXPR. If this function returns true, then it means that there are names for which we need to generate ASSERT_EXPRs. Those assertions are inserted by process_assert_insertions. TODO. Handle SWITCH_EXPR. */ static bool find_assert_locations (basic_block bb) { block_stmt_iterator si; tree last, phi; bool need_assert; basic_block son; if (TEST_BIT (blocks_visited, bb->index)) return false; SET_BIT (blocks_visited, bb->index); need_assert = false; /* Traverse all PHI nodes in BB marking used operands. */ for (phi = phi_nodes (bb); phi; phi = PHI_CHAIN (phi)) { use_operand_p arg_p; ssa_op_iter i; FOR_EACH_PHI_ARG (arg_p, phi, i, SSA_OP_USE) { tree arg = USE_FROM_PTR (arg_p); if (TREE_CODE (arg) == SSA_NAME) { gcc_assert (is_gimple_reg (PHI_RESULT (phi))); SET_BIT (found_in_subgraph, SSA_NAME_VERSION (arg)); } } } /* Traverse all the statements in BB marking used names and looking for statements that may infer assertions for their used operands. */ last = NULL_TREE; for (si = bsi_start (bb); !bsi_end_p (si); bsi_next (&si)) { tree stmt, op; ssa_op_iter i; stmt = bsi_stmt (si); /* See if we can derive an assertion for any of STMT's operands. */ FOR_EACH_SSA_TREE_OPERAND (op, stmt, i, SSA_OP_USE) { tree value; enum tree_code comp_code; /* Mark OP in bitmap FOUND_IN_SUBGRAPH. If STMT is inside the sub-graph of a conditional block, when we return from this recursive walk, our parent will use the FOUND_IN_SUBGRAPH bitset to determine if one of the operands it was looking for was present in the sub-graph. */ SET_BIT (found_in_subgraph, SSA_NAME_VERSION (op)); /* If OP is used only once, namely in this STMT, don't bother creating an ASSERT_EXPR for it. Such an ASSERT_EXPR would do nothing but increase compile time. Experiments show that with this simple check, we can save more than 20% of ASSERT_EXPRs. */ if (has_single_use (op)) continue; /* If OP is used in such a way that we can infer a value range for it, and we don't find a previous assertion for it, create a new assertion location node for OP. */ if (infer_value_range (stmt, op, &comp_code, &value)) { register_new_assert_for (op, comp_code, value, bb, NULL, si); need_assert = true; } } /* Remember the last statement of the block. */ last = stmt; } /* If BB's last statement is a conditional expression involving integer operands, recurse into each of the sub-graphs rooted at BB to determine if we need to add ASSERT_EXPRs. */ if (last && TREE_CODE (last) == COND_EXPR && !fp_predicate (COND_EXPR_COND (last)) && !ZERO_SSA_OPERANDS (last, SSA_OP_USE)) need_assert |= find_conditional_asserts (bb); /* Recurse into the dominator children of BB. */ for (son = first_dom_son (CDI_DOMINATORS, bb); son; son = next_dom_son (CDI_DOMINATORS, son)) need_assert |= find_assert_locations (son); return need_assert; } /* Create an ASSERT_EXPR for NAME and insert it in the location indicated by LOC. Return true if we made any edge insertions. */ static bool process_assert_insertions_for (tree name, assert_locus_t loc) { /* Build the comparison expression NAME_i COMP_CODE VAL. */ tree stmt, cond, assert_expr; edge_iterator ei; edge e; cond = build2 (loc->comp_code, boolean_type_node, name, loc->val); assert_expr = build_assert_expr_for (cond, name); if (loc->e) { /* We have been asked to insert the assertion on an edge. This is used only by COND_EXPR and SWITCH_EXPR assertions. */ #if defined ENABLE_CHECKING gcc_assert (TREE_CODE (bsi_stmt (loc->si)) == COND_EXPR || TREE_CODE (bsi_stmt (loc->si)) == SWITCH_EXPR); #endif bsi_insert_on_edge (loc->e, assert_expr); return true; } /* Otherwise, we can insert right after LOC->SI iff the statement must not be the last statement in the block. */ stmt = bsi_stmt (loc->si); if (!stmt_ends_bb_p (stmt)) { bsi_insert_after (&loc->si, assert_expr, BSI_SAME_STMT); return false; } /* If STMT must be the last statement in BB, we can only insert new assertions on the non-abnormal edge out of BB. Note that since STMT is not control flow, there may only be one non-abnormal edge out of BB. */ FOR_EACH_EDGE (e, ei, loc->bb->succs) if (!(e->flags & EDGE_ABNORMAL)) { bsi_insert_on_edge (e, assert_expr); return true; } gcc_unreachable (); } /* Process all the insertions registered for every name N_i registered in NEED_ASSERT_FOR. The list of assertions to be inserted are found in ASSERTS_FOR[i]. */ static void process_assert_insertions (void) { unsigned i; bitmap_iterator bi; bool update_edges_p = false; int num_asserts = 0; if (dump_file && (dump_flags & TDF_DETAILS)) dump_all_asserts (dump_file); EXECUTE_IF_SET_IN_BITMAP (need_assert_for, 0, i, bi) { assert_locus_t loc = asserts_for[i]; gcc_assert (loc); while (loc) { assert_locus_t next = loc->next; update_edges_p |= process_assert_insertions_for (ssa_name (i), loc); free (loc); loc = next; num_asserts++; } } if (update_edges_p) bsi_commit_edge_inserts (); if (dump_file && (dump_flags & TDF_STATS)) fprintf (dump_file, "\nNumber of ASSERT_EXPR expressions inserted: %d\n\n", num_asserts); } /* Traverse the flowgraph looking for conditional jumps to insert range expressions. These range expressions are meant to provide information to optimizations that need to reason in terms of value ranges. They will not be expanded into RTL. For instance, given: x = ... y = ... if (x < y) y = x - 2; else x = y + 3; this pass will transform the code into: x = ... y = ... if (x < y) { x = ASSERT_EXPR y = x - 2 } else { y = ASSERT_EXPR x = y + 3 } The idea is that once copy and constant propagation have run, other optimizations will be able to determine what ranges of values can 'x' take in different paths of the code, simply by checking the reaching definition of 'x'. */ static void insert_range_assertions (void) { edge e; edge_iterator ei; bool update_ssa_p; found_in_subgraph = sbitmap_alloc (num_ssa_names); sbitmap_zero (found_in_subgraph); blocks_visited = sbitmap_alloc (last_basic_block); sbitmap_zero (blocks_visited); need_assert_for = BITMAP_ALLOC (NULL); asserts_for = XNEWVEC (assert_locus_t, num_ssa_names); memset (asserts_for, 0, num_ssa_names * sizeof (assert_locus_t)); calculate_dominance_info (CDI_DOMINATORS); update_ssa_p = false; FOR_EACH_EDGE (e, ei, ENTRY_BLOCK_PTR->succs) if (find_assert_locations (e->dest)) update_ssa_p = true; if (update_ssa_p) { process_assert_insertions (); update_ssa (TODO_update_ssa_no_phi); } if (dump_file && (dump_flags & TDF_DETAILS)) { fprintf (dump_file, "\nSSA form after inserting ASSERT_EXPRs\n"); dump_function_to_file (current_function_decl, dump_file, dump_flags); } sbitmap_free (found_in_subgraph); free (asserts_for); BITMAP_FREE (need_assert_for); } /* Convert range assertion expressions into the implied copies and copy propagate away the copies. Doing the trivial copy propagation here avoids the need to run the full copy propagation pass after VRP. FIXME, this will eventually lead to copy propagation removing the names that had useful range information attached to them. For instance, if we had the assertion N_i = ASSERT_EXPR 3>, then N_i will have the range [3, +INF]. However, by converting the assertion into the implied copy operation N_i = N_j, we will then copy-propagate N_j into the uses of N_i and lose the range information. We may want to hold on to ASSERT_EXPRs a little while longer as the ranges could be used in things like jump threading. The problem with keeping ASSERT_EXPRs around is that passes after VRP need to handle them appropriately. Another approach would be to make the range information a first class property of the SSA_NAME so that it can be queried from any pass. This is made somewhat more complex by the need for multiple ranges to be associated with one SSA_NAME. */ static void remove_range_assertions (void) { basic_block bb; block_stmt_iterator si; /* Note that the BSI iterator bump happens at the bottom of the loop and no bump is necessary if we're removing the statement referenced by the current BSI. */ FOR_EACH_BB (bb) for (si = bsi_start (bb); !bsi_end_p (si);) { tree stmt = bsi_stmt (si); if (TREE_CODE (stmt) == MODIFY_EXPR && TREE_CODE (TREE_OPERAND (stmt, 1)) == ASSERT_EXPR) { tree rhs = TREE_OPERAND (stmt, 1); tree cond = fold (ASSERT_EXPR_COND (rhs)); use_operand_p use_p; imm_use_iterator iter; gcc_assert (cond != boolean_false_node); TREE_OPERAND (stmt, 1) = ASSERT_EXPR_VAR (rhs); update_stmt (stmt); /* The statement is now a copy. Propagate the RHS into every use of the LHS. */ FOR_EACH_IMM_USE_SAFE (use_p, iter, TREE_OPERAND (stmt, 0)) { SET_USE (use_p, ASSERT_EXPR_VAR (rhs)); update_stmt (USE_STMT (use_p)); } /* And finally, remove the copy, it is not needed. */ bsi_remove (&si, true); } else bsi_next (&si); } sbitmap_free (blocks_visited); } /* Return true if STMT is interesting for VRP. */ static bool stmt_interesting_for_vrp (tree stmt) { if (TREE_CODE (stmt) == PHI_NODE && is_gimple_reg (PHI_RESULT (stmt)) && (INTEGRAL_TYPE_P (TREE_TYPE (PHI_RESULT (stmt))) || POINTER_TYPE_P (TREE_TYPE (PHI_RESULT (stmt))))) return true; else if (TREE_CODE (stmt) == MODIFY_EXPR) { tree lhs = TREE_OPERAND (stmt, 0); if (TREE_CODE (lhs) == SSA_NAME && (INTEGRAL_TYPE_P (TREE_TYPE (lhs)) || POINTER_TYPE_P (TREE_TYPE (lhs))) && ZERO_SSA_OPERANDS (stmt, SSA_OP_ALL_VIRTUALS)) return true; } else if (TREE_CODE (stmt) == COND_EXPR || TREE_CODE (stmt) == SWITCH_EXPR) return true; return false; } /* Initialize local data structures for VRP. */ static void vrp_initialize (void) { basic_block bb; vr_value = XNEWVEC (value_range_t *, num_ssa_names); memset (vr_value, 0, num_ssa_names * sizeof (value_range_t *)); FOR_EACH_BB (bb) { block_stmt_iterator si; tree phi; for (phi = phi_nodes (bb); phi; phi = PHI_CHAIN (phi)) { if (!stmt_interesting_for_vrp (phi)) { tree lhs = PHI_RESULT (phi); set_value_range_to_varying (get_value_range (lhs)); DONT_SIMULATE_AGAIN (phi) = true; } else DONT_SIMULATE_AGAIN (phi) = false; } for (si = bsi_start (bb); !bsi_end_p (si); bsi_next (&si)) { tree stmt = bsi_stmt (si); if (!stmt_interesting_for_vrp (stmt)) { ssa_op_iter i; tree def; FOR_EACH_SSA_TREE_OPERAND (def, stmt, i, SSA_OP_DEF) set_value_range_to_varying (get_value_range (def)); DONT_SIMULATE_AGAIN (stmt) = true; } else { DONT_SIMULATE_AGAIN (stmt) = false; } } } } /* Visit assignment STMT. If it produces an interesting range, record the SSA name in *OUTPUT_P. */ static enum ssa_prop_result vrp_visit_assignment (tree stmt, tree *output_p) { tree lhs, rhs, def; ssa_op_iter iter; lhs = TREE_OPERAND (stmt, 0); rhs = TREE_OPERAND (stmt, 1); /* We only keep track of ranges in integral and pointer types. */ if (TREE_CODE (lhs) == SSA_NAME && (INTEGRAL_TYPE_P (TREE_TYPE (lhs)) || POINTER_TYPE_P (TREE_TYPE (lhs)))) { struct loop *l; value_range_t new_vr = { VR_UNDEFINED, NULL_TREE, NULL_TREE, NULL }; extract_range_from_expr (&new_vr, rhs); /* If STMT is inside a loop, we may be able to know something else about the range of LHS by examining scalar evolution information. */ if (current_loops && (l = loop_containing_stmt (stmt))) adjust_range_with_scev (&new_vr, l, stmt, lhs); if (update_value_range (lhs, &new_vr)) { *output_p = lhs; if (dump_file && (dump_flags & TDF_DETAILS)) { fprintf (dump_file, "Found new range for "); print_generic_expr (dump_file, lhs, 0); fprintf (dump_file, ": "); dump_value_range (dump_file, &new_vr); fprintf (dump_file, "\n\n"); } if (new_vr.type == VR_VARYING) return SSA_PROP_VARYING; return SSA_PROP_INTERESTING; } return SSA_PROP_NOT_INTERESTING; } /* Every other statement produces no useful ranges. */ FOR_EACH_SSA_TREE_OPERAND (def, stmt, iter, SSA_OP_DEF) set_value_range_to_varying (get_value_range (def)); return SSA_PROP_VARYING; } /* Compare all the value ranges for names equivalent to VAR with VAL using comparison code COMP. Return the same value returned by compare_range_with_value. */ static tree compare_name_with_value (enum tree_code comp, tree var, tree val) { bitmap_iterator bi; unsigned i; bitmap e; tree retval, t; t = retval = NULL_TREE; /* Get the set of equivalences for VAR. */ e = get_value_range (var)->equiv; /* Add VAR to its own set of equivalences so that VAR's value range is processed by this loop (otherwise, we would have to replicate the body of the loop just to check VAR's value range). */ bitmap_set_bit (e, SSA_NAME_VERSION (var)); EXECUTE_IF_SET_IN_BITMAP (e, 0, i, bi) { value_range_t equiv_vr = *(vr_value[i]); /* If name N_i does not have a valid range, use N_i as its own range. This allows us to compare against names that may have N_i in their ranges. */ if (equiv_vr.type == VR_VARYING || equiv_vr.type == VR_UNDEFINED) { equiv_vr.type = VR_RANGE; equiv_vr.min = ssa_name (i); equiv_vr.max = ssa_name (i); } t = compare_range_with_value (comp, &equiv_vr, val); if (t) { /* All the ranges should compare the same against VAL. */ gcc_assert (retval == NULL || t == retval); retval = t; } } /* Remove VAR from its own equivalence set. */ bitmap_clear_bit (e, SSA_NAME_VERSION (var)); if (retval) return retval; /* We couldn't find a non-NULL value for the predicate. */ return NULL_TREE; } /* Given a comparison code COMP and names N1 and N2, compare all the ranges equivalent to N1 against all the ranges equivalent to N2 to determine the value of N1 COMP N2. Return the same value returned by compare_ranges. */ static tree compare_names (enum tree_code comp, tree n1, tree n2) { tree t, retval; bitmap e1, e2; bitmap_iterator bi1, bi2; unsigned i1, i2; /* Compare the ranges of every name equivalent to N1 against the ranges of every name equivalent to N2. */ e1 = get_value_range (n1)->equiv; e2 = get_value_range (n2)->equiv; /* Add N1 and N2 to their own set of equivalences to avoid duplicating the body of the loop just to check N1 and N2 ranges. */ bitmap_set_bit (e1, SSA_NAME_VERSION (n1)); bitmap_set_bit (e2, SSA_NAME_VERSION (n2)); /* If the equivalence sets have a common intersection, then the two names can be compared without checking their ranges. */ if (bitmap_intersect_p (e1, e2)) { bitmap_clear_bit (e1, SSA_NAME_VERSION (n1)); bitmap_clear_bit (e2, SSA_NAME_VERSION (n2)); return (comp == EQ_EXPR || comp == GE_EXPR || comp == LE_EXPR) ? boolean_true_node : boolean_false_node; } /* Otherwise, compare all the equivalent ranges. First, add N1 and N2 to their own set of equivalences to avoid duplicating the body of the loop just to check N1 and N2 ranges. */ EXECUTE_IF_SET_IN_BITMAP (e1, 0, i1, bi1) { value_range_t vr1 = *(vr_value[i1]); /* If the range is VARYING or UNDEFINED, use the name itself. */ if (vr1.type == VR_VARYING || vr1.type == VR_UNDEFINED) { vr1.type = VR_RANGE; vr1.min = ssa_name (i1); vr1.max = ssa_name (i1); } t = retval = NULL_TREE; EXECUTE_IF_SET_IN_BITMAP (e2, 0, i2, bi2) { value_range_t vr2 = *(vr_value[i2]); if (vr2.type == VR_VARYING || vr2.type == VR_UNDEFINED) { vr2.type = VR_RANGE; vr2.min = ssa_name (i2); vr2.max = ssa_name (i2); } t = compare_ranges (comp, &vr1, &vr2); if (t) { /* All the ranges in the equivalent sets should compare the same. */ gcc_assert (retval == NULL || t == retval); retval = t; } } if (retval) { bitmap_clear_bit (e1, SSA_NAME_VERSION (n1)); bitmap_clear_bit (e2, SSA_NAME_VERSION (n2)); return retval; } } /* None of the equivalent ranges are useful in computing this comparison. */ bitmap_clear_bit (e1, SSA_NAME_VERSION (n1)); bitmap_clear_bit (e2, SSA_NAME_VERSION (n2)); return NULL_TREE; } /* Given a conditional predicate COND, try to determine if COND yields true or false based on the value ranges of its operands. Return BOOLEAN_TRUE_NODE if the conditional always evaluates to true, BOOLEAN_FALSE_NODE if the conditional always evaluates to false, and, NULL if the conditional cannot be evaluated at compile time. If USE_EQUIV_P is true, the ranges of all the names equivalent with the operands in COND are used when trying to compute its value. This is only used during final substitution. During propagation, we only check the range of each variable and not its equivalents. */ tree vrp_evaluate_conditional (tree cond, bool use_equiv_p) { gcc_assert (TREE_CODE (cond) == SSA_NAME || TREE_CODE_CLASS (TREE_CODE (cond)) == tcc_comparison); if (TREE_CODE (cond) == SSA_NAME) { value_range_t *vr; tree retval; if (use_equiv_p) retval = compare_name_with_value (NE_EXPR, cond, boolean_false_node); else { value_range_t *vr = get_value_range (cond); retval = compare_range_with_value (NE_EXPR, vr, boolean_false_node); } /* If COND has a known boolean range, return it. */ if (retval) return retval; /* Otherwise, if COND has a symbolic range of exactly one value, return it. */ vr = get_value_range (cond); if (vr->type == VR_RANGE && vr->min == vr->max) return vr->min; } else { tree op0 = TREE_OPERAND (cond, 0); tree op1 = TREE_OPERAND (cond, 1); /* We only deal with integral and pointer types. */ if (!INTEGRAL_TYPE_P (TREE_TYPE (op0)) && !POINTER_TYPE_P (TREE_TYPE (op0))) return NULL_TREE; if (use_equiv_p) { if (TREE_CODE (op0) == SSA_NAME && TREE_CODE (op1) == SSA_NAME) return compare_names (TREE_CODE (cond), op0, op1); else if (TREE_CODE (op0) == SSA_NAME) return compare_name_with_value (TREE_CODE (cond), op0, op1); else if (TREE_CODE (op1) == SSA_NAME) return compare_name_with_value ( swap_tree_comparison (TREE_CODE (cond)), op1, op0); } else { value_range_t *vr0, *vr1; vr0 = (TREE_CODE (op0) == SSA_NAME) ? get_value_range (op0) : NULL; vr1 = (TREE_CODE (op1) == SSA_NAME) ? get_value_range (op1) : NULL; if (vr0 && vr1) return compare_ranges (TREE_CODE (cond), vr0, vr1); else if (vr0 && vr1 == NULL) return compare_range_with_value (TREE_CODE (cond), vr0, op1); else if (vr0 == NULL && vr1) return compare_range_with_value ( swap_tree_comparison (TREE_CODE (cond)), vr1, op0); } } /* Anything else cannot be computed statically. */ return NULL_TREE; } /* Visit conditional statement STMT. If we can determine which edge will be taken out of STMT's basic block, record it in *TAKEN_EDGE_P and return SSA_PROP_INTERESTING. Otherwise, return SSA_PROP_VARYING. */ static enum ssa_prop_result vrp_visit_cond_stmt (tree stmt, edge *taken_edge_p) { tree cond, val; *taken_edge_p = NULL; /* FIXME. Handle SWITCH_EXPRs. But first, the assert pass needs to add ASSERT_EXPRs for them. */ if (TREE_CODE (stmt) == SWITCH_EXPR) return SSA_PROP_VARYING; cond = COND_EXPR_COND (stmt); if (dump_file && (dump_flags & TDF_DETAILS)) { tree use; ssa_op_iter i; fprintf (dump_file, "\nVisiting conditional with predicate: "); print_generic_expr (dump_file, cond, 0); fprintf (dump_file, "\nWith known ranges\n"); FOR_EACH_SSA_TREE_OPERAND (use, stmt, i, SSA_OP_USE) { fprintf (dump_file, "\t"); print_generic_expr (dump_file, use, 0); fprintf (dump_file, ": "); dump_value_range (dump_file, vr_value[SSA_NAME_VERSION (use)]); } fprintf (dump_file, "\n"); } /* Compute the value of the predicate COND by checking the known ranges of each of its operands. Note that we cannot evaluate all the equivalent ranges here because those ranges may not yet be final and with the current propagation strategy, we cannot determine when the value ranges of the names in the equivalence set have changed. For instance, given the following code fragment i_5 = PHI <8, i_13> ... i_14 = ASSERT_EXPR if (i_14 == 1) ... Assume that on the first visit to i_14, i_5 has the temporary range [8, 8] because the second argument to the PHI function is not yet executable. We derive the range ~[0, 0] for i_14 and the equivalence set { i_5 }. So, when we visit 'if (i_14 == 1)' for the first time, since i_14 is equivalent to the range [8, 8], we determine that the predicate is always false. On the next round of propagation, i_13 is determined to be VARYING, which causes i_5 to drop down to VARYING. So, another visit to i_14 is scheduled. In this second visit, we compute the exact same range and equivalence set for i_14, namely ~[0, 0] and { i_5 }. But we did not have the previous range for i_5 registered, so vrp_visit_assignment thinks that the range for i_14 has not changed. Therefore, the predicate 'if (i_14 == 1)' is not visited again, which stops propagation from visiting statements in the THEN clause of that if(). To properly fix this we would need to keep the previous range value for the names in the equivalence set. This way we would've discovered that from one visit to the other i_5 changed from range [8, 8] to VR_VARYING. However, fixing this apparent limitation may not be worth the additional checking. Testing on several code bases (GCC, DLV, MICO, TRAMP3D and SPEC2000) showed that doing this results in 4 more predicates folded in SPEC. */ val = vrp_evaluate_conditional (cond, false); if (val) *taken_edge_p = find_taken_edge (bb_for_stmt (stmt), val); if (dump_file && (dump_flags & TDF_DETAILS)) { fprintf (dump_file, "\nPredicate evaluates to: "); if (val == NULL_TREE) fprintf (dump_file, "DON'T KNOW\n"); else print_generic_stmt (dump_file, val, 0); } return (*taken_edge_p) ? SSA_PROP_INTERESTING : SSA_PROP_VARYING; } /* Evaluate statement STMT. If the statement produces a useful range, return SSA_PROP_INTERESTING and record the SSA name with the interesting range into *OUTPUT_P. If STMT is a conditional branch and we can determine its truth value, the taken edge is recorded in *TAKEN_EDGE_P. If STMT produces a varying value, return SSA_PROP_VARYING. */ static enum ssa_prop_result vrp_visit_stmt (tree stmt, edge *taken_edge_p, tree *output_p) { tree def; ssa_op_iter iter; stmt_ann_t ann; if (dump_file && (dump_flags & TDF_DETAILS)) { fprintf (dump_file, "\nVisiting statement:\n"); print_generic_stmt (dump_file, stmt, dump_flags); fprintf (dump_file, "\n"); } ann = stmt_ann (stmt); if (TREE_CODE (stmt) == MODIFY_EXPR && ZERO_SSA_OPERANDS (stmt, SSA_OP_ALL_VIRTUALS)) return vrp_visit_assignment (stmt, output_p); else if (TREE_CODE (stmt) == COND_EXPR || TREE_CODE (stmt) == SWITCH_EXPR) return vrp_visit_cond_stmt (stmt, taken_edge_p); /* All other statements produce nothing of interest for VRP, so mark their outputs varying and prevent further simulation. */ FOR_EACH_SSA_TREE_OPERAND (def, stmt, iter, SSA_OP_DEF) set_value_range_to_varying (get_value_range (def)); return SSA_PROP_VARYING; } /* Meet operation for value ranges. Given two value ranges VR0 and VR1, store in VR0 the result of meeting VR0 and VR1. The meeting rules are as follows: 1- If VR0 and VR1 have an empty intersection, set VR0 to VR_VARYING. 2- If VR0 and VR1 have a non-empty intersection, set VR0 to the union of VR0 and VR1. */ static void vrp_meet (value_range_t *vr0, value_range_t *vr1) { if (vr0->type == VR_UNDEFINED) { copy_value_range (vr0, vr1); return; } if (vr1->type == VR_UNDEFINED) { /* Nothing to do. VR0 already has the resulting range. */ return; } if (vr0->type == VR_VARYING) { /* Nothing to do. VR0 already has the resulting range. */ return; } if (vr1->type == VR_VARYING) { set_value_range_to_varying (vr0); return; } if (vr0->type == VR_RANGE && vr1->type == VR_RANGE) { /* If VR0 and VR1 have a non-empty intersection, compute the union of both ranges. */ if (value_ranges_intersect_p (vr0, vr1)) { int cmp; tree min, max; /* The lower limit of the new range is the minimum of the two ranges. If they cannot be compared, the result is VARYING. */ cmp = compare_values (vr0->min, vr1->min); if (cmp == 0 || cmp == 1) min = vr1->min; else if (cmp == -1) min = vr0->min; else { set_value_range_to_varying (vr0); return; } /* Similarly, the upper limit of the new range is the maximum of the two ranges. If they cannot be compared, the result is VARYING. */ cmp = compare_values (vr0->max, vr1->max); if (cmp == 0 || cmp == -1) max = vr1->max; else if (cmp == 1) max = vr0->max; else { set_value_range_to_varying (vr0); return; } /* The resulting set of equivalences is the intersection of the two sets. */ if (vr0->equiv && vr1->equiv && vr0->equiv != vr1->equiv) bitmap_and_into (vr0->equiv, vr1->equiv); else if (vr0->equiv && !vr1->equiv) bitmap_clear (vr0->equiv); set_value_range (vr0, vr0->type, min, max, vr0->equiv); } else goto no_meet; } else if (vr0->type == VR_ANTI_RANGE && vr1->type == VR_ANTI_RANGE) { /* Two anti-ranges meet only if they are both identical. */ if (compare_values (vr0->min, vr1->min) == 0 && compare_values (vr0->max, vr1->max) == 0 && compare_values (vr0->min, vr0->max) == 0) { /* The resulting set of equivalences is the intersection of the two sets. */ if (vr0->equiv && vr1->equiv && vr0->equiv != vr1->equiv) bitmap_and_into (vr0->equiv, vr1->equiv); else if (vr0->equiv && !vr1->equiv) bitmap_clear (vr0->equiv); } else goto no_meet; } else if (vr0->type == VR_ANTI_RANGE || vr1->type == VR_ANTI_RANGE) { /* A numeric range [VAL1, VAL2] and an anti-range ~[VAL3, VAL4] meet only if the ranges have an empty intersection. The result of the meet operation is the anti-range. */ if (!symbolic_range_p (vr0) && !symbolic_range_p (vr1) && !value_ranges_intersect_p (vr0, vr1)) { /* Copy most of VR1 into VR0. Don't copy VR1's equivalence set. We need to compute the intersection of the two equivalence sets. */ if (vr1->type == VR_ANTI_RANGE) set_value_range (vr0, vr1->type, vr1->min, vr1->max, vr0->equiv); /* The resulting set of equivalences is the intersection of the two sets. */ if (vr0->equiv && vr1->equiv && vr0->equiv != vr1->equiv) bitmap_and_into (vr0->equiv, vr1->equiv); else if (vr0->equiv && !vr1->equiv) bitmap_clear (vr0->equiv); } else goto no_meet; } else gcc_unreachable (); return; no_meet: /* The two range VR0 and VR1 do not meet. Before giving up and setting the result to VARYING, see if we can at least derive a useful anti-range. FIXME, all this nonsense about distinguishing anti-ranges from ranges is necessary because of the odd semantics of range_includes_zero_p and friends. */ if (!symbolic_range_p (vr0) && ((vr0->type == VR_RANGE && !range_includes_zero_p (vr0)) || (vr0->type == VR_ANTI_RANGE && range_includes_zero_p (vr0))) && !symbolic_range_p (vr1) && ((vr1->type == VR_RANGE && !range_includes_zero_p (vr1)) || (vr1->type == VR_ANTI_RANGE && range_includes_zero_p (vr1)))) { set_value_range_to_nonnull (vr0, TREE_TYPE (vr0->min)); /* Since this meet operation did not result from the meeting of two equivalent names, VR0 cannot have any equivalences. */ if (vr0->equiv) bitmap_clear (vr0->equiv); } else set_value_range_to_varying (vr0); } /* Visit all arguments for PHI node PHI that flow through executable edges. If a valid value range can be derived from all the incoming value ranges, set a new range for the LHS of PHI. */ static enum ssa_prop_result vrp_visit_phi_node (tree phi) { int i; tree lhs = PHI_RESULT (phi); value_range_t *lhs_vr = get_value_range (lhs); value_range_t vr_result = { VR_UNDEFINED, NULL_TREE, NULL_TREE, NULL }; copy_value_range (&vr_result, lhs_vr); if (dump_file && (dump_flags & TDF_DETAILS)) { fprintf (dump_file, "\nVisiting PHI node: "); print_generic_expr (dump_file, phi, dump_flags); } for (i = 0; i < PHI_NUM_ARGS (phi); i++) { edge e = PHI_ARG_EDGE (phi, i); if (dump_file && (dump_flags & TDF_DETAILS)) { fprintf (dump_file, "\n Argument #%d (%d -> %d %sexecutable)\n", i, e->src->index, e->dest->index, (e->flags & EDGE_EXECUTABLE) ? "" : "not "); } if (e->flags & EDGE_EXECUTABLE) { tree arg = PHI_ARG_DEF (phi, i); value_range_t vr_arg; if (TREE_CODE (arg) == SSA_NAME) vr_arg = *(get_value_range (arg)); else { vr_arg.type = VR_RANGE; vr_arg.min = arg; vr_arg.max = arg; vr_arg.equiv = NULL; } if (dump_file && (dump_flags & TDF_DETAILS)) { fprintf (dump_file, "\t"); print_generic_expr (dump_file, arg, dump_flags); fprintf (dump_file, "\n\tValue: "); dump_value_range (dump_file, &vr_arg); fprintf (dump_file, "\n"); } vrp_meet (&vr_result, &vr_arg); if (vr_result.type == VR_VARYING) break; } } if (vr_result.type == VR_VARYING) goto varying; /* To prevent infinite iterations in the algorithm, derive ranges when the new value is slightly bigger or smaller than the previous one. */ if (lhs_vr->type == VR_RANGE && vr_result.type == VR_RANGE) { if (!POINTER_TYPE_P (TREE_TYPE (lhs))) { int cmp_min = compare_values (lhs_vr->min, vr_result.min); int cmp_max = compare_values (lhs_vr->max, vr_result.max); /* If the new minimum is smaller or larger than the previous one, go all the way to -INF. In the first case, to avoid iterating millions of times to reach -INF, and in the other case to avoid infinite bouncing between different minimums. */ if (cmp_min > 0 || cmp_min < 0) vr_result.min = TYPE_MIN_VALUE (TREE_TYPE (vr_result.min)); /* Similarly, if the new maximum is smaller or larger than the previous one, go all the way to +INF. */ if (cmp_max < 0 || cmp_max > 0) vr_result.max = TYPE_MAX_VALUE (TREE_TYPE (vr_result.max)); /* If we ended up with a (-INF, +INF) range, set it to VARYING. */ if (vr_result.min == TYPE_MIN_VALUE (TREE_TYPE (vr_result.min)) && vr_result.max == TYPE_MAX_VALUE (TREE_TYPE (vr_result.max))) goto varying; } } /* If the new range is different than the previous value, keep iterating. */ if (update_value_range (lhs, &vr_result)) return SSA_PROP_INTERESTING; /* Nothing changed, don't add outgoing edges. */ return SSA_PROP_NOT_INTERESTING; /* No match found. Set the LHS to VARYING. */ varying: set_value_range_to_varying (lhs_vr); return SSA_PROP_VARYING; } /* Simplify a division or modulo operator to a right shift or bitwise and if the first operand is unsigned or is greater than zero and the second operand is an exact power of two. */ static void simplify_div_or_mod_using_ranges (tree stmt, tree rhs, enum tree_code rhs_code) { tree val = NULL; tree op = TREE_OPERAND (rhs, 0); value_range_t *vr = get_value_range (TREE_OPERAND (rhs, 0)); if (TYPE_UNSIGNED (TREE_TYPE (op))) { val = integer_one_node; } else { val = compare_range_with_value (GT_EXPR, vr, integer_zero_node); } if (val && integer_onep (val)) { tree t; tree op0 = TREE_OPERAND (rhs, 0); tree op1 = TREE_OPERAND (rhs, 1); if (rhs_code == TRUNC_DIV_EXPR) { t = build_int_cst (NULL_TREE, tree_log2 (op1)); t = build2 (RSHIFT_EXPR, TREE_TYPE (op0), op0, t); } else { t = build_int_cst (TREE_TYPE (op1), 1); t = int_const_binop (MINUS_EXPR, op1, t, 0); t = fold_convert (TREE_TYPE (op0), t); t = build2 (BIT_AND_EXPR, TREE_TYPE (op0), op0, t); } TREE_OPERAND (stmt, 1) = t; update_stmt (stmt); } } /* If the operand to an ABS_EXPR is >= 0, then eliminate the ABS_EXPR. If the operand is <= 0, then simplify the ABS_EXPR into a NEGATE_EXPR. */ static void simplify_abs_using_ranges (tree stmt, tree rhs) { tree val = NULL; tree op = TREE_OPERAND (rhs, 0); tree type = TREE_TYPE (op); value_range_t *vr = get_value_range (TREE_OPERAND (rhs, 0)); if (TYPE_UNSIGNED (type)) { val = integer_zero_node; } else if (vr) { val = compare_range_with_value (LE_EXPR, vr, integer_zero_node); if (!val) { val = compare_range_with_value (GE_EXPR, vr, integer_zero_node); if (val) { if (integer_zerop (val)) val = integer_one_node; else if (integer_onep (val)) val = integer_zero_node; } } if (val && (integer_onep (val) || integer_zerop (val))) { tree t; if (integer_onep (val)) t = build1 (NEGATE_EXPR, TREE_TYPE (op), op); else t = op; TREE_OPERAND (stmt, 1) = t; update_stmt (stmt); } } } /* We are comparing trees OP0 and OP1 using COND_CODE. OP0 has a known value range VR. If there is one and only one value which will satisfy the conditional, then return that value. Else return NULL. */ static tree test_for_singularity (enum tree_code cond_code, tree op0, tree op1, value_range_t *vr) { tree min = NULL; tree max = NULL; /* Extract minimum/maximum values which satisfy the the conditional as it was written. */ if (cond_code == LE_EXPR || cond_code == LT_EXPR) { min = TYPE_MIN_VALUE (TREE_TYPE (op0)); max = op1; if (cond_code == LT_EXPR) { tree one = build_int_cst (TREE_TYPE (op0), 1); max = fold_build2 (MINUS_EXPR, TREE_TYPE (op0), max, one); } } else if (cond_code == GE_EXPR || cond_code == GT_EXPR) { max = TYPE_MAX_VALUE (TREE_TYPE (op0)); min = op1; if (cond_code == GT_EXPR) { tree one = build_int_cst (TREE_TYPE (op0), 1); min = fold_build2 (PLUS_EXPR, TREE_TYPE (op0), min, one); } } /* Now refine the minimum and maximum values using any value range information we have for op0. */ if (min && max) { if (compare_values (vr->min, min) == -1) min = min; else min = vr->min; if (compare_values (vr->max, max) == 1) max = max; else max = vr->max; /* If the new min/max values have converged to a single value, then there is only one value which can satisfy the condition, return that value. */ if (operand_equal_p (min, max, 0) && is_gimple_min_invariant (min)) return min; } return NULL; } /* Simplify a conditional using a relational operator to an equality test if the range information indicates only one value can satisfy the original conditional. */ static void simplify_cond_using_ranges (tree stmt) { tree cond = COND_EXPR_COND (stmt); tree op0 = TREE_OPERAND (cond, 0); tree op1 = TREE_OPERAND (cond, 1); enum tree_code cond_code = TREE_CODE (cond); if (cond_code != NE_EXPR && cond_code != EQ_EXPR && TREE_CODE (op0) == SSA_NAME && INTEGRAL_TYPE_P (TREE_TYPE (op0)) && is_gimple_min_invariant (op1)) { value_range_t *vr = get_value_range (op0); /* If we have range information for OP0, then we might be able to simplify this conditional. */ if (vr->type == VR_RANGE) { tree new = test_for_singularity (cond_code, op0, op1, vr); if (new) { if (dump_file) { fprintf (dump_file, "Simplified relational "); print_generic_expr (dump_file, cond, 0); fprintf (dump_file, " into "); } COND_EXPR_COND (stmt) = build2 (EQ_EXPR, boolean_type_node, op0, new); update_stmt (stmt); if (dump_file) { print_generic_expr (dump_file, COND_EXPR_COND (stmt), 0); fprintf (dump_file, "\n"); } return; } /* Try again after inverting the condition. We only deal with integral types here, so no need to worry about issues with inverting FP comparisons. */ cond_code = invert_tree_comparison (cond_code, false); new = test_for_singularity (cond_code, op0, op1, vr); if (new) { if (dump_file) { fprintf (dump_file, "Simplified relational "); print_generic_expr (dump_file, cond, 0); fprintf (dump_file, " into "); } COND_EXPR_COND (stmt) = build2 (NE_EXPR, boolean_type_node, op0, new); update_stmt (stmt); if (dump_file) { print_generic_expr (dump_file, COND_EXPR_COND (stmt), 0); fprintf (dump_file, "\n"); } return; } } } } /* Simplify STMT using ranges if possible. */ void simplify_stmt_using_ranges (tree stmt) { if (TREE_CODE (stmt) == MODIFY_EXPR) { tree rhs = TREE_OPERAND (stmt, 1); enum tree_code rhs_code = TREE_CODE (rhs); /* Transform TRUNC_DIV_EXPR and TRUNC_MOD_EXPR into RSHIFT_EXPR and BIT_AND_EXPR respectively if the first operand is greater than zero and the second operand is an exact power of two. */ if ((rhs_code == TRUNC_DIV_EXPR || rhs_code == TRUNC_MOD_EXPR) && INTEGRAL_TYPE_P (TREE_TYPE (TREE_OPERAND (rhs, 0))) && integer_pow2p (TREE_OPERAND (rhs, 1))) simplify_div_or_mod_using_ranges (stmt, rhs, rhs_code); /* Transform ABS (X) into X or -X as appropriate. */ if (rhs_code == ABS_EXPR && TREE_CODE (TREE_OPERAND (rhs, 0)) == SSA_NAME && INTEGRAL_TYPE_P (TREE_TYPE (TREE_OPERAND (rhs, 0)))) simplify_abs_using_ranges (stmt, rhs); } else if (TREE_CODE (stmt) == COND_EXPR && COMPARISON_CLASS_P (COND_EXPR_COND (stmt))) { simplify_cond_using_ranges (stmt); } } /* Stack of dest,src equivalency pairs that need to be restored after each attempt to thread a block's incoming edge to an outgoing edge. A NULL entry is used to mark the end of pairs which need to be restored. */ static VEC(tree,heap) *stack; /* A trivial wrapper so that we can present the generic jump threading code with a simple API for simplifying statements. */ static tree simplify_stmt_for_jump_threading (tree stmt) { /* We only use VRP information to simplify conditionals. This is overly conservative, but it's unclear if doing more would be worth the compile time cost. */ if (TREE_CODE (stmt) != COND_EXPR) return NULL; return vrp_evaluate_conditional (COND_EXPR_COND (stmt), true); } /* Blocks which have more than one predecessor and more than one successor present jump threading opportunities. ie, when the block is reached from a specific predecessor, we may be able to determine which of the outgoing edges will be traversed. When this optimization applies, we are able to avoid conditionals at runtime and we may expose secondary optimization opportunities. This routine is effectively a driver for the generic jump threading code. It basically just presents the generic code with edges that may be suitable for jump threading. Unlike DOM, we do not iterate VRP if jump threading was successful. While iterating may expose new opportunities for VRP, it is expected those opportunities would be very limited and the compile time cost to expose those opportunities would be significant. As jump threading opportunities are discovered, they are registered for later realization. */ static void identify_jump_threads (void) { basic_block bb; tree dummy; /* Ugh. When substituting values earlier in this pass we can wipe the dominance information. So rebuild the dominator information as we need it within the jump threading code. */ calculate_dominance_info (CDI_DOMINATORS); /* We do not allow VRP information to be used for jump threading across a back edge in the CFG. Otherwise it becomes too difficult to avoid eliminating loop exit tests. Of course EDGE_DFS_BACK is not accurate at this time so we have to recompute it. */ mark_dfs_back_edges (); /* Allocate our unwinder stack to unwind any temporary equivalences that might be recorded. */ stack = VEC_alloc (tree, heap, 20); /* To avoid lots of silly node creation, we create a single conditional and just modify it in-place when attempting to thread jumps. */ dummy = build2 (EQ_EXPR, boolean_type_node, NULL, NULL); dummy = build3 (COND_EXPR, void_type_node, dummy, NULL, NULL); /* Walk through all the blocks finding those which present a potential jump threading opportunity. We could set this up as a dominator walker and record data during the walk, but I doubt it's worth the effort for the classes of jump threading opportunities we are trying to identify at this point in compilation. */ FOR_EACH_BB (bb) { tree last, cond; /* If the generic jump threading code does not find this block interesting, then there is nothing to do. */ if (! potentially_threadable_block (bb)) continue; /* We only care about blocks ending in a COND_EXPR. While there may be some value in handling SWITCH_EXPR here, I doubt it's terribly important. */ last = bsi_stmt (bsi_last (bb)); if (TREE_CODE (last) != COND_EXPR) continue; /* We're basically looking for any kind of conditional with integral type arguments. */ cond = COND_EXPR_COND (last); if ((TREE_CODE (cond) == SSA_NAME && INTEGRAL_TYPE_P (TREE_TYPE (cond))) || (COMPARISON_CLASS_P (cond) && TREE_CODE (TREE_OPERAND (cond, 0)) == SSA_NAME && INTEGRAL_TYPE_P (TREE_TYPE (TREE_OPERAND (cond, 0))) && (TREE_CODE (TREE_OPERAND (cond, 1)) == SSA_NAME || is_gimple_min_invariant (TREE_OPERAND (cond, 1))) && INTEGRAL_TYPE_P (TREE_TYPE (TREE_OPERAND (cond, 1))))) { edge_iterator ei; edge e; /* We've got a block with multiple predecessors and multiple successors which also ends in a suitable conditional. For each predecessor, see if we can thread it to a specific successor. */ FOR_EACH_EDGE (e, ei, bb->preds) { /* Do not thread across back edges or abnormal edges in the CFG. */ if (e->flags & (EDGE_DFS_BACK | EDGE_COMPLEX)) continue; thread_across_edge (dummy, e, true, &stack, simplify_stmt_for_jump_threading); } } } /* We do not actually update the CFG or SSA graphs at this point as ASSERT_EXPRs are still in the IL and cfg cleanup code does not yet handle ASSERT_EXPRs gracefully. */ } /* We identified all the jump threading opportunities earlier, but could not transform the CFG at that time. This routine transforms the CFG and arranges for the dominator tree to be rebuilt if necessary. Note the SSA graph update will occur during the normal TODO processing by the pass manager. */ static void finalize_jump_threads (void) { bool cfg_altered = false; cfg_altered = thread_through_all_blocks (); /* If we threaded jumps, then we need to recompute the dominance information, to safely do that we must clean up the CFG first. */ if (cfg_altered) { free_dominance_info (CDI_DOMINATORS); cleanup_tree_cfg (); calculate_dominance_info (CDI_DOMINATORS); } VEC_free (tree, heap, stack); } /* Traverse all the blocks folding conditionals with known ranges. */ static void vrp_finalize (void) { size_t i; prop_value_t *single_val_range; bool do_value_subst_p; if (dump_file) { fprintf (dump_file, "\nValue ranges after VRP:\n\n"); dump_all_value_ranges (dump_file); fprintf (dump_file, "\n"); } /* We may have ended with ranges that have exactly one value. Those values can be substituted as any other copy/const propagated value using substitute_and_fold. */ single_val_range = XNEWVEC (prop_value_t, num_ssa_names); memset (single_val_range, 0, num_ssa_names * sizeof (*single_val_range)); do_value_subst_p = false; for (i = 0; i < num_ssa_names; i++) if (vr_value[i] && vr_value[i]->type == VR_RANGE && vr_value[i]->min == vr_value[i]->max) { single_val_range[i].value = vr_value[i]->min; do_value_subst_p = true; } if (!do_value_subst_p) { /* We found no single-valued ranges, don't waste time trying to do single value substitution in substitute_and_fold. */ free (single_val_range); single_val_range = NULL; } substitute_and_fold (single_val_range, true); /* We must identify jump threading opportunities before we release the datastructures built by VRP. */ identify_jump_threads (); /* Free allocated memory. */ for (i = 0; i < num_ssa_names; i++) if (vr_value[i]) { BITMAP_FREE (vr_value[i]->equiv); free (vr_value[i]); } free (single_val_range); free (vr_value); } /* Main entry point to VRP (Value Range Propagation). This pass is loosely based on J. R. C. Patterson, ``Accurate Static Branch Prediction by Value Range Propagation,'' in SIGPLAN Conference on Programming Language Design and Implementation, pp. 67-78, 1995. Also available at http://citeseer.ist.psu.edu/patterson95accurate.html This is essentially an SSA-CCP pass modified to deal with ranges instead of constants. While propagating ranges, we may find that two or more SSA name have equivalent, though distinct ranges. For instance, 1 x_9 = p_3->a; 2 p_4 = ASSERT_EXPR 3 if (p_4 == q_2) 4 p_5 = ASSERT_EXPR ; 5 endif 6 if (q_2) In the code above, pointer p_5 has range [q_2, q_2], but from the code we can also determine that p_5 cannot be NULL and, if q_2 had a non-varying range, p_5's range should also be compatible with it. These equivalences are created by two expressions: ASSERT_EXPR and copy operations. Since p_5 is an assertion on p_4, and p_4 was the result of another assertion, then we can use the fact that p_5 and p_4 are equivalent when evaluating p_5's range. Together with value ranges, we also propagate these equivalences between names so that we can take advantage of information from multiple ranges when doing final replacement. Note that this equivalency relation is transitive but not symmetric. In the example above, p_5 is equivalent to p_4, q_2 and p_3, but we cannot assert that q_2 is equivalent to p_5 because q_2 may be used in contexts where that assertion does not hold (e.g., in line 6). TODO, the main difference between this pass and Patterson's is that we do not propagate edge probabilities. We only compute whether edges can be taken or not. That is, instead of having a spectrum of jump probabilities between 0 and 1, we only deal with 0, 1 and DON'T KNOW. In the future, it may be worthwhile to propagate probabilities to aid branch prediction. */ static void execute_vrp (void) { insert_range_assertions (); current_loops = loop_optimizer_init (LOOPS_NORMAL); if (current_loops) scev_initialize (current_loops); vrp_initialize (); ssa_propagate (vrp_visit_stmt, vrp_visit_phi_node); vrp_finalize (); if (current_loops) { scev_finalize (); loop_optimizer_finalize (current_loops); current_loops = NULL; } /* ASSERT_EXPRs must be removed before finalizing jump threads as finalizing jump threads calls the CFG cleanup code which does not properly handle ASSERT_EXPRs. */ remove_range_assertions (); /* If we exposed any new variables, go ahead and put them into SSA form now, before we handle jump threading. This simplifies interactions between rewriting of _DECL nodes into SSA form and rewriting SSA_NAME nodes into SSA form after block duplication and CFG manipulation. */ update_ssa (TODO_update_ssa); finalize_jump_threads (); } static bool gate_vrp (void) { return flag_tree_vrp != 0; } struct tree_opt_pass pass_vrp = { "vrp", /* name */ gate_vrp, /* gate */ execute_vrp, /* execute */ NULL, /* sub */ NULL, /* next */ 0, /* static_pass_number */ TV_TREE_VRP, /* tv_id */ PROP_ssa | PROP_alias, /* properties_required */ 0, /* properties_provided */ 0, /* properties_destroyed */ 0, /* todo_flags_start */ TODO_cleanup_cfg | TODO_ggc_collect | TODO_verify_ssa | TODO_dump_func | TODO_update_ssa, /* todo_flags_finish */ 0 /* letter */ };