/* Scalar evolution detector.
- Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008 Free Software
- Foundation, Inc.
+ Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009
+ Free Software Foundation, Inc.
Contributed by Sebastian Pop <s.pop@laposte.net>
This file is part of GCC.
basic block INSTANTIATED_BELOW, the value of VAR can be expressed
as CHREC. */
-struct scev_info_str GTY(())
-{
+struct GTY(()) scev_info_str {
basic_block instantiated_below;
tree var;
tree chrec;
case GIMPLE_SINGLE_RHS:
return follow_ssa_edge_expr (loop, stmt, gimple_assign_rhs1 (stmt),
halting_phi, evolution_of_loop, limit);
+ case GIMPLE_UNARY_RHS:
+ if (code == NOP_EXPR)
+ {
+ /* This assignment is under the form "a_1 = (cast) rhs. */
+ t_bool res
+ = follow_ssa_edge_expr (loop, stmt, gimple_assign_rhs1 (stmt),
+ halting_phi, evolution_of_loop, limit);
+ *evolution_of_loop = chrec_convert (type, *evolution_of_loop, stmt);
+ return res;
+ }
+ /* FALLTHRU */
+
default:
return t_false;
}
*evolution_of_loop = evolution_of_branch;
- /* If the phi node is just a copy, do not increase the limit. */
n = gimple_phi_num_args (condition_phi);
- if (n > 1)
- limit++;
-
for (i = 1; i < n; i++)
{
/* Quickly give up when the evolution of one of the branches is
if (*evolution_of_loop == chrec_dont_know)
return t_true;
+ /* Increase the limit by the PHI argument number to avoid exponential
+ time and memory complexity. */
res = follow_ssa_edge_in_condition_phi_branch (i, loop, condition_phi,
halting_phi,
&evolution_of_branch,
- init, limit);
+ init, limit + i);
if (res == t_false || res == t_dont_know)
return res;
if (init_cond == chrec_not_analyzed_yet)
init_cond = chrec_dont_know;
+ /* During early loop unrolling we do not have fully constant propagated IL.
+ Handle degenerate PHIs here to not miss important unrollings. */
+ if (TREE_CODE (init_cond) == SSA_NAME)
+ {
+ gimple def = SSA_NAME_DEF_STMT (init_cond);
+ tree res;
+ if (gimple_code (def) == GIMPLE_PHI
+ && (res = degenerate_phi_result (def)) != NULL_TREE
+ /* Only allow invariants here, otherwise we may break
+ loop-closed SSA form. */
+ && is_gimple_min_invariant (res))
+ init_cond = res;
+ }
+
if (dump_file && (dump_flags & TDF_DETAILS))
{
fprintf (dump_file, " (init_cond = ");
fold_convert (type, integer_minus_one_node));
break;
+ case BIT_NOT_EXPR:
+ /* Handle ~X as -1 - X. */
+ chrec1 = analyze_scalar_evolution (loop, rhs1);
+ chrec1 = chrec_convert (type, chrec1, at_stmt);
+ res = chrec_fold_minus (type,
+ fold_convert (type, integer_minus_one_node),
+ chrec1);
+ break;
+
case MULT_EXPR:
chrec1 = analyze_scalar_evolution (loop, rhs1);
chrec2 = analyze_scalar_evolution (loop, rhs2);
}
/* Analyze scalar evolution of use of VERSION in USE_LOOP with respect to
- WRTO_LOOP (which should be a superloop of both USE_LOOP and definition
- of VERSION).
+ WRTO_LOOP (which should be a superloop of USE_LOOP)
FOLDED_CASTS is set to true if resolve_mixers used
chrec_convert_aggressive (TODO -- not really, we are way too conservative
- at the moment in order to keep things simple). */
+ at the moment in order to keep things simple).
+
+ To illustrate the meaning of USE_LOOP and WRTO_LOOP, consider the following
+ example:
+
+ for (i = 0; i < 100; i++) -- loop 1
+ {
+ for (j = 0; j < 100; j++) -- loop 2
+ {
+ k1 = i;
+ k2 = j;
+
+ use2 (k1, k2);
+
+ for (t = 0; t < 100; t++) -- loop 3
+ use3 (k1, k2);
+
+ }
+ use1 (k1, k2);
+ }
+
+ Both k1 and k2 are invariants in loop3, thus
+ analyze_scalar_evolution_in_loop (loop3, loop3, k1) = k1
+ analyze_scalar_evolution_in_loop (loop3, loop3, k2) = k2
+
+ As they are invariant, it does not matter whether we consider their
+ usage in loop 3 or loop 2, hence
+ analyze_scalar_evolution_in_loop (loop2, loop3, k1) =
+ analyze_scalar_evolution_in_loop (loop2, loop2, k1) = i
+ analyze_scalar_evolution_in_loop (loop2, loop3, k2) =
+ analyze_scalar_evolution_in_loop (loop2, loop2, k2) = [0,+,1]_2
+
+ Similarly for their evolutions with respect to loop 1. The values of K2
+ in the use in loop 2 vary independently on loop 1, thus we cannot express
+ the evolution with respect to loop 1:
+ analyze_scalar_evolution_in_loop (loop1, loop3, k1) =
+ analyze_scalar_evolution_in_loop (loop1, loop2, k1) = [0,+,1]_1
+ analyze_scalar_evolution_in_loop (loop1, loop3, k2) =
+ analyze_scalar_evolution_in_loop (loop1, loop2, k2) = dont_know
+
+ The value of k2 in the use in loop 1 is known, though:
+ analyze_scalar_evolution_in_loop (loop1, loop1, k1) = [0,+,1]_1
+ analyze_scalar_evolution_in_loop (loop1, loop1, k2) = 100
+ */
static tree
analyze_scalar_evolution_in_loop (struct loop *wrto_loop, struct loop *use_loop,
bool val = false;
tree ev = version, tmp;
+ /* We cannot just do
+
+ tmp = analyze_scalar_evolution (use_loop, version);
+ ev = resolve_mixers (wrto_loop, tmp);
+
+ as resolve_mixers would query the scalar evolution with respect to
+ wrto_loop. For example, in the situation described in the function
+ comment, suppose that wrto_loop = loop1, use_loop = loop3 and
+ version = k2. Then
+
+ analyze_scalar_evolution (use_loop, version) = k2
+
+ and resolve_mixers (loop1, k2) finds that the value of k2 in loop 1
+ is 100, which is a wrong result, since we are interested in the
+ value in loop 3.
+
+ Instead, we need to proceed from use_loop to wrto_loop loop by loop,
+ each time checking that there is no evolution in the inner loop. */
+
if (folded_casts)
*folded_casts = false;
while (1)
PTR *slot;
pattern.var = version;
+ pattern.instantiated_below = instantiated_below;
slot = htab_find_slot (cache, &pattern, INSERT);
if (!*slot)
return chrec_convert (TREE_TYPE (chrec), op0, NULL);
+ case BIT_NOT_EXPR:
+ /* Handle ~X as -1 - X. */
+ op0 = instantiate_scev_1 (instantiate_below, evolution_loop,
+ TREE_OPERAND (chrec, 0),
+ fold_conversions, cache, size_expr);
+ if (op0 == chrec_dont_know)
+ return chrec_dont_know;
+
+ if (TREE_OPERAND (chrec, 0) != op0)
+ {
+ op0 = chrec_convert (type, op0, NULL);
+ chrec = chrec_fold_minus (type,
+ fold_convert (type,
+ integer_minus_one_node),
+ op0);
+ }
+ return chrec;
+
case SCEV_NOT_KNOWN:
return chrec_dont_know;
break;
}
- gcc_assert (!VL_EXP_CLASS_P (chrec));
+ if (VL_EXP_CLASS_P (chrec))
+ return chrec_dont_know;
+
switch (TREE_CODE_LENGTH (TREE_CODE (chrec)))
{
case 3:
}
}
-/* Checks whether OP behaves as a simple affine iv of LOOP in STMT and returns
- its base and step in IV if possible. If ALLOW_NONCONSTANT_STEP is true, we
- want step to be invariant in LOOP. Otherwise we require it to be an
- integer constant. IV->no_overflow is set to true if we are sure the iv cannot
- overflow (e.g. because it is computed in signed arithmetics). */
+/* Checks whether use of OP in USE_LOOP behaves as a simple affine iv with
+ respect to WRTO_LOOP and returns its base and step in IV if possible
+ (see analyze_scalar_evolution_in_loop for more details on USE_LOOP
+ and WRTO_LOOP). If ALLOW_NONCONSTANT_STEP is true, we want step to be
+ invariant in LOOP. Otherwise we require it to be an integer constant.
+
+ IV->no_overflow is set to true if we are sure the iv cannot overflow (e.g.
+ because it is computed in signed arithmetics). Consequently, adding an
+ induction variable
+
+ for (i = IV->base; ; i += IV->step)
+
+ is only safe if IV->no_overflow is false, or TYPE_OVERFLOW_UNDEFINED is
+ false for the type of the induction variable, or you can prove that i does
+ not wrap by some other argument. Otherwise, this might introduce undefined
+ behavior, and
+
+ for (i = iv->base; ; i = (type) ((unsigned type) i + (unsigned type) iv->step))
+
+ must be used instead. */
bool
-simple_iv (struct loop *loop, gimple stmt, tree op, affine_iv *iv,
- bool allow_nonconstant_step)
+simple_iv (struct loop *wrto_loop, struct loop *use_loop, tree op,
+ affine_iv *iv, bool allow_nonconstant_step)
{
- basic_block bb = gimple_bb (stmt);
tree type, ev;
bool folded_casts;
&& TREE_CODE (type) != POINTER_TYPE)
return false;
- ev = analyze_scalar_evolution_in_loop (loop, bb->loop_father, op,
+ ev = analyze_scalar_evolution_in_loop (wrto_loop, use_loop, op,
&folded_casts);
- if (chrec_contains_undetermined (ev))
+ if (chrec_contains_undetermined (ev)
+ || chrec_contains_symbols_defined_in_loop (ev, wrto_loop->num))
return false;
- if (tree_does_not_contain_chrecs (ev)
- && !chrec_contains_symbols_defined_in_loop (ev, loop->num))
+ if (tree_does_not_contain_chrecs (ev))
{
iv->base = ev;
iv->step = build_int_cst (TREE_TYPE (ev), 0);
}
if (TREE_CODE (ev) != POLYNOMIAL_CHREC
- || CHREC_VARIABLE (ev) != (unsigned) loop->num)
+ || CHREC_VARIABLE (ev) != (unsigned) wrto_loop->num)
return false;
iv->step = CHREC_RIGHT (ev);
- if (allow_nonconstant_step)
- {
- if (tree_contains_chrecs (iv->step, NULL)
- || chrec_contains_symbols_defined_in_loop (iv->step, loop->num))
- return false;
- }
- else if (TREE_CODE (iv->step) != INTEGER_CST)
+ if ((!allow_nonconstant_step && TREE_CODE (iv->step) != INTEGER_CST)
+ || tree_contains_chrecs (iv->step, NULL))
return false;
iv->base = CHREC_LEFT (ev);
- if (tree_contains_chrecs (iv->base, NULL)
- || chrec_contains_symbols_defined_in_loop (iv->base, loop->num))
+ if (tree_contains_chrecs (iv->base, NULL))
return false;
iv->no_overflow = !folded_casts && TYPE_OVERFLOW_UNDEFINED (type);
scalar_evolution_info = NULL;
}
+/* Returns true if the expression EXPR is considered to be too expensive
+ for scev_const_prop. */
+
+bool
+expression_expensive_p (tree expr)
+{
+ enum tree_code code;
+
+ if (is_gimple_val (expr))
+ return false;
+
+ code = TREE_CODE (expr);
+ if (code == TRUNC_DIV_EXPR
+ || code == CEIL_DIV_EXPR
+ || code == FLOOR_DIV_EXPR
+ || code == ROUND_DIV_EXPR
+ || code == TRUNC_MOD_EXPR
+ || code == CEIL_MOD_EXPR
+ || code == FLOOR_MOD_EXPR
+ || code == ROUND_MOD_EXPR
+ || code == EXACT_DIV_EXPR)
+ {
+ /* Division by power of two is usually cheap, so we allow it.
+ Forbid anything else. */
+ if (!integer_pow2p (TREE_OPERAND (expr, 1)))
+ return true;
+ }
+
+ switch (TREE_CODE_CLASS (code))
+ {
+ case tcc_binary:
+ case tcc_comparison:
+ if (expression_expensive_p (TREE_OPERAND (expr, 1)))
+ return true;
+
+ /* Fallthru. */
+ case tcc_unary:
+ return expression_expensive_p (TREE_OPERAND (expr, 0));
+
+ default:
+ return true;
+ }
+}
+
/* Replace ssa names for that scev can prove they are constant by the
appropriate constants. Also perform final value replacement in loops,
in case the replacement expressions are cheap.
continue;
niter = number_of_latch_executions (loop);
- /* We used to check here whether the computation of NITER is expensive,
- and avoided final value elimination if that is the case. The problem
- is that it is hard to evaluate whether the expression is too
- expensive, as we do not know what optimization opportunities the
- elimination of the final value may reveal. Therefore, we now
- eliminate the final values of induction variables unconditionally. */
if (niter == chrec_dont_know)
continue;
/* Moving the computation from the loop may prolong life range
of some ssa names, which may cause problems if they appear
on abnormal edges. */
- || contains_abnormal_ssa_name_p (def))
+ || contains_abnormal_ssa_name_p (def)
+ /* Do not emit expensive expressions. The rationale is that
+ when someone writes a code like
+
+ while (n > 45) n -= 45;
+
+ he probably knows that n is not large, and does not want it
+ to be turned into n %= 45. */
+ || expression_expensive_p (def))
{
gsi_next (&psi);
continue;
OSDN Git Service
