/* Induction variable canonicalization.
- Copyright (C) 2004, 2005 Free Software Foundation, Inc.
+ Copyright (C) 2004, 2005, 2007, 2008 Free Software Foundation, Inc.
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
+Free Software Foundation; either version 3, or (at your option) any
later version.
GCC is distributed in the hope that it will be useful, but WITHOUT
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. */
+along with GCC; see the file COPYING3. If not see
+<http://www.gnu.org/licenses/>. */
/* This pass detects the loops that iterate a constant number of times,
adds a canonical induction variable (step -1, tested against 0)
#include "params.h"
#include "flags.h"
#include "tree-inline.h"
+#include "target.h"
/* Specifies types of loops that may be unrolled. */
create_canonical_iv (struct loop *loop, edge exit, tree niter)
{
edge in;
- tree cond, type, var;
- block_stmt_iterator incr_at;
+ tree type, var;
+ gimple cond;
+ gimple_stmt_iterator incr_at;
enum tree_code cmp;
if (dump_file && (dump_flags & TDF_DETAILS))
niter = fold_build2 (PLUS_EXPR, type,
niter,
build_int_cst (type, 1));
- incr_at = bsi_last (in->src);
+ incr_at = gsi_last_bb (in->src);
create_iv (niter,
build_int_cst (type, -1),
NULL_TREE, loop,
&incr_at, false, NULL, &var);
cmp = (exit->flags & EDGE_TRUE_VALUE) ? EQ_EXPR : NE_EXPR;
- COND_EXPR_COND (cond) = build2 (cmp, boolean_type_node,
- var,
- build_int_cst (type, 0));
+ gimple_cond_set_code (cond, cmp);
+ gimple_cond_set_lhs (cond, var);
+ gimple_cond_set_rhs (cond, build_int_cst (type, 0));
update_stmt (cond);
}
-/* Computes an estimated number of insns in LOOP. */
+/* Computes an estimated number of insns in LOOP, weighted by WEIGHTS. */
unsigned
-tree_num_loop_insns (struct loop *loop)
+tree_num_loop_insns (struct loop *loop, eni_weights *weights)
{
basic_block *body = get_loop_body (loop);
- block_stmt_iterator bsi;
- unsigned size = 1, i;
+ gimple_stmt_iterator gsi;
+ unsigned size = 0, i;
for (i = 0; i < loop->num_nodes; i++)
- for (bsi = bsi_start (body[i]); !bsi_end_p (bsi); bsi_next (&bsi))
- size += estimate_num_insns (bsi_stmt (bsi));
+ for (gsi = gsi_start_bb (body[i]); !gsi_end_p (gsi); gsi_next (&gsi))
+ size += estimate_num_insns (gsi_stmt (gsi), weights);
free (body);
return size;
}
-/* Estimate number of insns of completely unrolled loop. We assume
- that the size of the unrolled loop is decreased in the
- following way (the numbers of insns are based on what
- estimate_num_insns returns for appropriate statements):
+/* Describe size of loop as detected by tree_estimate_loop_size. */
+struct loop_size
+{
+ /* Number of instructions in the loop. */
+ int overall;
+
+ /* Number of instructions that will be likely optimized out in
+ peeled iterations of loop (i.e. computation based on induction
+ variable where induction variable starts at known constant.) */
+ int eliminated_by_peeling;
+
+ /* Same statistics for last iteration of loop: it is smaller because
+ instructions after exit are not executed. */
+ int last_iteration;
+ int last_iteration_eliminated_by_peeling;
+};
- 1) exit condition gets removed (2 insns)
- 2) increment of the control variable gets removed (2 insns)
- 3) All remaining statements are likely to get simplified
- due to constant propagation. Hard to estimate; just
- as a heuristics we decrease the rest by 1/3.
+/* Return true if OP in STMT will be constant after peeling LOOP. */
- NINSNS is the number of insns in the loop before unrolling.
- NUNROLL is the number of times the loop is unrolled. */
+static bool
+constant_after_peeling (tree op, gimple stmt, struct loop *loop)
+{
+ affine_iv iv;
+
+ if (is_gimple_min_invariant (op))
+ return true;
+
+ /* We can still fold accesses to constant arrays when index is known. */
+ if (TREE_CODE (op) != SSA_NAME)
+ {
+ tree base = op;
+
+ /* First make fast look if we see constant array inside. */
+ while (handled_component_p (base))
+ base = TREE_OPERAND (base, 0);
+ if ((DECL_P (base)
+ && TREE_STATIC (base)
+ && TREE_READONLY (base)
+ && (DECL_INITIAL (base)
+ || (!DECL_EXTERNAL (base)
+ && targetm.binds_local_p (base))))
+ || CONSTANT_CLASS_P (base))
+ {
+ /* If so, see if we understand all the indices. */
+ base = op;
+ while (handled_component_p (base))
+ {
+ if (TREE_CODE (base) == ARRAY_REF
+ && !constant_after_peeling (TREE_OPERAND (base, 1), stmt, loop))
+ return false;
+ base = TREE_OPERAND (base, 0);
+ }
+ return true;
+ }
+ return false;
+ }
+
+ /* Induction variables are constants. */
+ if (!simple_iv (loop, loop_containing_stmt (stmt), op, &iv, false))
+ return false;
+ if (!is_gimple_min_invariant (iv.base))
+ return false;
+ if (!is_gimple_min_invariant (iv.step))
+ return false;
+ return true;
+}
+
+/* Computes an estimated number of insns in LOOP, weighted by WEIGHTS.
+ Return results in SIZE, estimate benefits for complete unrolling exiting by EXIT. */
+
+static void
+tree_estimate_loop_size (struct loop *loop, edge exit, struct loop_size *size)
+{
+ basic_block *body = get_loop_body (loop);
+ gimple_stmt_iterator gsi;
+ unsigned int i;
+ bool after_exit;
+
+ size->overall = 0;
+ size->eliminated_by_peeling = 0;
+ size->last_iteration = 0;
+ size->last_iteration_eliminated_by_peeling = 0;
+
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file, "Estimating sizes for loop %i\n", loop->num);
+ for (i = 0; i < loop->num_nodes; i++)
+ {
+ if (exit && body[i] != exit->src
+ && dominated_by_p (CDI_DOMINATORS, body[i], exit->src))
+ after_exit = true;
+ else
+ after_exit = false;
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file, " BB: %i, after_exit: %i\n", body[i]->index, after_exit);
+
+ for (gsi = gsi_start_bb (body[i]); !gsi_end_p (gsi); gsi_next (&gsi))
+ {
+ gimple stmt = gsi_stmt (gsi);
+ int num = estimate_num_insns (stmt, &eni_size_weights);
+ bool likely_eliminated = false;
+
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ {
+ fprintf (dump_file, " size: %3i ", num);
+ print_gimple_stmt (dump_file, gsi_stmt (gsi), 0, 0);
+ }
+
+ /* Look for reasons why we might optimize this stmt away. */
+
+ /* Exit conditional. */
+ if (body[i] == exit->src && stmt == last_stmt (exit->src))
+ {
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file, " Exit condition will be eliminated.\n");
+ likely_eliminated = true;
+ }
+ /* Sets of IV variables */
+ else if (gimple_code (stmt) == GIMPLE_ASSIGN
+ && constant_after_peeling (gimple_assign_lhs (stmt), stmt, loop))
+ {
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file, " Induction variable computation will"
+ " be folded away.\n");
+ likely_eliminated = true;
+ }
+ /* Assignments of IV variables. */
+ else if (gimple_code (stmt) == GIMPLE_ASSIGN
+ && TREE_CODE (gimple_assign_lhs (stmt)) == SSA_NAME
+ && constant_after_peeling (gimple_assign_rhs1 (stmt), stmt,loop)
+ && (gimple_assign_rhs_class (stmt) != GIMPLE_BINARY_RHS
+ || constant_after_peeling (gimple_assign_rhs2 (stmt),
+ stmt, loop)))
+ {
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file, " Constant expression will be folded away.\n");
+ likely_eliminated = true;
+ }
+ /* Conditionals. */
+ else if (gimple_code (stmt) == GIMPLE_COND
+ && constant_after_peeling (gimple_cond_lhs (stmt), stmt, loop)
+ && constant_after_peeling (gimple_cond_rhs (stmt), stmt, loop))
+ {
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file, " Constant conditional.\n");
+ likely_eliminated = true;
+ }
+
+ size->overall += num;
+ if (likely_eliminated)
+ size->eliminated_by_peeling += num;
+ if (!after_exit)
+ {
+ size->last_iteration += num;
+ if (likely_eliminated)
+ size->last_iteration_eliminated_by_peeling += num;
+ }
+ }
+ }
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file, "size: %i-%i, last_iteration: %i-%i\n", size->overall,
+ size->eliminated_by_peeling, size->last_iteration,
+ size->last_iteration_eliminated_by_peeling);
+
+ free (body);
+}
+
+/* Estimate number of insns of completely unrolled loop.
+ It is (NUNROLL + 1) * size of loop body with taking into account
+ the fact that in last copy everything after exit conditional
+ is dead and that some instructions will be eliminated after
+ peeling.
+
+ Loop body is likely going to simplify futher, this is difficult
+ to guess, we just decrease the result by 1/3. */
static unsigned HOST_WIDE_INT
-estimated_unrolled_size (unsigned HOST_WIDE_INT ninsns,
+estimated_unrolled_size (struct loop_size *size,
unsigned HOST_WIDE_INT nunroll)
{
- HOST_WIDE_INT unr_insns = 2 * ((HOST_WIDE_INT) ninsns - 4) / 3;
+ HOST_WIDE_INT unr_insns = ((nunroll)
+ * (HOST_WIDE_INT) (size->overall
+ - size->eliminated_by_peeling));
+ if (!nunroll)
+ unr_insns = 0;
+ unr_insns += size->last_iteration - size->last_iteration_eliminated_by_peeling;
+
+ unr_insns = unr_insns * 2 / 3;
if (unr_insns <= 0)
unr_insns = 1;
- unr_insns *= (nunroll + 1);
return unr_insns;
}
enum unroll_level ul)
{
unsigned HOST_WIDE_INT n_unroll, ninsns, max_unroll, unr_insns;
- tree old_cond, cond, dont_exit, do_exit;
+ gimple cond;
+ struct loop_size size;
if (loop->inner)
return false;
if (ul == UL_SINGLE_ITER)
return false;
- ninsns = tree_num_loop_insns (loop);
+ tree_estimate_loop_size (loop, exit, &size);
+ ninsns = size.overall;
- if (n_unroll * ninsns
- > (unsigned) PARAM_VALUE (PARAM_MAX_COMPLETELY_PEELED_INSNS))
- return false;
+ unr_insns = estimated_unrolled_size (&size, n_unroll);
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ {
+ fprintf (dump_file, " Loop size: %d\n", (int) ninsns);
+ fprintf (dump_file, " Estimated size after unrolling: %d\n",
+ (int) unr_insns);
+ }
- if (ul == UL_NO_GROWTH)
+ if (unr_insns > ninsns
+ && (unr_insns
+ > (unsigned) PARAM_VALUE (PARAM_MAX_COMPLETELY_PEELED_INSNS)))
{
- unr_insns = estimated_unrolled_size (ninsns, n_unroll);
-
if (dump_file && (dump_flags & TDF_DETAILS))
- {
- fprintf (dump_file, " Loop size: %d\n", (int) ninsns);
- fprintf (dump_file, " Estimated size after unrolling: %d\n",
- (int) unr_insns);
- }
-
- if (unr_insns > ninsns)
- {
- if (dump_file && (dump_flags & TDF_DETAILS))
- fprintf (dump_file, "Not unrolling loop %d:\n", loop->num);
- return false;
- }
+ fprintf (dump_file, "Not unrolling loop %d "
+ "(--param max-completely-peeled-insns limit reached).\n",
+ loop->num);
+ return false;
}
- }
- if (exit->flags & EDGE_TRUE_VALUE)
- {
- dont_exit = boolean_false_node;
- do_exit = boolean_true_node;
- }
- else
- {
- dont_exit = boolean_true_node;
- do_exit = boolean_false_node;
+ if (ul == UL_NO_GROWTH
+ && unr_insns > ninsns)
+ {
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file, "Not unrolling loop %d.\n", loop->num);
+ return false;
+ }
}
- cond = last_stmt (exit->src);
-
+
if (n_unroll)
{
sbitmap wont_exit;
- edge *edges_to_remove = XNEWVEC (edge, n_unroll);
- unsigned int n_to_remove = 0;
+ edge e;
+ unsigned i;
+ VEC (edge, heap) *to_remove = NULL;
- old_cond = COND_EXPR_COND (cond);
- COND_EXPR_COND (cond) = dont_exit;
- update_stmt (cond);
initialize_original_copy_tables ();
-
wont_exit = sbitmap_alloc (n_unroll + 1);
sbitmap_ones (wont_exit);
RESET_BIT (wont_exit, 0);
- if (!tree_duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
- n_unroll, wont_exit,
- exit, edges_to_remove,
- &n_to_remove,
- DLTHE_FLAG_UPDATE_FREQ
- | DLTHE_FLAG_COMPLETTE_PEEL))
+ if (!gimple_duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
+ n_unroll, wont_exit,
+ exit, &to_remove,
+ DLTHE_FLAG_UPDATE_FREQ
+ | DLTHE_FLAG_COMPLETTE_PEEL))
{
- COND_EXPR_COND (cond) = old_cond;
- update_stmt (cond);
free_original_copy_tables ();
free (wont_exit);
- free (edges_to_remove);
return false;
}
+
+ for (i = 0; VEC_iterate (edge, to_remove, i, e); i++)
+ {
+ bool ok = remove_path (e);
+ gcc_assert (ok);
+ }
+
+ VEC_free (edge, heap, to_remove);
free (wont_exit);
- free (edges_to_remove);
free_original_copy_tables ();
}
-
- COND_EXPR_COND (cond) = do_exit;
- update_stmt (cond);
+ cond = last_stmt (exit->src);
+ if (exit->flags & EDGE_TRUE_VALUE)
+ gimple_cond_make_true (cond);
+ else
+ gimple_cond_make_false (cond);
+ update_stmt (cond);
update_ssa (TODO_update_ssa);
if (dump_file && (dump_flags & TDF_DETAILS))
size of the code does not increase. */
unsigned int
-tree_unroll_loops_completely (bool may_increase_size)
+tree_unroll_loops_completely (bool may_increase_size, bool unroll_outer)
{
loop_iterator li;
struct loop *loop;
- bool changed = false;
+ bool changed;
enum unroll_level ul;
- FOR_EACH_LOOP (li, loop, 0)
+ do
{
- if (may_increase_size && maybe_hot_bb_p (loop->header))
- ul = UL_ALL;
- else
- ul = UL_NO_GROWTH;
- changed |= canonicalize_loop_induction_variables (loop,
- false, ul,
- !flag_tree_loop_ivcanon);
- }
+ changed = false;
- /* Clean up the information about numbers of iterations, since complete
- unrolling might have invalidated it. */
- scev_reset ();
+ FOR_EACH_LOOP (li, loop, LI_ONLY_INNERMOST)
+ {
+ if (may_increase_size && optimize_loop_for_speed_p (loop)
+ /* Unroll outermost loops only if asked to do so or they do
+ not cause code growth. */
+ && (unroll_outer
+ || loop_outer (loop_outer (loop))))
+ ul = UL_ALL;
+ else
+ ul = UL_NO_GROWTH;
+ changed |= canonicalize_loop_induction_variables
+ (loop, false, ul, !flag_tree_loop_ivcanon);
+ }
+
+ if (changed)
+ {
+ /* This will take care of removing completely unrolled loops
+ from the loop structures so we can continue unrolling now
+ innermost loops. */
+ if (cleanup_tree_cfg ())
+ update_ssa (TODO_update_ssa_only_virtuals);
+
+ /* Clean up the information about numbers of iterations, since
+ complete unrolling might have invalidated it. */
+ scev_reset ();
+ }
+ }
+ while (changed);
- if (changed)
- return TODO_cleanup_cfg;
return 0;
}
empty_loop_p (struct loop *loop)
{
edge exit;
- struct tree_niter_desc niter;
- tree phi, def;
basic_block *body;
- block_stmt_iterator bsi;
+ gimple_stmt_iterator gsi;
unsigned i;
- tree stmt;
/* If the loop has multiple exits, it is too hard for us to handle.
Similarly, if the exit is not dominating, we cannot determine
return false;
/* The loop must be finite. */
- if (!number_of_iterations_exit (loop, exit, &niter, false))
+ if (!finite_loop_p (loop))
return false;
/* Values of all loop exit phi nodes must be invariants. */
- for (phi = phi_nodes (exit->dest); phi; phi = PHI_CHAIN (phi))
+ for (gsi = gsi_start(phi_nodes (exit->dest)); !gsi_end_p (gsi); gsi_next (&gsi))
{
+ gimple phi = gsi_stmt (gsi);
+ tree def;
+
if (!is_gimple_reg (PHI_RESULT (phi)))
continue;
return false;
}
- for (bsi = bsi_start (body[i]); !bsi_end_p (bsi); bsi_next (&bsi))
+ for (gsi = gsi_start_bb (body[i]); !gsi_end_p (gsi); gsi_next (&gsi))
{
- stmt = bsi_stmt (bsi);
- if (!ZERO_SSA_OPERANDS (stmt, SSA_OP_VIRTUAL_DEFS)
- || stmt_ann (stmt)->has_volatile_ops)
+ gimple stmt = gsi_stmt (gsi);
+
+ if (gimple_vdef (stmt)
+ || gimple_has_volatile_ops (stmt))
{
free (body);
return false;
/* Also, asm statements and calls may have side effects and we
cannot change the number of times they are executed. */
- switch (TREE_CODE (stmt))
+ switch (gimple_code (stmt))
{
- case RETURN_EXPR:
- case GIMPLE_MODIFY_STMT:
- stmt = get_call_expr_in (stmt);
- if (!stmt)
- break;
-
- case CALL_EXPR:
- if (TREE_SIDE_EFFECTS (stmt))
+ case GIMPLE_CALL:
+ if (gimple_has_side_effects (stmt))
{
free (body);
return false;
}
break;
- case ASM_EXPR:
+ case GIMPLE_ASM:
/* We cannot remove volatile assembler. */
- if (ASM_VOLATILE_P (stmt))
+ if (gimple_asm_volatile_p (stmt))
{
free (body);
return false;
remove_empty_loop (struct loop *loop)
{
edge exit = single_dom_exit (loop), non_exit;
- tree cond_stmt = last_stmt (exit->src);
- tree do_exit;
+ gimple cond_stmt = last_stmt (exit->src);
basic_block *body;
unsigned n_before, freq_in, freq_h;
gcov_type exit_count = exit->count;
+ if (dump_file)
+ fprintf (dump_file, "Removing empty loop %d\n", loop->num);
+
non_exit = EDGE_SUCC (exit->src, 0);
if (non_exit == exit)
non_exit = EDGE_SUCC (exit->src, 1);
if (exit->flags & EDGE_TRUE_VALUE)
- do_exit = boolean_true_node;
+ gimple_cond_make_true (cond_stmt);
else
- do_exit = boolean_false_node;
-
- COND_EXPR_COND (cond_stmt) = do_exit;
+ gimple_cond_make_false (cond_stmt);
update_stmt (cond_stmt);
/* Let us set the probabilities of the edges coming from the exit block. */
}
return 0;
}
+