static void
update_phis_for_loop_copy (struct loop *orig_loop, struct loop *new_loop)
{
tree new_ssa_name;
gimple_stmt_iterator si_new, si_orig;
edge orig_loop_latch = loop_latch_edge (orig_loop);
edge orig_entry_e = loop_preheader_edge (orig_loop);
edge new_loop_entry_e = loop_preheader_edge (new_loop);
/* Scan the phis in the headers of the old and new loops
(they are organized in exactly the same order). */
for (si_new = gsi_start_phis (new_loop->header),
si_orig = gsi_start_phis (orig_loop->header);
!gsi_end_p (si_new) && !gsi_end_p (si_orig);
gsi_next (&si_new), gsi_next (&si_orig))
{
tree def;
source_location locus;
gimple phi_new = gsi_stmt (si_new);
gimple phi_orig = gsi_stmt (si_orig);
/* Add the first phi argument for the phi in NEW_LOOP (the one
associated with the entry of NEW_LOOP) */
def = PHI_ARG_DEF_FROM_EDGE (phi_orig, orig_entry_e);
locus = gimple_phi_arg_location_from_edge (phi_orig, orig_entry_e);
add_phi_arg (phi_new, def, new_loop_entry_e, locus);
/* Add the second phi argument for the phi in NEW_LOOP (the one
associated with the latch of NEW_LOOP) */
def = PHI_ARG_DEF_FROM_EDGE (phi_orig, orig_loop_latch);
locus = gimple_phi_arg_location_from_edge (phi_orig, orig_loop_latch);
if (TREE_CODE (def) == SSA_NAME)
{
new_ssa_name = get_current_def (def);
if (!new_ssa_name)
/* This only happens if there are no definitions inside the
loop. Use the phi_result in this case. */
new_ssa_name = PHI_RESULT (phi_new);
}
else
/* Could be an integer. */
new_ssa_name = def;
add_phi_arg (phi_new, new_ssa_name, loop_latch_edge (new_loop), locus);
}
}
static struct loop *
copy_loop_before (struct loop *loop)
{
struct loop *res;
edge preheader = loop_preheader_edge (loop);
if (!single_exit (loop))
return NULL;
initialize_original_copy_tables ();
res = slpeel_tree_duplicate_loop_to_edge_cfg (loop, preheader);
free_original_copy_tables ();
if (!res)
return NULL;
update_phis_for_loop_copy (loop, res);
rename_variables_in_loop (res);
return res;
}
static bool
try_unroll_loop_completely (struct loop *loop,
edge exit, tree niter,
enum unroll_level ul,
HOST_WIDE_INT maxiter,
location_t locus)
{
unsigned HOST_WIDE_INT n_unroll = 0, ninsns, unr_insns;
struct loop_size size;
bool n_unroll_found = false;
edge edge_to_cancel = NULL;
int report_flags = MSG_OPTIMIZED_LOCATIONS | TDF_RTL | TDF_DETAILS;
/* See if we proved number of iterations to be low constant.
EXIT is an edge that will be removed in all but last iteration of
the loop.
EDGE_TO_CACNEL is an edge that will be removed from the last iteration
of the unrolled sequence and is expected to make the final loop not
rolling.
If the number of execution of loop is determined by standard induction
variable test, then EXIT and EDGE_TO_CANCEL are the two edges leaving
from the iv test. */
if (tree_fits_uhwi_p (niter))
{
n_unroll = tree_to_uhwi (niter);
n_unroll_found = true;
edge_to_cancel = EDGE_SUCC (exit->src, 0);
if (edge_to_cancel == exit)
edge_to_cancel = EDGE_SUCC (exit->src, 1);
}
/* We do not know the number of iterations and thus we can not eliminate
the EXIT edge. */
else
exit = NULL;
/* See if we can improve our estimate by using recorded loop bounds. */
if (maxiter >= 0
&& (!n_unroll_found || (unsigned HOST_WIDE_INT)maxiter < n_unroll))
{
n_unroll = maxiter;
n_unroll_found = true;
/* Loop terminates before the IV variable test, so we can not
remove it in the last iteration. */
edge_to_cancel = NULL;
}
if (!n_unroll_found)
return false;
if (n_unroll > (unsigned) PARAM_VALUE (PARAM_MAX_COMPLETELY_PEEL_TIMES))
{
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "Not unrolling loop %d "
"(--param max-completely-peeled-times limit reached).\n",
loop->num);
return false;
}
if (!edge_to_cancel)
edge_to_cancel = loop_edge_to_cancel (loop);
if (n_unroll)
{
sbitmap wont_exit;
edge e;
unsigned i;
bool large;
vec<edge> to_remove = vNULL;
if (ul == UL_SINGLE_ITER)
return false;
large = tree_estimate_loop_size
(loop, exit, edge_to_cancel, &size,
PARAM_VALUE (PARAM_MAX_COMPLETELY_PEELED_INSNS));
ninsns = size.overall;
if (large)
{
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "Not unrolling loop %d: it is too large.\n",
loop->num);
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 the code is going to shrink, we don't need to be extra cautious
on guessing if the unrolling is going to be profitable. */
if (unr_insns
/* If there is IV variable that will become constant, we save
one instruction in the loop prologue we do not account
otherwise. */
<= ninsns + (size.constant_iv != false))
;
/* We unroll only inner loops, because we do not consider it profitable
otheriwse. We still can cancel loopback edge of not rolling loop;
this is always a good idea. */
else if (ul == UL_NO_GROWTH)
{
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "Not unrolling loop %d: size would grow.\n",
loop->num);
return false;
}
/* Outer loops tend to be less interesting candidates for complete
unrolling unless we can do a lot of propagation into the inner loop
body. For now we disable outer loop unrolling when the code would
grow. */
else if (loop->inner)
{
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "Not unrolling loop %d: "
"it is not innermost and code would grow.\n",
loop->num);
return false;
}
/* If there is call on a hot path through the loop, then
there is most probably not much to optimize. */
else if (size.num_non_pure_calls_on_hot_path)
{
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "Not unrolling loop %d: "
"contains call and code would grow.\n",
loop->num);
return false;
}
/* If there is pure/const call in the function, then we
can still optimize the unrolled loop body if it contains
some other interesting code than the calls and code
storing or cumulating the return value. */
else if (size.num_pure_calls_on_hot_path
/* One IV increment, one test, one ivtmp store
and one useful stmt. That is about minimal loop
doing pure call. */
&& (size.non_call_stmts_on_hot_path
<= 3 + size.num_pure_calls_on_hot_path))
{
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "Not unrolling loop %d: "
"contains just pure calls and code would grow.\n",
loop->num);
return false;
}
/* Complette unrolling is major win when control flow is removed and
one big basic block is created. If the loop contains control flow
the optimization may still be a win because of eliminating the loop
overhead but it also may blow the branch predictor tables.
Limit number of branches on the hot path through the peeled
sequence. */
else if (size.num_branches_on_hot_path * (int)n_unroll
> PARAM_VALUE (PARAM_MAX_PEEL_BRANCHES))
{
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "Not unrolling loop %d: "
" number of branches on hot path in the unrolled sequence"
" reach --param max-peel-branches limit.\n",
loop->num);
return false;
}
else if (unr_insns
> (unsigned) PARAM_VALUE (PARAM_MAX_COMPLETELY_PEELED_INSNS))
{
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "Not unrolling loop %d: "
"(--param max-completely-peeled-insns limit reached).\n",
loop->num);
return false;
}
dump_printf_loc (report_flags, locus,
"loop turned into non-loop; it never loops.\n");
initialize_original_copy_tables ();
wont_exit = sbitmap_alloc (n_unroll + 1);
bitmap_ones (wont_exit);
bitmap_clear_bit (wont_exit, 0);
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))
{
free_original_copy_tables ();
free (wont_exit);
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "Failed to duplicate the loop\n");
return false;
}
FOR_EACH_VEC_ELT (to_remove, i, e)
{
bool ok = remove_path (e);
gcc_assert (ok);
}
to_remove.release ();
free (wont_exit);
free_original_copy_tables ();
}
static bool
generate_builtin (struct loop *loop, bitmap partition, bool copy_p)
{
bool res = false;
unsigned i, x = 0;
basic_block *bbs;
gimple write = NULL;
tree op0, op1;
gimple_stmt_iterator bsi;
tree nb_iter = number_of_exit_cond_executions (loop);
if (!nb_iter || nb_iter == chrec_dont_know)
return false;
bbs = get_loop_body_in_dom_order (loop);
for (i = 0; i < loop->num_nodes; i++)
{
basic_block bb = bbs[i];
for (bsi = gsi_start_phis (bb); !gsi_end_p (bsi); gsi_next (&bsi))
x++;
for (bsi = gsi_start_bb (bb); !gsi_end_p (bsi); gsi_next (&bsi))
{
gimple stmt = gsi_stmt (bsi);
if (bitmap_bit_p (partition, x++)
&& is_gimple_assign (stmt)
&& !is_gimple_reg (gimple_assign_lhs (stmt)))
{
/* Don't generate the builtins when there are more than
one memory write. */
if (write != NULL)
goto end;
write = stmt;
}
}
}
if (!write)
goto end;
op0 = gimple_assign_lhs (write);
op1 = gimple_assign_rhs1 (write);
if (!(TREE_CODE (op0) == ARRAY_REF
|| TREE_CODE (op0) == INDIRECT_REF))
goto end;
/* The new statements will be placed before LOOP. */
bsi = gsi_last_bb (loop_preheader_edge (loop)->src);
if (gimple_assign_rhs_code (write) == INTEGER_CST
&& (integer_zerop (op1) || real_zerop (op1)))
res = generate_memset_zero (write, op0, nb_iter, bsi);
/* If this is the last partition for which we generate code, we have
to destroy the loop. */
if (res && !copy_p)
{
unsigned nbbs = loop->num_nodes;
basic_block src = loop_preheader_edge (loop)->src;
basic_block dest = single_exit (loop)->dest;
prop_phis (dest);
make_edge (src, dest, EDGE_FALLTHRU);
cancel_loop_tree (loop);
for (i = 0; i < nbbs; i++)
delete_basic_block (bbs[i]);
set_immediate_dominator (CDI_DOMINATORS, dest,
recompute_dominator (CDI_DOMINATORS, dest));
}
end:
free (bbs);
return res;
}
static struct loop *
unswitch_loop (struct loop *loop, basic_block unswitch_on, rtx cond, rtx cinsn)
{
edge entry, latch_edge, true_edge, false_edge, e;
basic_block switch_bb, unswitch_on_alt;
struct loop *nloop;
int irred_flag, prob;
rtx seq;
/* Some sanity checking. */
gcc_assert (flow_bb_inside_loop_p (loop, unswitch_on));
gcc_assert (EDGE_COUNT (unswitch_on->succs) == 2);
gcc_assert (just_once_each_iteration_p (loop, unswitch_on));
gcc_assert (!loop->inner);
gcc_assert (flow_bb_inside_loop_p (loop, EDGE_SUCC (unswitch_on, 0)->dest));
gcc_assert (flow_bb_inside_loop_p (loop, EDGE_SUCC (unswitch_on, 1)->dest));
entry = loop_preheader_edge (loop);
/* Make a copy. */
irred_flag = entry->flags & EDGE_IRREDUCIBLE_LOOP;
entry->flags &= ~EDGE_IRREDUCIBLE_LOOP;
if (!duplicate_loop_to_header_edge (loop, entry, 1,
NULL, NULL, NULL, 0))
return NULL;
entry->flags |= irred_flag;
/* Record the block with condition we unswitch on. */
unswitch_on_alt = get_bb_copy (unswitch_on);
true_edge = BRANCH_EDGE (unswitch_on_alt);
false_edge = FALLTHRU_EDGE (unswitch_on);
latch_edge = single_succ_edge (get_bb_copy (loop->latch));
/* Create a block with the condition. */
prob = true_edge->probability;
switch_bb = create_empty_bb (EXIT_BLOCK_PTR->prev_bb);
seq = compare_and_jump_seq (XEXP (cond, 0), XEXP (cond, 1), GET_CODE (cond),
block_label (true_edge->dest),
prob, cinsn);
emit_insn_after (seq, BB_END (switch_bb));
e = make_edge (switch_bb, true_edge->dest, 0);
e->probability = prob;
e->count = latch_edge->count * prob / REG_BR_PROB_BASE;
e = make_edge (switch_bb, FALLTHRU_EDGE (unswitch_on)->dest, EDGE_FALLTHRU);
e->probability = false_edge->probability;
e->count = latch_edge->count * (false_edge->probability) / REG_BR_PROB_BASE;
if (irred_flag)
{
switch_bb->flags |= BB_IRREDUCIBLE_LOOP;
EDGE_SUCC (switch_bb, 0)->flags |= EDGE_IRREDUCIBLE_LOOP;
EDGE_SUCC (switch_bb, 1)->flags |= EDGE_IRREDUCIBLE_LOOP;
}
else
{
switch_bb->flags &= ~BB_IRREDUCIBLE_LOOP;
EDGE_SUCC (switch_bb, 0)->flags &= ~EDGE_IRREDUCIBLE_LOOP;
EDGE_SUCC (switch_bb, 1)->flags &= ~EDGE_IRREDUCIBLE_LOOP;
}
/* Loopify from the copy of LOOP body, constructing the new loop. */
nloop = loopify (latch_edge,
single_pred_edge (get_bb_copy (loop->header)), switch_bb,
BRANCH_EDGE (switch_bb), FALLTHRU_EDGE (switch_bb), true,
prob, REG_BR_PROB_BASE - prob);
copy_loop_info (loop, nloop);
/* Remove branches that are now unreachable in new loops. */
remove_path (true_edge);
remove_path (false_edge);
/* Preserve the simple loop preheaders. */
split_edge (loop_preheader_edge (loop));
split_edge (loop_preheader_edge (nloop));
return nloop;
}
static void
copy_loop_headers (void)
{
struct loops *loops;
unsigned i;
struct loop *loop;
basic_block header;
edge exit;
basic_block *bbs;
unsigned n_bbs;
loops = loop_optimizer_init (dump_file);
if (!loops)
return;
rewrite_into_loop_closed_ssa ();
/* We do not try to keep the information about irreducible regions
up-to-date. */
loops->state &= ~LOOPS_HAVE_MARKED_IRREDUCIBLE_REGIONS;
#ifdef ENABLE_CHECKING
verify_loop_structure (loops);
#endif
bbs = xmalloc (sizeof (basic_block) * n_basic_blocks);
for (i = 1; i < loops->num; i++)
{
/* Copy at most 20 insns. */
int limit = 20;
loop = loops->parray[i];
if (!loop)
continue;
header = loop->header;
/* If the loop is already a do-while style one (either because it was
written as such, or because jump threading transformed it into one),
we might be in fact peeling the first iteration of the loop. This
in general is not a good idea. */
if (do_while_loop_p (loop))
continue;
/* Iterate the header copying up to limit; this takes care of the cases
like while (a && b) {...}, where we want to have both of the conditions
copied. TODO -- handle while (a || b) - like cases, by not requiring
the header to have just a single successor and copying up to
postdominator. */
exit = NULL;
n_bbs = 0;
while (should_duplicate_loop_header_p (header, loop, &limit))
{
/* Find a successor of header that is inside a loop; i.e. the new
header after the condition is copied. */
if (flow_bb_inside_loop_p (loop, EDGE_SUCC (header, 0)->dest))
exit = EDGE_SUCC (header, 0);
else
exit = EDGE_SUCC (header, 1);
bbs[n_bbs++] = header;
header = exit->dest;
}
if (!exit)
continue;
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file,
"Duplicating header of the loop %d up to edge %d->%d.\n",
loop->num, exit->src->index, exit->dest->index);
/* Ensure that the header will have just the latch as a predecessor
inside the loop. */
if (EDGE_COUNT (exit->dest->preds) > 1)
exit = EDGE_SUCC (loop_split_edge_with (exit, NULL), 0);
if (!tree_duplicate_sese_region (loop_preheader_edge (loop), exit,
bbs, n_bbs, NULL))
{
fprintf (dump_file, "Duplication failed.\n");
continue;
}
/* Ensure that the latch and the preheader is simple (we know that they
are not now, since there was the loop exit condition. */
loop_split_edge_with (loop_preheader_edge (loop), NULL);
loop_split_edge_with (loop_latch_edge (loop), NULL);
}
free (bbs);
#ifdef ENABLE_CHECKING
verify_loop_closed_ssa ();
#endif
loop_optimizer_finalize (loops, NULL);
}
static void
doloop_modify (struct loop *loop, struct niter_desc *desc,
rtx doloop_seq, rtx condition, rtx count)
{
rtx counter_reg;
rtx tmp, noloop = NULL_RTX;
rtx sequence;
rtx jump_insn;
rtx jump_label;
int nonneg = 0, irr;
bool increment_count;
basic_block loop_end = desc->out_edge->src;
enum machine_mode mode;
jump_insn = BB_END (loop_end);
if (dump_file)
{
fprintf (dump_file, "Doloop: Inserting doloop pattern (");
if (desc->const_iter)
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC, desc->niter);
else
fputs ("runtime", dump_file);
fputs (" iterations).\n", dump_file);
}
/* Discard original jump to continue loop. The original compare
result may still be live, so it cannot be discarded explicitly. */
delete_insn (jump_insn);
counter_reg = XEXP (condition, 0);
if (GET_CODE (counter_reg) == PLUS)
counter_reg = XEXP (counter_reg, 0);
mode = GET_MODE (counter_reg);
increment_count = false;
switch (GET_CODE (condition))
{
case NE:
/* Currently only NE tests against zero and one are supported. */
if (XEXP (condition, 1) == const1_rtx)
{
increment_count = true;
noloop = const1_rtx;
}
else if (XEXP (condition, 1) == const0_rtx)
noloop = const0_rtx;
else
abort ();
break;
case GE:
/* Currently only GE tests against zero are supported. */
if (XEXP (condition, 1) != const0_rtx)
abort ();
noloop = constm1_rtx;
/* The iteration count does not need incrementing for a GE test. */
increment_count = false;
/* Determine if the iteration counter will be non-negative.
Note that the maximum value loaded is iterations_max - 1. */
if (desc->niter_max
<= ((unsigned HOST_WIDEST_INT) 1
<< (GET_MODE_BITSIZE (mode) - 1)))
nonneg = 1;
break;
/* Abort if an invalid doloop pattern has been generated. */
default:
abort ();
}
if (increment_count)
count = simplify_gen_binary (PLUS, mode, count, const1_rtx);
/* Insert initialization of the count register into the loop header. */
start_sequence ();
tmp = force_operand (count, counter_reg);
convert_move (counter_reg, tmp, 1);
sequence = get_insns ();
end_sequence ();
emit_insn_after (sequence, BB_END (loop_preheader_edge (loop)->src));
if (desc->noloop_assumptions)
{
rtx ass = copy_rtx (desc->noloop_assumptions);
basic_block preheader = loop_preheader_edge (loop)->src;
basic_block set_zero
= loop_split_edge_with (loop_preheader_edge (loop), NULL_RTX);
basic_block new_preheader
= loop_split_edge_with (loop_preheader_edge (loop), NULL_RTX);
basic_block bb;
edge te;
gcov_type cnt;
/* Expand the condition testing the assumptions and if it does not pass,
reset the count register to 0. */
add_test (XEXP (ass, 0), preheader, set_zero);
EDGE_SUCC (preheader, 0)->flags &= ~EDGE_FALLTHRU;
cnt = EDGE_SUCC (preheader, 0)->count;
EDGE_SUCC (preheader, 0)->probability = 0;
EDGE_SUCC (preheader, 0)->count = 0;
irr = EDGE_SUCC (preheader, 0)->flags & EDGE_IRREDUCIBLE_LOOP;
te = make_edge (preheader, new_preheader, EDGE_FALLTHRU | irr);
te->probability = REG_BR_PROB_BASE;
te->count = cnt;
set_immediate_dominator (CDI_DOMINATORS, new_preheader, preheader);
set_zero->count = 0;
set_zero->frequency = 0;
for (ass = XEXP (ass, 1); ass; ass = XEXP (ass, 1))
{
bb = loop_split_edge_with (te, NULL_RTX);
te = EDGE_SUCC (bb, 0);
add_test (XEXP (ass, 0), bb, set_zero);
make_edge (bb, set_zero, irr);
}
start_sequence ();
convert_move (counter_reg, noloop, 0);
sequence = get_insns ();
end_sequence ();
emit_insn_after (sequence, BB_END (set_zero));
}
/* Some targets (eg, C4x) need to initialize special looping
registers. */
#ifdef HAVE_doloop_begin
{
rtx init;
unsigned level = get_loop_level (loop) + 1;
init = gen_doloop_begin (counter_reg,
desc->const_iter ? desc->niter_expr : const0_rtx,
desc->niter_max,
GEN_INT (level));
if (init)
{
start_sequence ();
emit_insn (init);
sequence = get_insns ();
end_sequence ();
emit_insn_after (sequence, BB_END (loop_preheader_edge (loop)->src));
}
}
#endif
/* Insert the new low-overhead looping insn. */
emit_jump_insn_after (doloop_seq, BB_END (loop_end));
jump_insn = BB_END (loop_end);
jump_label = block_label (desc->in_edge->dest);
JUMP_LABEL (jump_insn) = jump_label;
LABEL_NUSES (jump_label)++;
/* Ensure the right fallthru edge is marked, for case we have reversed
the condition. */
desc->in_edge->flags &= ~EDGE_FALLTHRU;
desc->out_edge->flags |= EDGE_FALLTHRU;
/* Add a REG_NONNEG note if the actual or estimated maximum number
of iterations is non-negative. */
if (nonneg)
{
REG_NOTES (jump_insn)
= gen_rtx_EXPR_LIST (REG_NONNEG, NULL_RTX, REG_NOTES (jump_insn));
}
}
static unsigned int
copy_loop_headers (void)
{
struct loops *loops;
unsigned i;
struct loop *loop;
basic_block header;
edge exit, entry;
basic_block *bbs, *copied_bbs;
unsigned n_bbs;
unsigned bbs_size;
loops = loop_optimizer_init (LOOPS_HAVE_PREHEADERS
| LOOPS_HAVE_SIMPLE_LATCHES);
if (!loops)
return 0;
#ifdef ENABLE_CHECKING
verify_loop_structure (loops);
#endif
bbs = XNEWVEC (basic_block, n_basic_blocks);
copied_bbs = XNEWVEC (basic_block, n_basic_blocks);
bbs_size = n_basic_blocks;
for (i = 1; i < loops->num; i++)
{
/* Copy at most 20 insns. */
int limit = 20;
loop = loops->parray[i];
if (!loop)
continue;
header = loop->header;
/* If the loop is already a do-while style one (either because it was
written as such, or because jump threading transformed it into one),
we might be in fact peeling the first iteration of the loop. This
in general is not a good idea. */
if (do_while_loop_p (loop))
continue;
/* Iterate the header copying up to limit; this takes care of the cases
like while (a && b) {...}, where we want to have both of the conditions
copied. TODO -- handle while (a || b) - like cases, by not requiring
the header to have just a single successor and copying up to
postdominator. */
exit = NULL;
n_bbs = 0;
while (should_duplicate_loop_header_p (header, loop, &limit))
{
/* Find a successor of header that is inside a loop; i.e. the new
header after the condition is copied. */
if (flow_bb_inside_loop_p (loop, EDGE_SUCC (header, 0)->dest))
exit = EDGE_SUCC (header, 0);
else
exit = EDGE_SUCC (header, 1);
bbs[n_bbs++] = header;
gcc_assert (bbs_size > n_bbs);
header = exit->dest;
}
if (!exit)
continue;
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file,
"Duplicating header of the loop %d up to edge %d->%d.\n",
loop->num, exit->src->index, exit->dest->index);
/* Ensure that the header will have just the latch as a predecessor
inside the loop. */
if (!single_pred_p (exit->dest))
exit = single_pred_edge (loop_split_edge_with (exit, NULL));
entry = loop_preheader_edge (loop);
if (!tree_duplicate_sese_region (entry, exit, bbs, n_bbs, copied_bbs))
{
fprintf (dump_file, "Duplication failed.\n");
continue;
}
/* If the loop has the form "for (i = j; i < j + 10; i++)" then
this copying can introduce a case where we rely on undefined
signed overflow to eliminate the preheader condition, because
we assume that "j < j + 10" is true. We don't want to warn
about that case for -Wstrict-overflow, because in general we
don't warn about overflow involving loops. Prevent the
warning by setting TREE_NO_WARNING. */
if (warn_strict_overflow > 0)
{
unsigned int i;
for (i = 0; i < n_bbs; ++i)
{
tree last;
last = last_stmt (copied_bbs[i]);
if (TREE_CODE (last) == COND_EXPR)
TREE_NO_WARNING (last) = 1;
}
}
/* Ensure that the latch and the preheader is simple (we know that they
are not now, since there was the loop exit condition. */
loop_split_edge_with (loop_preheader_edge (loop), NULL);
loop_split_edge_with (loop_latch_edge (loop), NULL);
}
free (bbs);
free (copied_bbs);
loop_optimizer_finalize (loops);
return 0;
}
static bool
generate_builtin (struct loop *loop, bitmap partition, bool copy_p)
{
bool res = false;
unsigned i, x = 0;
basic_block *bbs;
gimple write = NULL;
gimple_stmt_iterator bsi;
tree nb_iter = number_of_exit_cond_executions (loop);
if (!nb_iter || nb_iter == chrec_dont_know)
return false;
bbs = get_loop_body_in_dom_order (loop);
for (i = 0; i < loop->num_nodes; i++)
{
basic_block bb = bbs[i];
for (bsi = gsi_start_phis (bb); !gsi_end_p (bsi); gsi_next (&bsi))
x++;
for (bsi = gsi_start_bb (bb); !gsi_end_p (bsi); gsi_next (&bsi))
{
gimple stmt = gsi_stmt (bsi);
if (gimple_code (stmt) == GIMPLE_LABEL
|| is_gimple_debug (stmt))
continue;
if (!bitmap_bit_p (partition, x++))
continue;
/* If the stmt has uses outside of the loop fail. */
if (stmt_has_scalar_dependences_outside_loop (stmt))
goto end;
if (is_gimple_assign (stmt)
&& !is_gimple_reg (gimple_assign_lhs (stmt)))
{
/* Don't generate the builtins when there are more than
one memory write. */
if (write != NULL)
goto end;
write = stmt;
if (bb == loop->latch)
nb_iter = number_of_latch_executions (loop);
}
}
}
if (!stmt_with_adjacent_zero_store_dr_p (write))
goto end;
/* The new statements will be placed before LOOP. */
bsi = gsi_last_bb (loop_preheader_edge (loop)->src);
generate_memset_zero (write, gimple_assign_lhs (write), nb_iter, bsi);
res = true;
/* If this is the last partition for which we generate code, we have
to destroy the loop. */
if (!copy_p)
{
unsigned nbbs = loop->num_nodes;
edge exit = single_exit (loop);
basic_block src = loop_preheader_edge (loop)->src, dest = exit->dest;
redirect_edge_pred (exit, src);
exit->flags &= ~(EDGE_TRUE_VALUE|EDGE_FALSE_VALUE);
exit->flags |= EDGE_FALLTHRU;
cancel_loop_tree (loop);
rescan_loop_exit (exit, false, true);
for (i = 0; i < nbbs; i++)
delete_basic_block (bbs[i]);
set_immediate_dominator (CDI_DOMINATORS, dest,
recompute_dominator (CDI_DOMINATORS, dest));
}
end:
free (bbs);
return res;
}
