sbitmap
sbitmap_alloc_with_popcount (unsigned int n_elms)
{
sbitmap const bmap = sbitmap_alloc (n_elms);
bmap->popcount = XNEWVEC (unsigned char, bmap->size);
return bmap;
}
/* Splits superblocks. */
static void
break_superblocks (void)
{
sbitmap superblocks;
int i, need;
superblocks = sbitmap_alloc (n_basic_blocks);
sbitmap_zero (superblocks);
need = 0;
for (i = 0; i < n_basic_blocks; i++)
if (BASIC_BLOCK(i)->flags & BB_SUPERBLOCK)
{
BASIC_BLOCK(i)->flags &= ~BB_SUPERBLOCK;
SET_BIT (superblocks, i);
need = 1;
}
if (need)
{
rebuild_jump_labels (get_insns ());
find_many_sub_basic_blocks (superblocks);
}
free (superblocks);
}
/* Splits superblocks. */
void
break_superblocks (void)
{
sbitmap superblocks;
bool need = false;
basic_block bb;
superblocks = sbitmap_alloc (last_basic_block);
sbitmap_zero (superblocks);
FOR_EACH_BB (bb)
if (bb->flags & BB_SUPERBLOCK)
{
bb->flags &= ~BB_SUPERBLOCK;
SET_BIT (superblocks, bb->index);
need = true;
}
if (need)
{
rebuild_jump_labels (get_insns ());
find_many_sub_basic_blocks (superblocks);
}
free (superblocks);
}
static bool
reachable_at_most_once (basic_block va_arg_bb, basic_block va_start_bb)
{
vec<edge> stack = vNULL;
edge e;
edge_iterator ei;
sbitmap visited;
bool ret;
if (va_arg_bb == va_start_bb)
return true;
if (! dominated_by_p (CDI_DOMINATORS, va_arg_bb, va_start_bb))
return false;
visited = sbitmap_alloc (last_basic_block_for_fn (cfun));
bitmap_clear (visited);
ret = true;
FOR_EACH_EDGE (e, ei, va_arg_bb->preds)
stack.safe_push (e);
while (! stack.is_empty ())
{
basic_block src;
e = stack.pop ();
src = e->src;
if (e->flags & EDGE_COMPLEX)
{
ret = false;
break;
}
if (src == va_start_bb)
continue;
/* va_arg_bb can be executed more times than va_start_bb. */
if (src == va_arg_bb)
{
ret = false;
break;
}
gcc_assert (src != ENTRY_BLOCK_PTR_FOR_FN (cfun));
if (! bitmap_bit_p (visited, src->index))
{
bitmap_set_bit (visited, src->index);
FOR_EACH_EDGE (e, ei, src->preds)
stack.safe_push (e);
}
}
stack.release ();
sbitmap_free (visited);
return ret;
}
static bool
reachable_at_most_once (basic_block va_arg_bb, basic_block va_start_bb)
{
VEC (edge, heap) *stack = NULL;
edge e;
edge_iterator ei;
sbitmap visited;
bool ret;
if (va_arg_bb == va_start_bb)
return true;
if (! dominated_by_p (CDI_DOMINATORS, va_arg_bb, va_start_bb))
return false;
visited = sbitmap_alloc (last_basic_block);
sbitmap_zero (visited);
ret = true;
FOR_EACH_EDGE (e, ei, va_arg_bb->preds)
VEC_safe_push (edge, heap, stack, e);
while (! VEC_empty (edge, stack))
{
basic_block src;
e = VEC_pop (edge, stack);
src = e->src;
if (e->flags & EDGE_COMPLEX)
{
ret = false;
break;
}
if (src == va_start_bb)
continue;
/* va_arg_bb can be executed more times than va_start_bb. */
if (src == va_arg_bb)
{
ret = false;
break;
}
gcc_assert (src != ENTRY_BLOCK_PTR);
if (! TEST_BIT (visited, src->index))
{
SET_BIT (visited, src->index);
FOR_EACH_EDGE (e, ei, src->preds)
VEC_safe_push (edge, heap, stack, e);
}
}
VEC_free (edge, heap, stack);
sbitmap_free (visited);
return ret;
}
static unsigned int
copyprop_hardreg_forward (void)
{
struct value_data *all_vd;
basic_block bb;
sbitmap visited;
bool analyze_called = false;
all_vd = XNEWVEC (struct value_data, last_basic_block_for_fn (cfun));
visited = sbitmap_alloc (last_basic_block_for_fn (cfun));
bitmap_clear (visited);
if (MAY_HAVE_DEBUG_INSNS)
debug_insn_changes_pool
= create_alloc_pool ("debug insn changes pool",
sizeof (struct queued_debug_insn_change), 256);
FOR_EACH_BB_FN (bb, cfun)
{
bitmap_set_bit (visited, bb->index);
/* If a block has a single predecessor, that we've already
processed, begin with the value data that was live at
the end of the predecessor block. */
/* ??? Ought to use more intelligent queuing of blocks. */
if (single_pred_p (bb)
&& bitmap_bit_p (visited, single_pred (bb)->index)
&& ! (single_pred_edge (bb)->flags & (EDGE_ABNORMAL_CALL | EDGE_EH)))
{
all_vd[bb->index] = all_vd[single_pred (bb)->index];
if (all_vd[bb->index].n_debug_insn_changes)
{
unsigned int regno;
for (regno = 0; regno < FIRST_PSEUDO_REGISTER; regno++)
{
if (all_vd[bb->index].e[regno].debug_insn_changes)
{
all_vd[bb->index].e[regno].debug_insn_changes = NULL;
if (--all_vd[bb->index].n_debug_insn_changes == 0)
break;
}
}
}
}
else
init_value_data (all_vd + bb->index);
copyprop_hardreg_forward_1 (bb, all_vd + bb->index);
}
static void
compute_earliest (struct edge_list *edge_list, int n_exprs, sbitmap *antin,
sbitmap *antout, sbitmap *avout, sbitmap *kill,
sbitmap *earliest)
{
sbitmap difference, temp_bitmap;
int x, num_edges;
basic_block pred, succ;
num_edges = NUM_EDGES (edge_list);
difference = sbitmap_alloc (n_exprs);
temp_bitmap = sbitmap_alloc (n_exprs);
for (x = 0; x < num_edges; x++)
{
pred = INDEX_EDGE_PRED_BB (edge_list, x);
succ = INDEX_EDGE_SUCC_BB (edge_list, x);
if (pred == ENTRY_BLOCK_PTR_FOR_FN (cfun))
bitmap_copy (earliest[x], antin[succ->index]);
else
{
if (succ == EXIT_BLOCK_PTR_FOR_FN (cfun))
bitmap_clear (earliest[x]);
else
{
bitmap_and_compl (difference, antin[succ->index],
avout[pred->index]);
bitmap_not (temp_bitmap, antout[pred->index]);
bitmap_and_or (earliest[x], difference,
kill[pred->index], temp_bitmap);
}
}
}
sbitmap_free (temp_bitmap);
sbitmap_free (difference);
}
static void
compute_farthest (struct edge_list *edge_list, int n_exprs,
sbitmap *st_avout, sbitmap *st_avin, sbitmap *st_antin,
sbitmap *kill, sbitmap *farthest)
{
sbitmap difference, temp_bitmap;
int x, num_edges;
basic_block pred, succ;
num_edges = NUM_EDGES (edge_list);
difference = sbitmap_alloc (n_exprs);
temp_bitmap = sbitmap_alloc (n_exprs);
for (x = 0; x < num_edges; x++)
{
pred = INDEX_EDGE_PRED_BB (edge_list, x);
succ = INDEX_EDGE_SUCC_BB (edge_list, x);
if (succ == EXIT_BLOCK_PTR)
sbitmap_copy (farthest[x], st_avout[pred->index]);
else
{
if (pred == ENTRY_BLOCK_PTR)
sbitmap_zero (farthest[x]);
else
{
sbitmap_difference (difference, st_avout[pred->index],
st_antin[succ->index]);
sbitmap_not (temp_bitmap, st_avin[succ->index]);
sbitmap_a_and_b_or_c (farthest[x], difference,
kill[succ->index], temp_bitmap);
}
}
}
sbitmap_free (temp_bitmap);
sbitmap_free (difference);
}
basic_block *
blocks_in_phiopt_order (void)
{
basic_block x, y;
basic_block *order = XNEWVEC (basic_block, n_basic_blocks);
unsigned n = n_basic_blocks - NUM_FIXED_BLOCKS;
unsigned np, i;
sbitmap visited = sbitmap_alloc (last_basic_block);
#define MARK_VISITED(BB) (SET_BIT (visited, (BB)->index))
#define VISITED_P(BB) (TEST_BIT (visited, (BB)->index))
sbitmap_zero (visited);
MARK_VISITED (ENTRY_BLOCK_PTR);
FOR_EACH_BB (x)
{
if (VISITED_P (x))
continue;
/* Walk the predecessors of x as long as they have precisely one
predecessor and add them to the list, so that they get stored
after x. */
for (y = x, np = 1;
single_pred_p (y) && !VISITED_P (single_pred (y));
y = single_pred (y))
np++;
for (y = x, i = n - np;
single_pred_p (y) && !VISITED_P (single_pred (y));
y = single_pred (y), i++)
{
order[i] = y;
MARK_VISITED (y);
}
order[i] = y;
MARK_VISITED (y);
gcc_assert (i == n - 1);
n -= np;
}
sbitmap_free (visited);
gcc_assert (n == 0);
return order;
#undef MARK_VISITED
#undef VISITED_P
}
static elim_graph
new_elim_graph (int size)
{
elim_graph g = (elim_graph) xmalloc (sizeof (struct _elim_graph));
g->nodes = VEC_alloc (int, heap, 30);
g->const_dests = VEC_alloc (int, heap, 20);
g->const_copies = VEC_alloc (tree, heap, 20);
g->copy_locus = VEC_alloc (source_location, heap, 10);
g->edge_list = VEC_alloc (int, heap, 20);
g->edge_locus = VEC_alloc (source_location, heap, 10);
g->stack = VEC_alloc (int, heap, 30);
g->visited = sbitmap_alloc (size);
return g;
}
static void
dump_copy_of (FILE *file, tree var)
{
tree val;
sbitmap visited;
print_generic_expr (file, var, dump_flags);
if (TREE_CODE (var) != SSA_NAME)
return;
visited = sbitmap_alloc (num_ssa_names);
sbitmap_zero (visited);
SET_BIT (visited, SSA_NAME_VERSION (var));
fprintf (file, " copy-of chain: ");
val = var;
print_generic_expr (file, val, 0);
fprintf (file, " ");
while (copy_of[SSA_NAME_VERSION (val)].value)
{
fprintf (file, "-> ");
val = copy_of[SSA_NAME_VERSION (val)].value;
print_generic_expr (file, val, 0);
fprintf (file, " ");
if (TEST_BIT (visited, SSA_NAME_VERSION (val)))
break;
SET_BIT (visited, SSA_NAME_VERSION (val));
}
val = get_copy_of_val (var)->value;
if (val == NULL_TREE)
fprintf (file, "[UNDEFINED]");
else if (val != var)
fprintf (file, "[COPY]");
else
fprintf (file, "[NOT A COPY]");
sbitmap_free (visited);
}
static bool
tail_duplicate (void)
{
auto_vec<fibonacci_node<long, basic_block_def>*> blocks;
blocks.safe_grow_cleared (last_basic_block_for_fn (cfun));
basic_block *trace = XNEWVEC (basic_block, n_basic_blocks_for_fn (cfun));
int *counts = XNEWVEC (int, last_basic_block_for_fn (cfun));
int ninsns = 0, nduplicated = 0;
gcov_type weighted_insns = 0, traced_insns = 0;
fibonacci_heap<long, basic_block_def> heap (LONG_MIN);
gcov_type cover_insns;
int max_dup_insns;
basic_block bb;
bool changed = false;
/* Create an oversized sbitmap to reduce the chance that we need to
resize it. */
bb_seen = sbitmap_alloc (last_basic_block_for_fn (cfun) * 2);
bitmap_clear (bb_seen);
initialize_original_copy_tables ();
if (profile_info && flag_branch_probabilities)
probability_cutoff = PARAM_VALUE (TRACER_MIN_BRANCH_PROBABILITY_FEEDBACK);
else
probability_cutoff = PARAM_VALUE (TRACER_MIN_BRANCH_PROBABILITY);
probability_cutoff = REG_BR_PROB_BASE / 100 * probability_cutoff;
branch_ratio_cutoff =
(REG_BR_PROB_BASE / 100 * PARAM_VALUE (TRACER_MIN_BRANCH_RATIO));
FOR_EACH_BB_FN (bb, cfun)
{
int n = count_insns (bb);
if (!ignore_bb_p (bb))
blocks[bb->index] = heap.insert (-bb->frequency, bb);
counts [bb->index] = n;
ninsns += n;
weighted_insns += n * bb->frequency;
}
int
flow_loops_find (struct loops *loops, int flags)
{
int i;
int b;
int num_loops;
edge e;
sbitmap headers;
int *dfs_order;
int *rc_order;
basic_block header;
basic_block bb;
/* This function cannot be repeatedly called with different
flags to build up the loop information. The loop tree
must always be built if this function is called. */
if (! (flags & LOOP_TREE))
abort ();
memset (loops, 0, sizeof *loops);
/* Taking care of this degenerate case makes the rest of
this code simpler. */
if (n_basic_blocks == 0)
return 0;
dfs_order = NULL;
rc_order = NULL;
/* Join loops with shared headers. */
canonicalize_loop_headers ();
/* Compute the dominators. */
calculate_dominance_info (CDI_DOMINATORS);
/* Count the number of loop headers. This should be the
same as the number of natural loops. */
headers = sbitmap_alloc (last_basic_block);
sbitmap_zero (headers);
num_loops = 0;
FOR_EACH_BB (header)
{
int more_latches = 0;
header->loop_depth = 0;
/* If we have an abnormal predecessor, do not consider the
loop (not worth the problems). */
for (e = header->pred; e; e = e->pred_next)
if (e->flags & EDGE_ABNORMAL)
break;
if (e)
continue;
for (e = header->pred; e; e = e->pred_next)
{
basic_block latch = e->src;
if (e->flags & EDGE_ABNORMAL)
abort ();
/* Look for back edges where a predecessor is dominated
by this block. A natural loop has a single entry
node (header) that dominates all the nodes in the
loop. It also has single back edge to the header
from a latch node. */
if (latch != ENTRY_BLOCK_PTR
&& dominated_by_p (CDI_DOMINATORS, latch, header))
{
/* Shared headers should be eliminated by now. */
if (more_latches)
abort ();
more_latches = 1;
SET_BIT (headers, header->index);
num_loops++;
}
}
}
/* Allocate loop structures. */
loops->parray = xcalloc (num_loops + 1, sizeof (struct loop *));
/* Dummy loop containing whole function. */
loops->parray[0] = xcalloc (1, sizeof (struct loop));
loops->parray[0]->next = NULL;
loops->parray[0]->inner = NULL;
loops->parray[0]->outer = NULL;
loops->parray[0]->depth = 0;
loops->parray[0]->pred = NULL;
loops->parray[0]->num_nodes = n_basic_blocks + 2;
loops->parray[0]->latch = EXIT_BLOCK_PTR;
loops->parray[0]->header = ENTRY_BLOCK_PTR;
ENTRY_BLOCK_PTR->loop_father = loops->parray[0];
EXIT_BLOCK_PTR->loop_father = loops->parray[0];
loops->tree_root = loops->parray[0];
/* Find and record information about all the natural loops
in the CFG. */
loops->num = 1;
FOR_EACH_BB (bb)
bb->loop_father = loops->tree_root;
if (num_loops)
{
/* Compute depth first search order of the CFG so that outer
natural loops will be found before inner natural loops. */
dfs_order = xmalloc (n_basic_blocks * sizeof (int));
rc_order = xmalloc (n_basic_blocks * sizeof (int));
flow_depth_first_order_compute (dfs_order, rc_order);
/* Save CFG derived information to avoid recomputing it. */
loops->cfg.dfs_order = dfs_order;
loops->cfg.rc_order = rc_order;
num_loops = 1;
for (b = 0; b < n_basic_blocks; b++)
{
struct loop *loop;
/* Search the nodes of the CFG in reverse completion order
so that we can find outer loops first. */
if (!TEST_BIT (headers, rc_order[b]))
continue;
header = BASIC_BLOCK (rc_order[b]);
loop = loops->parray[num_loops] = xcalloc (1, sizeof (struct loop));
loop->header = header;
loop->num = num_loops;
num_loops++;
/* Look for the latch for this header block. */
for (e = header->pred; e; e = e->pred_next)
{
basic_block latch = e->src;
if (latch != ENTRY_BLOCK_PTR
&& dominated_by_p (CDI_DOMINATORS, latch, header))
{
loop->latch = latch;
break;
}
}
flow_loop_tree_node_add (header->loop_father, loop);
loop->num_nodes = flow_loop_nodes_find (loop->header, loop);
}
/* Assign the loop nesting depth and enclosed loop level for each
loop. */
loops->levels = flow_loops_level_compute (loops);
/* Scan the loops. */
for (i = 1; i < num_loops; i++)
flow_loop_scan (loops->parray[i], flags);
loops->num = num_loops;
}
else
{
free_dominance_info (CDI_DOMINATORS);
}
sbitmap_free (headers);
loops->state = 0;
#ifdef ENABLE_CHECKING
verify_flow_info ();
verify_loop_structure (loops);
#endif
return loops->num;
}
/* Checks that LOOPS are all right:
-- sizes of loops are all right
-- results of get_loop_body really belong to the loop
-- loop header have just single entry edge and single latch edge
-- loop latches have only single successor that is header of their loop
-- irreducible loops are correctly marked
*/
void
verify_loop_structure (struct loops *loops)
{
unsigned *sizes, i, j;
sbitmap irreds;
basic_block *bbs, bb;
struct loop *loop;
int err = 0;
edge e;
/* Check sizes. */
sizes = xcalloc (loops->num, sizeof (int));
sizes[0] = 2;
FOR_EACH_BB (bb)
for (loop = bb->loop_father; loop; loop = loop->outer)
sizes[loop->num]++;
for (i = 0; i < loops->num; i++)
{
if (!loops->parray[i])
continue;
if (loops->parray[i]->num_nodes != sizes[i])
{
error ("Size of loop %d should be %d, not %d.",
i, sizes[i], loops->parray[i]->num_nodes);
err = 1;
}
}
free (sizes);
/* Check get_loop_body. */
for (i = 1; i < loops->num; i++)
{
loop = loops->parray[i];
if (!loop)
continue;
bbs = get_loop_body (loop);
for (j = 0; j < loop->num_nodes; j++)
if (!flow_bb_inside_loop_p (loop, bbs[j]))
{
error ("Bb %d do not belong to loop %d.",
bbs[j]->index, i);
err = 1;
}
free (bbs);
}
/* Check headers and latches. */
for (i = 1; i < loops->num; i++)
{
loop = loops->parray[i];
if (!loop)
continue;
if ((loops->state & LOOPS_HAVE_PREHEADERS)
&& (!loop->header->pred->pred_next
|| loop->header->pred->pred_next->pred_next))
{
error ("Loop %d's header does not have exactly 2 entries.", i);
err = 1;
}
if (loops->state & LOOPS_HAVE_SIMPLE_LATCHES)
{
if (!loop->latch->succ
|| loop->latch->succ->succ_next)
{
error ("Loop %d's latch does not have exactly 1 successor.", i);
err = 1;
}
if (loop->latch->succ->dest != loop->header)
{
error ("Loop %d's latch does not have header as successor.", i);
err = 1;
}
if (loop->latch->loop_father != loop)
{
error ("Loop %d's latch does not belong directly to it.", i);
err = 1;
}
}
if (loop->header->loop_father != loop)
{
error ("Loop %d's header does not belong directly to it.", i);
err = 1;
}
if ((loops->state & LOOPS_HAVE_MARKED_IRREDUCIBLE_REGIONS)
&& (loop_latch_edge (loop)->flags & EDGE_IRREDUCIBLE_LOOP))
{
error ("Loop %d's latch is marked as part of irreducible region.", i);
err = 1;
}
}
/* Check irreducible loops. */
if (loops->state & LOOPS_HAVE_MARKED_IRREDUCIBLE_REGIONS)
{
/* Record old info. */
irreds = sbitmap_alloc (last_basic_block);
FOR_EACH_BB (bb)
{
if (bb->flags & BB_IRREDUCIBLE_LOOP)
SET_BIT (irreds, bb->index);
else
RESET_BIT (irreds, bb->index);
for (e = bb->succ; e; e = e->succ_next)
if (e->flags & EDGE_IRREDUCIBLE_LOOP)
e->flags |= EDGE_ALL_FLAGS + 1;
}
/* Recount it. */
mark_irreducible_loops (loops);
/* Compare. */
FOR_EACH_BB (bb)
{
if ((bb->flags & BB_IRREDUCIBLE_LOOP)
&& !TEST_BIT (irreds, bb->index))
{
error ("Basic block %d should be marked irreducible.", bb->index);
err = 1;
}
else if (!(bb->flags & BB_IRREDUCIBLE_LOOP)
&& TEST_BIT (irreds, bb->index))
{
error ("Basic block %d should not be marked irreducible.", bb->index);
err = 1;
}
for (e = bb->succ; e; e = e->succ_next)
{
if ((e->flags & EDGE_IRREDUCIBLE_LOOP)
&& !(e->flags & (EDGE_ALL_FLAGS + 1)))
{
error ("Edge from %d to %d should be marked irreducible.",
e->src->index, e->dest->index);
err = 1;
}
else if (!(e->flags & EDGE_IRREDUCIBLE_LOOP)
&& (e->flags & (EDGE_ALL_FLAGS + 1)))
{
error ("Edge from %d to %d should not be marked irreducible.",
e->src->index, e->dest->index);
err = 1;
}
e->flags &= ~(EDGE_ALL_FLAGS + 1);
}
}
free (irreds);
}
static void
tree_expand_cfg (void)
{
basic_block bb, init_block;
sbitmap blocks;
/* Some backends want to know that we are expanding to RTL. */
currently_expanding_to_rtl = 1;
/* Prepare the rtl middle end to start recording block changes. */
reset_block_changes ();
/* Expand the variables recorded during gimple lowering. */
expand_used_vars ();
#ifdef KEY
// Run expand_used_vars above to set DECL_SECTION_NAME. Bug 10876.
if (flag_spin_file)
return;
#endif
/* Set up parameters and prepare for return, for the function. */
expand_function_start (current_function_decl);
/* If this function is `main', emit a call to `__main'
to run global initializers, etc. */
if (DECL_NAME (current_function_decl)
&& MAIN_NAME_P (DECL_NAME (current_function_decl))
&& DECL_FILE_SCOPE_P (current_function_decl))
expand_main_function ();
/* Register rtl specific functions for cfg. */
rtl_register_cfg_hooks ();
init_block = construct_init_block ();
FOR_BB_BETWEEN (bb, init_block->next_bb, EXIT_BLOCK_PTR, next_bb)
bb = expand_gimple_basic_block (bb, dump_file);
construct_exit_block ();
/* We're done expanding trees to RTL. */
currently_expanding_to_rtl = 0;
/* Convert tree EH labels to RTL EH labels, and clean out any unreachable
EH regions. */
convert_from_eh_region_ranges ();
rebuild_jump_labels (get_insns ());
find_exception_handler_labels ();
blocks = sbitmap_alloc (last_basic_block);
sbitmap_ones (blocks);
find_many_sub_basic_blocks (blocks);
purge_all_dead_edges (0);
sbitmap_free (blocks);
compact_blocks ();
#ifdef ENABLE_CHECKING
verify_flow_info();
#endif
/* There's no need to defer outputting this function any more; we
know we want to output it. */
DECL_DEFER_OUTPUT (current_function_decl) = 0;
/* Now that we're done expanding trees to RTL, we shouldn't have any
more CONCATs anywhere. */
generating_concat_p = 0;
finalize_block_changes ();
if (dump_file)
{
fprintf (dump_file,
"\n\n;;\n;; Full RTL generated for this function:\n;;\n");
/* And the pass manager will dump RTL for us. */
}
}
static struct loop *
unswitch_loop (struct loops *loops, 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;
sbitmap zero_bitmap;
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;
zero_bitmap = sbitmap_alloc (2);
sbitmap_zero (zero_bitmap);
if (!duplicate_loop_to_header_edge (loop, entry, loops, 1,
zero_bitmap, NULL, NULL, NULL, 0))
{
sbitmap_free (zero_bitmap);
return NULL;
}
sbitmap_free (zero_bitmap);
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 (loops, latch_edge,
single_pred_edge (get_bb_copy (loop->header)), switch_bb,
BRANCH_EDGE (switch_bb), FALLTHRU_EDGE (switch_bb), true);
/* Remove branches that are now unreachable in new loops. */
remove_path (loops, true_edge);
remove_path (loops, false_edge);
/* One of created loops do not have to be subloop of the outer loop now,
so fix its placement in loop data structure. */
fix_loop_placement (loop);
fix_loop_placement (nloop);
/* Preserve the simple loop preheaders. */
loop_split_edge_with (loop_preheader_edge (loop), NULL_RTX);
loop_split_edge_with (loop_preheader_edge (nloop), NULL_RTX);
return nloop;
}
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
tail_duplicate (void)
{
fibnode_t *blocks = XCNEWVEC (fibnode_t, last_basic_block);
basic_block *trace = XNEWVEC (basic_block, n_basic_blocks);
int *counts = XNEWVEC (int, last_basic_block);
int ninsns = 0, nduplicated = 0;
gcov_type weighted_insns = 0, traced_insns = 0;
fibheap_t heap = fibheap_new ();
gcov_type cover_insns;
int max_dup_insns;
basic_block bb;
bool changed = false;
/* Create an oversized sbitmap to reduce the chance that we need to
resize it. */
bb_seen = sbitmap_alloc (last_basic_block * 2);
bitmap_clear (bb_seen);
initialize_original_copy_tables ();
if (profile_info && flag_branch_probabilities)
probability_cutoff = PARAM_VALUE (TRACER_MIN_BRANCH_PROBABILITY_FEEDBACK);
else
probability_cutoff = PARAM_VALUE (TRACER_MIN_BRANCH_PROBABILITY);
probability_cutoff = REG_BR_PROB_BASE / 100 * probability_cutoff;
branch_ratio_cutoff =
(REG_BR_PROB_BASE / 100 * PARAM_VALUE (TRACER_MIN_BRANCH_RATIO));
FOR_EACH_BB (bb)
{
int n = count_insns (bb);
if (!ignore_bb_p (bb))
blocks[bb->index] = fibheap_insert (heap, -bb->frequency,
bb);
counts [bb->index] = n;
ninsns += n;
weighted_insns += n * bb->frequency;
}
if (profile_info && flag_branch_probabilities)
cover_insns = PARAM_VALUE (TRACER_DYNAMIC_COVERAGE_FEEDBACK);
else
cover_insns = PARAM_VALUE (TRACER_DYNAMIC_COVERAGE);
cover_insns = (weighted_insns * cover_insns + 50) / 100;
max_dup_insns = (ninsns * PARAM_VALUE (TRACER_MAX_CODE_GROWTH) + 50) / 100;
while (traced_insns < cover_insns && nduplicated < max_dup_insns
&& !fibheap_empty (heap))
{
basic_block bb = (basic_block) fibheap_extract_min (heap);
int n, pos;
if (!bb)
break;
blocks[bb->index] = NULL;
if (ignore_bb_p (bb))
continue;
gcc_assert (!bb_seen_p (bb));
n = find_trace (bb, trace);
bb = trace[0];
traced_insns += bb->frequency * counts [bb->index];
if (blocks[bb->index])
{
fibheap_delete_node (heap, blocks[bb->index]);
blocks[bb->index] = NULL;
}
for (pos = 1; pos < n; pos++)
{
basic_block bb2 = trace[pos];
if (blocks[bb2->index])
{
fibheap_delete_node (heap, blocks[bb2->index]);
blocks[bb2->index] = NULL;
}
traced_insns += bb2->frequency * counts [bb2->index];
if (EDGE_COUNT (bb2->preds) > 1
&& can_duplicate_block_p (bb2)
/* We have the tendency to duplicate the loop header
of all do { } while loops. Do not do that - it is
not profitable and it might create a loop with multiple
entries or at least rotate the loop. */
&& (!current_loops
|| bb2->loop_father->header != bb2))
{
edge e;
basic_block copy;
nduplicated += counts [bb2->index];
e = find_edge (bb, bb2);
copy = duplicate_block (bb2, e, bb);
flush_pending_stmts (e);
add_phi_args_after_copy (©, 1, NULL);
/* Reconsider the original copy of block we've duplicated.
Removing the most common predecessor may make it to be
head. */
blocks[bb2->index] =
fibheap_insert (heap, -bb2->frequency, bb2);
if (dump_file)
fprintf (dump_file, "Duplicated %i as %i [%i]\n",
bb2->index, copy->index, copy->frequency);
bb2 = copy;
changed = true;
}
mark_bb_seen (bb2);
bb = bb2;
/* In case the trace became infrequent, stop duplicating. */
if (ignore_bb_p (bb))
break;
}
if (dump_file)
fprintf (dump_file, " covered now %.1f\n\n",
traced_insns * 100.0 / weighted_insns);
}
if (dump_file)
fprintf (dump_file, "Duplicated %i insns (%i%%)\n", nduplicated,
nduplicated * 100 / ninsns);
free_original_copy_tables ();
sbitmap_free (bb_seen);
free (blocks);
free (trace);
free (counts);
fibheap_delete (heap);
return changed;
}
static unsigned int
copyprop_hardreg_forward (void)
{
struct value_data *all_vd;
basic_block bb;
sbitmap visited;
bool analyze_called = false;
all_vd = XNEWVEC (struct value_data, last_basic_block);
visited = sbitmap_alloc (last_basic_block);
sbitmap_zero (visited);
if (MAY_HAVE_DEBUG_INSNS)
debug_insn_changes_pool
= create_alloc_pool ("debug insn changes pool",
sizeof (struct queued_debug_insn_change), 256);
FOR_EACH_BB (bb)
{
SET_BIT (visited, bb->index);
/* If a block has a single predecessor, that we've already
processed, begin with the value data that was live at
the end of the predecessor block. */
/* ??? Ought to use more intelligent queuing of blocks. */
if (single_pred_p (bb)
&& TEST_BIT (visited, single_pred (bb)->index)
&& ! (single_pred_edge (bb)->flags & (EDGE_ABNORMAL_CALL | EDGE_EH)))
{
all_vd[bb->index] = all_vd[single_pred (bb)->index];
if (all_vd[bb->index].n_debug_insn_changes)
{
unsigned int regno;
for (regno = 0; regno < FIRST_PSEUDO_REGISTER; regno++)
{
if (all_vd[bb->index].e[regno].debug_insn_changes)
{
all_vd[bb->index].e[regno].debug_insn_changes = NULL;
if (--all_vd[bb->index].n_debug_insn_changes == 0)
break;
}
}
}
}
else
init_value_data (all_vd + bb->index);
copyprop_hardreg_forward_1 (bb, all_vd + bb->index);
}
if (MAY_HAVE_DEBUG_INSNS)
{
FOR_EACH_BB (bb)
if (TEST_BIT (visited, bb->index)
&& all_vd[bb->index].n_debug_insn_changes)
{
unsigned int regno;
bitmap live;
if (!analyze_called)
{
df_analyze ();
analyze_called = true;
}
live = df_get_live_out (bb);
for (regno = 0; regno < FIRST_PSEUDO_REGISTER; regno++)
if (all_vd[bb->index].e[regno].debug_insn_changes)
{
if (REGNO_REG_SET_P (live, regno))
apply_debug_insn_changes (all_vd + bb->index, regno);
if (all_vd[bb->index].n_debug_insn_changes == 0)
break;
}
}
free_alloc_pool (debug_insn_changes_pool);
}
sbitmap_free (visited);
free (all_vd);
return 0;
}
