modulo-sched.c (get_sched_window): Use just one loop for predecessors and one loop for successors.

gcc/
	* modulo-sched.c (get_sched_window): Use just one loop for predecessors
	and one loop for successors.  Fix upper bound of memory range.

From-SVN: r177555
This commit is contained in:
Richard Sandiford 2011-08-08 09:26:54 +00:00 committed by Richard Sandiford
parent d855a67e7d
commit fe43febc8c
2 changed files with 114 additions and 198 deletions

View File

@ -1,3 +1,8 @@
2011-08-08 Richard Sandiford <richard.sandiford@linaro.org>
* modulo-sched.c (get_sched_window): Use just one loop for predecessors
and one loop for successors. Fix upper bound of memory range.
2011-08-06 Uros Bizjak <ubizjak@gmail.com> 2011-08-06 Uros Bizjak <ubizjak@gmail.com>
PR target/50001 PR target/50001

View File

@ -1630,9 +1630,11 @@ sms_schedule (void)
static int static int
get_sched_window (partial_schedule_ptr ps, ddg_node_ptr u_node, get_sched_window (partial_schedule_ptr ps, ddg_node_ptr u_node,
sbitmap sched_nodes, int ii, int *start_p, int *step_p, int *end_p) sbitmap sched_nodes, int ii, int *start_p, int *step_p,
int *end_p)
{ {
int start, step, end; int start, step, end;
int early_start, late_start;
ddg_edge_ptr e; ddg_edge_ptr e;
sbitmap psp = sbitmap_alloc (ps->g->num_nodes); sbitmap psp = sbitmap_alloc (ps->g->num_nodes);
sbitmap pss = sbitmap_alloc (ps->g->num_nodes); sbitmap pss = sbitmap_alloc (ps->g->num_nodes);
@ -1640,6 +1642,8 @@ get_sched_window (partial_schedule_ptr ps, ddg_node_ptr u_node,
sbitmap u_node_succs = NODE_SUCCESSORS (u_node); sbitmap u_node_succs = NODE_SUCCESSORS (u_node);
int psp_not_empty; int psp_not_empty;
int pss_not_empty; int pss_not_empty;
int count_preds;
int count_succs;
/* 1. compute sched window for u (start, end, step). */ /* 1. compute sched window for u (start, end, step). */
sbitmap_zero (psp); sbitmap_zero (psp);
@ -1647,214 +1651,121 @@ get_sched_window (partial_schedule_ptr ps, ddg_node_ptr u_node,
psp_not_empty = sbitmap_a_and_b_cg (psp, u_node_preds, sched_nodes); psp_not_empty = sbitmap_a_and_b_cg (psp, u_node_preds, sched_nodes);
pss_not_empty = sbitmap_a_and_b_cg (pss, u_node_succs, sched_nodes); pss_not_empty = sbitmap_a_and_b_cg (pss, u_node_succs, sched_nodes);
if (psp_not_empty && !pss_not_empty) /* We first compute a forward range (start <= end), then decide whether
{ to reverse it. */
int early_start = INT_MIN; early_start = INT_MIN;
late_start = INT_MAX;
start = INT_MIN;
end = INT_MAX;
step = 1;
end = INT_MAX; count_preds = 0;
for (e = u_node->in; e != 0; e = e->next_in) count_succs = 0;
{
ddg_node_ptr v_node = e->src;
if (dump_file) /* Calculate early_start and limit end. Both bounds are inclusive. */
{ if (psp_not_empty)
fprintf (dump_file, "\nProcessing edge: "); for (e = u_node->in; e != 0; e = e->next_in)
print_ddg_edge (dump_file, e); {
ddg_node_ptr v_node = e->src;
if (dump_file)
{
fprintf (dump_file, "\nProcessing edge: ");
print_ddg_edge (dump_file, e);
fprintf (dump_file,
"\nScheduling %d (%d) in psp_not_empty,"
" checking p %d (%d): ", u_node->cuid,
INSN_UID (u_node->insn), v_node->cuid, INSN_UID
(v_node->insn));
}
if (TEST_BIT (sched_nodes, v_node->cuid))
{
int p_st = SCHED_TIME (v_node);
early_start = MAX (early_start,
p_st + e->latency - (e->distance * ii));
if (e->data_type == MEM_DEP)
end = MIN (end, p_st + ii - 1);
if (e->type == TRUE_DEP && e->data_type == REG_DEP)
count_preds++;
if (dump_file)
fprintf (dump_file, fprintf (dump_file,
"\nScheduling %d (%d) in psp_not_empty," "pred st = %d; early_start = %d; latency: %d;"
" checking p %d (%d): ", u_node->cuid, " end: %d\n", p_st, early_start, e->latency, end);
INSN_UID (u_node->insn), v_node->cuid, INSN_UID
(v_node->insn));
}
if (TEST_BIT (sched_nodes, v_node->cuid)) }
{ else if (dump_file)
int p_st = SCHED_TIME (v_node); fprintf (dump_file, "the node is not scheduled\n");
}
early_start = /* Calculate late_start and limit start. Both bounds are inclusive. */
MAX (early_start, p_st + e->latency - (e->distance * ii)); if (pss_not_empty)
for (e = u_node->out; e != 0; e = e->next_out)
{
ddg_node_ptr v_node = e->dest;
if (dump_file) if (dump_file)
fprintf (dump_file, {
"pred st = %d; early_start = %d; latency: %d", fprintf (dump_file, "\nProcessing edge:");
p_st, early_start, e->latency); print_ddg_edge (dump_file, e);
fprintf (dump_file,
"\nScheduling %d (%d) in pss_not_empty,"
" checking s %d (%d): ", u_node->cuid,
INSN_UID (u_node->insn), v_node->cuid, INSN_UID
(v_node->insn));
}
if (e->data_type == MEM_DEP) if (TEST_BIT (sched_nodes, v_node->cuid))
end = MIN (end, SCHED_TIME (v_node) + ii - 1); {
} int s_st = SCHED_TIME (v_node);
else if (dump_file)
fprintf (dump_file, "the node is not scheduled\n");
}
start = early_start;
end = MIN (end, early_start + ii);
/* Schedule the node close to it's predecessors. */
step = 1;
if (dump_file) late_start = MIN (late_start,
fprintf (dump_file, s_st - e->latency + (e->distance * ii));
"\nScheduling %d (%d) in a window (%d..%d) with step %d\n",
u_node->cuid, INSN_UID (u_node->insn), start, end, step);
}
else if (!psp_not_empty && pss_not_empty) if (e->data_type == MEM_DEP)
start = MAX (start, s_st - ii + 1);
if (e->type == TRUE_DEP && e->data_type == REG_DEP)
count_succs++;
if (dump_file)
fprintf (dump_file,
"succ st = %d; late_start = %d; latency = %d;"
" start=%d", s_st, late_start, e->latency, start);
}
else if (dump_file)
fprintf (dump_file, "the node is not scheduled\n");
}
/* Get a target scheduling window no bigger than ii. */
if (early_start == INT_MIN && late_start == INT_MAX)
early_start = SCHED_ASAP (u_node);
else if (early_start == INT_MIN)
early_start = late_start - (ii - 1);
late_start = MIN (late_start, early_start + (ii - 1));
/* Apply memory dependence limits. */
start = MAX (start, early_start);
end = MIN (end, late_start);
/* If there are at least as many successors as predecessors, schedule the
node close to its successors. */
if (pss_not_empty && count_succs >= count_preds)
{ {
int late_start = INT_MAX; int tmp = end;
end = start;
end = INT_MIN; start = tmp;
for (e = u_node->out; e != 0; e = e->next_out)
{
ddg_node_ptr v_node = e->dest;
if (dump_file)
{
fprintf (dump_file, "\nProcessing edge:");
print_ddg_edge (dump_file, e);
fprintf (dump_file,
"\nScheduling %d (%d) in pss_not_empty,"
" checking s %d (%d): ", u_node->cuid,
INSN_UID (u_node->insn), v_node->cuid, INSN_UID
(v_node->insn));
}
if (TEST_BIT (sched_nodes, v_node->cuid))
{
int s_st = SCHED_TIME (v_node);
late_start = MIN (late_start,
s_st - e->latency + (e->distance * ii));
if (dump_file)
fprintf (dump_file,
"succ st = %d; late_start = %d; latency = %d",
s_st, late_start, e->latency);
if (e->data_type == MEM_DEP)
end = MAX (end, SCHED_TIME (v_node) - ii + 1);
if (dump_file)
fprintf (dump_file, "end = %d\n", end);
}
else if (dump_file)
fprintf (dump_file, "the node is not scheduled\n");
}
start = late_start;
end = MAX (end, late_start - ii);
/* Schedule the node close to it's successors. */
step = -1; step = -1;
if (dump_file)
fprintf (dump_file,
"\nScheduling %d (%d) in a window (%d..%d) with step %d\n",
u_node->cuid, INSN_UID (u_node->insn), start, end, step);
} }
else if (psp_not_empty && pss_not_empty) /* Now that we've finalized the window, make END an exclusive rather
{ than an inclusive bound. */
int early_start = INT_MIN; end += step;
int late_start = INT_MAX;
int count_preds = 0;
int count_succs = 0;
start = INT_MIN;
end = INT_MAX;
for (e = u_node->in; e != 0; e = e->next_in)
{
ddg_node_ptr v_node = e->src;
if (dump_file)
{
fprintf (dump_file, "\nProcessing edge:");
print_ddg_edge (dump_file, e);
fprintf (dump_file,
"\nScheduling %d (%d) in psp_pss_not_empty,"
" checking p %d (%d): ", u_node->cuid, INSN_UID
(u_node->insn), v_node->cuid, INSN_UID
(v_node->insn));
}
if (TEST_BIT (sched_nodes, v_node->cuid))
{
int p_st = SCHED_TIME (v_node);
early_start = MAX (early_start,
p_st + e->latency
- (e->distance * ii));
if (dump_file)
fprintf (dump_file,
"pred st = %d; early_start = %d; latency = %d",
p_st, early_start, e->latency);
if (e->type == TRUE_DEP && e->data_type == REG_DEP)
count_preds++;
if (e->data_type == MEM_DEP)
end = MIN (end, SCHED_TIME (v_node) + ii - 1);
}
else if (dump_file)
fprintf (dump_file, "the node is not scheduled\n");
}
for (e = u_node->out; e != 0; e = e->next_out)
{
ddg_node_ptr v_node = e->dest;
if (dump_file)
{
fprintf (dump_file, "\nProcessing edge:");
print_ddg_edge (dump_file, e);
fprintf (dump_file,
"\nScheduling %d (%d) in psp_pss_not_empty,"
" checking s %d (%d): ", u_node->cuid, INSN_UID
(u_node->insn), v_node->cuid, INSN_UID
(v_node->insn));
}
if (TEST_BIT (sched_nodes, v_node->cuid))
{
int s_st = SCHED_TIME (v_node);
late_start = MIN (late_start,
s_st - e->latency
+ (e->distance * ii));
if (dump_file)
fprintf (dump_file,
"succ st = %d; late_start = %d; latency = %d",
s_st, late_start, e->latency);
if (e->type == TRUE_DEP && e->data_type == REG_DEP)
count_succs++;
if (e->data_type == MEM_DEP)
start = MAX (start, SCHED_TIME (v_node) - ii + 1);
}
else if (dump_file)
fprintf (dump_file, "the node is not scheduled\n");
}
start = MAX (start, early_start);
end = MIN (end, MIN (early_start + ii, late_start + 1));
step = 1;
/* If there are more successors than predecessors schedule the
node close to it's successors. */
if (count_succs >= count_preds)
{
int old_start = start;
start = end - 1;
end = old_start - 1;
step = -1;
}
}
else /* psp is empty && pss is empty. */
{
start = SCHED_ASAP (u_node);
end = start + ii;
step = 1;
}
*start_p = start; *start_p = start;
*step_p = step; *step_p = step;
@ -1867,10 +1778,10 @@ get_sched_window (partial_schedule_ptr ps, ddg_node_ptr u_node,
if (dump_file) if (dump_file)
fprintf (dump_file, "\nEmpty window: start=%d, end=%d, step=%d\n", fprintf (dump_file, "\nEmpty window: start=%d, end=%d, step=%d\n",
start, end, step); start, end, step);
return -1; return -1;
} }
return 0; return 0;
} }
/* Calculate MUST_PRECEDE/MUST_FOLLOW bitmaps of U_NODE; which is the /* Calculate MUST_PRECEDE/MUST_FOLLOW bitmaps of U_NODE; which is the