/* Instruction scheduling pass. Selective scheduler and pipeliner.
- Copyright (C) 2006, 2007, 2008, 2009 Free Software Foundation, Inc.
+ Copyright (C) 2006, 2007, 2008, 2009, 2010 Free Software Foundation, Inc.
This file is part of GCC.
#include "system.h"
#include "coretypes.h"
#include "tm.h"
-#include "toplev.h"
-#include "rtl.h"
+#include "rtl-error.h"
#include "tm_p.h"
#include "hard-reg-set.h"
#include "regs.h"
#include "insn-config.h"
#include "insn-attr.h"
#include "except.h"
-#include "toplev.h"
#include "recog.h"
#include "params.h"
#include "target.h"
#include "langhooks.h"
#include "rtlhooks-def.h"
#include "output.h"
+#include "emit-rtl.h"
#ifdef INSN_SCHEDULING
#include "sel-sched-ir.h"
in_fallthru_bb_p (rtx insn, rtx succ)
{
basic_block bb = BLOCK_FOR_INSN (insn);
+ edge e;
if (bb == BLOCK_FOR_INSN (succ))
return true;
- if (find_fallthru_edge (bb))
- bb = find_fallthru_edge (bb)->dest;
+ e = find_fallthru_edge_from (bb);
+ if (e)
+ bb = e->dest;
else
return false;
if (GET_CODE (*cur_rtx) == SUBREG
&& REG_P (p->x)
- && REGNO (SUBREG_REG (*cur_rtx)) == REGNO (p->x))
+ && (!REG_P (SUBREG_REG (*cur_rtx))
+ || REGNO (SUBREG_REG (*cur_rtx)) == REGNO (p->x)))
{
/* ??? Do not support substituting regs inside subregs. In that case,
simplify_subreg will be called by validate_replace_rtx, and
/* Can't use regs which aren't saved by
the prologue. */
|| !TEST_HARD_REG_BIT (sel_hrd.regs_ever_used, cur_reg + i)
+ /* Can't use regs with non-null REG_BASE_VALUE, because adjusting
+ it affects aliasing globally and invalidates all AV sets. */
+ || get_reg_base_value (cur_reg + i)
#ifdef LEAF_REGISTERS
/* We can't use a non-leaf register if we're in a
leaf function. */
frame pointer, or we could not discover its class. */
if (fixed_regs[regno]
|| global_regs[regno]
-#if FRAME_POINTER_REGNUM != HARD_FRAME_POINTER_REGNUM
+#if !HARD_FRAME_POINTER_IS_FRAME_POINTER
|| (frame_pointer_needed && regno == HARD_FRAME_POINTER_REGNUM)
#else
|| (frame_pointer_needed && regno == FRAME_POINTER_REGNUM)
SET_HARD_REG_BIT (reg_rename_p->unavailable_hard_regs,
FRAME_POINTER_REGNUM + i);
-#if FRAME_POINTER_REGNUM != HARD_FRAME_POINTER_REGNUM
+#if !HARD_FRAME_POINTER_IS_FRAME_POINTER
for (i = hard_regno_nregs[HARD_FRAME_POINTER_REGNUM][Pmode]; i--;)
SET_HARD_REG_BIT (reg_rename_p->unavailable_hard_regs,
HARD_FRAME_POINTER_REGNUM + i);
if (recovery_block != NULL)
{
rtx twin_rtx;
- insn_t twin;
twin_rtx = copy_rtx (PATTERN (EXPR_INSN_RTX (c_expr)));
twin_rtx = create_insn_rtx_from_pattern (twin_rtx, NULL_RTX);
- twin = sel_gen_recovery_insn_from_rtx_after (twin_rtx,
- INSN_EXPR (orig_insn),
- INSN_SEQNO (insn),
- bb_note (recovery_block));
+ sel_gen_recovery_insn_from_rtx_after (twin_rtx,
+ INSN_EXPR (orig_insn),
+ INSN_SEQNO (insn),
+ bb_note (recovery_block));
}
/* If we've generated a data speculation check, make sure
EXPR_TARGET_AVAILABLE (expr) = false;
if (pti->type == TRANS_SPECULATION)
{
- ds_t ds;
-
- ds = EXPR_SPEC_DONE_DS (expr);
-
EXPR_SPEC_DONE_DS (expr) = pti->ds;
EXPR_NEEDS_SPEC_CHECK_P (expr) |= pti->needs_check;
}
/* Add insn to to the tail of current path. */
ilist_add (&p, insn);
- for (is = 0; VEC_iterate (rtx, sinfo->succs_ok, is, succ); is++)
+ FOR_EACH_VEC_ELT (rtx, sinfo->succs_ok, is, succ)
{
av_set_t succ_set;
VEC_index (int, sinfo->probs_ok, is),
sinfo->all_prob);
- if (sinfo->all_succs_n > 1
- && sinfo->all_succs_n == sinfo->succs_ok_n)
+ if (sinfo->all_succs_n > 1)
{
/* Find EXPR'es that came from *all* successors and save them
into expr_in_all_succ_branches. This set will be used later
/* Check liveness restrictions via hard way when there are more than
two successors. */
if (sinfo->succs_ok_n > 2)
- for (is = 0; VEC_iterate (rtx, sinfo->succs_ok, is, succ); is++)
+ FOR_EACH_VEC_ELT (rtx, sinfo->succs_ok, is, succ)
{
basic_block succ_bb = BLOCK_FOR_INSN (succ);
/* Finally, check liveness restrictions on paths leaving the region. */
if (sinfo->all_succs_n > sinfo->succs_ok_n)
- for (is = 0; VEC_iterate (rtx, sinfo->succs_other, is, succ); is++)
+ FOR_EACH_VEC_ELT (rtx, sinfo->succs_other, is, succ)
mark_unavailable_targets
(av1, NULL, BB_LV_SET (BLOCK_FOR_INSN (succ)));
vinsn_t vinsn;
int n;
- for (n = 0; VEC_iterate (vinsn_t, vinsn_vec, n, vinsn); n++)
+ FOR_EACH_VEC_ELT (vinsn_t, vinsn_vec, n, vinsn)
if (VINSN_SEPARABLE_P (vinsn))
{
if (vinsn_equal_p (vinsn, EXPR_VINSN (expr)))
vinsn_t vinsn;
int n;
- for (n = 0; VEC_iterate (vinsn_t, *vinsn_vec, n, vinsn); n++)
+ FOR_EACH_VEC_ELT (vinsn_t, *vinsn_vec, n, vinsn)
vinsn_detach (vinsn);
VEC_block_remove (vinsn_t, *vinsn_vec, 0, len);
}
}
/* Sort the vector. */
- qsort (VEC_address (expr_t, vec_av_set), VEC_length (expr_t, vec_av_set),
- sizeof (expr_t), sel_rank_for_schedule);
+ VEC_qsort (expr_t, vec_av_set, sel_rank_for_schedule);
/* We record maximal priority of insns in av set for current instruction
group. */
{
expr_t expr = VEC_index (expr_t, vec_av_set, n);
insn_t insn = EXPR_INSN_RTX (expr);
- char target_available;
+ signed char target_available;
bool is_orig_reg_p = true;
int need_cycles, new_prio;
gcc_assert (min_need_stall == 0);
/* Sort the vector. */
- qsort (VEC_address (expr_t, vec_av_set), VEC_length (expr_t, vec_av_set),
- sizeof (expr_t), sel_rank_for_schedule);
+ VEC_qsort (expr_t, vec_av_set, sel_rank_for_schedule);
if (sched_verbose >= 4)
{
sel_print ("Total ready exprs: %d, stalled: %d\n",
VEC_length (expr_t, vec_av_set), stalled);
sel_print ("Sorted av set (%d): ", VEC_length (expr_t, vec_av_set));
- for (n = 0; VEC_iterate (expr_t, vec_av_set, n, expr); n++)
+ FOR_EACH_VEC_ELT (expr_t, vec_av_set, n, expr)
dump_expr (expr);
sel_print ("\n");
}
sched_extend_ready_list (ready.n_ready);
}
- for (n = 0; VEC_iterate (expr_t, vec_av_set, n, expr); n++)
+ FOR_EACH_VEC_ELT (expr_t, vec_av_set, n, expr)
{
vinsn_t vi = EXPR_VINSN (expr);
insn_t insn = VINSN_INSN_RTX (vi);
calculate_privileged_insns (void)
{
expr_t cur_expr, min_spec_expr = NULL;
- insn_t cur_insn, min_spec_insn;
int privileged_n = 0, i;
for (i = 0; i < ready.n_ready; i++)
continue;
if (! min_spec_expr)
- {
- min_spec_insn = ready_element (&ready, i);
- min_spec_expr = find_expr_for_ready (i, true);
- }
+ min_spec_expr = find_expr_for_ready (i, true);
- cur_insn = ready_element (&ready, i);
cur_expr = find_expr_for_ready (i, true);
if (EXPR_SPEC (cur_expr) > EXPR_SPEC (min_spec_expr))
if (dfa_lookahead > 0)
{
cycle_issued_insns = FENCE_ISSUED_INSNS (fence);
+ /* TODO: pass equivalent of first_cycle_insn_p to max_issue (). */
can_issue = max_issue (&ready, privileged_n,
- FENCE_STATE (fence), index);
+ FENCE_STATE (fence), true, index);
if (sched_verbose >= 2)
sel_print ("max_issue: we can issue %d insns, already did %d insns\n",
can_issue, FENCE_ISSUED_INSNS (fence));
best = fill_ready_list (av_vliw_ptr, bnds, fence, pneed_stall);
if (best == NULL && ready.n_ready > 0)
{
- int privileged_n, index, avail_n;
+ int privileged_n, index;
can_issue_more = invoke_reorder_hooks (fence);
if (can_issue_more > 0)
scheduled due to liveness restrictions on its destination register.
In the future, we'd like to choose once and then just probe insns
in the order of their priority. */
- avail_n = invoke_dfa_lookahead_guard ();
+ invoke_dfa_lookahead_guard ();
privileged_n = calculate_privileged_insns ();
can_issue_more = choose_best_insn (fence, privileged_n, &index);
if (can_issue_more)
{
can_issue_more = invoke_aftermath_hooks (fence, EXPR_INSN_RTX (best),
can_issue_more);
- if (can_issue_more == 0)
+ if (targetm.sched.variable_issue
+ && can_issue_more == 0)
*pneed_stall = 1;
}
if (INSN_P (insn))
EXPR_ORIG_BB_INDEX (INSN_EXPR (insn)) = succ->index;
- if (bitmap_bit_p (code_motion_visited_blocks, new_bb->index))
- {
- bitmap_set_bit (code_motion_visited_blocks, succ->index);
- bitmap_clear_bit (code_motion_visited_blocks, new_bb->index);
- }
+ if (bitmap_clear_bit (code_motion_visited_blocks, new_bb->index))
+ bitmap_set_bit (code_motion_visited_blocks, succ->index);
gcc_assert (LABEL_P (BB_HEAD (new_bb))
&& LABEL_P (BB_HEAD (succ)));
move_cond_jump (rtx insn, bnd_t bnd)
{
edge ft_edge;
- basic_block block_from, block_next, block_new;
- rtx next, prev, link;
+ basic_block block_from, block_next, block_new, block_bnd, bb;
+ rtx next, prev, link, head;
- /* BLOCK_FROM holds basic block of the jump. */
block_from = BLOCK_FOR_INSN (insn);
+ block_bnd = BLOCK_FOR_INSN (BND_TO (bnd));
+ prev = BND_TO (bnd);
- /* Moving of jump should not cross any other jumps or
- beginnings of new basic blocks. */
- gcc_assert (block_from == BLOCK_FOR_INSN (BND_TO (bnd)));
+#ifdef ENABLE_CHECKING
+ /* Moving of jump should not cross any other jumps or beginnings of new
+ basic blocks. The only exception is when we move a jump through
+ mutually exclusive insns along fallthru edges. */
+ if (block_from != block_bnd)
+ {
+ bb = block_from;
+ for (link = PREV_INSN (insn); link != PREV_INSN (prev);
+ link = PREV_INSN (link))
+ {
+ if (INSN_P (link))
+ gcc_assert (sched_insns_conditions_mutex_p (insn, link));
+ if (BLOCK_FOR_INSN (link) && BLOCK_FOR_INSN (link) != bb)
+ {
+ gcc_assert (single_pred (bb) == BLOCK_FOR_INSN (link));
+ bb = BLOCK_FOR_INSN (link);
+ }
+ }
+ }
+#endif
/* Jump is moved to the boundary. */
- prev = BND_TO (bnd);
next = PREV_INSN (insn);
BND_TO (bnd) = insn;
- ft_edge = find_fallthru_edge (block_from);
+ ft_edge = find_fallthru_edge_from (block_from);
block_next = ft_edge->dest;
/* There must be a fallthrough block (or where should go
control flow in case of false jump predicate otherwise?). */
gcc_assert (block_new->next_bb == block_next
&& block_from->next_bb == block_new);
- gcc_assert (BB_END (block_from) == insn);
-
- /* Move all instructions except INSN from BLOCK_FROM to
- BLOCK_NEW. */
- for (link = prev; link != insn; link = NEXT_INSN (link))
+ /* Move all instructions except INSN to BLOCK_NEW. */
+ bb = block_bnd;
+ head = BB_HEAD (block_new);
+ while (bb != block_from->next_bb)
{
- EXPR_ORIG_BB_INDEX (INSN_EXPR (link)) = block_new->index;
- df_insn_change_bb (link, block_new);
- }
+ rtx from, to;
+ from = bb == block_bnd ? prev : sel_bb_head (bb);
+ to = bb == block_from ? next : sel_bb_end (bb);
+
+ /* The jump being moved can be the first insn in the block.
+ In this case we don't have to move anything in this block. */
+ if (NEXT_INSN (to) != from)
+ {
+ reorder_insns (from, to, head);
- /* Set correct basic block and instructions properties. */
- BB_END (block_new) = PREV_INSN (insn);
+ for (link = to; link != head; link = PREV_INSN (link))
+ EXPR_ORIG_BB_INDEX (INSN_EXPR (link)) = block_new->index;
+ head = to;
+ }
- NEXT_INSN (PREV_INSN (prev)) = insn;
- PREV_INSN (insn) = PREV_INSN (prev);
+ /* Cleanup possibly empty blocks left. */
+ block_next = bb->next_bb;
+ if (bb != block_from)
+ tidy_control_flow (bb, false);
+ bb = block_next;
+ }
/* Assert there is no jump to BLOCK_NEW, only fallthrough edge. */
gcc_assert (NOTE_INSN_BASIC_BLOCK_P (BB_HEAD (block_new)));
- PREV_INSN (prev) = BB_HEAD (block_new);
- NEXT_INSN (next) = NEXT_INSN (BB_HEAD (block_new));
- NEXT_INSN (BB_HEAD (block_new)) = prev;
- PREV_INSN (NEXT_INSN (next)) = next;
gcc_assert (!sel_bb_empty_p (block_from)
&& !sel_bb_empty_p (block_new));
int i;
insn_t insn;
- for (i = 0; VEC_iterate (insn_t, vec_temp_moveop_nops, i, insn); i++)
+ FOR_EACH_VEC_ELT (insn_t, vec_temp_moveop_nops, i, insn)
{
gcc_assert (INSN_NOP_P (insn));
return_nop_to_pool (insn, full_tidying);
blist_t *bnds_tailp1, *bndsp;
expr_t expr_vliw;
int need_stall;
- int was_stall = 0, scheduled_insns = 0, stall_iterations = 0;
+ int was_stall = 0, scheduled_insns = 0;
int max_insns = pipelining_p ? issue_rate : 2 * issue_rate;
int max_stall = pipelining_p ? 1 : 3;
bool last_insn_was_debug = false;
do
{
expr_vliw = find_best_expr (&av_vliw, bnds, fence, &need_stall);
- if (!expr_vliw && need_stall)
+ if (! expr_vliw && need_stall)
{
/* All expressions required a stall. Do not recompute av sets
as we'll get the same answer (modulo the insns between
the fence and its boundary, which will not be available for
- pipelining). */
- gcc_assert (! expr_vliw && stall_iterations < 2);
- was_stall++;
- /* If we are going to stall for too long, break to recompute av
+ pipelining).
+ If we are going to stall for too long, break to recompute av
sets and bring more insns for pipelining. */
+ was_stall++;
if (need_stall <= 3)
stall_for_cycles (fence, need_stall);
else
we still need to count it as an originator. */
bitmap_set_bit (current_originators, INSN_UID (insn));
- if (!bitmap_bit_p (current_copies, INSN_UID (insn)))
+ if (!bitmap_clear_bit (current_copies, INSN_UID (insn)))
{
/* Note that original block needs to be rescheduled, as we pulled an
instruction out of it. */
else if (INSN_UID (insn) < first_emitted_uid && !DEBUG_INSN_P (insn))
num_insns_scheduled++;
}
- else
- bitmap_clear_bit (current_copies, INSN_UID (insn));
/* For instructions we must immediately remove insn from the
stream, so subsequent update_data_sets () won't include this
bookkeeping generated for another fence or for another path in current
move_op. */
gcc_assert (res == 1
- || (res == 0
- && av_set_could_be_blocked_by_bookkeeping_p (orig_ops,
+ || (res == 0
+ && av_set_could_be_blocked_by_bookkeeping_p (orig_ops,
static_params))
- || res == -1);
+ || res == -1);
#endif
/* Merge data, clean up, etc. */
init_seqno_1 (succ, visited_bbs, blocks_to_reschedule);
}
+ else if (blocks_to_reschedule)
+ bitmap_set_bit (forced_ebb_heads, succ->index);
}
for (insn = BB_END (bb); insn != note; insn = PREV_INSN (insn))
/* Set hooks so that no newly generated insn will go out unnoticed. */
sel_register_cfg_hooks ();
- /* !!! We call target.sched.md_init () for the whole region, but we invoke
- targetm.sched.md_finish () for every ebb. */
- if (targetm.sched.md_init)
+ /* !!! We call target.sched.init () for the whole region, but we invoke
+ targetm.sched.finish () for every ebb. */
+ if (targetm.sched.init)
/* None of the arguments are actually used in any target. */
- targetm.sched.md_init (sched_dump, sched_verbose, -1);
+ targetm.sched.init (sched_dump, sched_verbose, -1);
first_emitted_uid = get_max_uid () + 1;
preheader_removed = false;
int last_clock = 0;
int haifa_last_clock = -1;
int haifa_clock = 0;
+ int issued_insns = 0;
insn_t insn;
- if (targetm.sched.md_init)
+ if (targetm.sched.init)
{
/* None of the arguments are actually used in any target.
NB: We should have md_reset () hook for cases like this. */
- targetm.sched.md_init (sched_dump, sched_verbose, -1);
+ targetm.sched.init (sched_dump, sched_verbose, -1);
}
state_reset (curr_state);
haifa_cost = cost;
after_stall = 1;
}
-
+ if (haifa_cost == 0
+ && issued_insns == issue_rate)
+ haifa_cost = 1;
if (haifa_cost > 0)
{
int i = 0;
while (haifa_cost--)
{
advance_state (curr_state);
+ issued_insns = 0;
i++;
if (sched_verbose >= 2)
}
haifa_clock += i;
+ if (sched_verbose >= 2)
+ sel_print ("haifa clock: %d\n", haifa_clock);
}
else
gcc_assert (haifa_cost == 0);
&sort_p))
{
advance_state (curr_state);
+ issued_insns = 0;
haifa_clock++;
if (sched_verbose >= 2)
{
sel_print ("advance_state (dfa_new_cycle)\n");
debug_state (curr_state);
+ sel_print ("haifa clock: %d\n", haifa_clock + 1);
}
}
if (real_insn)
{
cost = state_transition (curr_state, insn);
+ issued_insns++;
if (sched_verbose >= 2)
- debug_state (curr_state);
-
+ {
+ sel_print ("scheduled insn %d, clock %d\n", INSN_UID (insn),
+ haifa_clock + 1);
+ debug_state (curr_state);
+ }
gcc_assert (cost < 0);
}
if (reset_sched_cycles_p)
reset_sched_cycles_in_current_ebb ();
- if (targetm.sched.md_init)
- targetm.sched.md_init (sched_dump, sched_verbose, -1);
+ if (targetm.sched.init)
+ targetm.sched.init (sched_dump, sched_verbose, -1);
put_TImodes ();
- if (targetm.sched.md_finish)
+ if (targetm.sched.finish)
{
- targetm.sched.md_finish (sched_dump, sched_verbose);
+ targetm.sched.finish (sched_dump, sched_verbose);
/* Extend luids so that insns generated by the target will
get zero luid. */
/* Mark BB as head of the new ebb. */
bitmap_set_bit (forced_ebb_heads, bb->index);
- bitmap_clear_bit (blocks_to_reschedule, bb->index);
-
gcc_assert (fences == NULL);
init_fences (bb_note (bb));
OSDN Git Service
