/* A single instruction in the partial schedule. */
struct ps_insn
{
- /* The number of the ddg node whose instruction is being scheduled. */
+ /* Identifies the instruction to be scheduled. Values smaller than
+ the ddg's num_nodes refer directly to ddg nodes. A value of
+ X - num_nodes refers to register move X. */
int id;
/* The (absolute) cycle in which the PS instruction is scheduled.
};
+/* Information about a register move that has been added to a partial
+ schedule. */
+struct ps_reg_move_info
+{
+ /* The source of the move is defined by the ps_insn with id DEF.
+ The destination is used by the ps_insns with the ids in USES. */
+ int def;
+ sbitmap uses;
+
+ /* The original form of USES' instructions used OLD_REG, but they
+ should now use NEW_REG. */
+ rtx old_reg;
+ rtx new_reg;
+
+ /* The number of consecutive stages that the move occupies. */
+ int num_consecutive_stages;
+
+ /* An instruction that sets NEW_REG to the correct value. The first
+ move associated with DEF will have an rhs of OLD_REG; later moves
+ use the result of the previous move. */
+ rtx insn;
+};
+
+typedef struct ps_reg_move_info ps_reg_move_info;
+DEF_VEC_O (ps_reg_move_info);
+DEF_VEC_ALLOC_O (ps_reg_move_info, heap);
+
/* Holds the partial schedule as an array of II rows. Each entry of the
array points to a linked list of PS_INSNs, which represents the
instructions that are scheduled for that row. */
/* rows[i] points to linked list of insns scheduled in row i (0<=i<ii). */
ps_insn_ptr *rows;
+ /* All the moves added for this partial schedule. Index X has
+ a ps_insn id of X + g->num_nodes. */
+ VEC (ps_reg_move_info, heap) *reg_moves;
+
/* rows_length[i] holds the number of instructions in the row.
It is used only (as an optimization) to back off quickly from
trying to schedule a node in a full row; that is, to avoid running
static void permute_partial_schedule (partial_schedule_ptr, rtx);
static void generate_prolog_epilog (partial_schedule_ptr, struct loop *,
rtx, rtx);
-static void duplicate_insns_of_cycles (partial_schedule_ptr,
- int, int, int, rtx);
static int calculate_stage_count (partial_schedule_ptr, int);
static void calculate_must_precede_follow (ddg_node_ptr, int, int,
int, int, sbitmap, sbitmap, sbitmap);
#define NODE_ASAP(node) ((node)->aux.count)
-#define SCHED_PARAMS(x) (&node_sched_params[x])
+#define SCHED_PARAMS(x) VEC_index (node_sched_params, node_sched_param_vec, x)
#define SCHED_TIME(x) (SCHED_PARAMS (x)->time)
-#define SCHED_FIRST_REG_MOVE(x) (SCHED_PARAMS (x)->first_reg_move)
-#define SCHED_NREG_MOVES(x) (SCHED_PARAMS (x)->nreg_moves)
#define SCHED_ROW(x) (SCHED_PARAMS (x)->row)
#define SCHED_STAGE(x) (SCHED_PARAMS (x)->stage)
#define SCHED_COLUMN(x) (SCHED_PARAMS (x)->column)
{
int time; /* The absolute scheduling cycle. */
- /* The following field (first_reg_move) is a pointer to the first
- register-move instruction added to handle the modulo-variable-expansion
- of the register defined by this node. This register-move copies the
- original register defined by the node. */
- rtx first_reg_move;
-
- /* The number of register-move instructions added, immediately preceding
- first_reg_move. */
- int nreg_moves;
-
int row; /* Holds time % ii. */
int stage; /* Holds time / ii. */
int column;
} *node_sched_params_ptr;
+typedef struct node_sched_params node_sched_params;
+DEF_VEC_O (node_sched_params);
+DEF_VEC_ALLOC_O (node_sched_params, heap);
\f
/* The following three functions are copied from the current scheduler
code in order to use sched_analyze() for computing the dependencies.
0
};
+/* Partial schedule instruction ID in PS is a register move. Return
+ information about it. */
+static struct ps_reg_move_info *
+ps_reg_move (partial_schedule_ptr ps, int id)
+{
+ gcc_checking_assert (id >= ps->g->num_nodes);
+ return VEC_index (ps_reg_move_info, ps->reg_moves, id - ps->g->num_nodes);
+}
+
/* Return the rtl instruction that is being scheduled by partial schedule
instruction ID, which belongs to schedule PS. */
static rtx
ps_rtl_insn (partial_schedule_ptr ps, int id)
{
- return ps->g->nodes[id].insn;
+ if (id < ps->g->num_nodes)
+ return ps->g->nodes[id].insn;
+ else
+ return ps_reg_move (ps, id)->insn;
}
-/* Return the first instruction in the original (unscheduled) loop that
- was associated with ps_rtl_insn (PS, ID). If the instruction had
- some notes before it, this is the first of those notes. */
+/* Partial schedule instruction ID, which belongs to PS, occured in
+ the original (unscheduled) loop. Return the first instruction
+ in the loop that was associated with ps_rtl_insn (PS, ID).
+ If the instruction had some notes before it, this is the first
+ of those notes. */
static rtx
ps_first_note (partial_schedule_ptr ps, int id)
{
+ gcc_assert (id < ps->g->num_nodes);
return ps->g->nodes[id].first_note;
}
+/* Return the number of consecutive stages that are occupied by
+ partial schedule instruction ID in PS. */
+static int
+ps_num_consecutive_stages (partial_schedule_ptr ps, int id)
+{
+ if (id < ps->g->num_nodes)
+ return 1;
+ else
+ return ps_reg_move (ps, id)->num_consecutive_stages;
+}
+
/* Given HEAD and TAIL which are the first and last insns in a loop;
return the register which controls the loop. Return zero if it has
more than one occurrence in the loop besides the control part or the
}
-/* Points to the array that contains the sched data for each node. */
-static node_sched_params_ptr node_sched_params;
+/* A vector that contains the sched data for each ps_insn. */
+static VEC (node_sched_params, heap) *node_sched_param_vec;
/* Allocate sched_params for each node and initialize it. */
static void
set_node_sched_params (ddg_ptr g)
{
- node_sched_params = XCNEWVEC (struct node_sched_params, g->num_nodes);
+ VEC_truncate (node_sched_params, node_sched_param_vec, 0);
+ VEC_safe_grow_cleared (node_sched_params, heap,
+ node_sched_param_vec, g->num_nodes);
}
+/* Make sure that node_sched_param_vec has an entry for every move in PS. */
+static void
+extend_node_sched_params (partial_schedule_ptr ps)
+{
+ VEC_safe_grow_cleared (node_sched_params, heap, node_sched_param_vec,
+ ps->g->num_nodes + VEC_length (ps_reg_move_info,
+ ps->reg_moves));
+}
+
+/* Update the sched_params (time, row and stage) for node U using the II,
+ the CYCLE of U and MIN_CYCLE.
+ We're not simply taking the following
+ SCHED_STAGE (u) = CALC_STAGE_COUNT (SCHED_TIME (u), min_cycle, ii);
+ because the stages may not be aligned on cycle 0. */
static void
-print_node_sched_params (FILE *file, int num_nodes, ddg_ptr g)
+update_node_sched_params (int u, int ii, int cycle, int min_cycle)
+{
+ int sc_until_cycle_zero;
+ int stage;
+
+ SCHED_TIME (u) = cycle;
+ SCHED_ROW (u) = SMODULO (cycle, ii);
+
+ /* The calculation of stage count is done adding the number
+ of stages before cycle zero and after cycle zero. */
+ sc_until_cycle_zero = CALC_STAGE_COUNT (-1, min_cycle, ii);
+
+ if (SCHED_TIME (u) < 0)
+ {
+ stage = CALC_STAGE_COUNT (-1, SCHED_TIME (u), ii);
+ SCHED_STAGE (u) = sc_until_cycle_zero - stage;
+ }
+ else
+ {
+ stage = CALC_STAGE_COUNT (SCHED_TIME (u), 0, ii);
+ SCHED_STAGE (u) = sc_until_cycle_zero + stage - 1;
+ }
+}
+
+static void
+print_node_sched_params (FILE *file, int num_nodes, partial_schedule_ptr ps)
{
int i;
for (i = 0; i < num_nodes; i++)
{
node_sched_params_ptr nsp = SCHED_PARAMS (i);
- rtx reg_move = nsp->first_reg_move;
- int j;
fprintf (file, "Node = %d; INSN = %d\n", i,
- (INSN_UID (g->nodes[i].insn)));
- fprintf (file, " asap = %d:\n", NODE_ASAP (&g->nodes[i]));
+ INSN_UID (ps_rtl_insn (ps, i)));
+ fprintf (file, " asap = %d:\n", NODE_ASAP (&ps->g->nodes[i]));
fprintf (file, " time = %d:\n", nsp->time);
- fprintf (file, " nreg_moves = %d:\n", nsp->nreg_moves);
- for (j = 0; j < nsp->nreg_moves; j++)
+ fprintf (file, " stage = %d:\n", nsp->stage);
+ }
+}
+
+/* Set SCHED_COLUMN for each instruction in row ROW of PS. */
+static void
+set_columns_for_row (partial_schedule_ptr ps, int row)
+{
+ ps_insn_ptr cur_insn;
+ int column;
+
+ column = 0;
+ for (cur_insn = ps->rows[row]; cur_insn; cur_insn = cur_insn->next_in_row)
+ SCHED_COLUMN (cur_insn->id) = column++;
+}
+
+/* Set SCHED_COLUMN for each instruction in PS. */
+static void
+set_columns_for_ps (partial_schedule_ptr ps)
+{
+ int row;
+
+ for (row = 0; row < ps->ii; row++)
+ set_columns_for_row (ps, row);
+}
+
+/* Try to schedule the move with ps_insn identifier I_REG_MOVE in PS.
+ Its single predecessor has already been scheduled, as has its
+ ddg node successors. (The move may have also another move as its
+ successor, in which case that successor will be scheduled later.)
+
+ The move is part of a chain that satisfies register dependencies
+ between a producing ddg node and various consuming ddg nodes.
+ If some of these dependencies have a distance of 1 (meaning that
+ the use is upward-exposed) then DISTANCE1_USES is nonnull and
+ contains the set of uses with distance-1 dependencies.
+ DISTANCE1_USES is null otherwise.
+
+ MUST_FOLLOW is a scratch bitmap that is big enough to hold
+ all current ps_insn ids.
+
+ Return true on success. */
+static bool
+schedule_reg_move (partial_schedule_ptr ps, int i_reg_move,
+ sbitmap distance1_uses, sbitmap must_follow)
+{
+ unsigned int u;
+ int this_time, this_distance, this_start, this_end, this_latency;
+ int start, end, c, ii;
+ sbitmap_iterator sbi;
+ ps_reg_move_info *move;
+ rtx this_insn;
+ ps_insn_ptr psi;
+
+ move = ps_reg_move (ps, i_reg_move);
+ ii = ps->ii;
+ if (dump_file)
+ {
+ fprintf (dump_file, "Scheduling register move INSN %d; ii = %d"
+ ", min cycle = %d\n\n", INSN_UID (move->insn), ii,
+ PS_MIN_CYCLE (ps));
+ print_rtl_single (dump_file, move->insn);
+ fprintf (dump_file, "\n%11s %11s %5s\n", "start", "end", "time");
+ fprintf (dump_file, "=========== =========== =====\n");
+ }
+
+ start = INT_MIN;
+ end = INT_MAX;
+
+ /* For dependencies of distance 1 between a producer ddg node A
+ and consumer ddg node B, we have a chain of dependencies:
+
+ A --(T,L1,1)--> M1 --(T,L2,0)--> M2 ... --(T,Ln,0)--> B
+
+ where Mi is the ith move. For dependencies of distance 0 between
+ a producer ddg node A and consumer ddg node C, we have a chain of
+ dependencies:
+
+ A --(T,L1',0)--> M1' --(T,L2',0)--> M2' ... --(T,Ln',0)--> C
+
+ where Mi' occupies the same position as Mi but occurs a stage later.
+ We can only schedule each move once, so if we have both types of
+ chain, we model the second as:
+
+ A --(T,L1',1)--> M1 --(T,L2',0)--> M2 ... --(T,Ln',-1)--> C
+
+ First handle the dependencies between the previously-scheduled
+ predecessor and the move. */
+ this_insn = ps_rtl_insn (ps, move->def);
+ this_latency = insn_latency (this_insn, move->insn);
+ this_distance = distance1_uses && move->def < ps->g->num_nodes ? 1 : 0;
+ this_time = SCHED_TIME (move->def) - this_distance * ii;
+ this_start = this_time + this_latency;
+ this_end = this_time + ii;
+ if (dump_file)
+ fprintf (dump_file, "%11d %11d %5d %d --(T,%d,%d)--> %d\n",
+ this_start, this_end, SCHED_TIME (move->def),
+ INSN_UID (this_insn), this_latency, this_distance,
+ INSN_UID (move->insn));
+
+ if (start < this_start)
+ start = this_start;
+ if (end > this_end)
+ end = this_end;
+
+ /* Handle the dependencies between the move and previously-scheduled
+ successors. */
+ EXECUTE_IF_SET_IN_SBITMAP (move->uses, 0, u, sbi)
+ {
+ this_insn = ps_rtl_insn (ps, u);
+ this_latency = insn_latency (move->insn, this_insn);
+ if (distance1_uses && !TEST_BIT (distance1_uses, u))
+ this_distance = -1;
+ else
+ this_distance = 0;
+ this_time = SCHED_TIME (u) + this_distance * ii;
+ this_start = this_time - ii;
+ this_end = this_time - this_latency;
+ if (dump_file)
+ fprintf (dump_file, "%11d %11d %5d %d --(T,%d,%d)--> %d\n",
+ this_start, this_end, SCHED_TIME (u), INSN_UID (move->insn),
+ this_latency, this_distance, INSN_UID (this_insn));
+
+ if (start < this_start)
+ start = this_start;
+ if (end > this_end)
+ end = this_end;
+ }
+
+ if (dump_file)
+ {
+ fprintf (dump_file, "----------- ----------- -----\n");
+ fprintf (dump_file, "%11d %11d %5s %s\n", start, end, "", "(max, min)");
+ }
+
+ sbitmap_zero (must_follow);
+ SET_BIT (must_follow, move->def);
+
+ start = MAX (start, end - (ii - 1));
+ for (c = end; c >= start; c--)
+ {
+ psi = ps_add_node_check_conflicts (ps, i_reg_move, c,
+ move->uses, must_follow);
+ if (psi)
{
- fprintf (file, " reg_move = ");
- print_rtl_single (file, reg_move);
- reg_move = PREV_INSN (reg_move);
+ update_node_sched_params (i_reg_move, ii, c, PS_MIN_CYCLE (ps));
+ if (dump_file)
+ fprintf (dump_file, "\nScheduled register move INSN %d at"
+ " time %d, row %d\n\n", INSN_UID (move->insn), c,
+ SCHED_ROW (i_reg_move));
+ return true;
}
}
+
+ if (dump_file)
+ fprintf (dump_file, "\nNo available slot\n\n");
+
+ return false;
}
/*
nreg_moves = ----------------------------------- + 1 - { dependence.
ii { 1 if not.
*/
-static void
-generate_reg_moves (partial_schedule_ptr ps, bool rescan)
+static bool
+schedule_reg_moves (partial_schedule_ptr ps)
{
ddg_ptr g = ps->g;
int ii = ps->ii;
ddg_node_ptr u = &g->nodes[i];
ddg_edge_ptr e;
int nreg_moves = 0, i_reg_move;
- sbitmap *uses_of_defs;
- rtx last_reg_move;
rtx prev_reg, old_reg;
+ int first_move;
+ int distances[2];
+ sbitmap must_follow;
+ sbitmap distance1_uses;
rtx set = single_set (u->insn);
/* Skip instructions that do not set a register. */
/* Compute the number of reg_moves needed for u, by looking at life
ranges started at u (excluding self-loops). */
+ distances[0] = distances[1] = false;
for (e = u->out; e; e = e->next_out)
if (e->type == TRUE_DEP && e->dest != e->src)
{
gcc_assert (!autoinc_var_is_used_p (u->insn, e->dest->insn));
}
+ if (nreg_moves4e)
+ {
+ gcc_assert (e->distance < 2);
+ distances[e->distance] = true;
+ }
nreg_moves = MAX (nreg_moves, nreg_moves4e);
}
if (nreg_moves == 0)
continue;
+ /* Create NREG_MOVES register moves. */
+ first_move = VEC_length (ps_reg_move_info, ps->reg_moves);
+ VEC_safe_grow_cleared (ps_reg_move_info, heap, ps->reg_moves,
+ first_move + nreg_moves);
+ extend_node_sched_params (ps);
+
+ /* Record the moves associated with this node. */
+ first_move += ps->g->num_nodes;
+
+ /* Generate each move. */
+ old_reg = prev_reg = SET_DEST (single_set (u->insn));
+ for (i_reg_move = 0; i_reg_move < nreg_moves; i_reg_move++)
+ {
+ ps_reg_move_info *move = ps_reg_move (ps, first_move + i_reg_move);
+
+ move->def = i_reg_move > 0 ? first_move + i_reg_move - 1 : i;
+ move->uses = sbitmap_alloc (first_move + nreg_moves);
+ move->old_reg = old_reg;
+ move->new_reg = gen_reg_rtx (GET_MODE (prev_reg));
+ move->num_consecutive_stages = distances[0] && distances[1] ? 2 : 1;
+ move->insn = gen_move_insn (move->new_reg, copy_rtx (prev_reg));
+ sbitmap_zero (move->uses);
+
+ prev_reg = move->new_reg;
+ }
+
+ distance1_uses = distances[1] ? sbitmap_alloc (g->num_nodes) : NULL;
+
/* Every use of the register defined by node may require a different
copy of this register, depending on the time the use is scheduled.
- Set a bitmap vector, telling which nodes use each copy of this
- register. */
- uses_of_defs = sbitmap_vector_alloc (nreg_moves, g->num_nodes);
- sbitmap_vector_zero (uses_of_defs, nreg_moves);
+ Record which uses require which move results. */
for (e = u->out; e; e = e->next_out)
if (e->type == TRUE_DEP && e->dest != e->src)
{
dest_copy--;
if (dest_copy)
- SET_BIT (uses_of_defs[dest_copy - 1], e->dest->cuid);
- }
+ {
+ ps_reg_move_info *move;
- /* Now generate the reg_moves, attaching relevant uses to them. */
- SCHED_NREG_MOVES (i) = nreg_moves;
- old_reg = prev_reg = copy_rtx (SET_DEST (single_set (u->insn)));
- /* Insert the reg-moves right before the notes which precede
- the insn they relates to. */
- last_reg_move = u->first_note;
+ move = ps_reg_move (ps, first_move + dest_copy - 1);
+ SET_BIT (move->uses, e->dest->cuid);
+ if (e->distance == 1)
+ SET_BIT (distance1_uses, e->dest->cuid);
+ }
+ }
+ must_follow = sbitmap_alloc (first_move + nreg_moves);
for (i_reg_move = 0; i_reg_move < nreg_moves; i_reg_move++)
- {
- unsigned int i_use = 0;
- rtx new_reg = gen_reg_rtx (GET_MODE (prev_reg));
- rtx reg_move = gen_move_insn (new_reg, prev_reg);
- sbitmap_iterator sbi;
-
- add_insn_before (reg_move, last_reg_move, NULL);
- last_reg_move = reg_move;
-
- if (!SCHED_FIRST_REG_MOVE (i))
- SCHED_FIRST_REG_MOVE (i) = reg_move;
-
- EXECUTE_IF_SET_IN_SBITMAP (uses_of_defs[i_reg_move], 0, i_use, sbi)
- {
- replace_rtx (g->nodes[i_use].insn, old_reg, new_reg);
- if (rescan)
- df_insn_rescan (g->nodes[i_use].insn);
- }
-
- prev_reg = new_reg;
- }
- sbitmap_vector_free (uses_of_defs);
+ if (!schedule_reg_move (ps, first_move + i_reg_move,
+ distance1_uses, must_follow))
+ break;
+ sbitmap_free (must_follow);
+ if (distance1_uses)
+ sbitmap_free (distance1_uses);
+ if (i_reg_move < nreg_moves)
+ return false;
}
+ return true;
}
-/* Update the sched_params (time, row and stage) for node U using the II,
- the CYCLE of U and MIN_CYCLE.
- We're not simply taking the following
- SCHED_STAGE (u) = CALC_STAGE_COUNT (SCHED_TIME (u), min_cycle, ii);
- because the stages may not be aligned on cycle 0. */
+/* Emit the moves associatied with PS. Apply the substitutions
+ associated with them. */
static void
-update_node_sched_params (int u, int ii, int cycle, int min_cycle)
+apply_reg_moves (partial_schedule_ptr ps)
{
- int sc_until_cycle_zero;
- int stage;
-
- SCHED_TIME (u) = cycle;
- SCHED_ROW (u) = SMODULO (cycle, ii);
-
- /* The calculation of stage count is done adding the number
- of stages before cycle zero and after cycle zero. */
- sc_until_cycle_zero = CALC_STAGE_COUNT (-1, min_cycle, ii);
+ ps_reg_move_info *move;
+ int i;
- if (SCHED_TIME (u) < 0)
- {
- stage = CALC_STAGE_COUNT (-1, SCHED_TIME (u), ii);
- SCHED_STAGE (u) = sc_until_cycle_zero - stage;
- }
- else
+ FOR_EACH_VEC_ELT (ps_reg_move_info, ps->reg_moves, i, move)
{
- stage = CALC_STAGE_COUNT (SCHED_TIME (u), 0, ii);
- SCHED_STAGE (u) = sc_until_cycle_zero + stage - 1;
+ unsigned int i_use;
+ sbitmap_iterator sbi;
+
+ EXECUTE_IF_SET_IN_SBITMAP (move->uses, 0, i_use, sbi)
+ {
+ replace_rtx (ps->g->nodes[i_use].insn, move->old_reg, move->new_reg);
+ df_insn_rescan (ps->g->nodes[i_use].insn);
+ }
}
}
}
}
-/* Set SCHED_COLUMN of each node according to its position in PS. */
-static void
-set_columns_for_ps (partial_schedule_ptr ps)
-{
- int row;
-
- for (row = 0; row < ps->ii; row++)
- {
- ps_insn_ptr cur_insn = ps->rows[row];
- int column = 0;
-
- for (; cur_insn; cur_insn = cur_insn->next_in_row)
- SCHED_COLUMN (cur_insn->id) = column++;
- }
-}
-
/* Permute the insns according to their order in PS, from row 0 to
row ii-1, and position them right before LAST. This schedules
the insns of the loop kernel. */
rtx insn = ps_rtl_insn (ps, ps_ij->id);
if (PREV_INSN (last) != insn)
- reorder_insns_nobb (ps_first_note (ps, ps_ij->id), insn,
- PREV_INSN (last));
+ {
+ if (ps_ij->id < ps->g->num_nodes)
+ reorder_insns_nobb (ps_first_note (ps, ps_ij->id), insn,
+ PREV_INSN (last));
+ else
+ add_insn_before (insn, last, NULL);
+ }
}
}
static void
duplicate_insns_of_cycles (partial_schedule_ptr ps, int from_stage,
- int to_stage, int for_prolog, rtx count_reg)
+ int to_stage, rtx count_reg)
{
int row;
ps_insn_ptr ps_ij;
for (ps_ij = ps->rows[row]; ps_ij; ps_ij = ps_ij->next_in_row)
{
int u = ps_ij->id;
- int j, i_reg_moves;
- rtx reg_move = NULL_RTX;
+ int first_u, last_u;
rtx u_insn;
/* Do not duplicate any insn which refers to count_reg as it
|| JUMP_P (u_insn))
continue;
- if (for_prolog)
- {
- /* SCHED_STAGE (u) >= from_stage == 0. Generate increasing
- number of reg_moves starting with the second occurrence of
- u, which is generated if its SCHED_STAGE <= to_stage. */
- i_reg_moves = to_stage - SCHED_STAGE (u) + 1;
- i_reg_moves = MAX (i_reg_moves, 0);
- i_reg_moves = MIN (i_reg_moves, SCHED_NREG_MOVES (u));
-
- /* The reg_moves start from the *first* reg_move backwards. */
- if (i_reg_moves)
- {
- reg_move = SCHED_FIRST_REG_MOVE (u);
- for (j = 1; j < i_reg_moves; j++)
- reg_move = PREV_INSN (reg_move);
- }
- }
- else /* It's for the epilog. */
+ first_u = SCHED_STAGE (u);
+ last_u = first_u + ps_num_consecutive_stages (ps, u) - 1;
+ if (from_stage <= last_u && to_stage >= first_u)
{
- /* SCHED_STAGE (u) <= to_stage. Generate all reg_moves,
- starting to decrease one stage after u no longer occurs;
- that is, generate all reg_moves until
- SCHED_STAGE (u) == from_stage - 1. */
- i_reg_moves = (SCHED_NREG_MOVES (u)
- - (from_stage - SCHED_STAGE (u) - 1));
- i_reg_moves = MAX (i_reg_moves, 0);
- i_reg_moves = MIN (i_reg_moves, SCHED_NREG_MOVES (u));
-
- /* The reg_moves start from the *last* reg_move forwards. */
- if (i_reg_moves)
- {
- reg_move = SCHED_FIRST_REG_MOVE (u);
- for (j = 1; j < SCHED_NREG_MOVES (u); j++)
- reg_move = PREV_INSN (reg_move);
- }
+ if (u < ps->g->num_nodes)
+ duplicate_insn_chain (ps_first_note (ps, u), u_insn);
+ else
+ emit_insn (copy_rtx (PATTERN (u_insn)));
}
-
- for (j = 0; j < i_reg_moves; j++, reg_move = NEXT_INSN (reg_move))
- emit_insn (copy_rtx (PATTERN (reg_move)));
- if (SCHED_STAGE (u) >= from_stage
- && SCHED_STAGE (u) <= to_stage)
- duplicate_insn_chain (ps_first_note (ps, u), u_insn);
}
}
}
for (i = 0; i < last_stage; i++)
- duplicate_insns_of_cycles (ps, 0, i, 1, count_reg);
+ duplicate_insns_of_cycles (ps, 0, i, count_reg);
/* Put the prolog on the entry edge. */
e = loop_preheader_edge (loop);
split_edge_and_insert (e, get_insns ());
+ if (!flag_resched_modulo_sched)
+ e->dest->flags |= BB_DISABLE_SCHEDULE;
end_sequence ();
start_sequence ();
for (i = 0; i < last_stage; i++)
- duplicate_insns_of_cycles (ps, i + 1, last_stage, 0, count_reg);
+ duplicate_insns_of_cycles (ps, i + 1, last_stage, count_reg);
/* Put the epilogue on the exit edge. */
gcc_assert (single_exit (loop));
e = single_exit (loop);
split_edge_and_insert (e, get_insns ());
+ if (!flag_resched_modulo_sched)
+ e->dest->flags |= BB_DISABLE_SCHEDULE;
+
end_sequence ();
}
+/* Mark LOOP as software pipelined so the later
+ scheduling passes don't touch it. */
+static void
+mark_loop_unsched (struct loop *loop)
+{
+ unsigned i;
+ basic_block *bbs = get_loop_body (loop);
+
+ for (i = 0; i < loop->num_nodes; i++)
+ bbs[i]->flags |= BB_DISABLE_SCHEDULE;
+}
+
/* Return true if all the BBs of the loop are empty except the
loop header. */
static bool
{
rtx head, tail;
rtx count_reg, count_init;
- int mii, rec_mii;
- unsigned stage_count = 0;
+ int mii, rec_mii, stage_count, min_cycle;
HOST_WIDEST_INT loop_count = 0;
- bool opt_sc_p = false;
+ bool opt_sc_p;
if (! (g = g_arr[loop->num]))
continue;
fprintf (dump_file, "SMS iis %d %d %d (rec_mii, mii, maxii)\n",
rec_mii, mii, maxii);
- set_node_sched_params (g);
-
- ps = sms_schedule_by_order (g, mii, maxii, node_order);
-
- if (ps)
+ for (;;)
{
- /* Try to achieve optimized SC by normalizing the partial
- schedule (having the cycles start from cycle zero).
- The branch location must be placed in row ii-1 in the
- final scheduling. If failed, shift all instructions to
- position the branch in row ii-1. */
- opt_sc_p = optimize_sc (ps, g);
- if (opt_sc_p)
- stage_count = calculate_stage_count (ps, 0);
- else
+ set_node_sched_params (g);
+
+ stage_count = 0;
+ opt_sc_p = false;
+ ps = sms_schedule_by_order (g, mii, maxii, node_order);
+
+ if (ps)
{
- /* Bring the branch to cycle ii-1. */
- int amount = SCHED_TIME (g->closing_branch->cuid) - (ps->ii - 1);
-
- if (dump_file)
- fprintf (dump_file, "SMS schedule branch at cycle ii-1\n");
-
- stage_count = calculate_stage_count (ps, amount);
+ /* Try to achieve optimized SC by normalizing the partial
+ schedule (having the cycles start from cycle zero).
+ The branch location must be placed in row ii-1 in the
+ final scheduling. If failed, shift all instructions to
+ position the branch in row ii-1. */
+ opt_sc_p = optimize_sc (ps, g);
+ if (opt_sc_p)
+ stage_count = calculate_stage_count (ps, 0);
+ else
+ {
+ /* Bring the branch to cycle ii-1. */
+ int amount = (SCHED_TIME (g->closing_branch->cuid)
+ - (ps->ii - 1));
+
+ if (dump_file)
+ fprintf (dump_file, "SMS schedule branch at cycle ii-1\n");
+
+ stage_count = calculate_stage_count (ps, amount);
+ }
+
+ gcc_assert (stage_count >= 1);
}
-
- gcc_assert (stage_count >= 1);
- PS_STAGE_COUNT (ps) = stage_count;
- }
-
- /* The default value of PARAM_SMS_MIN_SC is 2 as stage count of
- 1 means that there is no interleaving between iterations thus
- we let the scheduling passes do the job in this case. */
- if (stage_count < (unsigned) PARAM_VALUE (PARAM_SMS_MIN_SC)
- || (count_init && (loop_count <= stage_count))
- || (flag_branch_probabilities && (trip_count <= stage_count)))
- {
- if (dump_file)
+
+ /* The default value of PARAM_SMS_MIN_SC is 2 as stage count of
+ 1 means that there is no interleaving between iterations thus
+ we let the scheduling passes do the job in this case. */
+ if (stage_count < PARAM_VALUE (PARAM_SMS_MIN_SC)
+ || (count_init && (loop_count <= stage_count))
+ || (flag_branch_probabilities && (trip_count <= stage_count)))
{
- fprintf (dump_file, "SMS failed... \n");
- fprintf (dump_file, "SMS sched-failed (stage-count=%d, loop-count=", stage_count);
- fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC, loop_count);
- fprintf (dump_file, ", trip-count=");
- fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC, trip_count);
- fprintf (dump_file, ")\n");
+ if (dump_file)
+ {
+ fprintf (dump_file, "SMS failed... \n");
+ fprintf (dump_file, "SMS sched-failed (stage-count=%d,"
+ " loop-count=", stage_count);
+ fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC, loop_count);
+ fprintf (dump_file, ", trip-count=");
+ fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC, trip_count);
+ fprintf (dump_file, ")\n");
+ }
+ break;
}
- }
- else
- {
+
if (!opt_sc_p)
{
/* Rotate the partial schedule to have the branch in row ii-1. */
set_columns_for_ps (ps);
+ min_cycle = PS_MIN_CYCLE (ps) - SMODULO (PS_MIN_CYCLE (ps), ps->ii);
+ if (!schedule_reg_moves (ps))
+ {
+ mii = ps->ii + 1;
+ free_partial_schedule (ps);
+ continue;
+ }
+
+ /* Moves that handle incoming values might have been added
+ to a new first stage. Bump the stage count if so.
+
+ ??? Perhaps we could consider rotating the schedule here
+ instead? */
+ if (PS_MIN_CYCLE (ps) < min_cycle)
+ {
+ reset_sched_times (ps, 0);
+ stage_count++;
+ }
+
+ /* The stage count should now be correct without rotation. */
+ gcc_checking_assert (stage_count == calculate_stage_count (ps, 0));
+ PS_STAGE_COUNT (ps) = stage_count;
+
canon_loop (loop);
if (dump_file)
permute_partial_schedule (ps, g->closing_branch->first_note);
/* Mark this loop as software pipelined so the later
- scheduling passes doesn't touch it. */
+ scheduling passes don't touch it. */
if (! flag_resched_modulo_sched)
- g->bb->flags |= BB_DISABLE_SCHEDULE;
+ mark_loop_unsched (loop);
+
/* The life-info is not valid any more. */
df_set_bb_dirty (g->bb);
- generate_reg_moves (ps, true);
+ apply_reg_moves (ps);
if (dump_file)
- print_node_sched_params (dump_file, g->num_nodes, g);
+ print_node_sched_params (dump_file, g->num_nodes, ps);
/* Generate prolog and epilog. */
generate_prolog_epilog (ps, loop, count_reg, count_init);
+ break;
}
free_partial_schedule (ps);
- free (node_sched_params);
+ VEC_free (node_sched_params, heap, node_sched_param_vec);
free (node_order);
free_ddg (g);
}
41. endif
42. compute epilogue & prologue
43. finish - succeeded to schedule
-*/
+
+ ??? The algorithm restricts the scheduling window to II cycles.
+ In rare cases, it may be better to allow windows of II+1 cycles.
+ The window would then start and end on the same row, but with
+ different "must precede" and "must follow" requirements. */
/* A limit on the number of cycles that resource conflicts can span. ??? Should
be provided by DFA, and be dependent on the type of insn scheduled. Currently
partial_schedule_ptr ps = XNEW (struct partial_schedule);
ps->rows = (ps_insn_ptr *) xcalloc (ii, sizeof (ps_insn_ptr));
ps->rows_length = (int *) xcalloc (ii, sizeof (int));
+ ps->reg_moves = NULL;
ps->ii = ii;
ps->history = history;
ps->min_cycle = INT_MAX;
static void
free_partial_schedule (partial_schedule_ptr ps)
{
+ ps_reg_move_info *move;
+ unsigned int i;
+
if (!ps)
return;
+
+ FOR_EACH_VEC_ELT (ps_reg_move_info, ps->reg_moves, i, move)
+ sbitmap_free (move->uses);
+ VEC_free (ps_reg_move_info, heap, ps->reg_moves);
+
free_ps_insns (ps);
free (ps->rows);
free (ps->rows_length);