/* Swing Modulo Scheduling implementation.
|
/* Swing Modulo Scheduling implementation.
|
Copyright (C) 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012
|
Copyright (C) 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012
|
Free Software Foundation, Inc.
|
Free Software Foundation, Inc.
|
Contributed by Ayal Zaks and Mustafa Hagog <zaks,mustafa@il.ibm.com>
|
Contributed by Ayal Zaks and Mustafa Hagog <zaks,mustafa@il.ibm.com>
|
|
|
This file is part of GCC.
|
This file is part of GCC.
|
|
|
GCC is free software; you can redistribute it and/or modify it under
|
GCC is free software; you can redistribute it and/or modify it under
|
the terms of the GNU General Public License as published by the Free
|
the terms of the GNU General Public License as published by the Free
|
Software Foundation; either version 3, or (at your option) any later
|
Software Foundation; either version 3, or (at your option) any later
|
version.
|
version.
|
|
|
GCC is distributed in the hope that it will be useful, but WITHOUT ANY
|
GCC is distributed in the hope that it will be useful, but WITHOUT ANY
|
WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
for more details.
|
for more details.
|
|
|
You should have received a copy of the GNU General Public License
|
You should have received a copy of the GNU General Public License
|
along with GCC; see the file COPYING3. If not see
|
along with GCC; see the file COPYING3. If not see
|
<http://www.gnu.org/licenses/>. */
|
<http://www.gnu.org/licenses/>. */
|
|
|
|
|
#include "config.h"
|
#include "config.h"
|
#include "system.h"
|
#include "system.h"
|
#include "coretypes.h"
|
#include "coretypes.h"
|
#include "tm.h"
|
#include "tm.h"
|
#include "diagnostic-core.h"
|
#include "diagnostic-core.h"
|
#include "rtl.h"
|
#include "rtl.h"
|
#include "tm_p.h"
|
#include "tm_p.h"
|
#include "hard-reg-set.h"
|
#include "hard-reg-set.h"
|
#include "regs.h"
|
#include "regs.h"
|
#include "function.h"
|
#include "function.h"
|
#include "flags.h"
|
#include "flags.h"
|
#include "insn-config.h"
|
#include "insn-config.h"
|
#include "insn-attr.h"
|
#include "insn-attr.h"
|
#include "except.h"
|
#include "except.h"
|
#include "recog.h"
|
#include "recog.h"
|
#include "sched-int.h"
|
#include "sched-int.h"
|
#include "target.h"
|
#include "target.h"
|
#include "cfglayout.h"
|
#include "cfglayout.h"
|
#include "cfgloop.h"
|
#include "cfgloop.h"
|
#include "cfghooks.h"
|
#include "cfghooks.h"
|
#include "expr.h"
|
#include "expr.h"
|
#include "params.h"
|
#include "params.h"
|
#include "gcov-io.h"
|
#include "gcov-io.h"
|
#include "ddg.h"
|
#include "ddg.h"
|
#include "timevar.h"
|
#include "timevar.h"
|
#include "tree-pass.h"
|
#include "tree-pass.h"
|
#include "dbgcnt.h"
|
#include "dbgcnt.h"
|
#include "df.h"
|
#include "df.h"
|
|
|
#ifdef INSN_SCHEDULING
|
#ifdef INSN_SCHEDULING
|
|
|
/* This file contains the implementation of the Swing Modulo Scheduler,
|
/* This file contains the implementation of the Swing Modulo Scheduler,
|
described in the following references:
|
described in the following references:
|
[1] J. Llosa, A. Gonzalez, E. Ayguade, M. Valero., and J. Eckhardt.
|
[1] J. Llosa, A. Gonzalez, E. Ayguade, M. Valero., and J. Eckhardt.
|
Lifetime--sensitive modulo scheduling in a production environment.
|
Lifetime--sensitive modulo scheduling in a production environment.
|
IEEE Trans. on Comps., 50(3), March 2001
|
IEEE Trans. on Comps., 50(3), March 2001
|
[2] J. Llosa, A. Gonzalez, E. Ayguade, and M. Valero.
|
[2] J. Llosa, A. Gonzalez, E. Ayguade, and M. Valero.
|
Swing Modulo Scheduling: A Lifetime Sensitive Approach.
|
Swing Modulo Scheduling: A Lifetime Sensitive Approach.
|
PACT '96 , pages 80-87, October 1996 (Boston - Massachusetts - USA).
|
PACT '96 , pages 80-87, October 1996 (Boston - Massachusetts - USA).
|
|
|
The basic structure is:
|
The basic structure is:
|
1. Build a data-dependence graph (DDG) for each loop.
|
1. Build a data-dependence graph (DDG) for each loop.
|
2. Use the DDG to order the insns of a loop (not in topological order
|
2. Use the DDG to order the insns of a loop (not in topological order
|
necessarily, but rather) trying to place each insn after all its
|
necessarily, but rather) trying to place each insn after all its
|
predecessors _or_ after all its successors.
|
predecessors _or_ after all its successors.
|
3. Compute MII: a lower bound on the number of cycles to schedule the loop.
|
3. Compute MII: a lower bound on the number of cycles to schedule the loop.
|
4. Use the ordering to perform list-scheduling of the loop:
|
4. Use the ordering to perform list-scheduling of the loop:
|
1. Set II = MII. We will try to schedule the loop within II cycles.
|
1. Set II = MII. We will try to schedule the loop within II cycles.
|
2. Try to schedule the insns one by one according to the ordering.
|
2. Try to schedule the insns one by one according to the ordering.
|
For each insn compute an interval of cycles by considering already-
|
For each insn compute an interval of cycles by considering already-
|
scheduled preds and succs (and associated latencies); try to place
|
scheduled preds and succs (and associated latencies); try to place
|
the insn in the cycles of this window checking for potential
|
the insn in the cycles of this window checking for potential
|
resource conflicts (using the DFA interface).
|
resource conflicts (using the DFA interface).
|
Note: this is different from the cycle-scheduling of schedule_insns;
|
Note: this is different from the cycle-scheduling of schedule_insns;
|
here the insns are not scheduled monotonically top-down (nor bottom-
|
here the insns are not scheduled monotonically top-down (nor bottom-
|
up).
|
up).
|
3. If failed in scheduling all insns - bump II++ and try again, unless
|
3. If failed in scheduling all insns - bump II++ and try again, unless
|
II reaches an upper bound MaxII, in which case report failure.
|
II reaches an upper bound MaxII, in which case report failure.
|
5. If we succeeded in scheduling the loop within II cycles, we now
|
5. If we succeeded in scheduling the loop within II cycles, we now
|
generate prolog and epilog, decrease the counter of the loop, and
|
generate prolog and epilog, decrease the counter of the loop, and
|
perform modulo variable expansion for live ranges that span more than
|
perform modulo variable expansion for live ranges that span more than
|
II cycles (i.e. use register copies to prevent a def from overwriting
|
II cycles (i.e. use register copies to prevent a def from overwriting
|
itself before reaching the use).
|
itself before reaching the use).
|
|
|
SMS works with countable loops (1) whose control part can be easily
|
SMS works with countable loops (1) whose control part can be easily
|
decoupled from the rest of the loop and (2) whose loop count can
|
decoupled from the rest of the loop and (2) whose loop count can
|
be easily adjusted. This is because we peel a constant number of
|
be easily adjusted. This is because we peel a constant number of
|
iterations into a prologue and epilogue for which we want to avoid
|
iterations into a prologue and epilogue for which we want to avoid
|
emitting the control part, and a kernel which is to iterate that
|
emitting the control part, and a kernel which is to iterate that
|
constant number of iterations less than the original loop. So the
|
constant number of iterations less than the original loop. So the
|
control part should be a set of insns clearly identified and having
|
control part should be a set of insns clearly identified and having
|
its own iv, not otherwise used in the loop (at-least for now), which
|
its own iv, not otherwise used in the loop (at-least for now), which
|
initializes a register before the loop to the number of iterations.
|
initializes a register before the loop to the number of iterations.
|
Currently SMS relies on the do-loop pattern to recognize such loops,
|
Currently SMS relies on the do-loop pattern to recognize such loops,
|
where (1) the control part comprises of all insns defining and/or
|
where (1) the control part comprises of all insns defining and/or
|
using a certain 'count' register and (2) the loop count can be
|
using a certain 'count' register and (2) the loop count can be
|
adjusted by modifying this register prior to the loop.
|
adjusted by modifying this register prior to the loop.
|
TODO: Rely on cfgloop analysis instead. */
|
TODO: Rely on cfgloop analysis instead. */
|
|
|
/* This page defines partial-schedule structures and functions for
|
/* This page defines partial-schedule structures and functions for
|
modulo scheduling. */
|
modulo scheduling. */
|
|
|
typedef struct partial_schedule *partial_schedule_ptr;
|
typedef struct partial_schedule *partial_schedule_ptr;
|
typedef struct ps_insn *ps_insn_ptr;
|
typedef struct ps_insn *ps_insn_ptr;
|
|
|
/* The minimum (absolute) cycle that a node of ps was scheduled in. */
|
/* The minimum (absolute) cycle that a node of ps was scheduled in. */
|
#define PS_MIN_CYCLE(ps) (((partial_schedule_ptr)(ps))->min_cycle)
|
#define PS_MIN_CYCLE(ps) (((partial_schedule_ptr)(ps))->min_cycle)
|
|
|
/* The maximum (absolute) cycle that a node of ps was scheduled in. */
|
/* The maximum (absolute) cycle that a node of ps was scheduled in. */
|
#define PS_MAX_CYCLE(ps) (((partial_schedule_ptr)(ps))->max_cycle)
|
#define PS_MAX_CYCLE(ps) (((partial_schedule_ptr)(ps))->max_cycle)
|
|
|
/* Perform signed modulo, always returning a non-negative value. */
|
/* Perform signed modulo, always returning a non-negative value. */
|
#define SMODULO(x,y) ((x) % (y) < 0 ? ((x) % (y) + (y)) : (x) % (y))
|
#define SMODULO(x,y) ((x) % (y) < 0 ? ((x) % (y) + (y)) : (x) % (y))
|
|
|
/* The number of different iterations the nodes in ps span, assuming
|
/* The number of different iterations the nodes in ps span, assuming
|
the stage boundaries are placed efficiently. */
|
the stage boundaries are placed efficiently. */
|
#define CALC_STAGE_COUNT(max_cycle,min_cycle,ii) ((max_cycle - min_cycle \
|
#define CALC_STAGE_COUNT(max_cycle,min_cycle,ii) ((max_cycle - min_cycle \
|
+ 1 + ii - 1) / ii)
|
+ 1 + ii - 1) / ii)
|
/* The stage count of ps. */
|
/* The stage count of ps. */
|
#define PS_STAGE_COUNT(ps) (((partial_schedule_ptr)(ps))->stage_count)
|
#define PS_STAGE_COUNT(ps) (((partial_schedule_ptr)(ps))->stage_count)
|
|
|
/* A single instruction in the partial schedule. */
|
/* A single instruction in the partial schedule. */
|
struct ps_insn
|
struct ps_insn
|
{
|
{
|
/* Identifies the instruction to be scheduled. Values smaller than
|
/* Identifies the instruction to be scheduled. Values smaller than
|
the ddg's num_nodes refer directly to ddg nodes. A value of
|
the ddg's num_nodes refer directly to ddg nodes. A value of
|
X - num_nodes refers to register move X. */
|
X - num_nodes refers to register move X. */
|
int id;
|
int id;
|
|
|
/* The (absolute) cycle in which the PS instruction is scheduled.
|
/* The (absolute) cycle in which the PS instruction is scheduled.
|
Same as SCHED_TIME (node). */
|
Same as SCHED_TIME (node). */
|
int cycle;
|
int cycle;
|
|
|
/* The next/prev PS_INSN in the same row. */
|
/* The next/prev PS_INSN in the same row. */
|
ps_insn_ptr next_in_row,
|
ps_insn_ptr next_in_row,
|
prev_in_row;
|
prev_in_row;
|
|
|
};
|
};
|
|
|
/* Information about a register move that has been added to a partial
|
/* Information about a register move that has been added to a partial
|
schedule. */
|
schedule. */
|
struct ps_reg_move_info
|
struct ps_reg_move_info
|
{
|
{
|
/* The source of the move is defined by the ps_insn with id DEF.
|
/* 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. */
|
The destination is used by the ps_insns with the ids in USES. */
|
int def;
|
int def;
|
sbitmap uses;
|
sbitmap uses;
|
|
|
/* The original form of USES' instructions used OLD_REG, but they
|
/* The original form of USES' instructions used OLD_REG, but they
|
should now use NEW_REG. */
|
should now use NEW_REG. */
|
rtx old_reg;
|
rtx old_reg;
|
rtx new_reg;
|
rtx new_reg;
|
|
|
/* The number of consecutive stages that the move occupies. */
|
/* The number of consecutive stages that the move occupies. */
|
int num_consecutive_stages;
|
int num_consecutive_stages;
|
|
|
/* An instruction that sets NEW_REG to the correct value. The first
|
/* 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
|
move associated with DEF will have an rhs of OLD_REG; later moves
|
use the result of the previous move. */
|
use the result of the previous move. */
|
rtx insn;
|
rtx insn;
|
};
|
};
|
|
|
typedef struct ps_reg_move_info ps_reg_move_info;
|
typedef struct ps_reg_move_info ps_reg_move_info;
|
DEF_VEC_O (ps_reg_move_info);
|
DEF_VEC_O (ps_reg_move_info);
|
DEF_VEC_ALLOC_O (ps_reg_move_info, heap);
|
DEF_VEC_ALLOC_O (ps_reg_move_info, heap);
|
|
|
/* Holds the partial schedule as an array of II rows. Each entry of the
|
/* 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
|
array points to a linked list of PS_INSNs, which represents the
|
instructions that are scheduled for that row. */
|
instructions that are scheduled for that row. */
|
struct partial_schedule
|
struct partial_schedule
|
{
|
{
|
int ii; /* Number of rows in the partial schedule. */
|
int ii; /* Number of rows in the partial schedule. */
|
int history; /* Threshold for conflict checking using DFA. */
|
int history; /* Threshold for conflict checking using DFA. */
|
|
|
/* rows[i] points to linked list of insns scheduled in row i (0<=i<ii). */
|
/* rows[i] points to linked list of insns scheduled in row i (0<=i<ii). */
|
ps_insn_ptr *rows;
|
ps_insn_ptr *rows;
|
|
|
/* All the moves added for this partial schedule. Index X has
|
/* All the moves added for this partial schedule. Index X has
|
a ps_insn id of X + g->num_nodes. */
|
a ps_insn id of X + g->num_nodes. */
|
VEC (ps_reg_move_info, heap) *reg_moves;
|
VEC (ps_reg_move_info, heap) *reg_moves;
|
|
|
/* rows_length[i] holds the number of instructions in the row.
|
/* rows_length[i] holds the number of instructions in the row.
|
It is used only (as an optimization) to back off quickly from
|
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
|
trying to schedule a node in a full row; that is, to avoid running
|
through futile DFA state transitions. */
|
through futile DFA state transitions. */
|
int *rows_length;
|
int *rows_length;
|
|
|
/* The earliest absolute cycle of an insn in the partial schedule. */
|
/* The earliest absolute cycle of an insn in the partial schedule. */
|
int min_cycle;
|
int min_cycle;
|
|
|
/* The latest absolute cycle of an insn in the partial schedule. */
|
/* The latest absolute cycle of an insn in the partial schedule. */
|
int max_cycle;
|
int max_cycle;
|
|
|
ddg_ptr g; /* The DDG of the insns in the partial schedule. */
|
ddg_ptr g; /* The DDG of the insns in the partial schedule. */
|
|
|
int stage_count; /* The stage count of the partial schedule. */
|
int stage_count; /* The stage count of the partial schedule. */
|
};
|
};
|
|
|
|
|
static partial_schedule_ptr create_partial_schedule (int ii, ddg_ptr, int history);
|
static partial_schedule_ptr create_partial_schedule (int ii, ddg_ptr, int history);
|
static void free_partial_schedule (partial_schedule_ptr);
|
static void free_partial_schedule (partial_schedule_ptr);
|
static void reset_partial_schedule (partial_schedule_ptr, int new_ii);
|
static void reset_partial_schedule (partial_schedule_ptr, int new_ii);
|
void print_partial_schedule (partial_schedule_ptr, FILE *);
|
void print_partial_schedule (partial_schedule_ptr, FILE *);
|
static void verify_partial_schedule (partial_schedule_ptr, sbitmap);
|
static void verify_partial_schedule (partial_schedule_ptr, sbitmap);
|
static ps_insn_ptr ps_add_node_check_conflicts (partial_schedule_ptr,
|
static ps_insn_ptr ps_add_node_check_conflicts (partial_schedule_ptr,
|
int, int, sbitmap, sbitmap);
|
int, int, sbitmap, sbitmap);
|
static void rotate_partial_schedule (partial_schedule_ptr, int);
|
static void rotate_partial_schedule (partial_schedule_ptr, int);
|
void set_row_column_for_ps (partial_schedule_ptr);
|
void set_row_column_for_ps (partial_schedule_ptr);
|
static void ps_insert_empty_row (partial_schedule_ptr, int, sbitmap);
|
static void ps_insert_empty_row (partial_schedule_ptr, int, sbitmap);
|
static int compute_split_row (sbitmap, int, int, int, ddg_node_ptr);
|
static int compute_split_row (sbitmap, int, int, int, ddg_node_ptr);
|
|
|
|
|
/* This page defines constants and structures for the modulo scheduling
|
/* This page defines constants and structures for the modulo scheduling
|
driver. */
|
driver. */
|
|
|
static int sms_order_nodes (ddg_ptr, int, int *, int *);
|
static int sms_order_nodes (ddg_ptr, int, int *, int *);
|
static void set_node_sched_params (ddg_ptr);
|
static void set_node_sched_params (ddg_ptr);
|
static partial_schedule_ptr sms_schedule_by_order (ddg_ptr, int, int, int *);
|
static partial_schedule_ptr sms_schedule_by_order (ddg_ptr, int, int, int *);
|
static void permute_partial_schedule (partial_schedule_ptr, rtx);
|
static void permute_partial_schedule (partial_schedule_ptr, rtx);
|
static void generate_prolog_epilog (partial_schedule_ptr, struct loop *,
|
static void generate_prolog_epilog (partial_schedule_ptr, struct loop *,
|
rtx, rtx);
|
rtx, rtx);
|
static int calculate_stage_count (partial_schedule_ptr, int);
|
static int calculate_stage_count (partial_schedule_ptr, int);
|
static void calculate_must_precede_follow (ddg_node_ptr, int, int,
|
static void calculate_must_precede_follow (ddg_node_ptr, int, int,
|
int, int, sbitmap, sbitmap, sbitmap);
|
int, int, sbitmap, sbitmap, sbitmap);
|
static int get_sched_window (partial_schedule_ptr, ddg_node_ptr,
|
static int get_sched_window (partial_schedule_ptr, ddg_node_ptr,
|
sbitmap, int, int *, int *, int *);
|
sbitmap, int, int *, int *, int *);
|
static bool try_scheduling_node_in_cycle (partial_schedule_ptr, int, int,
|
static bool try_scheduling_node_in_cycle (partial_schedule_ptr, int, int,
|
sbitmap, int *, sbitmap, sbitmap);
|
sbitmap, int *, sbitmap, sbitmap);
|
static void remove_node_from_ps (partial_schedule_ptr, ps_insn_ptr);
|
static void remove_node_from_ps (partial_schedule_ptr, ps_insn_ptr);
|
|
|
#define NODE_ASAP(node) ((node)->aux.count)
|
#define NODE_ASAP(node) ((node)->aux.count)
|
|
|
#define SCHED_PARAMS(x) VEC_index (node_sched_params, node_sched_param_vec, 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_TIME(x) (SCHED_PARAMS (x)->time)
|
#define SCHED_ROW(x) (SCHED_PARAMS (x)->row)
|
#define SCHED_ROW(x) (SCHED_PARAMS (x)->row)
|
#define SCHED_STAGE(x) (SCHED_PARAMS (x)->stage)
|
#define SCHED_STAGE(x) (SCHED_PARAMS (x)->stage)
|
#define SCHED_COLUMN(x) (SCHED_PARAMS (x)->column)
|
#define SCHED_COLUMN(x) (SCHED_PARAMS (x)->column)
|
|
|
/* The scheduling parameters held for each node. */
|
/* The scheduling parameters held for each node. */
|
typedef struct node_sched_params
|
typedef struct node_sched_params
|
{
|
{
|
int time; /* The absolute scheduling cycle. */
|
int time; /* The absolute scheduling cycle. */
|
|
|
int row; /* Holds time % ii. */
|
int row; /* Holds time % ii. */
|
int stage; /* Holds time / ii. */
|
int stage; /* Holds time / ii. */
|
|
|
/* The column of a node inside the ps. If nodes u, v are on the same row,
|
/* The column of a node inside the ps. If nodes u, v are on the same row,
|
u will precede v if column (u) < column (v). */
|
u will precede v if column (u) < column (v). */
|
int column;
|
int column;
|
} *node_sched_params_ptr;
|
} *node_sched_params_ptr;
|
|
|
typedef struct node_sched_params node_sched_params;
|
typedef struct node_sched_params node_sched_params;
|
DEF_VEC_O (node_sched_params);
|
DEF_VEC_O (node_sched_params);
|
DEF_VEC_ALLOC_O (node_sched_params, heap);
|
DEF_VEC_ALLOC_O (node_sched_params, heap);
|
|
|
/* The following three functions are copied from the current scheduler
|
/* The following three functions are copied from the current scheduler
|
code in order to use sched_analyze() for computing the dependencies.
|
code in order to use sched_analyze() for computing the dependencies.
|
They are used when initializing the sched_info structure. */
|
They are used when initializing the sched_info structure. */
|
static const char *
|
static const char *
|
sms_print_insn (const_rtx insn, int aligned ATTRIBUTE_UNUSED)
|
sms_print_insn (const_rtx insn, int aligned ATTRIBUTE_UNUSED)
|
{
|
{
|
static char tmp[80];
|
static char tmp[80];
|
|
|
sprintf (tmp, "i%4d", INSN_UID (insn));
|
sprintf (tmp, "i%4d", INSN_UID (insn));
|
return tmp;
|
return tmp;
|
}
|
}
|
|
|
static void
|
static void
|
compute_jump_reg_dependencies (rtx insn ATTRIBUTE_UNUSED,
|
compute_jump_reg_dependencies (rtx insn ATTRIBUTE_UNUSED,
|
regset used ATTRIBUTE_UNUSED)
|
regset used ATTRIBUTE_UNUSED)
|
{
|
{
|
}
|
}
|
|
|
static struct common_sched_info_def sms_common_sched_info;
|
static struct common_sched_info_def sms_common_sched_info;
|
|
|
static struct sched_deps_info_def sms_sched_deps_info =
|
static struct sched_deps_info_def sms_sched_deps_info =
|
{
|
{
|
compute_jump_reg_dependencies,
|
compute_jump_reg_dependencies,
|
NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
|
NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
|
NULL,
|
NULL,
|
0, 0, 0
|
0, 0, 0
|
};
|
};
|
|
|
static struct haifa_sched_info sms_sched_info =
|
static struct haifa_sched_info sms_sched_info =
|
{
|
{
|
NULL,
|
NULL,
|
NULL,
|
NULL,
|
NULL,
|
NULL,
|
NULL,
|
NULL,
|
NULL,
|
NULL,
|
sms_print_insn,
|
sms_print_insn,
|
NULL,
|
NULL,
|
NULL, /* insn_finishes_block_p */
|
NULL, /* insn_finishes_block_p */
|
NULL, NULL,
|
NULL, NULL,
|
NULL, NULL,
|
NULL, NULL,
|
0, 0,
|
0, 0,
|
|
|
NULL, NULL, NULL, NULL,
|
NULL, NULL, NULL, NULL,
|
NULL, NULL,
|
NULL, NULL,
|
0
|
0
|
};
|
};
|
|
|
/* Partial schedule instruction ID in PS is a register move. Return
|
/* Partial schedule instruction ID in PS is a register move. Return
|
information about it. */
|
information about it. */
|
static struct ps_reg_move_info *
|
static struct ps_reg_move_info *
|
ps_reg_move (partial_schedule_ptr ps, int id)
|
ps_reg_move (partial_schedule_ptr ps, int id)
|
{
|
{
|
gcc_checking_assert (id >= ps->g->num_nodes);
|
gcc_checking_assert (id >= ps->g->num_nodes);
|
return VEC_index (ps_reg_move_info, ps->reg_moves, 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
|
/* Return the rtl instruction that is being scheduled by partial schedule
|
instruction ID, which belongs to schedule PS. */
|
instruction ID, which belongs to schedule PS. */
|
static rtx
|
static rtx
|
ps_rtl_insn (partial_schedule_ptr ps, int id)
|
ps_rtl_insn (partial_schedule_ptr ps, int id)
|
{
|
{
|
if (id < ps->g->num_nodes)
|
if (id < ps->g->num_nodes)
|
return ps->g->nodes[id].insn;
|
return ps->g->nodes[id].insn;
|
else
|
else
|
return ps_reg_move (ps, id)->insn;
|
return ps_reg_move (ps, id)->insn;
|
}
|
}
|
|
|
/* Partial schedule instruction ID, which belongs to PS, occured in
|
/* Partial schedule instruction ID, which belongs to PS, occured in
|
the original (unscheduled) loop. Return the first instruction
|
the original (unscheduled) loop. Return the first instruction
|
in the loop that was associated with ps_rtl_insn (PS, ID).
|
in the loop that was associated with ps_rtl_insn (PS, ID).
|
If the instruction had some notes before it, this is the first
|
If the instruction had some notes before it, this is the first
|
of those notes. */
|
of those notes. */
|
static rtx
|
static rtx
|
ps_first_note (partial_schedule_ptr ps, int id)
|
ps_first_note (partial_schedule_ptr ps, int id)
|
{
|
{
|
gcc_assert (id < ps->g->num_nodes);
|
gcc_assert (id < ps->g->num_nodes);
|
return ps->g->nodes[id].first_note;
|
return ps->g->nodes[id].first_note;
|
}
|
}
|
|
|
/* Return the number of consecutive stages that are occupied by
|
/* Return the number of consecutive stages that are occupied by
|
partial schedule instruction ID in PS. */
|
partial schedule instruction ID in PS. */
|
static int
|
static int
|
ps_num_consecutive_stages (partial_schedule_ptr ps, int id)
|
ps_num_consecutive_stages (partial_schedule_ptr ps, int id)
|
{
|
{
|
if (id < ps->g->num_nodes)
|
if (id < ps->g->num_nodes)
|
return 1;
|
return 1;
|
else
|
else
|
return ps_reg_move (ps, id)->num_consecutive_stages;
|
return ps_reg_move (ps, id)->num_consecutive_stages;
|
}
|
}
|
|
|
/* Given HEAD and TAIL which are the first and last insns in a loop;
|
/* 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
|
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
|
more than one occurrence in the loop besides the control part or the
|
do-loop pattern is not of the form we expect. */
|
do-loop pattern is not of the form we expect. */
|
static rtx
|
static rtx
|
doloop_register_get (rtx head ATTRIBUTE_UNUSED, rtx tail ATTRIBUTE_UNUSED)
|
doloop_register_get (rtx head ATTRIBUTE_UNUSED, rtx tail ATTRIBUTE_UNUSED)
|
{
|
{
|
#ifdef HAVE_doloop_end
|
#ifdef HAVE_doloop_end
|
rtx reg, condition, insn, first_insn_not_to_check;
|
rtx reg, condition, insn, first_insn_not_to_check;
|
|
|
if (!JUMP_P (tail))
|
if (!JUMP_P (tail))
|
return NULL_RTX;
|
return NULL_RTX;
|
|
|
/* TODO: Free SMS's dependence on doloop_condition_get. */
|
/* TODO: Free SMS's dependence on doloop_condition_get. */
|
condition = doloop_condition_get (tail);
|
condition = doloop_condition_get (tail);
|
if (! condition)
|
if (! condition)
|
return NULL_RTX;
|
return NULL_RTX;
|
|
|
if (REG_P (XEXP (condition, 0)))
|
if (REG_P (XEXP (condition, 0)))
|
reg = XEXP (condition, 0);
|
reg = XEXP (condition, 0);
|
else if (GET_CODE (XEXP (condition, 0)) == PLUS
|
else if (GET_CODE (XEXP (condition, 0)) == PLUS
|
&& REG_P (XEXP (XEXP (condition, 0), 0)))
|
&& REG_P (XEXP (XEXP (condition, 0), 0)))
|
reg = XEXP (XEXP (condition, 0), 0);
|
reg = XEXP (XEXP (condition, 0), 0);
|
else
|
else
|
gcc_unreachable ();
|
gcc_unreachable ();
|
|
|
/* Check that the COUNT_REG has no other occurrences in the loop
|
/* Check that the COUNT_REG has no other occurrences in the loop
|
until the decrement. We assume the control part consists of
|
until the decrement. We assume the control part consists of
|
either a single (parallel) branch-on-count or a (non-parallel)
|
either a single (parallel) branch-on-count or a (non-parallel)
|
branch immediately preceded by a single (decrement) insn. */
|
branch immediately preceded by a single (decrement) insn. */
|
first_insn_not_to_check = (GET_CODE (PATTERN (tail)) == PARALLEL ? tail
|
first_insn_not_to_check = (GET_CODE (PATTERN (tail)) == PARALLEL ? tail
|
: prev_nondebug_insn (tail));
|
: prev_nondebug_insn (tail));
|
|
|
for (insn = head; insn != first_insn_not_to_check; insn = NEXT_INSN (insn))
|
for (insn = head; insn != first_insn_not_to_check; insn = NEXT_INSN (insn))
|
if (!DEBUG_INSN_P (insn) && reg_mentioned_p (reg, insn))
|
if (!DEBUG_INSN_P (insn) && reg_mentioned_p (reg, insn))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "SMS count_reg found ");
|
fprintf (dump_file, "SMS count_reg found ");
|
print_rtl_single (dump_file, reg);
|
print_rtl_single (dump_file, reg);
|
fprintf (dump_file, " outside control in insn:\n");
|
fprintf (dump_file, " outside control in insn:\n");
|
print_rtl_single (dump_file, insn);
|
print_rtl_single (dump_file, insn);
|
}
|
}
|
|
|
return NULL_RTX;
|
return NULL_RTX;
|
}
|
}
|
|
|
return reg;
|
return reg;
|
#else
|
#else
|
return NULL_RTX;
|
return NULL_RTX;
|
#endif
|
#endif
|
}
|
}
|
|
|
/* Check if COUNT_REG is set to a constant in the PRE_HEADER block, so
|
/* Check if COUNT_REG is set to a constant in the PRE_HEADER block, so
|
that the number of iterations is a compile-time constant. If so,
|
that the number of iterations is a compile-time constant. If so,
|
return the rtx that sets COUNT_REG to a constant, and set COUNT to
|
return the rtx that sets COUNT_REG to a constant, and set COUNT to
|
this constant. Otherwise return 0. */
|
this constant. Otherwise return 0. */
|
static rtx
|
static rtx
|
const_iteration_count (rtx count_reg, basic_block pre_header,
|
const_iteration_count (rtx count_reg, basic_block pre_header,
|
HOST_WIDEST_INT * count)
|
HOST_WIDEST_INT * count)
|
{
|
{
|
rtx insn;
|
rtx insn;
|
rtx head, tail;
|
rtx head, tail;
|
|
|
if (! pre_header)
|
if (! pre_header)
|
return NULL_RTX;
|
return NULL_RTX;
|
|
|
get_ebb_head_tail (pre_header, pre_header, &head, &tail);
|
get_ebb_head_tail (pre_header, pre_header, &head, &tail);
|
|
|
for (insn = tail; insn != PREV_INSN (head); insn = PREV_INSN (insn))
|
for (insn = tail; insn != PREV_INSN (head); insn = PREV_INSN (insn))
|
if (NONDEBUG_INSN_P (insn) && single_set (insn) &&
|
if (NONDEBUG_INSN_P (insn) && single_set (insn) &&
|
rtx_equal_p (count_reg, SET_DEST (single_set (insn))))
|
rtx_equal_p (count_reg, SET_DEST (single_set (insn))))
|
{
|
{
|
rtx pat = single_set (insn);
|
rtx pat = single_set (insn);
|
|
|
if (CONST_INT_P (SET_SRC (pat)))
|
if (CONST_INT_P (SET_SRC (pat)))
|
{
|
{
|
*count = INTVAL (SET_SRC (pat));
|
*count = INTVAL (SET_SRC (pat));
|
return insn;
|
return insn;
|
}
|
}
|
|
|
return NULL_RTX;
|
return NULL_RTX;
|
}
|
}
|
|
|
return NULL_RTX;
|
return NULL_RTX;
|
}
|
}
|
|
|
/* A very simple resource-based lower bound on the initiation interval.
|
/* A very simple resource-based lower bound on the initiation interval.
|
??? Improve the accuracy of this bound by considering the
|
??? Improve the accuracy of this bound by considering the
|
utilization of various units. */
|
utilization of various units. */
|
static int
|
static int
|
res_MII (ddg_ptr g)
|
res_MII (ddg_ptr g)
|
{
|
{
|
if (targetm.sched.sms_res_mii)
|
if (targetm.sched.sms_res_mii)
|
return targetm.sched.sms_res_mii (g);
|
return targetm.sched.sms_res_mii (g);
|
|
|
return ((g->num_nodes - g->num_debug) / issue_rate);
|
return ((g->num_nodes - g->num_debug) / issue_rate);
|
}
|
}
|
|
|
|
|
/* A vector that contains the sched data for each ps_insn. */
|
/* A vector that contains the sched data for each ps_insn. */
|
static VEC (node_sched_params, heap) *node_sched_param_vec;
|
static VEC (node_sched_params, heap) *node_sched_param_vec;
|
|
|
/* Allocate sched_params for each node and initialize it. */
|
/* Allocate sched_params for each node and initialize it. */
|
static void
|
static void
|
set_node_sched_params (ddg_ptr g)
|
set_node_sched_params (ddg_ptr g)
|
{
|
{
|
VEC_truncate (node_sched_params, node_sched_param_vec, 0);
|
VEC_truncate (node_sched_params, node_sched_param_vec, 0);
|
VEC_safe_grow_cleared (node_sched_params, heap,
|
VEC_safe_grow_cleared (node_sched_params, heap,
|
node_sched_param_vec, g->num_nodes);
|
node_sched_param_vec, g->num_nodes);
|
}
|
}
|
|
|
/* Make sure that node_sched_param_vec has an entry for every move in PS. */
|
/* Make sure that node_sched_param_vec has an entry for every move in PS. */
|
static void
|
static void
|
extend_node_sched_params (partial_schedule_ptr ps)
|
extend_node_sched_params (partial_schedule_ptr ps)
|
{
|
{
|
VEC_safe_grow_cleared (node_sched_params, heap, node_sched_param_vec,
|
VEC_safe_grow_cleared (node_sched_params, heap, node_sched_param_vec,
|
ps->g->num_nodes + VEC_length (ps_reg_move_info,
|
ps->g->num_nodes + VEC_length (ps_reg_move_info,
|
ps->reg_moves));
|
ps->reg_moves));
|
}
|
}
|
|
|
/* Update the sched_params (time, row and stage) for node U using the II,
|
/* Update the sched_params (time, row and stage) for node U using the II,
|
the CYCLE of U and MIN_CYCLE.
|
the CYCLE of U and MIN_CYCLE.
|
We're not simply taking the following
|
We're not simply taking the following
|
SCHED_STAGE (u) = CALC_STAGE_COUNT (SCHED_TIME (u), min_cycle, ii);
|
SCHED_STAGE (u) = CALC_STAGE_COUNT (SCHED_TIME (u), min_cycle, ii);
|
because the stages may not be aligned on cycle 0. */
|
because the stages may not be aligned on cycle 0. */
|
static void
|
static void
|
update_node_sched_params (int u, int ii, int cycle, int min_cycle)
|
update_node_sched_params (int u, int ii, int cycle, int min_cycle)
|
{
|
{
|
int sc_until_cycle_zero;
|
int sc_until_cycle_zero;
|
int stage;
|
int stage;
|
|
|
SCHED_TIME (u) = cycle;
|
SCHED_TIME (u) = cycle;
|
SCHED_ROW (u) = SMODULO (cycle, ii);
|
SCHED_ROW (u) = SMODULO (cycle, ii);
|
|
|
/* The calculation of stage count is done adding the number
|
/* The calculation of stage count is done adding the number
|
of stages before cycle zero and after cycle zero. */
|
of stages before cycle zero and after cycle zero. */
|
sc_until_cycle_zero = CALC_STAGE_COUNT (-1, min_cycle, ii);
|
sc_until_cycle_zero = CALC_STAGE_COUNT (-1, min_cycle, ii);
|
|
|
if (SCHED_TIME (u) < 0)
|
if (SCHED_TIME (u) < 0)
|
{
|
{
|
stage = CALC_STAGE_COUNT (-1, SCHED_TIME (u), ii);
|
stage = CALC_STAGE_COUNT (-1, SCHED_TIME (u), ii);
|
SCHED_STAGE (u) = sc_until_cycle_zero - stage;
|
SCHED_STAGE (u) = sc_until_cycle_zero - stage;
|
}
|
}
|
else
|
else
|
{
|
{
|
stage = CALC_STAGE_COUNT (SCHED_TIME (u), 0, ii);
|
stage = CALC_STAGE_COUNT (SCHED_TIME (u), 0, ii);
|
SCHED_STAGE (u) = sc_until_cycle_zero + stage - 1;
|
SCHED_STAGE (u) = sc_until_cycle_zero + stage - 1;
|
}
|
}
|
}
|
}
|
|
|
static void
|
static void
|
print_node_sched_params (FILE *file, int num_nodes, partial_schedule_ptr ps)
|
print_node_sched_params (FILE *file, int num_nodes, partial_schedule_ptr ps)
|
{
|
{
|
int i;
|
int i;
|
|
|
if (! file)
|
if (! file)
|
return;
|
return;
|
for (i = 0; i < num_nodes; i++)
|
for (i = 0; i < num_nodes; i++)
|
{
|
{
|
node_sched_params_ptr nsp = SCHED_PARAMS (i);
|
node_sched_params_ptr nsp = SCHED_PARAMS (i);
|
|
|
fprintf (file, "Node = %d; INSN = %d\n", i,
|
fprintf (file, "Node = %d; INSN = %d\n", i,
|
INSN_UID (ps_rtl_insn (ps, i)));
|
INSN_UID (ps_rtl_insn (ps, i)));
|
fprintf (file, " asap = %d:\n", NODE_ASAP (&ps->g->nodes[i]));
|
fprintf (file, " asap = %d:\n", NODE_ASAP (&ps->g->nodes[i]));
|
fprintf (file, " time = %d:\n", nsp->time);
|
fprintf (file, " time = %d:\n", nsp->time);
|
fprintf (file, " stage = %d:\n", nsp->stage);
|
fprintf (file, " stage = %d:\n", nsp->stage);
|
}
|
}
|
}
|
}
|
|
|
/* Set SCHED_COLUMN for each instruction in row ROW of PS. */
|
/* Set SCHED_COLUMN for each instruction in row ROW of PS. */
|
static void
|
static void
|
set_columns_for_row (partial_schedule_ptr ps, int row)
|
set_columns_for_row (partial_schedule_ptr ps, int row)
|
{
|
{
|
ps_insn_ptr cur_insn;
|
ps_insn_ptr cur_insn;
|
int column;
|
int column;
|
|
|
column = 0;
|
column = 0;
|
for (cur_insn = ps->rows[row]; cur_insn; cur_insn = cur_insn->next_in_row)
|
for (cur_insn = ps->rows[row]; cur_insn; cur_insn = cur_insn->next_in_row)
|
SCHED_COLUMN (cur_insn->id) = column++;
|
SCHED_COLUMN (cur_insn->id) = column++;
|
}
|
}
|
|
|
/* Set SCHED_COLUMN for each instruction in PS. */
|
/* Set SCHED_COLUMN for each instruction in PS. */
|
static void
|
static void
|
set_columns_for_ps (partial_schedule_ptr ps)
|
set_columns_for_ps (partial_schedule_ptr ps)
|
{
|
{
|
int row;
|
int row;
|
|
|
for (row = 0; row < ps->ii; row++)
|
for (row = 0; row < ps->ii; row++)
|
set_columns_for_row (ps, row);
|
set_columns_for_row (ps, row);
|
}
|
}
|
|
|
/* Try to schedule the move with ps_insn identifier I_REG_MOVE in PS.
|
/* Try to schedule the move with ps_insn identifier I_REG_MOVE in PS.
|
Its single predecessor has already been scheduled, as has its
|
Its single predecessor has already been scheduled, as has its
|
ddg node successors. (The move may have also another move as its
|
ddg node successors. (The move may have also another move as its
|
successor, in which case that successor will be scheduled later.)
|
successor, in which case that successor will be scheduled later.)
|
|
|
The move is part of a chain that satisfies register dependencies
|
The move is part of a chain that satisfies register dependencies
|
between a producing ddg node and various consuming ddg nodes.
|
between a producing ddg node and various consuming ddg nodes.
|
If some of these dependencies have a distance of 1 (meaning that
|
If some of these dependencies have a distance of 1 (meaning that
|
the use is upward-exposed) then DISTANCE1_USES is nonnull and
|
the use is upward-exposed) then DISTANCE1_USES is nonnull and
|
contains the set of uses with distance-1 dependencies.
|
contains the set of uses with distance-1 dependencies.
|
DISTANCE1_USES is null otherwise.
|
DISTANCE1_USES is null otherwise.
|
|
|
MUST_FOLLOW is a scratch bitmap that is big enough to hold
|
MUST_FOLLOW is a scratch bitmap that is big enough to hold
|
all current ps_insn ids.
|
all current ps_insn ids.
|
|
|
Return true on success. */
|
Return true on success. */
|
static bool
|
static bool
|
schedule_reg_move (partial_schedule_ptr ps, int i_reg_move,
|
schedule_reg_move (partial_schedule_ptr ps, int i_reg_move,
|
sbitmap distance1_uses, sbitmap must_follow)
|
sbitmap distance1_uses, sbitmap must_follow)
|
{
|
{
|
unsigned int u;
|
unsigned int u;
|
int this_time, this_distance, this_start, this_end, this_latency;
|
int this_time, this_distance, this_start, this_end, this_latency;
|
int start, end, c, ii;
|
int start, end, c, ii;
|
sbitmap_iterator sbi;
|
sbitmap_iterator sbi;
|
ps_reg_move_info *move;
|
ps_reg_move_info *move;
|
rtx this_insn;
|
rtx this_insn;
|
ps_insn_ptr psi;
|
ps_insn_ptr psi;
|
|
|
move = ps_reg_move (ps, i_reg_move);
|
move = ps_reg_move (ps, i_reg_move);
|
ii = ps->ii;
|
ii = ps->ii;
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "Scheduling register move INSN %d; ii = %d"
|
fprintf (dump_file, "Scheduling register move INSN %d; ii = %d"
|
", min cycle = %d\n\n", INSN_UID (move->insn), ii,
|
", min cycle = %d\n\n", INSN_UID (move->insn), ii,
|
PS_MIN_CYCLE (ps));
|
PS_MIN_CYCLE (ps));
|
print_rtl_single (dump_file, move->insn);
|
print_rtl_single (dump_file, move->insn);
|
fprintf (dump_file, "\n%11s %11s %5s\n", "start", "end", "time");
|
fprintf (dump_file, "\n%11s %11s %5s\n", "start", "end", "time");
|
fprintf (dump_file, "=========== =========== =====\n");
|
fprintf (dump_file, "=========== =========== =====\n");
|
}
|
}
|
|
|
start = INT_MIN;
|
start = INT_MIN;
|
end = INT_MAX;
|
end = INT_MAX;
|
|
|
/* For dependencies of distance 1 between a producer ddg node A
|
/* For dependencies of distance 1 between a producer ddg node A
|
and consumer ddg node B, we have a chain of dependencies:
|
and consumer ddg node B, we have a chain of dependencies:
|
|
|
A --(T,L1,1)--> M1 --(T,L2,0)--> M2 ... --(T,Ln,0)--> B
|
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
|
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
|
a producer ddg node A and consumer ddg node C, we have a chain of
|
dependencies:
|
dependencies:
|
|
|
A --(T,L1',0)--> M1' --(T,L2',0)--> M2' ... --(T,Ln',0)--> C
|
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.
|
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
|
We can only schedule each move once, so if we have both types of
|
chain, we model the second as:
|
chain, we model the second as:
|
|
|
A --(T,L1',1)--> M1 --(T,L2',0)--> M2 ... --(T,Ln',-1)--> C
|
A --(T,L1',1)--> M1 --(T,L2',0)--> M2 ... --(T,Ln',-1)--> C
|
|
|
First handle the dependencies between the previously-scheduled
|
First handle the dependencies between the previously-scheduled
|
predecessor and the move. */
|
predecessor and the move. */
|
this_insn = ps_rtl_insn (ps, move->def);
|
this_insn = ps_rtl_insn (ps, move->def);
|
this_latency = insn_latency (this_insn, move->insn);
|
this_latency = insn_latency (this_insn, move->insn);
|
this_distance = distance1_uses && move->def < ps->g->num_nodes ? 1 : 0;
|
this_distance = distance1_uses && move->def < ps->g->num_nodes ? 1 : 0;
|
this_time = SCHED_TIME (move->def) - this_distance * ii;
|
this_time = SCHED_TIME (move->def) - this_distance * ii;
|
this_start = this_time + this_latency;
|
this_start = this_time + this_latency;
|
this_end = this_time + ii;
|
this_end = this_time + ii;
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "%11d %11d %5d %d --(T,%d,%d)--> %d\n",
|
fprintf (dump_file, "%11d %11d %5d %d --(T,%d,%d)--> %d\n",
|
this_start, this_end, SCHED_TIME (move->def),
|
this_start, this_end, SCHED_TIME (move->def),
|
INSN_UID (this_insn), this_latency, this_distance,
|
INSN_UID (this_insn), this_latency, this_distance,
|
INSN_UID (move->insn));
|
INSN_UID (move->insn));
|
|
|
if (start < this_start)
|
if (start < this_start)
|
start = this_start;
|
start = this_start;
|
if (end > this_end)
|
if (end > this_end)
|
end = this_end;
|
end = this_end;
|
|
|
/* Handle the dependencies between the move and previously-scheduled
|
/* Handle the dependencies between the move and previously-scheduled
|
successors. */
|
successors. */
|
EXECUTE_IF_SET_IN_SBITMAP (move->uses, 0, u, sbi)
|
EXECUTE_IF_SET_IN_SBITMAP (move->uses, 0, u, sbi)
|
{
|
{
|
this_insn = ps_rtl_insn (ps, u);
|
this_insn = ps_rtl_insn (ps, u);
|
this_latency = insn_latency (move->insn, this_insn);
|
this_latency = insn_latency (move->insn, this_insn);
|
if (distance1_uses && !TEST_BIT (distance1_uses, u))
|
if (distance1_uses && !TEST_BIT (distance1_uses, u))
|
this_distance = -1;
|
this_distance = -1;
|
else
|
else
|
this_distance = 0;
|
this_distance = 0;
|
this_time = SCHED_TIME (u) + this_distance * ii;
|
this_time = SCHED_TIME (u) + this_distance * ii;
|
this_start = this_time - ii;
|
this_start = this_time - ii;
|
this_end = this_time - this_latency;
|
this_end = this_time - this_latency;
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "%11d %11d %5d %d --(T,%d,%d)--> %d\n",
|
fprintf (dump_file, "%11d %11d %5d %d --(T,%d,%d)--> %d\n",
|
this_start, this_end, SCHED_TIME (u), INSN_UID (move->insn),
|
this_start, this_end, SCHED_TIME (u), INSN_UID (move->insn),
|
this_latency, this_distance, INSN_UID (this_insn));
|
this_latency, this_distance, INSN_UID (this_insn));
|
|
|
if (start < this_start)
|
if (start < this_start)
|
start = this_start;
|
start = this_start;
|
if (end > this_end)
|
if (end > this_end)
|
end = this_end;
|
end = this_end;
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "----------- ----------- -----\n");
|
fprintf (dump_file, "----------- ----------- -----\n");
|
fprintf (dump_file, "%11d %11d %5s %s\n", start, end, "", "(max, min)");
|
fprintf (dump_file, "%11d %11d %5s %s\n", start, end, "", "(max, min)");
|
}
|
}
|
|
|
sbitmap_zero (must_follow);
|
sbitmap_zero (must_follow);
|
SET_BIT (must_follow, move->def);
|
SET_BIT (must_follow, move->def);
|
|
|
start = MAX (start, end - (ii - 1));
|
start = MAX (start, end - (ii - 1));
|
for (c = end; c >= start; c--)
|
for (c = end; c >= start; c--)
|
{
|
{
|
psi = ps_add_node_check_conflicts (ps, i_reg_move, c,
|
psi = ps_add_node_check_conflicts (ps, i_reg_move, c,
|
move->uses, must_follow);
|
move->uses, must_follow);
|
if (psi)
|
if (psi)
|
{
|
{
|
update_node_sched_params (i_reg_move, ii, c, PS_MIN_CYCLE (ps));
|
update_node_sched_params (i_reg_move, ii, c, PS_MIN_CYCLE (ps));
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "\nScheduled register move INSN %d at"
|
fprintf (dump_file, "\nScheduled register move INSN %d at"
|
" time %d, row %d\n\n", INSN_UID (move->insn), c,
|
" time %d, row %d\n\n", INSN_UID (move->insn), c,
|
SCHED_ROW (i_reg_move));
|
SCHED_ROW (i_reg_move));
|
return true;
|
return true;
|
}
|
}
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "\nNo available slot\n\n");
|
fprintf (dump_file, "\nNo available slot\n\n");
|
|
|
return false;
|
return false;
|
}
|
}
|
|
|
/*
|
/*
|
Breaking intra-loop register anti-dependences:
|
Breaking intra-loop register anti-dependences:
|
Each intra-loop register anti-dependence implies a cross-iteration true
|
Each intra-loop register anti-dependence implies a cross-iteration true
|
dependence of distance 1. Therefore, we can remove such false dependencies
|
dependence of distance 1. Therefore, we can remove such false dependencies
|
and figure out if the partial schedule broke them by checking if (for a
|
and figure out if the partial schedule broke them by checking if (for a
|
true-dependence of distance 1): SCHED_TIME (def) < SCHED_TIME (use) and
|
true-dependence of distance 1): SCHED_TIME (def) < SCHED_TIME (use) and
|
if so generate a register move. The number of such moves is equal to:
|
if so generate a register move. The number of such moves is equal to:
|
SCHED_TIME (use) - SCHED_TIME (def) { 0 broken
|
SCHED_TIME (use) - SCHED_TIME (def) { 0 broken
|
nreg_moves = ----------------------------------- + 1 - { dependence.
|
nreg_moves = ----------------------------------- + 1 - { dependence.
|
ii { 1 if not.
|
ii { 1 if not.
|
*/
|
*/
|
static bool
|
static bool
|
schedule_reg_moves (partial_schedule_ptr ps)
|
schedule_reg_moves (partial_schedule_ptr ps)
|
{
|
{
|
ddg_ptr g = ps->g;
|
ddg_ptr g = ps->g;
|
int ii = ps->ii;
|
int ii = ps->ii;
|
int i;
|
int i;
|
|
|
for (i = 0; i < g->num_nodes; i++)
|
for (i = 0; i < g->num_nodes; i++)
|
{
|
{
|
ddg_node_ptr u = &g->nodes[i];
|
ddg_node_ptr u = &g->nodes[i];
|
ddg_edge_ptr e;
|
ddg_edge_ptr e;
|
int nreg_moves = 0, i_reg_move;
|
int nreg_moves = 0, i_reg_move;
|
rtx prev_reg, old_reg;
|
rtx prev_reg, old_reg;
|
int first_move;
|
int first_move;
|
int distances[2];
|
int distances[2];
|
sbitmap must_follow;
|
sbitmap must_follow;
|
sbitmap distance1_uses;
|
sbitmap distance1_uses;
|
rtx set = single_set (u->insn);
|
rtx set = single_set (u->insn);
|
|
|
/* Skip instructions that do not set a register. */
|
/* Skip instructions that do not set a register. */
|
if ((set && !REG_P (SET_DEST (set))))
|
if ((set && !REG_P (SET_DEST (set))))
|
continue;
|
continue;
|
|
|
/* Compute the number of reg_moves needed for u, by looking at life
|
/* Compute the number of reg_moves needed for u, by looking at life
|
ranges started at u (excluding self-loops). */
|
ranges started at u (excluding self-loops). */
|
distances[0] = distances[1] = false;
|
distances[0] = distances[1] = false;
|
for (e = u->out; e; e = e->next_out)
|
for (e = u->out; e; e = e->next_out)
|
if (e->type == TRUE_DEP && e->dest != e->src)
|
if (e->type == TRUE_DEP && e->dest != e->src)
|
{
|
{
|
int nreg_moves4e = (SCHED_TIME (e->dest->cuid)
|
int nreg_moves4e = (SCHED_TIME (e->dest->cuid)
|
- SCHED_TIME (e->src->cuid)) / ii;
|
- SCHED_TIME (e->src->cuid)) / ii;
|
|
|
if (e->distance == 1)
|
if (e->distance == 1)
|
nreg_moves4e = (SCHED_TIME (e->dest->cuid)
|
nreg_moves4e = (SCHED_TIME (e->dest->cuid)
|
- SCHED_TIME (e->src->cuid) + ii) / ii;
|
- SCHED_TIME (e->src->cuid) + ii) / ii;
|
|
|
/* If dest precedes src in the schedule of the kernel, then dest
|
/* If dest precedes src in the schedule of the kernel, then dest
|
will read before src writes and we can save one reg_copy. */
|
will read before src writes and we can save one reg_copy. */
|
if (SCHED_ROW (e->dest->cuid) == SCHED_ROW (e->src->cuid)
|
if (SCHED_ROW (e->dest->cuid) == SCHED_ROW (e->src->cuid)
|
&& SCHED_COLUMN (e->dest->cuid) < SCHED_COLUMN (e->src->cuid))
|
&& SCHED_COLUMN (e->dest->cuid) < SCHED_COLUMN (e->src->cuid))
|
nreg_moves4e--;
|
nreg_moves4e--;
|
|
|
if (nreg_moves4e >= 1)
|
if (nreg_moves4e >= 1)
|
{
|
{
|
/* !single_set instructions are not supported yet and
|
/* !single_set instructions are not supported yet and
|
thus we do not except to encounter them in the loop
|
thus we do not except to encounter them in the loop
|
except from the doloop part. For the latter case
|
except from the doloop part. For the latter case
|
we assume no regmoves are generated as the doloop
|
we assume no regmoves are generated as the doloop
|
instructions are tied to the branch with an edge. */
|
instructions are tied to the branch with an edge. */
|
gcc_assert (set);
|
gcc_assert (set);
|
/* If the instruction contains auto-inc register then
|
/* If the instruction contains auto-inc register then
|
validate that the regmov is being generated for the
|
validate that the regmov is being generated for the
|
target regsiter rather then the inc'ed register. */
|
target regsiter rather then the inc'ed register. */
|
gcc_assert (!autoinc_var_is_used_p (u->insn, e->dest->insn));
|
gcc_assert (!autoinc_var_is_used_p (u->insn, e->dest->insn));
|
}
|
}
|
|
|
if (nreg_moves4e)
|
if (nreg_moves4e)
|
{
|
{
|
gcc_assert (e->distance < 2);
|
gcc_assert (e->distance < 2);
|
distances[e->distance] = true;
|
distances[e->distance] = true;
|
}
|
}
|
nreg_moves = MAX (nreg_moves, nreg_moves4e);
|
nreg_moves = MAX (nreg_moves, nreg_moves4e);
|
}
|
}
|
|
|
if (nreg_moves == 0)
|
if (nreg_moves == 0)
|
continue;
|
continue;
|
|
|
/* Create NREG_MOVES register moves. */
|
/* Create NREG_MOVES register moves. */
|
first_move = VEC_length (ps_reg_move_info, ps->reg_moves);
|
first_move = VEC_length (ps_reg_move_info, ps->reg_moves);
|
VEC_safe_grow_cleared (ps_reg_move_info, heap, ps->reg_moves,
|
VEC_safe_grow_cleared (ps_reg_move_info, heap, ps->reg_moves,
|
first_move + nreg_moves);
|
first_move + nreg_moves);
|
extend_node_sched_params (ps);
|
extend_node_sched_params (ps);
|
|
|
/* Record the moves associated with this node. */
|
/* Record the moves associated with this node. */
|
first_move += ps->g->num_nodes;
|
first_move += ps->g->num_nodes;
|
|
|
/* Generate each move. */
|
/* Generate each move. */
|
old_reg = prev_reg = SET_DEST (single_set (u->insn));
|
old_reg = prev_reg = SET_DEST (single_set (u->insn));
|
for (i_reg_move = 0; i_reg_move < nreg_moves; i_reg_move++)
|
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);
|
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->def = i_reg_move > 0 ? first_move + i_reg_move - 1 : i;
|
move->uses = sbitmap_alloc (first_move + nreg_moves);
|
move->uses = sbitmap_alloc (first_move + nreg_moves);
|
move->old_reg = old_reg;
|
move->old_reg = old_reg;
|
move->new_reg = gen_reg_rtx (GET_MODE (prev_reg));
|
move->new_reg = gen_reg_rtx (GET_MODE (prev_reg));
|
move->num_consecutive_stages = distances[0] && distances[1] ? 2 : 1;
|
move->num_consecutive_stages = distances[0] && distances[1] ? 2 : 1;
|
move->insn = gen_move_insn (move->new_reg, copy_rtx (prev_reg));
|
move->insn = gen_move_insn (move->new_reg, copy_rtx (prev_reg));
|
sbitmap_zero (move->uses);
|
sbitmap_zero (move->uses);
|
|
|
prev_reg = move->new_reg;
|
prev_reg = move->new_reg;
|
}
|
}
|
|
|
distance1_uses = distances[1] ? sbitmap_alloc (g->num_nodes) : NULL;
|
distance1_uses = distances[1] ? sbitmap_alloc (g->num_nodes) : NULL;
|
|
|
/* Every use of the register defined by node may require a different
|
/* Every use of the register defined by node may require a different
|
copy of this register, depending on the time the use is scheduled.
|
copy of this register, depending on the time the use is scheduled.
|
Record which uses require which move results. */
|
Record which uses require which move results. */
|
for (e = u->out; e; e = e->next_out)
|
for (e = u->out; e; e = e->next_out)
|
if (e->type == TRUE_DEP && e->dest != e->src)
|
if (e->type == TRUE_DEP && e->dest != e->src)
|
{
|
{
|
int dest_copy = (SCHED_TIME (e->dest->cuid)
|
int dest_copy = (SCHED_TIME (e->dest->cuid)
|
- SCHED_TIME (e->src->cuid)) / ii;
|
- SCHED_TIME (e->src->cuid)) / ii;
|
|
|
if (e->distance == 1)
|
if (e->distance == 1)
|
dest_copy = (SCHED_TIME (e->dest->cuid)
|
dest_copy = (SCHED_TIME (e->dest->cuid)
|
- SCHED_TIME (e->src->cuid) + ii) / ii;
|
- SCHED_TIME (e->src->cuid) + ii) / ii;
|
|
|
if (SCHED_ROW (e->dest->cuid) == SCHED_ROW (e->src->cuid)
|
if (SCHED_ROW (e->dest->cuid) == SCHED_ROW (e->src->cuid)
|
&& SCHED_COLUMN (e->dest->cuid) < SCHED_COLUMN (e->src->cuid))
|
&& SCHED_COLUMN (e->dest->cuid) < SCHED_COLUMN (e->src->cuid))
|
dest_copy--;
|
dest_copy--;
|
|
|
if (dest_copy)
|
if (dest_copy)
|
{
|
{
|
ps_reg_move_info *move;
|
ps_reg_move_info *move;
|
|
|
move = ps_reg_move (ps, first_move + dest_copy - 1);
|
move = ps_reg_move (ps, first_move + dest_copy - 1);
|
SET_BIT (move->uses, e->dest->cuid);
|
SET_BIT (move->uses, e->dest->cuid);
|
if (e->distance == 1)
|
if (e->distance == 1)
|
SET_BIT (distance1_uses, e->dest->cuid);
|
SET_BIT (distance1_uses, e->dest->cuid);
|
}
|
}
|
}
|
}
|
|
|
must_follow = sbitmap_alloc (first_move + nreg_moves);
|
must_follow = sbitmap_alloc (first_move + nreg_moves);
|
for (i_reg_move = 0; i_reg_move < nreg_moves; i_reg_move++)
|
for (i_reg_move = 0; i_reg_move < nreg_moves; i_reg_move++)
|
if (!schedule_reg_move (ps, first_move + i_reg_move,
|
if (!schedule_reg_move (ps, first_move + i_reg_move,
|
distance1_uses, must_follow))
|
distance1_uses, must_follow))
|
break;
|
break;
|
sbitmap_free (must_follow);
|
sbitmap_free (must_follow);
|
if (distance1_uses)
|
if (distance1_uses)
|
sbitmap_free (distance1_uses);
|
sbitmap_free (distance1_uses);
|
if (i_reg_move < nreg_moves)
|
if (i_reg_move < nreg_moves)
|
return false;
|
return false;
|
}
|
}
|
return true;
|
return true;
|
}
|
}
|
|
|
/* Emit the moves associatied with PS. Apply the substitutions
|
/* Emit the moves associatied with PS. Apply the substitutions
|
associated with them. */
|
associated with them. */
|
static void
|
static void
|
apply_reg_moves (partial_schedule_ptr ps)
|
apply_reg_moves (partial_schedule_ptr ps)
|
{
|
{
|
ps_reg_move_info *move;
|
ps_reg_move_info *move;
|
int i;
|
int i;
|
|
|
FOR_EACH_VEC_ELT (ps_reg_move_info, ps->reg_moves, i, move)
|
FOR_EACH_VEC_ELT (ps_reg_move_info, ps->reg_moves, i, move)
|
{
|
{
|
unsigned int i_use;
|
unsigned int i_use;
|
sbitmap_iterator sbi;
|
sbitmap_iterator sbi;
|
|
|
EXECUTE_IF_SET_IN_SBITMAP (move->uses, 0, i_use, 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);
|
replace_rtx (ps->g->nodes[i_use].insn, move->old_reg, move->new_reg);
|
df_insn_rescan (ps->g->nodes[i_use].insn);
|
df_insn_rescan (ps->g->nodes[i_use].insn);
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
/* Bump the SCHED_TIMEs of all nodes by AMOUNT. Set the values of
|
/* Bump the SCHED_TIMEs of all nodes by AMOUNT. Set the values of
|
SCHED_ROW and SCHED_STAGE. Instruction scheduled on cycle AMOUNT
|
SCHED_ROW and SCHED_STAGE. Instruction scheduled on cycle AMOUNT
|
will move to cycle zero. */
|
will move to cycle zero. */
|
static void
|
static void
|
reset_sched_times (partial_schedule_ptr ps, int amount)
|
reset_sched_times (partial_schedule_ptr ps, int amount)
|
{
|
{
|
int row;
|
int row;
|
int ii = ps->ii;
|
int ii = ps->ii;
|
ps_insn_ptr crr_insn;
|
ps_insn_ptr crr_insn;
|
|
|
for (row = 0; row < ii; row++)
|
for (row = 0; row < ii; row++)
|
for (crr_insn = ps->rows[row]; crr_insn; crr_insn = crr_insn->next_in_row)
|
for (crr_insn = ps->rows[row]; crr_insn; crr_insn = crr_insn->next_in_row)
|
{
|
{
|
int u = crr_insn->id;
|
int u = crr_insn->id;
|
int normalized_time = SCHED_TIME (u) - amount;
|
int normalized_time = SCHED_TIME (u) - amount;
|
int new_min_cycle = PS_MIN_CYCLE (ps) - amount;
|
int new_min_cycle = PS_MIN_CYCLE (ps) - amount;
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
/* Print the scheduling times after the rotation. */
|
/* Print the scheduling times after the rotation. */
|
rtx insn = ps_rtl_insn (ps, u);
|
rtx insn = ps_rtl_insn (ps, u);
|
|
|
fprintf (dump_file, "crr_insn->node=%d (insn id %d), "
|
fprintf (dump_file, "crr_insn->node=%d (insn id %d), "
|
"crr_insn->cycle=%d, min_cycle=%d", u,
|
"crr_insn->cycle=%d, min_cycle=%d", u,
|
INSN_UID (insn), normalized_time, new_min_cycle);
|
INSN_UID (insn), normalized_time, new_min_cycle);
|
if (JUMP_P (insn))
|
if (JUMP_P (insn))
|
fprintf (dump_file, " (branch)");
|
fprintf (dump_file, " (branch)");
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
}
|
}
|
|
|
gcc_assert (SCHED_TIME (u) >= ps->min_cycle);
|
gcc_assert (SCHED_TIME (u) >= ps->min_cycle);
|
gcc_assert (SCHED_TIME (u) <= ps->max_cycle);
|
gcc_assert (SCHED_TIME (u) <= ps->max_cycle);
|
|
|
crr_insn->cycle = normalized_time;
|
crr_insn->cycle = normalized_time;
|
update_node_sched_params (u, ii, normalized_time, new_min_cycle);
|
update_node_sched_params (u, ii, normalized_time, new_min_cycle);
|
}
|
}
|
}
|
}
|
|
|
/* Permute the insns according to their order in PS, from row 0 to
|
/* Permute the insns according to their order in PS, from row 0 to
|
row ii-1, and position them right before LAST. This schedules
|
row ii-1, and position them right before LAST. This schedules
|
the insns of the loop kernel. */
|
the insns of the loop kernel. */
|
static void
|
static void
|
permute_partial_schedule (partial_schedule_ptr ps, rtx last)
|
permute_partial_schedule (partial_schedule_ptr ps, rtx last)
|
{
|
{
|
int ii = ps->ii;
|
int ii = ps->ii;
|
int row;
|
int row;
|
ps_insn_ptr ps_ij;
|
ps_insn_ptr ps_ij;
|
|
|
for (row = 0; row < ii ; row++)
|
for (row = 0; row < ii ; row++)
|
for (ps_ij = ps->rows[row]; ps_ij; ps_ij = ps_ij->next_in_row)
|
for (ps_ij = ps->rows[row]; ps_ij; ps_ij = ps_ij->next_in_row)
|
{
|
{
|
rtx insn = ps_rtl_insn (ps, ps_ij->id);
|
rtx insn = ps_rtl_insn (ps, ps_ij->id);
|
|
|
if (PREV_INSN (last) != insn)
|
if (PREV_INSN (last) != insn)
|
{
|
{
|
if (ps_ij->id < ps->g->num_nodes)
|
if (ps_ij->id < ps->g->num_nodes)
|
reorder_insns_nobb (ps_first_note (ps, ps_ij->id), insn,
|
reorder_insns_nobb (ps_first_note (ps, ps_ij->id), insn,
|
PREV_INSN (last));
|
PREV_INSN (last));
|
else
|
else
|
add_insn_before (insn, last, NULL);
|
add_insn_before (insn, last, NULL);
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
/* Set bitmaps TMP_FOLLOW and TMP_PRECEDE to MUST_FOLLOW and MUST_PRECEDE
|
/* Set bitmaps TMP_FOLLOW and TMP_PRECEDE to MUST_FOLLOW and MUST_PRECEDE
|
respectively only if cycle C falls on the border of the scheduling
|
respectively only if cycle C falls on the border of the scheduling
|
window boundaries marked by START and END cycles. STEP is the
|
window boundaries marked by START and END cycles. STEP is the
|
direction of the window. */
|
direction of the window. */
|
static inline void
|
static inline void
|
set_must_precede_follow (sbitmap *tmp_follow, sbitmap must_follow,
|
set_must_precede_follow (sbitmap *tmp_follow, sbitmap must_follow,
|
sbitmap *tmp_precede, sbitmap must_precede, int c,
|
sbitmap *tmp_precede, sbitmap must_precede, int c,
|
int start, int end, int step)
|
int start, int end, int step)
|
{
|
{
|
*tmp_precede = NULL;
|
*tmp_precede = NULL;
|
*tmp_follow = NULL;
|
*tmp_follow = NULL;
|
|
|
if (c == start)
|
if (c == start)
|
{
|
{
|
if (step == 1)
|
if (step == 1)
|
*tmp_precede = must_precede;
|
*tmp_precede = must_precede;
|
else /* step == -1. */
|
else /* step == -1. */
|
*tmp_follow = must_follow;
|
*tmp_follow = must_follow;
|
}
|
}
|
if (c == end - step)
|
if (c == end - step)
|
{
|
{
|
if (step == 1)
|
if (step == 1)
|
*tmp_follow = must_follow;
|
*tmp_follow = must_follow;
|
else /* step == -1. */
|
else /* step == -1. */
|
*tmp_precede = must_precede;
|
*tmp_precede = must_precede;
|
}
|
}
|
|
|
}
|
}
|
|
|
/* Return True if the branch can be moved to row ii-1 while
|
/* Return True if the branch can be moved to row ii-1 while
|
normalizing the partial schedule PS to start from cycle zero and thus
|
normalizing the partial schedule PS to start from cycle zero and thus
|
optimize the SC. Otherwise return False. */
|
optimize the SC. Otherwise return False. */
|
static bool
|
static bool
|
optimize_sc (partial_schedule_ptr ps, ddg_ptr g)
|
optimize_sc (partial_schedule_ptr ps, ddg_ptr g)
|
{
|
{
|
int amount = PS_MIN_CYCLE (ps);
|
int amount = PS_MIN_CYCLE (ps);
|
sbitmap sched_nodes = sbitmap_alloc (g->num_nodes);
|
sbitmap sched_nodes = sbitmap_alloc (g->num_nodes);
|
int start, end, step;
|
int start, end, step;
|
int ii = ps->ii;
|
int ii = ps->ii;
|
bool ok = false;
|
bool ok = false;
|
int stage_count, stage_count_curr;
|
int stage_count, stage_count_curr;
|
|
|
/* Compare the SC after normalization and SC after bringing the branch
|
/* Compare the SC after normalization and SC after bringing the branch
|
to row ii-1. If they are equal just bail out. */
|
to row ii-1. If they are equal just bail out. */
|
stage_count = calculate_stage_count (ps, amount);
|
stage_count = calculate_stage_count (ps, amount);
|
stage_count_curr =
|
stage_count_curr =
|
calculate_stage_count (ps, SCHED_TIME (g->closing_branch->cuid) - (ii - 1));
|
calculate_stage_count (ps, SCHED_TIME (g->closing_branch->cuid) - (ii - 1));
|
|
|
if (stage_count == stage_count_curr)
|
if (stage_count == stage_count_curr)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "SMS SC already optimized.\n");
|
fprintf (dump_file, "SMS SC already optimized.\n");
|
|
|
ok = false;
|
ok = false;
|
goto clear;
|
goto clear;
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "SMS Trying to optimize branch location\n");
|
fprintf (dump_file, "SMS Trying to optimize branch location\n");
|
fprintf (dump_file, "SMS partial schedule before trial:\n");
|
fprintf (dump_file, "SMS partial schedule before trial:\n");
|
print_partial_schedule (ps, dump_file);
|
print_partial_schedule (ps, dump_file);
|
}
|
}
|
|
|
/* First, normalize the partial scheduling. */
|
/* First, normalize the partial scheduling. */
|
reset_sched_times (ps, amount);
|
reset_sched_times (ps, amount);
|
rotate_partial_schedule (ps, amount);
|
rotate_partial_schedule (ps, amount);
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file,
|
fprintf (dump_file,
|
"SMS partial schedule after normalization (ii, %d, SC %d):\n",
|
"SMS partial schedule after normalization (ii, %d, SC %d):\n",
|
ii, stage_count);
|
ii, stage_count);
|
print_partial_schedule (ps, dump_file);
|
print_partial_schedule (ps, dump_file);
|
}
|
}
|
|
|
if (SMODULO (SCHED_TIME (g->closing_branch->cuid), ii) == ii - 1)
|
if (SMODULO (SCHED_TIME (g->closing_branch->cuid), ii) == ii - 1)
|
{
|
{
|
ok = true;
|
ok = true;
|
goto clear;
|
goto clear;
|
}
|
}
|
|
|
sbitmap_ones (sched_nodes);
|
sbitmap_ones (sched_nodes);
|
|
|
/* Calculate the new placement of the branch. It should be in row
|
/* Calculate the new placement of the branch. It should be in row
|
ii-1 and fall into it's scheduling window. */
|
ii-1 and fall into it's scheduling window. */
|
if (get_sched_window (ps, g->closing_branch, sched_nodes, ii, &start,
|
if (get_sched_window (ps, g->closing_branch, sched_nodes, ii, &start,
|
&step, &end) == 0)
|
&step, &end) == 0)
|
{
|
{
|
bool success;
|
bool success;
|
ps_insn_ptr next_ps_i;
|
ps_insn_ptr next_ps_i;
|
int branch_cycle = SCHED_TIME (g->closing_branch->cuid);
|
int branch_cycle = SCHED_TIME (g->closing_branch->cuid);
|
int row = SMODULO (branch_cycle, ps->ii);
|
int row = SMODULO (branch_cycle, ps->ii);
|
int num_splits = 0;
|
int num_splits = 0;
|
sbitmap must_precede, must_follow, tmp_precede, tmp_follow;
|
sbitmap must_precede, must_follow, tmp_precede, tmp_follow;
|
int c;
|
int c;
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "\nTrying to schedule node %d "
|
fprintf (dump_file, "\nTrying to schedule node %d "
|
"INSN = %d in (%d .. %d) step %d\n",
|
"INSN = %d in (%d .. %d) step %d\n",
|
g->closing_branch->cuid,
|
g->closing_branch->cuid,
|
(INSN_UID (g->closing_branch->insn)), start, end, step);
|
(INSN_UID (g->closing_branch->insn)), start, end, step);
|
|
|
gcc_assert ((step > 0 && start < end) || (step < 0 && start > end));
|
gcc_assert ((step > 0 && start < end) || (step < 0 && start > end));
|
if (step == 1)
|
if (step == 1)
|
{
|
{
|
c = start + ii - SMODULO (start, ii) - 1;
|
c = start + ii - SMODULO (start, ii) - 1;
|
gcc_assert (c >= start);
|
gcc_assert (c >= start);
|
if (c >= end)
|
if (c >= end)
|
{
|
{
|
ok = false;
|
ok = false;
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
"SMS failed to schedule branch at cycle: %d\n", c);
|
"SMS failed to schedule branch at cycle: %d\n", c);
|
goto clear;
|
goto clear;
|
}
|
}
|
}
|
}
|
else
|
else
|
{
|
{
|
c = start - SMODULO (start, ii) - 1;
|
c = start - SMODULO (start, ii) - 1;
|
gcc_assert (c <= start);
|
gcc_assert (c <= start);
|
|
|
if (c <= end)
|
if (c <= end)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
"SMS failed to schedule branch at cycle: %d\n", c);
|
"SMS failed to schedule branch at cycle: %d\n", c);
|
ok = false;
|
ok = false;
|
goto clear;
|
goto clear;
|
}
|
}
|
}
|
}
|
|
|
must_precede = sbitmap_alloc (g->num_nodes);
|
must_precede = sbitmap_alloc (g->num_nodes);
|
must_follow = sbitmap_alloc (g->num_nodes);
|
must_follow = sbitmap_alloc (g->num_nodes);
|
|
|
/* Try to schedule the branch is it's new cycle. */
|
/* Try to schedule the branch is it's new cycle. */
|
calculate_must_precede_follow (g->closing_branch, start, end,
|
calculate_must_precede_follow (g->closing_branch, start, end,
|
step, ii, sched_nodes,
|
step, ii, sched_nodes,
|
must_precede, must_follow);
|
must_precede, must_follow);
|
|
|
set_must_precede_follow (&tmp_follow, must_follow, &tmp_precede,
|
set_must_precede_follow (&tmp_follow, must_follow, &tmp_precede,
|
must_precede, c, start, end, step);
|
must_precede, c, start, end, step);
|
|
|
/* Find the element in the partial schedule related to the closing
|
/* Find the element in the partial schedule related to the closing
|
branch so we can remove it from it's current cycle. */
|
branch so we can remove it from it's current cycle. */
|
for (next_ps_i = ps->rows[row];
|
for (next_ps_i = ps->rows[row];
|
next_ps_i; next_ps_i = next_ps_i->next_in_row)
|
next_ps_i; next_ps_i = next_ps_i->next_in_row)
|
if (next_ps_i->id == g->closing_branch->cuid)
|
if (next_ps_i->id == g->closing_branch->cuid)
|
break;
|
break;
|
|
|
remove_node_from_ps (ps, next_ps_i);
|
remove_node_from_ps (ps, next_ps_i);
|
success =
|
success =
|
try_scheduling_node_in_cycle (ps, g->closing_branch->cuid, c,
|
try_scheduling_node_in_cycle (ps, g->closing_branch->cuid, c,
|
sched_nodes, &num_splits,
|
sched_nodes, &num_splits,
|
tmp_precede, tmp_follow);
|
tmp_precede, tmp_follow);
|
gcc_assert (num_splits == 0);
|
gcc_assert (num_splits == 0);
|
if (!success)
|
if (!success)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
"SMS failed to schedule branch at cycle: %d, "
|
"SMS failed to schedule branch at cycle: %d, "
|
"bringing it back to cycle %d\n", c, branch_cycle);
|
"bringing it back to cycle %d\n", c, branch_cycle);
|
|
|
/* The branch was failed to be placed in row ii - 1.
|
/* The branch was failed to be placed in row ii - 1.
|
Put it back in it's original place in the partial
|
Put it back in it's original place in the partial
|
schedualing. */
|
schedualing. */
|
set_must_precede_follow (&tmp_follow, must_follow, &tmp_precede,
|
set_must_precede_follow (&tmp_follow, must_follow, &tmp_precede,
|
must_precede, branch_cycle, start, end,
|
must_precede, branch_cycle, start, end,
|
step);
|
step);
|
success =
|
success =
|
try_scheduling_node_in_cycle (ps, g->closing_branch->cuid,
|
try_scheduling_node_in_cycle (ps, g->closing_branch->cuid,
|
branch_cycle, sched_nodes,
|
branch_cycle, sched_nodes,
|
&num_splits, tmp_precede,
|
&num_splits, tmp_precede,
|
tmp_follow);
|
tmp_follow);
|
gcc_assert (success && (num_splits == 0));
|
gcc_assert (success && (num_splits == 0));
|
ok = false;
|
ok = false;
|
}
|
}
|
else
|
else
|
{
|
{
|
/* The branch is placed in row ii - 1. */
|
/* The branch is placed in row ii - 1. */
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
"SMS success in moving branch to cycle %d\n", c);
|
"SMS success in moving branch to cycle %d\n", c);
|
|
|
update_node_sched_params (g->closing_branch->cuid, ii, c,
|
update_node_sched_params (g->closing_branch->cuid, ii, c,
|
PS_MIN_CYCLE (ps));
|
PS_MIN_CYCLE (ps));
|
ok = true;
|
ok = true;
|
}
|
}
|
|
|
free (must_precede);
|
free (must_precede);
|
free (must_follow);
|
free (must_follow);
|
}
|
}
|
|
|
clear:
|
clear:
|
free (sched_nodes);
|
free (sched_nodes);
|
return ok;
|
return ok;
|
}
|
}
|
|
|
static void
|
static void
|
duplicate_insns_of_cycles (partial_schedule_ptr ps, int from_stage,
|
duplicate_insns_of_cycles (partial_schedule_ptr ps, int from_stage,
|
int to_stage, rtx count_reg)
|
int to_stage, rtx count_reg)
|
{
|
{
|
int row;
|
int row;
|
ps_insn_ptr ps_ij;
|
ps_insn_ptr ps_ij;
|
|
|
for (row = 0; row < ps->ii; row++)
|
for (row = 0; row < ps->ii; row++)
|
for (ps_ij = ps->rows[row]; ps_ij; ps_ij = ps_ij->next_in_row)
|
for (ps_ij = ps->rows[row]; ps_ij; ps_ij = ps_ij->next_in_row)
|
{
|
{
|
int u = ps_ij->id;
|
int u = ps_ij->id;
|
int first_u, last_u;
|
int first_u, last_u;
|
rtx u_insn;
|
rtx u_insn;
|
|
|
/* Do not duplicate any insn which refers to count_reg as it
|
/* Do not duplicate any insn which refers to count_reg as it
|
belongs to the control part.
|
belongs to the control part.
|
The closing branch is scheduled as well and thus should
|
The closing branch is scheduled as well and thus should
|
be ignored.
|
be ignored.
|
TODO: This should be done by analyzing the control part of
|
TODO: This should be done by analyzing the control part of
|
the loop. */
|
the loop. */
|
u_insn = ps_rtl_insn (ps, u);
|
u_insn = ps_rtl_insn (ps, u);
|
if (reg_mentioned_p (count_reg, u_insn)
|
if (reg_mentioned_p (count_reg, u_insn)
|
|| JUMP_P (u_insn))
|
|| JUMP_P (u_insn))
|
continue;
|
continue;
|
|
|
first_u = SCHED_STAGE (u);
|
first_u = SCHED_STAGE (u);
|
last_u = first_u + ps_num_consecutive_stages (ps, u) - 1;
|
last_u = first_u + ps_num_consecutive_stages (ps, u) - 1;
|
if (from_stage <= last_u && to_stage >= first_u)
|
if (from_stage <= last_u && to_stage >= first_u)
|
{
|
{
|
if (u < ps->g->num_nodes)
|
if (u < ps->g->num_nodes)
|
duplicate_insn_chain (ps_first_note (ps, u), u_insn);
|
duplicate_insn_chain (ps_first_note (ps, u), u_insn);
|
else
|
else
|
emit_insn (copy_rtx (PATTERN (u_insn)));
|
emit_insn (copy_rtx (PATTERN (u_insn)));
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
|
|
/* Generate the instructions (including reg_moves) for prolog & epilog. */
|
/* Generate the instructions (including reg_moves) for prolog & epilog. */
|
static void
|
static void
|
generate_prolog_epilog (partial_schedule_ptr ps, struct loop *loop,
|
generate_prolog_epilog (partial_schedule_ptr ps, struct loop *loop,
|
rtx count_reg, rtx count_init)
|
rtx count_reg, rtx count_init)
|
{
|
{
|
int i;
|
int i;
|
int last_stage = PS_STAGE_COUNT (ps) - 1;
|
int last_stage = PS_STAGE_COUNT (ps) - 1;
|
edge e;
|
edge e;
|
|
|
/* Generate the prolog, inserting its insns on the loop-entry edge. */
|
/* Generate the prolog, inserting its insns on the loop-entry edge. */
|
start_sequence ();
|
start_sequence ();
|
|
|
if (!count_init)
|
if (!count_init)
|
{
|
{
|
/* Generate instructions at the beginning of the prolog to
|
/* Generate instructions at the beginning of the prolog to
|
adjust the loop count by STAGE_COUNT. If loop count is constant
|
adjust the loop count by STAGE_COUNT. If loop count is constant
|
(count_init), this constant is adjusted by STAGE_COUNT in
|
(count_init), this constant is adjusted by STAGE_COUNT in
|
generate_prolog_epilog function. */
|
generate_prolog_epilog function. */
|
rtx sub_reg = NULL_RTX;
|
rtx sub_reg = NULL_RTX;
|
|
|
sub_reg = expand_simple_binop (GET_MODE (count_reg), MINUS,
|
sub_reg = expand_simple_binop (GET_MODE (count_reg), MINUS,
|
count_reg, GEN_INT (last_stage),
|
count_reg, GEN_INT (last_stage),
|
count_reg, 1, OPTAB_DIRECT);
|
count_reg, 1, OPTAB_DIRECT);
|
gcc_assert (REG_P (sub_reg));
|
gcc_assert (REG_P (sub_reg));
|
if (REGNO (sub_reg) != REGNO (count_reg))
|
if (REGNO (sub_reg) != REGNO (count_reg))
|
emit_move_insn (count_reg, sub_reg);
|
emit_move_insn (count_reg, sub_reg);
|
}
|
}
|
|
|
for (i = 0; i < last_stage; i++)
|
for (i = 0; i < last_stage; i++)
|
duplicate_insns_of_cycles (ps, 0, i, count_reg);
|
duplicate_insns_of_cycles (ps, 0, i, count_reg);
|
|
|
/* Put the prolog on the entry edge. */
|
/* Put the prolog on the entry edge. */
|
e = loop_preheader_edge (loop);
|
e = loop_preheader_edge (loop);
|
split_edge_and_insert (e, get_insns ());
|
split_edge_and_insert (e, get_insns ());
|
if (!flag_resched_modulo_sched)
|
if (!flag_resched_modulo_sched)
|
e->dest->flags |= BB_DISABLE_SCHEDULE;
|
e->dest->flags |= BB_DISABLE_SCHEDULE;
|
|
|
end_sequence ();
|
end_sequence ();
|
|
|
/* Generate the epilog, inserting its insns on the loop-exit edge. */
|
/* Generate the epilog, inserting its insns on the loop-exit edge. */
|
start_sequence ();
|
start_sequence ();
|
|
|
for (i = 0; i < last_stage; i++)
|
for (i = 0; i < last_stage; i++)
|
duplicate_insns_of_cycles (ps, i + 1, last_stage, count_reg);
|
duplicate_insns_of_cycles (ps, i + 1, last_stage, count_reg);
|
|
|
/* Put the epilogue on the exit edge. */
|
/* Put the epilogue on the exit edge. */
|
gcc_assert (single_exit (loop));
|
gcc_assert (single_exit (loop));
|
e = single_exit (loop);
|
e = single_exit (loop);
|
split_edge_and_insert (e, get_insns ());
|
split_edge_and_insert (e, get_insns ());
|
if (!flag_resched_modulo_sched)
|
if (!flag_resched_modulo_sched)
|
e->dest->flags |= BB_DISABLE_SCHEDULE;
|
e->dest->flags |= BB_DISABLE_SCHEDULE;
|
|
|
end_sequence ();
|
end_sequence ();
|
}
|
}
|
|
|
/* Mark LOOP as software pipelined so the later
|
/* Mark LOOP as software pipelined so the later
|
scheduling passes don't touch it. */
|
scheduling passes don't touch it. */
|
static void
|
static void
|
mark_loop_unsched (struct loop *loop)
|
mark_loop_unsched (struct loop *loop)
|
{
|
{
|
unsigned i;
|
unsigned i;
|
basic_block *bbs = get_loop_body (loop);
|
basic_block *bbs = get_loop_body (loop);
|
|
|
for (i = 0; i < loop->num_nodes; i++)
|
for (i = 0; i < loop->num_nodes; i++)
|
bbs[i]->flags |= BB_DISABLE_SCHEDULE;
|
bbs[i]->flags |= BB_DISABLE_SCHEDULE;
|
|
|
free (bbs);
|
free (bbs);
|
}
|
}
|
|
|
/* Return true if all the BBs of the loop are empty except the
|
/* Return true if all the BBs of the loop are empty except the
|
loop header. */
|
loop header. */
|
static bool
|
static bool
|
loop_single_full_bb_p (struct loop *loop)
|
loop_single_full_bb_p (struct loop *loop)
|
{
|
{
|
unsigned i;
|
unsigned i;
|
basic_block *bbs = get_loop_body (loop);
|
basic_block *bbs = get_loop_body (loop);
|
|
|
for (i = 0; i < loop->num_nodes ; i++)
|
for (i = 0; i < loop->num_nodes ; i++)
|
{
|
{
|
rtx head, tail;
|
rtx head, tail;
|
bool empty_bb = true;
|
bool empty_bb = true;
|
|
|
if (bbs[i] == loop->header)
|
if (bbs[i] == loop->header)
|
continue;
|
continue;
|
|
|
/* Make sure that basic blocks other than the header
|
/* Make sure that basic blocks other than the header
|
have only notes labels or jumps. */
|
have only notes labels or jumps. */
|
get_ebb_head_tail (bbs[i], bbs[i], &head, &tail);
|
get_ebb_head_tail (bbs[i], bbs[i], &head, &tail);
|
for (; head != NEXT_INSN (tail); head = NEXT_INSN (head))
|
for (; head != NEXT_INSN (tail); head = NEXT_INSN (head))
|
{
|
{
|
if (NOTE_P (head) || LABEL_P (head)
|
if (NOTE_P (head) || LABEL_P (head)
|
|| (INSN_P (head) && (DEBUG_INSN_P (head) || JUMP_P (head))))
|
|| (INSN_P (head) && (DEBUG_INSN_P (head) || JUMP_P (head))))
|
continue;
|
continue;
|
empty_bb = false;
|
empty_bb = false;
|
break;
|
break;
|
}
|
}
|
|
|
if (! empty_bb)
|
if (! empty_bb)
|
{
|
{
|
free (bbs);
|
free (bbs);
|
return false;
|
return false;
|
}
|
}
|
}
|
}
|
free (bbs);
|
free (bbs);
|
return true;
|
return true;
|
}
|
}
|
|
|
/* Dump file:line from INSN's location info to dump_file. */
|
/* Dump file:line from INSN's location info to dump_file. */
|
|
|
static void
|
static void
|
dump_insn_locator (rtx insn)
|
dump_insn_locator (rtx insn)
|
{
|
{
|
if (dump_file && INSN_LOCATOR (insn))
|
if (dump_file && INSN_LOCATOR (insn))
|
{
|
{
|
const char *file = insn_file (insn);
|
const char *file = insn_file (insn);
|
if (file)
|
if (file)
|
fprintf (dump_file, " %s:%i", file, insn_line (insn));
|
fprintf (dump_file, " %s:%i", file, insn_line (insn));
|
}
|
}
|
}
|
}
|
|
|
/* A simple loop from SMS point of view; it is a loop that is composed of
|
/* A simple loop from SMS point of view; it is a loop that is composed of
|
either a single basic block or two BBs - a header and a latch. */
|
either a single basic block or two BBs - a header and a latch. */
|
#define SIMPLE_SMS_LOOP_P(loop) ((loop->num_nodes < 3 ) \
|
#define SIMPLE_SMS_LOOP_P(loop) ((loop->num_nodes < 3 ) \
|
&& (EDGE_COUNT (loop->latch->preds) == 1) \
|
&& (EDGE_COUNT (loop->latch->preds) == 1) \
|
&& (EDGE_COUNT (loop->latch->succs) == 1))
|
&& (EDGE_COUNT (loop->latch->succs) == 1))
|
|
|
/* Return true if the loop is in its canonical form and false if not.
|
/* Return true if the loop is in its canonical form and false if not.
|
i.e. SIMPLE_SMS_LOOP_P and have one preheader block, and single exit. */
|
i.e. SIMPLE_SMS_LOOP_P and have one preheader block, and single exit. */
|
static bool
|
static bool
|
loop_canon_p (struct loop *loop)
|
loop_canon_p (struct loop *loop)
|
{
|
{
|
|
|
if (loop->inner || !loop_outer (loop))
|
if (loop->inner || !loop_outer (loop))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "SMS loop inner or !loop_outer\n");
|
fprintf (dump_file, "SMS loop inner or !loop_outer\n");
|
return false;
|
return false;
|
}
|
}
|
|
|
if (!single_exit (loop))
|
if (!single_exit (loop))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
rtx insn = BB_END (loop->header);
|
rtx insn = BB_END (loop->header);
|
|
|
fprintf (dump_file, "SMS loop many exits");
|
fprintf (dump_file, "SMS loop many exits");
|
dump_insn_locator (insn);
|
dump_insn_locator (insn);
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
}
|
}
|
return false;
|
return false;
|
}
|
}
|
|
|
if (! SIMPLE_SMS_LOOP_P (loop) && ! loop_single_full_bb_p (loop))
|
if (! SIMPLE_SMS_LOOP_P (loop) && ! loop_single_full_bb_p (loop))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
rtx insn = BB_END (loop->header);
|
rtx insn = BB_END (loop->header);
|
|
|
fprintf (dump_file, "SMS loop many BBs.");
|
fprintf (dump_file, "SMS loop many BBs.");
|
dump_insn_locator (insn);
|
dump_insn_locator (insn);
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
}
|
}
|
return false;
|
return false;
|
}
|
}
|
|
|
return true;
|
return true;
|
}
|
}
|
|
|
/* If there are more than one entry for the loop,
|
/* If there are more than one entry for the loop,
|
make it one by splitting the first entry edge and
|
make it one by splitting the first entry edge and
|
redirecting the others to the new BB. */
|
redirecting the others to the new BB. */
|
static void
|
static void
|
canon_loop (struct loop *loop)
|
canon_loop (struct loop *loop)
|
{
|
{
|
edge e;
|
edge e;
|
edge_iterator i;
|
edge_iterator i;
|
|
|
/* Avoid annoying special cases of edges going to exit
|
/* Avoid annoying special cases of edges going to exit
|
block. */
|
block. */
|
FOR_EACH_EDGE (e, i, EXIT_BLOCK_PTR->preds)
|
FOR_EACH_EDGE (e, i, EXIT_BLOCK_PTR->preds)
|
if ((e->flags & EDGE_FALLTHRU) && (EDGE_COUNT (e->src->succs) > 1))
|
if ((e->flags & EDGE_FALLTHRU) && (EDGE_COUNT (e->src->succs) > 1))
|
split_edge (e);
|
split_edge (e);
|
|
|
if (loop->latch == loop->header
|
if (loop->latch == loop->header
|
|| EDGE_COUNT (loop->latch->succs) > 1)
|
|| EDGE_COUNT (loop->latch->succs) > 1)
|
{
|
{
|
FOR_EACH_EDGE (e, i, loop->header->preds)
|
FOR_EACH_EDGE (e, i, loop->header->preds)
|
if (e->src == loop->latch)
|
if (e->src == loop->latch)
|
break;
|
break;
|
split_edge (e);
|
split_edge (e);
|
}
|
}
|
}
|
}
|
|
|
/* Setup infos. */
|
/* Setup infos. */
|
static void
|
static void
|
setup_sched_infos (void)
|
setup_sched_infos (void)
|
{
|
{
|
memcpy (&sms_common_sched_info, &haifa_common_sched_info,
|
memcpy (&sms_common_sched_info, &haifa_common_sched_info,
|
sizeof (sms_common_sched_info));
|
sizeof (sms_common_sched_info));
|
sms_common_sched_info.sched_pass_id = SCHED_SMS_PASS;
|
sms_common_sched_info.sched_pass_id = SCHED_SMS_PASS;
|
common_sched_info = &sms_common_sched_info;
|
common_sched_info = &sms_common_sched_info;
|
|
|
sched_deps_info = &sms_sched_deps_info;
|
sched_deps_info = &sms_sched_deps_info;
|
current_sched_info = &sms_sched_info;
|
current_sched_info = &sms_sched_info;
|
}
|
}
|
|
|
/* Probability in % that the sms-ed loop rolls enough so that optimized
|
/* Probability in % that the sms-ed loop rolls enough so that optimized
|
version may be entered. Just a guess. */
|
version may be entered. Just a guess. */
|
#define PROB_SMS_ENOUGH_ITERATIONS 80
|
#define PROB_SMS_ENOUGH_ITERATIONS 80
|
|
|
/* Used to calculate the upper bound of ii. */
|
/* Used to calculate the upper bound of ii. */
|
#define MAXII_FACTOR 2
|
#define MAXII_FACTOR 2
|
|
|
/* Main entry point, perform SMS scheduling on the loops of the function
|
/* Main entry point, perform SMS scheduling on the loops of the function
|
that consist of single basic blocks. */
|
that consist of single basic blocks. */
|
static void
|
static void
|
sms_schedule (void)
|
sms_schedule (void)
|
{
|
{
|
rtx insn;
|
rtx insn;
|
ddg_ptr *g_arr, g;
|
ddg_ptr *g_arr, g;
|
int * node_order;
|
int * node_order;
|
int maxii, max_asap;
|
int maxii, max_asap;
|
loop_iterator li;
|
loop_iterator li;
|
partial_schedule_ptr ps;
|
partial_schedule_ptr ps;
|
basic_block bb = NULL;
|
basic_block bb = NULL;
|
struct loop *loop;
|
struct loop *loop;
|
basic_block condition_bb = NULL;
|
basic_block condition_bb = NULL;
|
edge latch_edge;
|
edge latch_edge;
|
gcov_type trip_count = 0;
|
gcov_type trip_count = 0;
|
|
|
loop_optimizer_init (LOOPS_HAVE_PREHEADERS
|
loop_optimizer_init (LOOPS_HAVE_PREHEADERS
|
| LOOPS_HAVE_RECORDED_EXITS);
|
| LOOPS_HAVE_RECORDED_EXITS);
|
if (number_of_loops () <= 1)
|
if (number_of_loops () <= 1)
|
{
|
{
|
loop_optimizer_finalize ();
|
loop_optimizer_finalize ();
|
return; /* There are no loops to schedule. */
|
return; /* There are no loops to schedule. */
|
}
|
}
|
|
|
/* Initialize issue_rate. */
|
/* Initialize issue_rate. */
|
if (targetm.sched.issue_rate)
|
if (targetm.sched.issue_rate)
|
{
|
{
|
int temp = reload_completed;
|
int temp = reload_completed;
|
|
|
reload_completed = 1;
|
reload_completed = 1;
|
issue_rate = targetm.sched.issue_rate ();
|
issue_rate = targetm.sched.issue_rate ();
|
reload_completed = temp;
|
reload_completed = temp;
|
}
|
}
|
else
|
else
|
issue_rate = 1;
|
issue_rate = 1;
|
|
|
/* Initialize the scheduler. */
|
/* Initialize the scheduler. */
|
setup_sched_infos ();
|
setup_sched_infos ();
|
haifa_sched_init ();
|
haifa_sched_init ();
|
|
|
/* Allocate memory to hold the DDG array one entry for each loop.
|
/* Allocate memory to hold the DDG array one entry for each loop.
|
We use loop->num as index into this array. */
|
We use loop->num as index into this array. */
|
g_arr = XCNEWVEC (ddg_ptr, number_of_loops ());
|
g_arr = XCNEWVEC (ddg_ptr, number_of_loops ());
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "\n\nSMS analysis phase\n");
|
fprintf (dump_file, "\n\nSMS analysis phase\n");
|
fprintf (dump_file, "===================\n\n");
|
fprintf (dump_file, "===================\n\n");
|
}
|
}
|
|
|
/* Build DDGs for all the relevant loops and hold them in G_ARR
|
/* Build DDGs for all the relevant loops and hold them in G_ARR
|
indexed by the loop index. */
|
indexed by the loop index. */
|
FOR_EACH_LOOP (li, loop, 0)
|
FOR_EACH_LOOP (li, loop, 0)
|
{
|
{
|
rtx head, tail;
|
rtx head, tail;
|
rtx count_reg;
|
rtx count_reg;
|
|
|
/* For debugging. */
|
/* For debugging. */
|
if (dbg_cnt (sms_sched_loop) == false)
|
if (dbg_cnt (sms_sched_loop) == false)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "SMS reached max limit... \n");
|
fprintf (dump_file, "SMS reached max limit... \n");
|
|
|
break;
|
break;
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
rtx insn = BB_END (loop->header);
|
rtx insn = BB_END (loop->header);
|
|
|
fprintf (dump_file, "SMS loop num: %d", loop->num);
|
fprintf (dump_file, "SMS loop num: %d", loop->num);
|
dump_insn_locator (insn);
|
dump_insn_locator (insn);
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
}
|
}
|
|
|
if (! loop_canon_p (loop))
|
if (! loop_canon_p (loop))
|
continue;
|
continue;
|
|
|
if (! loop_single_full_bb_p (loop))
|
if (! loop_single_full_bb_p (loop))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "SMS not loop_single_full_bb_p\n");
|
fprintf (dump_file, "SMS not loop_single_full_bb_p\n");
|
continue;
|
continue;
|
}
|
}
|
|
|
bb = loop->header;
|
bb = loop->header;
|
|
|
get_ebb_head_tail (bb, bb, &head, &tail);
|
get_ebb_head_tail (bb, bb, &head, &tail);
|
latch_edge = loop_latch_edge (loop);
|
latch_edge = loop_latch_edge (loop);
|
gcc_assert (single_exit (loop));
|
gcc_assert (single_exit (loop));
|
if (single_exit (loop)->count)
|
if (single_exit (loop)->count)
|
trip_count = latch_edge->count / single_exit (loop)->count;
|
trip_count = latch_edge->count / single_exit (loop)->count;
|
|
|
/* Perform SMS only on loops that their average count is above threshold. */
|
/* Perform SMS only on loops that their average count is above threshold. */
|
|
|
if ( latch_edge->count
|
if ( latch_edge->count
|
&& (latch_edge->count < single_exit (loop)->count * SMS_LOOP_AVERAGE_COUNT_THRESHOLD))
|
&& (latch_edge->count < single_exit (loop)->count * SMS_LOOP_AVERAGE_COUNT_THRESHOLD))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
dump_insn_locator (tail);
|
dump_insn_locator (tail);
|
fprintf (dump_file, "\nSMS single-bb-loop\n");
|
fprintf (dump_file, "\nSMS single-bb-loop\n");
|
if (profile_info && flag_branch_probabilities)
|
if (profile_info && flag_branch_probabilities)
|
{
|
{
|
fprintf (dump_file, "SMS loop-count ");
|
fprintf (dump_file, "SMS loop-count ");
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC,
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC,
|
(HOST_WIDEST_INT) bb->count);
|
(HOST_WIDEST_INT) bb->count);
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "SMS trip-count ");
|
fprintf (dump_file, "SMS trip-count ");
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC,
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC,
|
(HOST_WIDEST_INT) trip_count);
|
(HOST_WIDEST_INT) trip_count);
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "SMS profile-sum-max ");
|
fprintf (dump_file, "SMS profile-sum-max ");
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC,
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC,
|
(HOST_WIDEST_INT) profile_info->sum_max);
|
(HOST_WIDEST_INT) profile_info->sum_max);
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
}
|
}
|
}
|
}
|
continue;
|
continue;
|
}
|
}
|
|
|
/* Make sure this is a doloop. */
|
/* Make sure this is a doloop. */
|
if ( !(count_reg = doloop_register_get (head, tail)))
|
if ( !(count_reg = doloop_register_get (head, tail)))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "SMS doloop_register_get failed\n");
|
fprintf (dump_file, "SMS doloop_register_get failed\n");
|
continue;
|
continue;
|
}
|
}
|
|
|
/* Don't handle BBs with calls or barriers
|
/* Don't handle BBs with calls or barriers
|
or !single_set with the exception of instructions that include
|
or !single_set with the exception of instructions that include
|
count_reg---these instructions are part of the control part
|
count_reg---these instructions are part of the control part
|
that do-loop recognizes.
|
that do-loop recognizes.
|
??? Should handle insns defining subregs. */
|
??? Should handle insns defining subregs. */
|
for (insn = head; insn != NEXT_INSN (tail); insn = NEXT_INSN (insn))
|
for (insn = head; insn != NEXT_INSN (tail); insn = NEXT_INSN (insn))
|
{
|
{
|
rtx set;
|
rtx set;
|
|
|
if (CALL_P (insn)
|
if (CALL_P (insn)
|
|| BARRIER_P (insn)
|
|| BARRIER_P (insn)
|
|| (NONDEBUG_INSN_P (insn) && !JUMP_P (insn)
|
|| (NONDEBUG_INSN_P (insn) && !JUMP_P (insn)
|
&& !single_set (insn) && GET_CODE (PATTERN (insn)) != USE
|
&& !single_set (insn) && GET_CODE (PATTERN (insn)) != USE
|
&& !reg_mentioned_p (count_reg, insn))
|
&& !reg_mentioned_p (count_reg, insn))
|
|| (INSN_P (insn) && (set = single_set (insn))
|
|| (INSN_P (insn) && (set = single_set (insn))
|
&& GET_CODE (SET_DEST (set)) == SUBREG))
|
&& GET_CODE (SET_DEST (set)) == SUBREG))
|
break;
|
break;
|
}
|
}
|
|
|
if (insn != NEXT_INSN (tail))
|
if (insn != NEXT_INSN (tail))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
if (CALL_P (insn))
|
if (CALL_P (insn))
|
fprintf (dump_file, "SMS loop-with-call\n");
|
fprintf (dump_file, "SMS loop-with-call\n");
|
else if (BARRIER_P (insn))
|
else if (BARRIER_P (insn))
|
fprintf (dump_file, "SMS loop-with-barrier\n");
|
fprintf (dump_file, "SMS loop-with-barrier\n");
|
else if ((NONDEBUG_INSN_P (insn) && !JUMP_P (insn)
|
else if ((NONDEBUG_INSN_P (insn) && !JUMP_P (insn)
|
&& !single_set (insn) && GET_CODE (PATTERN (insn)) != USE))
|
&& !single_set (insn) && GET_CODE (PATTERN (insn)) != USE))
|
fprintf (dump_file, "SMS loop-with-not-single-set\n");
|
fprintf (dump_file, "SMS loop-with-not-single-set\n");
|
else
|
else
|
fprintf (dump_file, "SMS loop with subreg in lhs\n");
|
fprintf (dump_file, "SMS loop with subreg in lhs\n");
|
print_rtl_single (dump_file, insn);
|
print_rtl_single (dump_file, insn);
|
}
|
}
|
|
|
continue;
|
continue;
|
}
|
}
|
|
|
/* Always schedule the closing branch with the rest of the
|
/* Always schedule the closing branch with the rest of the
|
instructions. The branch is rotated to be in row ii-1 at the
|
instructions. The branch is rotated to be in row ii-1 at the
|
end of the scheduling procedure to make sure it's the last
|
end of the scheduling procedure to make sure it's the last
|
instruction in the iteration. */
|
instruction in the iteration. */
|
if (! (g = create_ddg (bb, 1)))
|
if (! (g = create_ddg (bb, 1)))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "SMS create_ddg failed\n");
|
fprintf (dump_file, "SMS create_ddg failed\n");
|
continue;
|
continue;
|
}
|
}
|
|
|
g_arr[loop->num] = g;
|
g_arr[loop->num] = g;
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "...OK\n");
|
fprintf (dump_file, "...OK\n");
|
|
|
}
|
}
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "\nSMS transformation phase\n");
|
fprintf (dump_file, "\nSMS transformation phase\n");
|
fprintf (dump_file, "=========================\n\n");
|
fprintf (dump_file, "=========================\n\n");
|
}
|
}
|
|
|
/* We don't want to perform SMS on new loops - created by versioning. */
|
/* We don't want to perform SMS on new loops - created by versioning. */
|
FOR_EACH_LOOP (li, loop, 0)
|
FOR_EACH_LOOP (li, loop, 0)
|
{
|
{
|
rtx head, tail;
|
rtx head, tail;
|
rtx count_reg, count_init;
|
rtx count_reg, count_init;
|
int mii, rec_mii, stage_count, min_cycle;
|
int mii, rec_mii, stage_count, min_cycle;
|
HOST_WIDEST_INT loop_count = 0;
|
HOST_WIDEST_INT loop_count = 0;
|
bool opt_sc_p;
|
bool opt_sc_p;
|
|
|
if (! (g = g_arr[loop->num]))
|
if (! (g = g_arr[loop->num]))
|
continue;
|
continue;
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
rtx insn = BB_END (loop->header);
|
rtx insn = BB_END (loop->header);
|
|
|
fprintf (dump_file, "SMS loop num: %d", loop->num);
|
fprintf (dump_file, "SMS loop num: %d", loop->num);
|
dump_insn_locator (insn);
|
dump_insn_locator (insn);
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
|
|
print_ddg (dump_file, g);
|
print_ddg (dump_file, g);
|
}
|
}
|
|
|
get_ebb_head_tail (loop->header, loop->header, &head, &tail);
|
get_ebb_head_tail (loop->header, loop->header, &head, &tail);
|
|
|
latch_edge = loop_latch_edge (loop);
|
latch_edge = loop_latch_edge (loop);
|
gcc_assert (single_exit (loop));
|
gcc_assert (single_exit (loop));
|
if (single_exit (loop)->count)
|
if (single_exit (loop)->count)
|
trip_count = latch_edge->count / single_exit (loop)->count;
|
trip_count = latch_edge->count / single_exit (loop)->count;
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
dump_insn_locator (tail);
|
dump_insn_locator (tail);
|
fprintf (dump_file, "\nSMS single-bb-loop\n");
|
fprintf (dump_file, "\nSMS single-bb-loop\n");
|
if (profile_info && flag_branch_probabilities)
|
if (profile_info && flag_branch_probabilities)
|
{
|
{
|
fprintf (dump_file, "SMS loop-count ");
|
fprintf (dump_file, "SMS loop-count ");
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC,
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC,
|
(HOST_WIDEST_INT) bb->count);
|
(HOST_WIDEST_INT) bb->count);
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "SMS profile-sum-max ");
|
fprintf (dump_file, "SMS profile-sum-max ");
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC,
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC,
|
(HOST_WIDEST_INT) profile_info->sum_max);
|
(HOST_WIDEST_INT) profile_info->sum_max);
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
}
|
}
|
fprintf (dump_file, "SMS doloop\n");
|
fprintf (dump_file, "SMS doloop\n");
|
fprintf (dump_file, "SMS built-ddg %d\n", g->num_nodes);
|
fprintf (dump_file, "SMS built-ddg %d\n", g->num_nodes);
|
fprintf (dump_file, "SMS num-loads %d\n", g->num_loads);
|
fprintf (dump_file, "SMS num-loads %d\n", g->num_loads);
|
fprintf (dump_file, "SMS num-stores %d\n", g->num_stores);
|
fprintf (dump_file, "SMS num-stores %d\n", g->num_stores);
|
}
|
}
|
|
|
|
|
/* In case of th loop have doloop register it gets special
|
/* In case of th loop have doloop register it gets special
|
handling. */
|
handling. */
|
count_init = NULL_RTX;
|
count_init = NULL_RTX;
|
if ((count_reg = doloop_register_get (head, tail)))
|
if ((count_reg = doloop_register_get (head, tail)))
|
{
|
{
|
basic_block pre_header;
|
basic_block pre_header;
|
|
|
pre_header = loop_preheader_edge (loop)->src;
|
pre_header = loop_preheader_edge (loop)->src;
|
count_init = const_iteration_count (count_reg, pre_header,
|
count_init = const_iteration_count (count_reg, pre_header,
|
&loop_count);
|
&loop_count);
|
}
|
}
|
gcc_assert (count_reg);
|
gcc_assert (count_reg);
|
|
|
if (dump_file && count_init)
|
if (dump_file && count_init)
|
{
|
{
|
fprintf (dump_file, "SMS const-doloop ");
|
fprintf (dump_file, "SMS const-doloop ");
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC,
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC,
|
loop_count);
|
loop_count);
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
}
|
}
|
|
|
node_order = XNEWVEC (int, g->num_nodes);
|
node_order = XNEWVEC (int, g->num_nodes);
|
|
|
mii = 1; /* Need to pass some estimate of mii. */
|
mii = 1; /* Need to pass some estimate of mii. */
|
rec_mii = sms_order_nodes (g, mii, node_order, &max_asap);
|
rec_mii = sms_order_nodes (g, mii, node_order, &max_asap);
|
mii = MAX (res_MII (g), rec_mii);
|
mii = MAX (res_MII (g), rec_mii);
|
maxii = MAX (max_asap, MAXII_FACTOR * mii);
|
maxii = MAX (max_asap, MAXII_FACTOR * mii);
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "SMS iis %d %d %d (rec_mii, mii, maxii)\n",
|
fprintf (dump_file, "SMS iis %d %d %d (rec_mii, mii, maxii)\n",
|
rec_mii, mii, maxii);
|
rec_mii, mii, maxii);
|
|
|
for (;;)
|
for (;;)
|
{
|
{
|
set_node_sched_params (g);
|
set_node_sched_params (g);
|
|
|
stage_count = 0;
|
stage_count = 0;
|
opt_sc_p = false;
|
opt_sc_p = false;
|
ps = sms_schedule_by_order (g, mii, maxii, node_order);
|
ps = sms_schedule_by_order (g, mii, maxii, node_order);
|
|
|
if (ps)
|
if (ps)
|
{
|
{
|
/* Try to achieve optimized SC by normalizing the partial
|
/* Try to achieve optimized SC by normalizing the partial
|
schedule (having the cycles start from cycle zero).
|
schedule (having the cycles start from cycle zero).
|
The branch location must be placed in row ii-1 in the
|
The branch location must be placed in row ii-1 in the
|
final scheduling. If failed, shift all instructions to
|
final scheduling. If failed, shift all instructions to
|
position the branch in row ii-1. */
|
position the branch in row ii-1. */
|
opt_sc_p = optimize_sc (ps, g);
|
opt_sc_p = optimize_sc (ps, g);
|
if (opt_sc_p)
|
if (opt_sc_p)
|
stage_count = calculate_stage_count (ps, 0);
|
stage_count = calculate_stage_count (ps, 0);
|
else
|
else
|
{
|
{
|
/* Bring the branch to cycle ii-1. */
|
/* Bring the branch to cycle ii-1. */
|
int amount = (SCHED_TIME (g->closing_branch->cuid)
|
int amount = (SCHED_TIME (g->closing_branch->cuid)
|
- (ps->ii - 1));
|
- (ps->ii - 1));
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "SMS schedule branch at cycle ii-1\n");
|
fprintf (dump_file, "SMS schedule branch at cycle ii-1\n");
|
|
|
stage_count = calculate_stage_count (ps, amount);
|
stage_count = calculate_stage_count (ps, amount);
|
}
|
}
|
|
|
gcc_assert (stage_count >= 1);
|
gcc_assert (stage_count >= 1);
|
}
|
}
|
|
|
/* The default value of PARAM_SMS_MIN_SC is 2 as stage count of
|
/* The default value of PARAM_SMS_MIN_SC is 2 as stage count of
|
1 means that there is no interleaving between iterations thus
|
1 means that there is no interleaving between iterations thus
|
we let the scheduling passes do the job in this case. */
|
we let the scheduling passes do the job in this case. */
|
if (stage_count < PARAM_VALUE (PARAM_SMS_MIN_SC)
|
if (stage_count < PARAM_VALUE (PARAM_SMS_MIN_SC)
|
|| (count_init && (loop_count <= stage_count))
|
|| (count_init && (loop_count <= stage_count))
|
|| (flag_branch_probabilities && (trip_count <= stage_count)))
|
|| (flag_branch_probabilities && (trip_count <= stage_count)))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "SMS failed... \n");
|
fprintf (dump_file, "SMS failed... \n");
|
fprintf (dump_file, "SMS sched-failed (stage-count=%d,"
|
fprintf (dump_file, "SMS sched-failed (stage-count=%d,"
|
" loop-count=", stage_count);
|
" loop-count=", stage_count);
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC, loop_count);
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC, loop_count);
|
fprintf (dump_file, ", trip-count=");
|
fprintf (dump_file, ", trip-count=");
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC, trip_count);
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC, trip_count);
|
fprintf (dump_file, ")\n");
|
fprintf (dump_file, ")\n");
|
}
|
}
|
break;
|
break;
|
}
|
}
|
|
|
if (!opt_sc_p)
|
if (!opt_sc_p)
|
{
|
{
|
/* Rotate the partial schedule to have the branch in row ii-1. */
|
/* Rotate the partial schedule to have the branch in row ii-1. */
|
int amount = SCHED_TIME (g->closing_branch->cuid) - (ps->ii - 1);
|
int amount = SCHED_TIME (g->closing_branch->cuid) - (ps->ii - 1);
|
|
|
reset_sched_times (ps, amount);
|
reset_sched_times (ps, amount);
|
rotate_partial_schedule (ps, amount);
|
rotate_partial_schedule (ps, amount);
|
}
|
}
|
|
|
set_columns_for_ps (ps);
|
set_columns_for_ps (ps);
|
|
|
min_cycle = PS_MIN_CYCLE (ps) - SMODULO (PS_MIN_CYCLE (ps), ps->ii);
|
min_cycle = PS_MIN_CYCLE (ps) - SMODULO (PS_MIN_CYCLE (ps), ps->ii);
|
if (!schedule_reg_moves (ps))
|
if (!schedule_reg_moves (ps))
|
{
|
{
|
mii = ps->ii + 1;
|
mii = ps->ii + 1;
|
free_partial_schedule (ps);
|
free_partial_schedule (ps);
|
continue;
|
continue;
|
}
|
}
|
|
|
/* Moves that handle incoming values might have been added
|
/* Moves that handle incoming values might have been added
|
to a new first stage. Bump the stage count if so.
|
to a new first stage. Bump the stage count if so.
|
|
|
??? Perhaps we could consider rotating the schedule here
|
??? Perhaps we could consider rotating the schedule here
|
instead? */
|
instead? */
|
if (PS_MIN_CYCLE (ps) < min_cycle)
|
if (PS_MIN_CYCLE (ps) < min_cycle)
|
{
|
{
|
reset_sched_times (ps, 0);
|
reset_sched_times (ps, 0);
|
stage_count++;
|
stage_count++;
|
}
|
}
|
|
|
/* The stage count should now be correct without rotation. */
|
/* The stage count should now be correct without rotation. */
|
gcc_checking_assert (stage_count == calculate_stage_count (ps, 0));
|
gcc_checking_assert (stage_count == calculate_stage_count (ps, 0));
|
PS_STAGE_COUNT (ps) = stage_count;
|
PS_STAGE_COUNT (ps) = stage_count;
|
|
|
canon_loop (loop);
|
canon_loop (loop);
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
dump_insn_locator (tail);
|
dump_insn_locator (tail);
|
fprintf (dump_file, " SMS succeeded %d %d (with ii, sc)\n",
|
fprintf (dump_file, " SMS succeeded %d %d (with ii, sc)\n",
|
ps->ii, stage_count);
|
ps->ii, stage_count);
|
print_partial_schedule (ps, dump_file);
|
print_partial_schedule (ps, dump_file);
|
}
|
}
|
|
|
/* case the BCT count is not known , Do loop-versioning */
|
/* case the BCT count is not known , Do loop-versioning */
|
if (count_reg && ! count_init)
|
if (count_reg && ! count_init)
|
{
|
{
|
rtx comp_rtx = gen_rtx_fmt_ee (GT, VOIDmode, count_reg,
|
rtx comp_rtx = gen_rtx_fmt_ee (GT, VOIDmode, count_reg,
|
GEN_INT(stage_count));
|
GEN_INT(stage_count));
|
unsigned prob = (PROB_SMS_ENOUGH_ITERATIONS
|
unsigned prob = (PROB_SMS_ENOUGH_ITERATIONS
|
* REG_BR_PROB_BASE) / 100;
|
* REG_BR_PROB_BASE) / 100;
|
|
|
loop_version (loop, comp_rtx, &condition_bb,
|
loop_version (loop, comp_rtx, &condition_bb,
|
prob, prob, REG_BR_PROB_BASE - prob,
|
prob, prob, REG_BR_PROB_BASE - prob,
|
true);
|
true);
|
}
|
}
|
|
|
/* Set new iteration count of loop kernel. */
|
/* Set new iteration count of loop kernel. */
|
if (count_reg && count_init)
|
if (count_reg && count_init)
|
SET_SRC (single_set (count_init)) = GEN_INT (loop_count
|
SET_SRC (single_set (count_init)) = GEN_INT (loop_count
|
- stage_count + 1);
|
- stage_count + 1);
|
|
|
/* Now apply the scheduled kernel to the RTL of the loop. */
|
/* Now apply the scheduled kernel to the RTL of the loop. */
|
permute_partial_schedule (ps, g->closing_branch->first_note);
|
permute_partial_schedule (ps, g->closing_branch->first_note);
|
|
|
/* Mark this loop as software pipelined so the later
|
/* Mark this loop as software pipelined so the later
|
scheduling passes don't touch it. */
|
scheduling passes don't touch it. */
|
if (! flag_resched_modulo_sched)
|
if (! flag_resched_modulo_sched)
|
mark_loop_unsched (loop);
|
mark_loop_unsched (loop);
|
|
|
/* The life-info is not valid any more. */
|
/* The life-info is not valid any more. */
|
df_set_bb_dirty (g->bb);
|
df_set_bb_dirty (g->bb);
|
|
|
apply_reg_moves (ps);
|
apply_reg_moves (ps);
|
if (dump_file)
|
if (dump_file)
|
print_node_sched_params (dump_file, g->num_nodes, ps);
|
print_node_sched_params (dump_file, g->num_nodes, ps);
|
/* Generate prolog and epilog. */
|
/* Generate prolog and epilog. */
|
generate_prolog_epilog (ps, loop, count_reg, count_init);
|
generate_prolog_epilog (ps, loop, count_reg, count_init);
|
break;
|
break;
|
}
|
}
|
|
|
free_partial_schedule (ps);
|
free_partial_schedule (ps);
|
VEC_free (node_sched_params, heap, node_sched_param_vec);
|
VEC_free (node_sched_params, heap, node_sched_param_vec);
|
free (node_order);
|
free (node_order);
|
free_ddg (g);
|
free_ddg (g);
|
}
|
}
|
|
|
free (g_arr);
|
free (g_arr);
|
|
|
/* Release scheduler data, needed until now because of DFA. */
|
/* Release scheduler data, needed until now because of DFA. */
|
haifa_sched_finish ();
|
haifa_sched_finish ();
|
loop_optimizer_finalize ();
|
loop_optimizer_finalize ();
|
}
|
}
|
|
|
/* The SMS scheduling algorithm itself
|
/* The SMS scheduling algorithm itself
|
-----------------------------------
|
-----------------------------------
|
Input: 'O' an ordered list of insns of a loop.
|
Input: 'O' an ordered list of insns of a loop.
|
Output: A scheduling of the loop - kernel, prolog, and epilogue.
|
Output: A scheduling of the loop - kernel, prolog, and epilogue.
|
|
|
'Q' is the empty Set
|
'Q' is the empty Set
|
'PS' is the partial schedule; it holds the currently scheduled nodes with
|
'PS' is the partial schedule; it holds the currently scheduled nodes with
|
their cycle/slot.
|
their cycle/slot.
|
'PSP' previously scheduled predecessors.
|
'PSP' previously scheduled predecessors.
|
'PSS' previously scheduled successors.
|
'PSS' previously scheduled successors.
|
't(u)' the cycle where u is scheduled.
|
't(u)' the cycle where u is scheduled.
|
'l(u)' is the latency of u.
|
'l(u)' is the latency of u.
|
'd(v,u)' is the dependence distance from v to u.
|
'd(v,u)' is the dependence distance from v to u.
|
'ASAP(u)' the earliest time at which u could be scheduled as computed in
|
'ASAP(u)' the earliest time at which u could be scheduled as computed in
|
the node ordering phase.
|
the node ordering phase.
|
'check_hardware_resources_conflicts(u, PS, c)'
|
'check_hardware_resources_conflicts(u, PS, c)'
|
run a trace around cycle/slot through DFA model
|
run a trace around cycle/slot through DFA model
|
to check resource conflicts involving instruction u
|
to check resource conflicts involving instruction u
|
at cycle c given the partial schedule PS.
|
at cycle c given the partial schedule PS.
|
'add_to_partial_schedule_at_time(u, PS, c)'
|
'add_to_partial_schedule_at_time(u, PS, c)'
|
Add the node/instruction u to the partial schedule
|
Add the node/instruction u to the partial schedule
|
PS at time c.
|
PS at time c.
|
'calculate_register_pressure(PS)'
|
'calculate_register_pressure(PS)'
|
Given a schedule of instructions, calculate the register
|
Given a schedule of instructions, calculate the register
|
pressure it implies. One implementation could be the
|
pressure it implies. One implementation could be the
|
maximum number of overlapping live ranges.
|
maximum number of overlapping live ranges.
|
'maxRP' The maximum allowed register pressure, it is usually derived from the number
|
'maxRP' The maximum allowed register pressure, it is usually derived from the number
|
registers available in the hardware.
|
registers available in the hardware.
|
|
|
1. II = MII.
|
1. II = MII.
|
2. PS = empty list
|
2. PS = empty list
|
3. for each node u in O in pre-computed order
|
3. for each node u in O in pre-computed order
|
4. if (PSP(u) != Q && PSS(u) == Q) then
|
4. if (PSP(u) != Q && PSS(u) == Q) then
|
5. Early_start(u) = max ( t(v) + l(v) - d(v,u)*II ) over all every v in PSP(u).
|
5. Early_start(u) = max ( t(v) + l(v) - d(v,u)*II ) over all every v in PSP(u).
|
6. start = Early_start; end = Early_start + II - 1; step = 1
|
6. start = Early_start; end = Early_start + II - 1; step = 1
|
11. else if (PSP(u) == Q && PSS(u) != Q) then
|
11. else if (PSP(u) == Q && PSS(u) != Q) then
|
12. Late_start(u) = min ( t(v) - l(v) + d(v,u)*II ) over all every v in PSS(u).
|
12. Late_start(u) = min ( t(v) - l(v) + d(v,u)*II ) over all every v in PSS(u).
|
13. start = Late_start; end = Late_start - II + 1; step = -1
|
13. start = Late_start; end = Late_start - II + 1; step = -1
|
14. else if (PSP(u) != Q && PSS(u) != Q) then
|
14. else if (PSP(u) != Q && PSS(u) != Q) then
|
15. Early_start(u) = max ( t(v) + l(v) - d(v,u)*II ) over all every v in PSP(u).
|
15. Early_start(u) = max ( t(v) + l(v) - d(v,u)*II ) over all every v in PSP(u).
|
16. Late_start(u) = min ( t(v) - l(v) + d(v,u)*II ) over all every v in PSS(u).
|
16. Late_start(u) = min ( t(v) - l(v) + d(v,u)*II ) over all every v in PSS(u).
|
17. start = Early_start;
|
17. start = Early_start;
|
18. end = min(Early_start + II - 1 , Late_start);
|
18. end = min(Early_start + II - 1 , Late_start);
|
19. step = 1
|
19. step = 1
|
20. else "if (PSP(u) == Q && PSS(u) == Q)"
|
20. else "if (PSP(u) == Q && PSS(u) == Q)"
|
21. start = ASAP(u); end = start + II - 1; step = 1
|
21. start = ASAP(u); end = start + II - 1; step = 1
|
22. endif
|
22. endif
|
|
|
23. success = false
|
23. success = false
|
24. for (c = start ; c != end ; c += step)
|
24. for (c = start ; c != end ; c += step)
|
25. if check_hardware_resources_conflicts(u, PS, c) then
|
25. if check_hardware_resources_conflicts(u, PS, c) then
|
26. add_to_partial_schedule_at_time(u, PS, c)
|
26. add_to_partial_schedule_at_time(u, PS, c)
|
27. success = true
|
27. success = true
|
28. break
|
28. break
|
29. endif
|
29. endif
|
30. endfor
|
30. endfor
|
31. if (success == false) then
|
31. if (success == false) then
|
32. II = II + 1
|
32. II = II + 1
|
33. if (II > maxII) then
|
33. if (II > maxII) then
|
34. finish - failed to schedule
|
34. finish - failed to schedule
|
35. endif
|
35. endif
|
36. goto 2.
|
36. goto 2.
|
37. endif
|
37. endif
|
38. endfor
|
38. endfor
|
39. if (calculate_register_pressure(PS) > maxRP) then
|
39. if (calculate_register_pressure(PS) > maxRP) then
|
40. goto 32.
|
40. goto 32.
|
41. endif
|
41. endif
|
42. compute epilogue & prologue
|
42. compute epilogue & prologue
|
43. finish - succeeded to schedule
|
43. finish - succeeded to schedule
|
|
|
??? The algorithm restricts the scheduling window to II cycles.
|
??? The algorithm restricts the scheduling window to II cycles.
|
In rare cases, it may be better to allow windows of II+1 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
|
The window would then start and end on the same row, but with
|
different "must precede" and "must follow" requirements. */
|
different "must precede" and "must follow" requirements. */
|
|
|
/* A limit on the number of cycles that resource conflicts can span. ??? Should
|
/* 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
|
be provided by DFA, and be dependent on the type of insn scheduled. Currently
|
set to 0 to save compile time. */
|
set to 0 to save compile time. */
|
#define DFA_HISTORY SMS_DFA_HISTORY
|
#define DFA_HISTORY SMS_DFA_HISTORY
|
|
|
/* A threshold for the number of repeated unsuccessful attempts to insert
|
/* A threshold for the number of repeated unsuccessful attempts to insert
|
an empty row, before we flush the partial schedule and start over. */
|
an empty row, before we flush the partial schedule and start over. */
|
#define MAX_SPLIT_NUM 10
|
#define MAX_SPLIT_NUM 10
|
/* Given the partial schedule PS, this function calculates and returns the
|
/* Given the partial schedule PS, this function calculates and returns the
|
cycles in which we can schedule the node with the given index I.
|
cycles in which we can schedule the node with the given index I.
|
NOTE: Here we do the backtracking in SMS, in some special cases. We have
|
NOTE: Here we do the backtracking in SMS, in some special cases. We have
|
noticed that there are several cases in which we fail to SMS the loop
|
noticed that there are several cases in which we fail to SMS the loop
|
because the sched window of a node is empty due to tight data-deps. In
|
because the sched window of a node is empty due to tight data-deps. In
|
such cases we want to unschedule some of the predecessors/successors
|
such cases we want to unschedule some of the predecessors/successors
|
until we get non-empty scheduling window. It returns -1 if the
|
until we get non-empty scheduling window. It returns -1 if the
|
scheduling window is empty and zero otherwise. */
|
scheduling window is empty and zero otherwise. */
|
|
|
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,
|
sbitmap sched_nodes, int ii, int *start_p, int *step_p,
|
int *end_p)
|
int *end_p)
|
{
|
{
|
int start, step, end;
|
int start, step, end;
|
int early_start, late_start;
|
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);
|
sbitmap u_node_preds = NODE_PREDECESSORS (u_node);
|
sbitmap u_node_preds = NODE_PREDECESSORS (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_preds;
|
int count_succs;
|
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);
|
sbitmap_zero (pss);
|
sbitmap_zero (pss);
|
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);
|
|
|
/* We first compute a forward range (start <= end), then decide whether
|
/* We first compute a forward range (start <= end), then decide whether
|
to reverse it. */
|
to reverse it. */
|
early_start = INT_MIN;
|
early_start = INT_MIN;
|
late_start = INT_MAX;
|
late_start = INT_MAX;
|
start = INT_MIN;
|
start = INT_MIN;
|
end = INT_MAX;
|
end = INT_MAX;
|
step = 1;
|
step = 1;
|
|
|
count_preds = 0;
|
count_preds = 0;
|
count_succs = 0;
|
count_succs = 0;
|
|
|
if (dump_file && (psp_not_empty || pss_not_empty))
|
if (dump_file && (psp_not_empty || pss_not_empty))
|
{
|
{
|
fprintf (dump_file, "\nAnalyzing dependencies for node %d (INSN %d)"
|
fprintf (dump_file, "\nAnalyzing dependencies for node %d (INSN %d)"
|
"; ii = %d\n\n", u_node->cuid, INSN_UID (u_node->insn), ii);
|
"; ii = %d\n\n", u_node->cuid, INSN_UID (u_node->insn), ii);
|
fprintf (dump_file, "%11s %11s %11s %11s %5s\n",
|
fprintf (dump_file, "%11s %11s %11s %11s %5s\n",
|
"start", "early start", "late start", "end", "time");
|
"start", "early start", "late start", "end", "time");
|
fprintf (dump_file, "=========== =========== =========== ==========="
|
fprintf (dump_file, "=========== =========== =========== ==========="
|
" =====\n");
|
" =====\n");
|
}
|
}
|
/* Calculate early_start and limit end. Both bounds are inclusive. */
|
/* Calculate early_start and limit end. Both bounds are inclusive. */
|
if (psp_not_empty)
|
if (psp_not_empty)
|
for (e = u_node->in; e != 0; e = e->next_in)
|
for (e = u_node->in; e != 0; e = e->next_in)
|
{
|
{
|
int v = e->src->cuid;
|
int v = e->src->cuid;
|
|
|
if (TEST_BIT (sched_nodes, v))
|
if (TEST_BIT (sched_nodes, v))
|
{
|
{
|
int p_st = SCHED_TIME (v);
|
int p_st = SCHED_TIME (v);
|
int earliest = p_st + e->latency - (e->distance * ii);
|
int earliest = p_st + e->latency - (e->distance * ii);
|
int latest = (e->data_type == MEM_DEP ? p_st + ii - 1 : INT_MAX);
|
int latest = (e->data_type == MEM_DEP ? p_st + ii - 1 : INT_MAX);
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "%11s %11d %11s %11d %5d",
|
fprintf (dump_file, "%11s %11d %11s %11d %5d",
|
"", earliest, "", latest, p_st);
|
"", earliest, "", latest, p_st);
|
print_ddg_edge (dump_file, e);
|
print_ddg_edge (dump_file, e);
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
}
|
}
|
|
|
early_start = MAX (early_start, earliest);
|
early_start = MAX (early_start, earliest);
|
end = MIN (end, latest);
|
end = MIN (end, latest);
|
|
|
if (e->type == TRUE_DEP && e->data_type == REG_DEP)
|
if (e->type == TRUE_DEP && e->data_type == REG_DEP)
|
count_preds++;
|
count_preds++;
|
}
|
}
|
}
|
}
|
|
|
/* Calculate late_start and limit start. Both bounds are inclusive. */
|
/* Calculate late_start and limit start. Both bounds are inclusive. */
|
if (pss_not_empty)
|
if (pss_not_empty)
|
for (e = u_node->out; e != 0; e = e->next_out)
|
for (e = u_node->out; e != 0; e = e->next_out)
|
{
|
{
|
int v = e->dest->cuid;
|
int v = e->dest->cuid;
|
|
|
if (TEST_BIT (sched_nodes, v))
|
if (TEST_BIT (sched_nodes, v))
|
{
|
{
|
int s_st = SCHED_TIME (v);
|
int s_st = SCHED_TIME (v);
|
int earliest = (e->data_type == MEM_DEP ? s_st - ii + 1 : INT_MIN);
|
int earliest = (e->data_type == MEM_DEP ? s_st - ii + 1 : INT_MIN);
|
int latest = s_st - e->latency + (e->distance * ii);
|
int latest = s_st - e->latency + (e->distance * ii);
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "%11d %11s %11d %11s %5d",
|
fprintf (dump_file, "%11d %11s %11d %11s %5d",
|
earliest, "", latest, "", s_st);
|
earliest, "", latest, "", s_st);
|
print_ddg_edge (dump_file, e);
|
print_ddg_edge (dump_file, e);
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
}
|
}
|
|
|
start = MAX (start, earliest);
|
start = MAX (start, earliest);
|
late_start = MIN (late_start, latest);
|
late_start = MIN (late_start, latest);
|
|
|
if (e->type == TRUE_DEP && e->data_type == REG_DEP)
|
if (e->type == TRUE_DEP && e->data_type == REG_DEP)
|
count_succs++;
|
count_succs++;
|
}
|
}
|
}
|
}
|
|
|
if (dump_file && (psp_not_empty || pss_not_empty))
|
if (dump_file && (psp_not_empty || pss_not_empty))
|
{
|
{
|
fprintf (dump_file, "----------- ----------- ----------- -----------"
|
fprintf (dump_file, "----------- ----------- ----------- -----------"
|
" -----\n");
|
" -----\n");
|
fprintf (dump_file, "%11d %11d %11d %11d %5s %s\n",
|
fprintf (dump_file, "%11d %11d %11d %11d %5s %s\n",
|
start, early_start, late_start, end, "",
|
start, early_start, late_start, end, "",
|
"(max, max, min, min)");
|
"(max, max, min, min)");
|
}
|
}
|
|
|
/* Get a target scheduling window no bigger than ii. */
|
/* Get a target scheduling window no bigger than ii. */
|
if (early_start == INT_MIN && late_start == INT_MAX)
|
if (early_start == INT_MIN && late_start == INT_MAX)
|
early_start = NODE_ASAP (u_node);
|
early_start = NODE_ASAP (u_node);
|
else if (early_start == INT_MIN)
|
else if (early_start == INT_MIN)
|
early_start = late_start - (ii - 1);
|
early_start = late_start - (ii - 1);
|
late_start = MIN (late_start, early_start + (ii - 1));
|
late_start = MIN (late_start, early_start + (ii - 1));
|
|
|
/* Apply memory dependence limits. */
|
/* Apply memory dependence limits. */
|
start = MAX (start, early_start);
|
start = MAX (start, early_start);
|
end = MIN (end, late_start);
|
end = MIN (end, late_start);
|
|
|
if (dump_file && (psp_not_empty || pss_not_empty))
|
if (dump_file && (psp_not_empty || pss_not_empty))
|
fprintf (dump_file, "%11s %11d %11d %11s %5s final window\n",
|
fprintf (dump_file, "%11s %11d %11d %11s %5s final window\n",
|
"", start, end, "", "");
|
"", start, end, "", "");
|
|
|
/* If there are at least as many successors as predecessors, schedule the
|
/* If there are at least as many successors as predecessors, schedule the
|
node close to its successors. */
|
node close to its successors. */
|
if (pss_not_empty && count_succs >= count_preds)
|
if (pss_not_empty && count_succs >= count_preds)
|
{
|
{
|
int tmp = end;
|
int tmp = end;
|
end = start;
|
end = start;
|
start = tmp;
|
start = tmp;
|
step = -1;
|
step = -1;
|
}
|
}
|
|
|
/* Now that we've finalized the window, make END an exclusive rather
|
/* Now that we've finalized the window, make END an exclusive rather
|
than an inclusive bound. */
|
than an inclusive bound. */
|
end += step;
|
end += step;
|
|
|
*start_p = start;
|
*start_p = start;
|
*step_p = step;
|
*step_p = step;
|
*end_p = end;
|
*end_p = end;
|
sbitmap_free (psp);
|
sbitmap_free (psp);
|
sbitmap_free (pss);
|
sbitmap_free (pss);
|
|
|
if ((start >= end && step == 1) || (start <= end && step == -1))
|
if ((start >= end && step == 1) || (start <= end && step == -1))
|
{
|
{
|
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
|
node currently been scheduled. At the end of the calculation
|
node currently been scheduled. At the end of the calculation
|
MUST_PRECEDE/MUST_FOLLOW contains all predecessors/successors of
|
MUST_PRECEDE/MUST_FOLLOW contains all predecessors/successors of
|
U_NODE which are (1) already scheduled in the first/last row of
|
U_NODE which are (1) already scheduled in the first/last row of
|
U_NODE's scheduling window, (2) whose dependence inequality with U
|
U_NODE's scheduling window, (2) whose dependence inequality with U
|
becomes an equality when U is scheduled in this same row, and (3)
|
becomes an equality when U is scheduled in this same row, and (3)
|
whose dependence latency is zero.
|
whose dependence latency is zero.
|
|
|
The first and last rows are calculated using the following parameters:
|
The first and last rows are calculated using the following parameters:
|
START/END rows - The cycles that begins/ends the traversal on the window;
|
START/END rows - The cycles that begins/ends the traversal on the window;
|
searching for an empty cycle to schedule U_NODE.
|
searching for an empty cycle to schedule U_NODE.
|
STEP - The direction in which we traverse the window.
|
STEP - The direction in which we traverse the window.
|
II - The initiation interval. */
|
II - The initiation interval. */
|
|
|
static void
|
static void
|
calculate_must_precede_follow (ddg_node_ptr u_node, int start, int end,
|
calculate_must_precede_follow (ddg_node_ptr u_node, int start, int end,
|
int step, int ii, sbitmap sched_nodes,
|
int step, int ii, sbitmap sched_nodes,
|
sbitmap must_precede, sbitmap must_follow)
|
sbitmap must_precede, sbitmap must_follow)
|
{
|
{
|
ddg_edge_ptr e;
|
ddg_edge_ptr e;
|
int first_cycle_in_window, last_cycle_in_window;
|
int first_cycle_in_window, last_cycle_in_window;
|
|
|
gcc_assert (must_precede && must_follow);
|
gcc_assert (must_precede && must_follow);
|
|
|
/* Consider the following scheduling window:
|
/* Consider the following scheduling window:
|
{first_cycle_in_window, first_cycle_in_window+1, ...,
|
{first_cycle_in_window, first_cycle_in_window+1, ...,
|
last_cycle_in_window}. If step is 1 then the following will be
|
last_cycle_in_window}. If step is 1 then the following will be
|
the order we traverse the window: {start=first_cycle_in_window,
|
the order we traverse the window: {start=first_cycle_in_window,
|
first_cycle_in_window+1, ..., end=last_cycle_in_window+1},
|
first_cycle_in_window+1, ..., end=last_cycle_in_window+1},
|
or {start=last_cycle_in_window, last_cycle_in_window-1, ...,
|
or {start=last_cycle_in_window, last_cycle_in_window-1, ...,
|
end=first_cycle_in_window-1} if step is -1. */
|
end=first_cycle_in_window-1} if step is -1. */
|
first_cycle_in_window = (step == 1) ? start : end - step;
|
first_cycle_in_window = (step == 1) ? start : end - step;
|
last_cycle_in_window = (step == 1) ? end - step : start;
|
last_cycle_in_window = (step == 1) ? end - step : start;
|
|
|
sbitmap_zero (must_precede);
|
sbitmap_zero (must_precede);
|
sbitmap_zero (must_follow);
|
sbitmap_zero (must_follow);
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "\nmust_precede: ");
|
fprintf (dump_file, "\nmust_precede: ");
|
|
|
/* Instead of checking if:
|
/* Instead of checking if:
|
(SMODULO (SCHED_TIME (e->src), ii) == first_row_in_window)
|
(SMODULO (SCHED_TIME (e->src), ii) == first_row_in_window)
|
&& ((SCHED_TIME (e->src) + e->latency - (e->distance * ii)) ==
|
&& ((SCHED_TIME (e->src) + e->latency - (e->distance * ii)) ==
|
first_cycle_in_window)
|
first_cycle_in_window)
|
&& e->latency == 0
|
&& e->latency == 0
|
we use the fact that latency is non-negative:
|
we use the fact that latency is non-negative:
|
SCHED_TIME (e->src) - (e->distance * ii) <=
|
SCHED_TIME (e->src) - (e->distance * ii) <=
|
SCHED_TIME (e->src) + e->latency - (e->distance * ii)) <=
|
SCHED_TIME (e->src) + e->latency - (e->distance * ii)) <=
|
first_cycle_in_window
|
first_cycle_in_window
|
and check only if
|
and check only if
|
SCHED_TIME (e->src) - (e->distance * ii) == first_cycle_in_window */
|
SCHED_TIME (e->src) - (e->distance * ii) == first_cycle_in_window */
|
for (e = u_node->in; e != 0; e = e->next_in)
|
for (e = u_node->in; e != 0; e = e->next_in)
|
if (TEST_BIT (sched_nodes, e->src->cuid)
|
if (TEST_BIT (sched_nodes, e->src->cuid)
|
&& ((SCHED_TIME (e->src->cuid) - (e->distance * ii)) ==
|
&& ((SCHED_TIME (e->src->cuid) - (e->distance * ii)) ==
|
first_cycle_in_window))
|
first_cycle_in_window))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "%d ", e->src->cuid);
|
fprintf (dump_file, "%d ", e->src->cuid);
|
|
|
SET_BIT (must_precede, e->src->cuid);
|
SET_BIT (must_precede, e->src->cuid);
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "\nmust_follow: ");
|
fprintf (dump_file, "\nmust_follow: ");
|
|
|
/* Instead of checking if:
|
/* Instead of checking if:
|
(SMODULO (SCHED_TIME (e->dest), ii) == last_row_in_window)
|
(SMODULO (SCHED_TIME (e->dest), ii) == last_row_in_window)
|
&& ((SCHED_TIME (e->dest) - e->latency + (e->distance * ii)) ==
|
&& ((SCHED_TIME (e->dest) - e->latency + (e->distance * ii)) ==
|
last_cycle_in_window)
|
last_cycle_in_window)
|
&& e->latency == 0
|
&& e->latency == 0
|
we use the fact that latency is non-negative:
|
we use the fact that latency is non-negative:
|
SCHED_TIME (e->dest) + (e->distance * ii) >=
|
SCHED_TIME (e->dest) + (e->distance * ii) >=
|
SCHED_TIME (e->dest) - e->latency + (e->distance * ii)) >=
|
SCHED_TIME (e->dest) - e->latency + (e->distance * ii)) >=
|
last_cycle_in_window
|
last_cycle_in_window
|
and check only if
|
and check only if
|
SCHED_TIME (e->dest) + (e->distance * ii) == last_cycle_in_window */
|
SCHED_TIME (e->dest) + (e->distance * ii) == last_cycle_in_window */
|
for (e = u_node->out; e != 0; e = e->next_out)
|
for (e = u_node->out; e != 0; e = e->next_out)
|
if (TEST_BIT (sched_nodes, e->dest->cuid)
|
if (TEST_BIT (sched_nodes, e->dest->cuid)
|
&& ((SCHED_TIME (e->dest->cuid) + (e->distance * ii)) ==
|
&& ((SCHED_TIME (e->dest->cuid) + (e->distance * ii)) ==
|
last_cycle_in_window))
|
last_cycle_in_window))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "%d ", e->dest->cuid);
|
fprintf (dump_file, "%d ", e->dest->cuid);
|
|
|
SET_BIT (must_follow, e->dest->cuid);
|
SET_BIT (must_follow, e->dest->cuid);
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
}
|
}
|
|
|
/* Return 1 if U_NODE can be scheduled in CYCLE. Use the following
|
/* Return 1 if U_NODE can be scheduled in CYCLE. Use the following
|
parameters to decide if that's possible:
|
parameters to decide if that's possible:
|
PS - The partial schedule.
|
PS - The partial schedule.
|
U - The serial number of U_NODE.
|
U - The serial number of U_NODE.
|
NUM_SPLITS - The number of row splits made so far.
|
NUM_SPLITS - The number of row splits made so far.
|
MUST_PRECEDE - The nodes that must precede U_NODE. (only valid at
|
MUST_PRECEDE - The nodes that must precede U_NODE. (only valid at
|
the first row of the scheduling window)
|
the first row of the scheduling window)
|
MUST_FOLLOW - The nodes that must follow U_NODE. (only valid at the
|
MUST_FOLLOW - The nodes that must follow U_NODE. (only valid at the
|
last row of the scheduling window) */
|
last row of the scheduling window) */
|
|
|
static bool
|
static bool
|
try_scheduling_node_in_cycle (partial_schedule_ptr ps,
|
try_scheduling_node_in_cycle (partial_schedule_ptr ps,
|
int u, int cycle, sbitmap sched_nodes,
|
int u, int cycle, sbitmap sched_nodes,
|
int *num_splits, sbitmap must_precede,
|
int *num_splits, sbitmap must_precede,
|
sbitmap must_follow)
|
sbitmap must_follow)
|
{
|
{
|
ps_insn_ptr psi;
|
ps_insn_ptr psi;
|
bool success = 0;
|
bool success = 0;
|
|
|
verify_partial_schedule (ps, sched_nodes);
|
verify_partial_schedule (ps, sched_nodes);
|
psi = ps_add_node_check_conflicts (ps, u, cycle, must_precede, must_follow);
|
psi = ps_add_node_check_conflicts (ps, u, cycle, must_precede, must_follow);
|
if (psi)
|
if (psi)
|
{
|
{
|
SCHED_TIME (u) = cycle;
|
SCHED_TIME (u) = cycle;
|
SET_BIT (sched_nodes, u);
|
SET_BIT (sched_nodes, u);
|
success = 1;
|
success = 1;
|
*num_splits = 0;
|
*num_splits = 0;
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "Scheduled w/o split in %d\n", cycle);
|
fprintf (dump_file, "Scheduled w/o split in %d\n", cycle);
|
|
|
}
|
}
|
|
|
return success;
|
return success;
|
}
|
}
|
|
|
/* This function implements the scheduling algorithm for SMS according to the
|
/* This function implements the scheduling algorithm for SMS according to the
|
above algorithm. */
|
above algorithm. */
|
static partial_schedule_ptr
|
static partial_schedule_ptr
|
sms_schedule_by_order (ddg_ptr g, int mii, int maxii, int *nodes_order)
|
sms_schedule_by_order (ddg_ptr g, int mii, int maxii, int *nodes_order)
|
{
|
{
|
int ii = mii;
|
int ii = mii;
|
int i, c, success, num_splits = 0;
|
int i, c, success, num_splits = 0;
|
int flush_and_start_over = true;
|
int flush_and_start_over = true;
|
int num_nodes = g->num_nodes;
|
int num_nodes = g->num_nodes;
|
int start, end, step; /* Place together into one struct? */
|
int start, end, step; /* Place together into one struct? */
|
sbitmap sched_nodes = sbitmap_alloc (num_nodes);
|
sbitmap sched_nodes = sbitmap_alloc (num_nodes);
|
sbitmap must_precede = sbitmap_alloc (num_nodes);
|
sbitmap must_precede = sbitmap_alloc (num_nodes);
|
sbitmap must_follow = sbitmap_alloc (num_nodes);
|
sbitmap must_follow = sbitmap_alloc (num_nodes);
|
sbitmap tobe_scheduled = sbitmap_alloc (num_nodes);
|
sbitmap tobe_scheduled = sbitmap_alloc (num_nodes);
|
|
|
partial_schedule_ptr ps = create_partial_schedule (ii, g, DFA_HISTORY);
|
partial_schedule_ptr ps = create_partial_schedule (ii, g, DFA_HISTORY);
|
|
|
sbitmap_ones (tobe_scheduled);
|
sbitmap_ones (tobe_scheduled);
|
sbitmap_zero (sched_nodes);
|
sbitmap_zero (sched_nodes);
|
|
|
while (flush_and_start_over && (ii < maxii))
|
while (flush_and_start_over && (ii < maxii))
|
{
|
{
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "Starting with ii=%d\n", ii);
|
fprintf (dump_file, "Starting with ii=%d\n", ii);
|
flush_and_start_over = false;
|
flush_and_start_over = false;
|
sbitmap_zero (sched_nodes);
|
sbitmap_zero (sched_nodes);
|
|
|
for (i = 0; i < num_nodes; i++)
|
for (i = 0; i < num_nodes; i++)
|
{
|
{
|
int u = nodes_order[i];
|
int u = nodes_order[i];
|
ddg_node_ptr u_node = &ps->g->nodes[u];
|
ddg_node_ptr u_node = &ps->g->nodes[u];
|
rtx insn = u_node->insn;
|
rtx insn = u_node->insn;
|
|
|
if (!NONDEBUG_INSN_P (insn))
|
if (!NONDEBUG_INSN_P (insn))
|
{
|
{
|
RESET_BIT (tobe_scheduled, u);
|
RESET_BIT (tobe_scheduled, u);
|
continue;
|
continue;
|
}
|
}
|
|
|
if (TEST_BIT (sched_nodes, u))
|
if (TEST_BIT (sched_nodes, u))
|
continue;
|
continue;
|
|
|
/* Try to get non-empty scheduling window. */
|
/* Try to get non-empty scheduling window. */
|
success = 0;
|
success = 0;
|
if (get_sched_window (ps, u_node, sched_nodes, ii, &start,
|
if (get_sched_window (ps, u_node, sched_nodes, ii, &start,
|
&step, &end) == 0)
|
&step, &end) == 0)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "\nTrying to schedule node %d "
|
fprintf (dump_file, "\nTrying to schedule node %d "
|
"INSN = %d in (%d .. %d) step %d\n", u, (INSN_UID
|
"INSN = %d in (%d .. %d) step %d\n", u, (INSN_UID
|
(g->nodes[u].insn)), start, end, step);
|
(g->nodes[u].insn)), start, end, step);
|
|
|
gcc_assert ((step > 0 && start < end)
|
gcc_assert ((step > 0 && start < end)
|
|| (step < 0 && start > end));
|
|| (step < 0 && start > end));
|
|
|
calculate_must_precede_follow (u_node, start, end, step, ii,
|
calculate_must_precede_follow (u_node, start, end, step, ii,
|
sched_nodes, must_precede,
|
sched_nodes, must_precede,
|
must_follow);
|
must_follow);
|
|
|
for (c = start; c != end; c += step)
|
for (c = start; c != end; c += step)
|
{
|
{
|
sbitmap tmp_precede, tmp_follow;
|
sbitmap tmp_precede, tmp_follow;
|
|
|
set_must_precede_follow (&tmp_follow, must_follow,
|
set_must_precede_follow (&tmp_follow, must_follow,
|
&tmp_precede, must_precede,
|
&tmp_precede, must_precede,
|
c, start, end, step);
|
c, start, end, step);
|
success =
|
success =
|
try_scheduling_node_in_cycle (ps, u, c,
|
try_scheduling_node_in_cycle (ps, u, c,
|
sched_nodes,
|
sched_nodes,
|
&num_splits, tmp_precede,
|
&num_splits, tmp_precede,
|
tmp_follow);
|
tmp_follow);
|
if (success)
|
if (success)
|
break;
|
break;
|
}
|
}
|
|
|
verify_partial_schedule (ps, sched_nodes);
|
verify_partial_schedule (ps, sched_nodes);
|
}
|
}
|
if (!success)
|
if (!success)
|
{
|
{
|
int split_row;
|
int split_row;
|
|
|
if (ii++ == maxii)
|
if (ii++ == maxii)
|
break;
|
break;
|
|
|
if (num_splits >= MAX_SPLIT_NUM)
|
if (num_splits >= MAX_SPLIT_NUM)
|
{
|
{
|
num_splits = 0;
|
num_splits = 0;
|
flush_and_start_over = true;
|
flush_and_start_over = true;
|
verify_partial_schedule (ps, sched_nodes);
|
verify_partial_schedule (ps, sched_nodes);
|
reset_partial_schedule (ps, ii);
|
reset_partial_schedule (ps, ii);
|
verify_partial_schedule (ps, sched_nodes);
|
verify_partial_schedule (ps, sched_nodes);
|
break;
|
break;
|
}
|
}
|
|
|
num_splits++;
|
num_splits++;
|
/* The scheduling window is exclusive of 'end'
|
/* The scheduling window is exclusive of 'end'
|
whereas compute_split_window() expects an inclusive,
|
whereas compute_split_window() expects an inclusive,
|
ordered range. */
|
ordered range. */
|
if (step == 1)
|
if (step == 1)
|
split_row = compute_split_row (sched_nodes, start, end - 1,
|
split_row = compute_split_row (sched_nodes, start, end - 1,
|
ps->ii, u_node);
|
ps->ii, u_node);
|
else
|
else
|
split_row = compute_split_row (sched_nodes, end + 1, start,
|
split_row = compute_split_row (sched_nodes, end + 1, start,
|
ps->ii, u_node);
|
ps->ii, u_node);
|
|
|
ps_insert_empty_row (ps, split_row, sched_nodes);
|
ps_insert_empty_row (ps, split_row, sched_nodes);
|
i--; /* Go back and retry node i. */
|
i--; /* Go back and retry node i. */
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "num_splits=%d\n", num_splits);
|
fprintf (dump_file, "num_splits=%d\n", num_splits);
|
}
|
}
|
|
|
/* ??? If (success), check register pressure estimates. */
|
/* ??? If (success), check register pressure estimates. */
|
} /* Continue with next node. */
|
} /* Continue with next node. */
|
} /* While flush_and_start_over. */
|
} /* While flush_and_start_over. */
|
if (ii >= maxii)
|
if (ii >= maxii)
|
{
|
{
|
free_partial_schedule (ps);
|
free_partial_schedule (ps);
|
ps = NULL;
|
ps = NULL;
|
}
|
}
|
else
|
else
|
gcc_assert (sbitmap_equal (tobe_scheduled, sched_nodes));
|
gcc_assert (sbitmap_equal (tobe_scheduled, sched_nodes));
|
|
|
sbitmap_free (sched_nodes);
|
sbitmap_free (sched_nodes);
|
sbitmap_free (must_precede);
|
sbitmap_free (must_precede);
|
sbitmap_free (must_follow);
|
sbitmap_free (must_follow);
|
sbitmap_free (tobe_scheduled);
|
sbitmap_free (tobe_scheduled);
|
|
|
return ps;
|
return ps;
|
}
|
}
|
|
|
/* This function inserts a new empty row into PS at the position
|
/* This function inserts a new empty row into PS at the position
|
according to SPLITROW, keeping all already scheduled instructions
|
according to SPLITROW, keeping all already scheduled instructions
|
intact and updating their SCHED_TIME and cycle accordingly. */
|
intact and updating their SCHED_TIME and cycle accordingly. */
|
static void
|
static void
|
ps_insert_empty_row (partial_schedule_ptr ps, int split_row,
|
ps_insert_empty_row (partial_schedule_ptr ps, int split_row,
|
sbitmap sched_nodes)
|
sbitmap sched_nodes)
|
{
|
{
|
ps_insn_ptr crr_insn;
|
ps_insn_ptr crr_insn;
|
ps_insn_ptr *rows_new;
|
ps_insn_ptr *rows_new;
|
int ii = ps->ii;
|
int ii = ps->ii;
|
int new_ii = ii + 1;
|
int new_ii = ii + 1;
|
int row;
|
int row;
|
int *rows_length_new;
|
int *rows_length_new;
|
|
|
verify_partial_schedule (ps, sched_nodes);
|
verify_partial_schedule (ps, sched_nodes);
|
|
|
/* We normalize sched_time and rotate ps to have only non-negative sched
|
/* We normalize sched_time and rotate ps to have only non-negative sched
|
times, for simplicity of updating cycles after inserting new row. */
|
times, for simplicity of updating cycles after inserting new row. */
|
split_row -= ps->min_cycle;
|
split_row -= ps->min_cycle;
|
split_row = SMODULO (split_row, ii);
|
split_row = SMODULO (split_row, ii);
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "split_row=%d\n", split_row);
|
fprintf (dump_file, "split_row=%d\n", split_row);
|
|
|
reset_sched_times (ps, PS_MIN_CYCLE (ps));
|
reset_sched_times (ps, PS_MIN_CYCLE (ps));
|
rotate_partial_schedule (ps, PS_MIN_CYCLE (ps));
|
rotate_partial_schedule (ps, PS_MIN_CYCLE (ps));
|
|
|
rows_new = (ps_insn_ptr *) xcalloc (new_ii, sizeof (ps_insn_ptr));
|
rows_new = (ps_insn_ptr *) xcalloc (new_ii, sizeof (ps_insn_ptr));
|
rows_length_new = (int *) xcalloc (new_ii, sizeof (int));
|
rows_length_new = (int *) xcalloc (new_ii, sizeof (int));
|
for (row = 0; row < split_row; row++)
|
for (row = 0; row < split_row; row++)
|
{
|
{
|
rows_new[row] = ps->rows[row];
|
rows_new[row] = ps->rows[row];
|
rows_length_new[row] = ps->rows_length[row];
|
rows_length_new[row] = ps->rows_length[row];
|
ps->rows[row] = NULL;
|
ps->rows[row] = NULL;
|
for (crr_insn = rows_new[row];
|
for (crr_insn = rows_new[row];
|
crr_insn; crr_insn = crr_insn->next_in_row)
|
crr_insn; crr_insn = crr_insn->next_in_row)
|
{
|
{
|
int u = crr_insn->id;
|
int u = crr_insn->id;
|
int new_time = SCHED_TIME (u) + (SCHED_TIME (u) / ii);
|
int new_time = SCHED_TIME (u) + (SCHED_TIME (u) / ii);
|
|
|
SCHED_TIME (u) = new_time;
|
SCHED_TIME (u) = new_time;
|
crr_insn->cycle = new_time;
|
crr_insn->cycle = new_time;
|
SCHED_ROW (u) = new_time % new_ii;
|
SCHED_ROW (u) = new_time % new_ii;
|
SCHED_STAGE (u) = new_time / new_ii;
|
SCHED_STAGE (u) = new_time / new_ii;
|
}
|
}
|
|
|
}
|
}
|
|
|
rows_new[split_row] = NULL;
|
rows_new[split_row] = NULL;
|
|
|
for (row = split_row; row < ii; row++)
|
for (row = split_row; row < ii; row++)
|
{
|
{
|
rows_new[row + 1] = ps->rows[row];
|
rows_new[row + 1] = ps->rows[row];
|
rows_length_new[row + 1] = ps->rows_length[row];
|
rows_length_new[row + 1] = ps->rows_length[row];
|
ps->rows[row] = NULL;
|
ps->rows[row] = NULL;
|
for (crr_insn = rows_new[row + 1];
|
for (crr_insn = rows_new[row + 1];
|
crr_insn; crr_insn = crr_insn->next_in_row)
|
crr_insn; crr_insn = crr_insn->next_in_row)
|
{
|
{
|
int u = crr_insn->id;
|
int u = crr_insn->id;
|
int new_time = SCHED_TIME (u) + (SCHED_TIME (u) / ii) + 1;
|
int new_time = SCHED_TIME (u) + (SCHED_TIME (u) / ii) + 1;
|
|
|
SCHED_TIME (u) = new_time;
|
SCHED_TIME (u) = new_time;
|
crr_insn->cycle = new_time;
|
crr_insn->cycle = new_time;
|
SCHED_ROW (u) = new_time % new_ii;
|
SCHED_ROW (u) = new_time % new_ii;
|
SCHED_STAGE (u) = new_time / new_ii;
|
SCHED_STAGE (u) = new_time / new_ii;
|
}
|
}
|
}
|
}
|
|
|
/* Updating ps. */
|
/* Updating ps. */
|
ps->min_cycle = ps->min_cycle + ps->min_cycle / ii
|
ps->min_cycle = ps->min_cycle + ps->min_cycle / ii
|
+ (SMODULO (ps->min_cycle, ii) >= split_row ? 1 : 0);
|
+ (SMODULO (ps->min_cycle, ii) >= split_row ? 1 : 0);
|
ps->max_cycle = ps->max_cycle + ps->max_cycle / ii
|
ps->max_cycle = ps->max_cycle + ps->max_cycle / ii
|
+ (SMODULO (ps->max_cycle, ii) >= split_row ? 1 : 0);
|
+ (SMODULO (ps->max_cycle, ii) >= split_row ? 1 : 0);
|
free (ps->rows);
|
free (ps->rows);
|
ps->rows = rows_new;
|
ps->rows = rows_new;
|
free (ps->rows_length);
|
free (ps->rows_length);
|
ps->rows_length = rows_length_new;
|
ps->rows_length = rows_length_new;
|
ps->ii = new_ii;
|
ps->ii = new_ii;
|
gcc_assert (ps->min_cycle >= 0);
|
gcc_assert (ps->min_cycle >= 0);
|
|
|
verify_partial_schedule (ps, sched_nodes);
|
verify_partial_schedule (ps, sched_nodes);
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "min_cycle=%d, max_cycle=%d\n", ps->min_cycle,
|
fprintf (dump_file, "min_cycle=%d, max_cycle=%d\n", ps->min_cycle,
|
ps->max_cycle);
|
ps->max_cycle);
|
}
|
}
|
|
|
/* Given U_NODE which is the node that failed to be scheduled; LOW and
|
/* Given U_NODE which is the node that failed to be scheduled; LOW and
|
UP which are the boundaries of it's scheduling window; compute using
|
UP which are the boundaries of it's scheduling window; compute using
|
SCHED_NODES and II a row in the partial schedule that can be split
|
SCHED_NODES and II a row in the partial schedule that can be split
|
which will separate a critical predecessor from a critical successor
|
which will separate a critical predecessor from a critical successor
|
thereby expanding the window, and return it. */
|
thereby expanding the window, and return it. */
|
static int
|
static int
|
compute_split_row (sbitmap sched_nodes, int low, int up, int ii,
|
compute_split_row (sbitmap sched_nodes, int low, int up, int ii,
|
ddg_node_ptr u_node)
|
ddg_node_ptr u_node)
|
{
|
{
|
ddg_edge_ptr e;
|
ddg_edge_ptr e;
|
int lower = INT_MIN, upper = INT_MAX;
|
int lower = INT_MIN, upper = INT_MAX;
|
int crit_pred = -1;
|
int crit_pred = -1;
|
int crit_succ = -1;
|
int crit_succ = -1;
|
int crit_cycle;
|
int crit_cycle;
|
|
|
for (e = u_node->in; e != 0; e = e->next_in)
|
for (e = u_node->in; e != 0; e = e->next_in)
|
{
|
{
|
int v = e->src->cuid;
|
int v = e->src->cuid;
|
|
|
if (TEST_BIT (sched_nodes, v)
|
if (TEST_BIT (sched_nodes, v)
|
&& (low == SCHED_TIME (v) + e->latency - (e->distance * ii)))
|
&& (low == SCHED_TIME (v) + e->latency - (e->distance * ii)))
|
if (SCHED_TIME (v) > lower)
|
if (SCHED_TIME (v) > lower)
|
{
|
{
|
crit_pred = v;
|
crit_pred = v;
|
lower = SCHED_TIME (v);
|
lower = SCHED_TIME (v);
|
}
|
}
|
}
|
}
|
|
|
if (crit_pred >= 0)
|
if (crit_pred >= 0)
|
{
|
{
|
crit_cycle = SCHED_TIME (crit_pred) + 1;
|
crit_cycle = SCHED_TIME (crit_pred) + 1;
|
return SMODULO (crit_cycle, ii);
|
return SMODULO (crit_cycle, ii);
|
}
|
}
|
|
|
for (e = u_node->out; e != 0; e = e->next_out)
|
for (e = u_node->out; e != 0; e = e->next_out)
|
{
|
{
|
int v = e->dest->cuid;
|
int v = e->dest->cuid;
|
|
|
if (TEST_BIT (sched_nodes, v)
|
if (TEST_BIT (sched_nodes, v)
|
&& (up == SCHED_TIME (v) - e->latency + (e->distance * ii)))
|
&& (up == SCHED_TIME (v) - e->latency + (e->distance * ii)))
|
if (SCHED_TIME (v) < upper)
|
if (SCHED_TIME (v) < upper)
|
{
|
{
|
crit_succ = v;
|
crit_succ = v;
|
upper = SCHED_TIME (v);
|
upper = SCHED_TIME (v);
|
}
|
}
|
}
|
}
|
|
|
if (crit_succ >= 0)
|
if (crit_succ >= 0)
|
{
|
{
|
crit_cycle = SCHED_TIME (crit_succ);
|
crit_cycle = SCHED_TIME (crit_succ);
|
return SMODULO (crit_cycle, ii);
|
return SMODULO (crit_cycle, ii);
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "Both crit_pred and crit_succ are NULL\n");
|
fprintf (dump_file, "Both crit_pred and crit_succ are NULL\n");
|
|
|
return SMODULO ((low + up + 1) / 2, ii);
|
return SMODULO ((low + up + 1) / 2, ii);
|
}
|
}
|
|
|
static void
|
static void
|
verify_partial_schedule (partial_schedule_ptr ps, sbitmap sched_nodes)
|
verify_partial_schedule (partial_schedule_ptr ps, sbitmap sched_nodes)
|
{
|
{
|
int row;
|
int row;
|
ps_insn_ptr crr_insn;
|
ps_insn_ptr crr_insn;
|
|
|
for (row = 0; row < ps->ii; row++)
|
for (row = 0; row < ps->ii; row++)
|
{
|
{
|
int length = 0;
|
int length = 0;
|
|
|
for (crr_insn = ps->rows[row]; crr_insn; crr_insn = crr_insn->next_in_row)
|
for (crr_insn = ps->rows[row]; crr_insn; crr_insn = crr_insn->next_in_row)
|
{
|
{
|
int u = crr_insn->id;
|
int u = crr_insn->id;
|
|
|
length++;
|
length++;
|
gcc_assert (TEST_BIT (sched_nodes, u));
|
gcc_assert (TEST_BIT (sched_nodes, u));
|
/* ??? Test also that all nodes of sched_nodes are in ps, perhaps by
|
/* ??? Test also that all nodes of sched_nodes are in ps, perhaps by
|
popcount (sched_nodes) == number of insns in ps. */
|
popcount (sched_nodes) == number of insns in ps. */
|
gcc_assert (SCHED_TIME (u) >= ps->min_cycle);
|
gcc_assert (SCHED_TIME (u) >= ps->min_cycle);
|
gcc_assert (SCHED_TIME (u) <= ps->max_cycle);
|
gcc_assert (SCHED_TIME (u) <= ps->max_cycle);
|
}
|
}
|
|
|
gcc_assert (ps->rows_length[row] == length);
|
gcc_assert (ps->rows_length[row] == length);
|
}
|
}
|
}
|
}
|
|
|
|
|
/* This page implements the algorithm for ordering the nodes of a DDG
|
/* This page implements the algorithm for ordering the nodes of a DDG
|
for modulo scheduling, activated through the
|
for modulo scheduling, activated through the
|
"int sms_order_nodes (ddg_ptr, int mii, int * result)" API. */
|
"int sms_order_nodes (ddg_ptr, int mii, int * result)" API. */
|
|
|
#define ORDER_PARAMS(x) ((struct node_order_params *) (x)->aux.info)
|
#define ORDER_PARAMS(x) ((struct node_order_params *) (x)->aux.info)
|
#define ASAP(x) (ORDER_PARAMS ((x))->asap)
|
#define ASAP(x) (ORDER_PARAMS ((x))->asap)
|
#define ALAP(x) (ORDER_PARAMS ((x))->alap)
|
#define ALAP(x) (ORDER_PARAMS ((x))->alap)
|
#define HEIGHT(x) (ORDER_PARAMS ((x))->height)
|
#define HEIGHT(x) (ORDER_PARAMS ((x))->height)
|
#define MOB(x) (ALAP ((x)) - ASAP ((x)))
|
#define MOB(x) (ALAP ((x)) - ASAP ((x)))
|
#define DEPTH(x) (ASAP ((x)))
|
#define DEPTH(x) (ASAP ((x)))
|
|
|
typedef struct node_order_params * nopa;
|
typedef struct node_order_params * nopa;
|
|
|
static void order_nodes_of_sccs (ddg_all_sccs_ptr, int * result);
|
static void order_nodes_of_sccs (ddg_all_sccs_ptr, int * result);
|
static int order_nodes_in_scc (ddg_ptr, sbitmap, sbitmap, int*, int);
|
static int order_nodes_in_scc (ddg_ptr, sbitmap, sbitmap, int*, int);
|
static nopa calculate_order_params (ddg_ptr, int, int *);
|
static nopa calculate_order_params (ddg_ptr, int, int *);
|
static int find_max_asap (ddg_ptr, sbitmap);
|
static int find_max_asap (ddg_ptr, sbitmap);
|
static int find_max_hv_min_mob (ddg_ptr, sbitmap);
|
static int find_max_hv_min_mob (ddg_ptr, sbitmap);
|
static int find_max_dv_min_mob (ddg_ptr, sbitmap);
|
static int find_max_dv_min_mob (ddg_ptr, sbitmap);
|
|
|
enum sms_direction {BOTTOMUP, TOPDOWN};
|
enum sms_direction {BOTTOMUP, TOPDOWN};
|
|
|
struct node_order_params
|
struct node_order_params
|
{
|
{
|
int asap;
|
int asap;
|
int alap;
|
int alap;
|
int height;
|
int height;
|
};
|
};
|
|
|
/* Check if NODE_ORDER contains a permutation of 0 .. NUM_NODES-1. */
|
/* Check if NODE_ORDER contains a permutation of 0 .. NUM_NODES-1. */
|
static void
|
static void
|
check_nodes_order (int *node_order, int num_nodes)
|
check_nodes_order (int *node_order, int num_nodes)
|
{
|
{
|
int i;
|
int i;
|
sbitmap tmp = sbitmap_alloc (num_nodes);
|
sbitmap tmp = sbitmap_alloc (num_nodes);
|
|
|
sbitmap_zero (tmp);
|
sbitmap_zero (tmp);
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "SMS final nodes order: \n");
|
fprintf (dump_file, "SMS final nodes order: \n");
|
|
|
for (i = 0; i < num_nodes; i++)
|
for (i = 0; i < num_nodes; i++)
|
{
|
{
|
int u = node_order[i];
|
int u = node_order[i];
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "%d ", u);
|
fprintf (dump_file, "%d ", u);
|
gcc_assert (u < num_nodes && u >= 0 && !TEST_BIT (tmp, u));
|
gcc_assert (u < num_nodes && u >= 0 && !TEST_BIT (tmp, u));
|
|
|
SET_BIT (tmp, u);
|
SET_BIT (tmp, u);
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
|
|
sbitmap_free (tmp);
|
sbitmap_free (tmp);
|
}
|
}
|
|
|
/* Order the nodes of G for scheduling and pass the result in
|
/* Order the nodes of G for scheduling and pass the result in
|
NODE_ORDER. Also set aux.count of each node to ASAP.
|
NODE_ORDER. Also set aux.count of each node to ASAP.
|
Put maximal ASAP to PMAX_ASAP. Return the recMII for the given DDG. */
|
Put maximal ASAP to PMAX_ASAP. Return the recMII for the given DDG. */
|
static int
|
static int
|
sms_order_nodes (ddg_ptr g, int mii, int * node_order, int *pmax_asap)
|
sms_order_nodes (ddg_ptr g, int mii, int * node_order, int *pmax_asap)
|
{
|
{
|
int i;
|
int i;
|
int rec_mii = 0;
|
int rec_mii = 0;
|
ddg_all_sccs_ptr sccs = create_ddg_all_sccs (g);
|
ddg_all_sccs_ptr sccs = create_ddg_all_sccs (g);
|
|
|
nopa nops = calculate_order_params (g, mii, pmax_asap);
|
nopa nops = calculate_order_params (g, mii, pmax_asap);
|
|
|
if (dump_file)
|
if (dump_file)
|
print_sccs (dump_file, sccs, g);
|
print_sccs (dump_file, sccs, g);
|
|
|
order_nodes_of_sccs (sccs, node_order);
|
order_nodes_of_sccs (sccs, node_order);
|
|
|
if (sccs->num_sccs > 0)
|
if (sccs->num_sccs > 0)
|
/* First SCC has the largest recurrence_length. */
|
/* First SCC has the largest recurrence_length. */
|
rec_mii = sccs->sccs[0]->recurrence_length;
|
rec_mii = sccs->sccs[0]->recurrence_length;
|
|
|
/* Save ASAP before destroying node_order_params. */
|
/* Save ASAP before destroying node_order_params. */
|
for (i = 0; i < g->num_nodes; i++)
|
for (i = 0; i < g->num_nodes; i++)
|
{
|
{
|
ddg_node_ptr v = &g->nodes[i];
|
ddg_node_ptr v = &g->nodes[i];
|
v->aux.count = ASAP (v);
|
v->aux.count = ASAP (v);
|
}
|
}
|
|
|
free (nops);
|
free (nops);
|
free_ddg_all_sccs (sccs);
|
free_ddg_all_sccs (sccs);
|
check_nodes_order (node_order, g->num_nodes);
|
check_nodes_order (node_order, g->num_nodes);
|
|
|
return rec_mii;
|
return rec_mii;
|
}
|
}
|
|
|
static void
|
static void
|
order_nodes_of_sccs (ddg_all_sccs_ptr all_sccs, int * node_order)
|
order_nodes_of_sccs (ddg_all_sccs_ptr all_sccs, int * node_order)
|
{
|
{
|
int i, pos = 0;
|
int i, pos = 0;
|
ddg_ptr g = all_sccs->ddg;
|
ddg_ptr g = all_sccs->ddg;
|
int num_nodes = g->num_nodes;
|
int num_nodes = g->num_nodes;
|
sbitmap prev_sccs = sbitmap_alloc (num_nodes);
|
sbitmap prev_sccs = sbitmap_alloc (num_nodes);
|
sbitmap on_path = sbitmap_alloc (num_nodes);
|
sbitmap on_path = sbitmap_alloc (num_nodes);
|
sbitmap tmp = sbitmap_alloc (num_nodes);
|
sbitmap tmp = sbitmap_alloc (num_nodes);
|
sbitmap ones = sbitmap_alloc (num_nodes);
|
sbitmap ones = sbitmap_alloc (num_nodes);
|
|
|
sbitmap_zero (prev_sccs);
|
sbitmap_zero (prev_sccs);
|
sbitmap_ones (ones);
|
sbitmap_ones (ones);
|
|
|
/* Perform the node ordering starting from the SCC with the highest recMII.
|
/* Perform the node ordering starting from the SCC with the highest recMII.
|
For each SCC order the nodes according to their ASAP/ALAP/HEIGHT etc. */
|
For each SCC order the nodes according to their ASAP/ALAP/HEIGHT etc. */
|
for (i = 0; i < all_sccs->num_sccs; i++)
|
for (i = 0; i < all_sccs->num_sccs; i++)
|
{
|
{
|
ddg_scc_ptr scc = all_sccs->sccs[i];
|
ddg_scc_ptr scc = all_sccs->sccs[i];
|
|
|
/* Add nodes on paths from previous SCCs to the current SCC. */
|
/* Add nodes on paths from previous SCCs to the current SCC. */
|
find_nodes_on_paths (on_path, g, prev_sccs, scc->nodes);
|
find_nodes_on_paths (on_path, g, prev_sccs, scc->nodes);
|
sbitmap_a_or_b (tmp, scc->nodes, on_path);
|
sbitmap_a_or_b (tmp, scc->nodes, on_path);
|
|
|
/* Add nodes on paths from the current SCC to previous SCCs. */
|
/* Add nodes on paths from the current SCC to previous SCCs. */
|
find_nodes_on_paths (on_path, g, scc->nodes, prev_sccs);
|
find_nodes_on_paths (on_path, g, scc->nodes, prev_sccs);
|
sbitmap_a_or_b (tmp, tmp, on_path);
|
sbitmap_a_or_b (tmp, tmp, on_path);
|
|
|
/* Remove nodes of previous SCCs from current extended SCC. */
|
/* Remove nodes of previous SCCs from current extended SCC. */
|
sbitmap_difference (tmp, tmp, prev_sccs);
|
sbitmap_difference (tmp, tmp, prev_sccs);
|
|
|
pos = order_nodes_in_scc (g, prev_sccs, tmp, node_order, pos);
|
pos = order_nodes_in_scc (g, prev_sccs, tmp, node_order, pos);
|
/* Above call to order_nodes_in_scc updated prev_sccs |= tmp. */
|
/* Above call to order_nodes_in_scc updated prev_sccs |= tmp. */
|
}
|
}
|
|
|
/* Handle the remaining nodes that do not belong to any scc. Each call
|
/* Handle the remaining nodes that do not belong to any scc. Each call
|
to order_nodes_in_scc handles a single connected component. */
|
to order_nodes_in_scc handles a single connected component. */
|
while (pos < g->num_nodes)
|
while (pos < g->num_nodes)
|
{
|
{
|
sbitmap_difference (tmp, ones, prev_sccs);
|
sbitmap_difference (tmp, ones, prev_sccs);
|
pos = order_nodes_in_scc (g, prev_sccs, tmp, node_order, pos);
|
pos = order_nodes_in_scc (g, prev_sccs, tmp, node_order, pos);
|
}
|
}
|
sbitmap_free (prev_sccs);
|
sbitmap_free (prev_sccs);
|
sbitmap_free (on_path);
|
sbitmap_free (on_path);
|
sbitmap_free (tmp);
|
sbitmap_free (tmp);
|
sbitmap_free (ones);
|
sbitmap_free (ones);
|
}
|
}
|
|
|
/* MII is needed if we consider backarcs (that do not close recursive cycles). */
|
/* MII is needed if we consider backarcs (that do not close recursive cycles). */
|
static struct node_order_params *
|
static struct node_order_params *
|
calculate_order_params (ddg_ptr g, int mii ATTRIBUTE_UNUSED, int *pmax_asap)
|
calculate_order_params (ddg_ptr g, int mii ATTRIBUTE_UNUSED, int *pmax_asap)
|
{
|
{
|
int u;
|
int u;
|
int max_asap;
|
int max_asap;
|
int num_nodes = g->num_nodes;
|
int num_nodes = g->num_nodes;
|
ddg_edge_ptr e;
|
ddg_edge_ptr e;
|
/* Allocate a place to hold ordering params for each node in the DDG. */
|
/* Allocate a place to hold ordering params for each node in the DDG. */
|
nopa node_order_params_arr;
|
nopa node_order_params_arr;
|
|
|
/* Initialize of ASAP/ALAP/HEIGHT to zero. */
|
/* Initialize of ASAP/ALAP/HEIGHT to zero. */
|
node_order_params_arr = (nopa) xcalloc (num_nodes,
|
node_order_params_arr = (nopa) xcalloc (num_nodes,
|
sizeof (struct node_order_params));
|
sizeof (struct node_order_params));
|
|
|
/* Set the aux pointer of each node to point to its order_params structure. */
|
/* Set the aux pointer of each node to point to its order_params structure. */
|
for (u = 0; u < num_nodes; u++)
|
for (u = 0; u < num_nodes; u++)
|
g->nodes[u].aux.info = &node_order_params_arr[u];
|
g->nodes[u].aux.info = &node_order_params_arr[u];
|
|
|
/* Disregarding a backarc from each recursive cycle to obtain a DAG,
|
/* Disregarding a backarc from each recursive cycle to obtain a DAG,
|
calculate ASAP, ALAP, mobility, distance, and height for each node
|
calculate ASAP, ALAP, mobility, distance, and height for each node
|
in the dependence (direct acyclic) graph. */
|
in the dependence (direct acyclic) graph. */
|
|
|
/* We assume that the nodes in the array are in topological order. */
|
/* We assume that the nodes in the array are in topological order. */
|
|
|
max_asap = 0;
|
max_asap = 0;
|
for (u = 0; u < num_nodes; u++)
|
for (u = 0; u < num_nodes; u++)
|
{
|
{
|
ddg_node_ptr u_node = &g->nodes[u];
|
ddg_node_ptr u_node = &g->nodes[u];
|
|
|
ASAP (u_node) = 0;
|
ASAP (u_node) = 0;
|
for (e = u_node->in; e; e = e->next_in)
|
for (e = u_node->in; e; e = e->next_in)
|
if (e->distance == 0)
|
if (e->distance == 0)
|
ASAP (u_node) = MAX (ASAP (u_node),
|
ASAP (u_node) = MAX (ASAP (u_node),
|
ASAP (e->src) + e->latency);
|
ASAP (e->src) + e->latency);
|
max_asap = MAX (max_asap, ASAP (u_node));
|
max_asap = MAX (max_asap, ASAP (u_node));
|
}
|
}
|
|
|
for (u = num_nodes - 1; u > -1; u--)
|
for (u = num_nodes - 1; u > -1; u--)
|
{
|
{
|
ddg_node_ptr u_node = &g->nodes[u];
|
ddg_node_ptr u_node = &g->nodes[u];
|
|
|
ALAP (u_node) = max_asap;
|
ALAP (u_node) = max_asap;
|
HEIGHT (u_node) = 0;
|
HEIGHT (u_node) = 0;
|
for (e = u_node->out; e; e = e->next_out)
|
for (e = u_node->out; e; e = e->next_out)
|
if (e->distance == 0)
|
if (e->distance == 0)
|
{
|
{
|
ALAP (u_node) = MIN (ALAP (u_node),
|
ALAP (u_node) = MIN (ALAP (u_node),
|
ALAP (e->dest) - e->latency);
|
ALAP (e->dest) - e->latency);
|
HEIGHT (u_node) = MAX (HEIGHT (u_node),
|
HEIGHT (u_node) = MAX (HEIGHT (u_node),
|
HEIGHT (e->dest) + e->latency);
|
HEIGHT (e->dest) + e->latency);
|
}
|
}
|
}
|
}
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "\nOrder params\n");
|
fprintf (dump_file, "\nOrder params\n");
|
for (u = 0; u < num_nodes; u++)
|
for (u = 0; u < num_nodes; u++)
|
{
|
{
|
ddg_node_ptr u_node = &g->nodes[u];
|
ddg_node_ptr u_node = &g->nodes[u];
|
|
|
fprintf (dump_file, "node %d, ASAP: %d, ALAP: %d, HEIGHT: %d\n", u,
|
fprintf (dump_file, "node %d, ASAP: %d, ALAP: %d, HEIGHT: %d\n", u,
|
ASAP (u_node), ALAP (u_node), HEIGHT (u_node));
|
ASAP (u_node), ALAP (u_node), HEIGHT (u_node));
|
}
|
}
|
}
|
}
|
|
|
*pmax_asap = max_asap;
|
*pmax_asap = max_asap;
|
return node_order_params_arr;
|
return node_order_params_arr;
|
}
|
}
|
|
|
static int
|
static int
|
find_max_asap (ddg_ptr g, sbitmap nodes)
|
find_max_asap (ddg_ptr g, sbitmap nodes)
|
{
|
{
|
unsigned int u = 0;
|
unsigned int u = 0;
|
int max_asap = -1;
|
int max_asap = -1;
|
int result = -1;
|
int result = -1;
|
sbitmap_iterator sbi;
|
sbitmap_iterator sbi;
|
|
|
EXECUTE_IF_SET_IN_SBITMAP (nodes, 0, u, sbi)
|
EXECUTE_IF_SET_IN_SBITMAP (nodes, 0, u, sbi)
|
{
|
{
|
ddg_node_ptr u_node = &g->nodes[u];
|
ddg_node_ptr u_node = &g->nodes[u];
|
|
|
if (max_asap < ASAP (u_node))
|
if (max_asap < ASAP (u_node))
|
{
|
{
|
max_asap = ASAP (u_node);
|
max_asap = ASAP (u_node);
|
result = u;
|
result = u;
|
}
|
}
|
}
|
}
|
return result;
|
return result;
|
}
|
}
|
|
|
static int
|
static int
|
find_max_hv_min_mob (ddg_ptr g, sbitmap nodes)
|
find_max_hv_min_mob (ddg_ptr g, sbitmap nodes)
|
{
|
{
|
unsigned int u = 0;
|
unsigned int u = 0;
|
int max_hv = -1;
|
int max_hv = -1;
|
int min_mob = INT_MAX;
|
int min_mob = INT_MAX;
|
int result = -1;
|
int result = -1;
|
sbitmap_iterator sbi;
|
sbitmap_iterator sbi;
|
|
|
EXECUTE_IF_SET_IN_SBITMAP (nodes, 0, u, sbi)
|
EXECUTE_IF_SET_IN_SBITMAP (nodes, 0, u, sbi)
|
{
|
{
|
ddg_node_ptr u_node = &g->nodes[u];
|
ddg_node_ptr u_node = &g->nodes[u];
|
|
|
if (max_hv < HEIGHT (u_node))
|
if (max_hv < HEIGHT (u_node))
|
{
|
{
|
max_hv = HEIGHT (u_node);
|
max_hv = HEIGHT (u_node);
|
min_mob = MOB (u_node);
|
min_mob = MOB (u_node);
|
result = u;
|
result = u;
|
}
|
}
|
else if ((max_hv == HEIGHT (u_node))
|
else if ((max_hv == HEIGHT (u_node))
|
&& (min_mob > MOB (u_node)))
|
&& (min_mob > MOB (u_node)))
|
{
|
{
|
min_mob = MOB (u_node);
|
min_mob = MOB (u_node);
|
result = u;
|
result = u;
|
}
|
}
|
}
|
}
|
return result;
|
return result;
|
}
|
}
|
|
|
static int
|
static int
|
find_max_dv_min_mob (ddg_ptr g, sbitmap nodes)
|
find_max_dv_min_mob (ddg_ptr g, sbitmap nodes)
|
{
|
{
|
unsigned int u = 0;
|
unsigned int u = 0;
|
int max_dv = -1;
|
int max_dv = -1;
|
int min_mob = INT_MAX;
|
int min_mob = INT_MAX;
|
int result = -1;
|
int result = -1;
|
sbitmap_iterator sbi;
|
sbitmap_iterator sbi;
|
|
|
EXECUTE_IF_SET_IN_SBITMAP (nodes, 0, u, sbi)
|
EXECUTE_IF_SET_IN_SBITMAP (nodes, 0, u, sbi)
|
{
|
{
|
ddg_node_ptr u_node = &g->nodes[u];
|
ddg_node_ptr u_node = &g->nodes[u];
|
|
|
if (max_dv < DEPTH (u_node))
|
if (max_dv < DEPTH (u_node))
|
{
|
{
|
max_dv = DEPTH (u_node);
|
max_dv = DEPTH (u_node);
|
min_mob = MOB (u_node);
|
min_mob = MOB (u_node);
|
result = u;
|
result = u;
|
}
|
}
|
else if ((max_dv == DEPTH (u_node))
|
else if ((max_dv == DEPTH (u_node))
|
&& (min_mob > MOB (u_node)))
|
&& (min_mob > MOB (u_node)))
|
{
|
{
|
min_mob = MOB (u_node);
|
min_mob = MOB (u_node);
|
result = u;
|
result = u;
|
}
|
}
|
}
|
}
|
return result;
|
return result;
|
}
|
}
|
|
|
/* Places the nodes of SCC into the NODE_ORDER array starting
|
/* Places the nodes of SCC into the NODE_ORDER array starting
|
at position POS, according to the SMS ordering algorithm.
|
at position POS, according to the SMS ordering algorithm.
|
NODES_ORDERED (in&out parameter) holds the bitset of all nodes in
|
NODES_ORDERED (in&out parameter) holds the bitset of all nodes in
|
the NODE_ORDER array, starting from position zero. */
|
the NODE_ORDER array, starting from position zero. */
|
static int
|
static int
|
order_nodes_in_scc (ddg_ptr g, sbitmap nodes_ordered, sbitmap scc,
|
order_nodes_in_scc (ddg_ptr g, sbitmap nodes_ordered, sbitmap scc,
|
int * node_order, int pos)
|
int * node_order, int pos)
|
{
|
{
|
enum sms_direction dir;
|
enum sms_direction dir;
|
int num_nodes = g->num_nodes;
|
int num_nodes = g->num_nodes;
|
sbitmap workset = sbitmap_alloc (num_nodes);
|
sbitmap workset = sbitmap_alloc (num_nodes);
|
sbitmap tmp = sbitmap_alloc (num_nodes);
|
sbitmap tmp = sbitmap_alloc (num_nodes);
|
sbitmap zero_bitmap = sbitmap_alloc (num_nodes);
|
sbitmap zero_bitmap = sbitmap_alloc (num_nodes);
|
sbitmap predecessors = sbitmap_alloc (num_nodes);
|
sbitmap predecessors = sbitmap_alloc (num_nodes);
|
sbitmap successors = sbitmap_alloc (num_nodes);
|
sbitmap successors = sbitmap_alloc (num_nodes);
|
|
|
sbitmap_zero (predecessors);
|
sbitmap_zero (predecessors);
|
find_predecessors (predecessors, g, nodes_ordered);
|
find_predecessors (predecessors, g, nodes_ordered);
|
|
|
sbitmap_zero (successors);
|
sbitmap_zero (successors);
|
find_successors (successors, g, nodes_ordered);
|
find_successors (successors, g, nodes_ordered);
|
|
|
sbitmap_zero (tmp);
|
sbitmap_zero (tmp);
|
if (sbitmap_a_and_b_cg (tmp, predecessors, scc))
|
if (sbitmap_a_and_b_cg (tmp, predecessors, scc))
|
{
|
{
|
sbitmap_copy (workset, tmp);
|
sbitmap_copy (workset, tmp);
|
dir = BOTTOMUP;
|
dir = BOTTOMUP;
|
}
|
}
|
else if (sbitmap_a_and_b_cg (tmp, successors, scc))
|
else if (sbitmap_a_and_b_cg (tmp, successors, scc))
|
{
|
{
|
sbitmap_copy (workset, tmp);
|
sbitmap_copy (workset, tmp);
|
dir = TOPDOWN;
|
dir = TOPDOWN;
|
}
|
}
|
else
|
else
|
{
|
{
|
int u;
|
int u;
|
|
|
sbitmap_zero (workset);
|
sbitmap_zero (workset);
|
if ((u = find_max_asap (g, scc)) >= 0)
|
if ((u = find_max_asap (g, scc)) >= 0)
|
SET_BIT (workset, u);
|
SET_BIT (workset, u);
|
dir = BOTTOMUP;
|
dir = BOTTOMUP;
|
}
|
}
|
|
|
sbitmap_zero (zero_bitmap);
|
sbitmap_zero (zero_bitmap);
|
while (!sbitmap_equal (workset, zero_bitmap))
|
while (!sbitmap_equal (workset, zero_bitmap))
|
{
|
{
|
int v;
|
int v;
|
ddg_node_ptr v_node;
|
ddg_node_ptr v_node;
|
sbitmap v_node_preds;
|
sbitmap v_node_preds;
|
sbitmap v_node_succs;
|
sbitmap v_node_succs;
|
|
|
if (dir == TOPDOWN)
|
if (dir == TOPDOWN)
|
{
|
{
|
while (!sbitmap_equal (workset, zero_bitmap))
|
while (!sbitmap_equal (workset, zero_bitmap))
|
{
|
{
|
v = find_max_hv_min_mob (g, workset);
|
v = find_max_hv_min_mob (g, workset);
|
v_node = &g->nodes[v];
|
v_node = &g->nodes[v];
|
node_order[pos++] = v;
|
node_order[pos++] = v;
|
v_node_succs = NODE_SUCCESSORS (v_node);
|
v_node_succs = NODE_SUCCESSORS (v_node);
|
sbitmap_a_and_b (tmp, v_node_succs, scc);
|
sbitmap_a_and_b (tmp, v_node_succs, scc);
|
|
|
/* Don't consider the already ordered successors again. */
|
/* Don't consider the already ordered successors again. */
|
sbitmap_difference (tmp, tmp, nodes_ordered);
|
sbitmap_difference (tmp, tmp, nodes_ordered);
|
sbitmap_a_or_b (workset, workset, tmp);
|
sbitmap_a_or_b (workset, workset, tmp);
|
RESET_BIT (workset, v);
|
RESET_BIT (workset, v);
|
SET_BIT (nodes_ordered, v);
|
SET_BIT (nodes_ordered, v);
|
}
|
}
|
dir = BOTTOMUP;
|
dir = BOTTOMUP;
|
sbitmap_zero (predecessors);
|
sbitmap_zero (predecessors);
|
find_predecessors (predecessors, g, nodes_ordered);
|
find_predecessors (predecessors, g, nodes_ordered);
|
sbitmap_a_and_b (workset, predecessors, scc);
|
sbitmap_a_and_b (workset, predecessors, scc);
|
}
|
}
|
else
|
else
|
{
|
{
|
while (!sbitmap_equal (workset, zero_bitmap))
|
while (!sbitmap_equal (workset, zero_bitmap))
|
{
|
{
|
v = find_max_dv_min_mob (g, workset);
|
v = find_max_dv_min_mob (g, workset);
|
v_node = &g->nodes[v];
|
v_node = &g->nodes[v];
|
node_order[pos++] = v;
|
node_order[pos++] = v;
|
v_node_preds = NODE_PREDECESSORS (v_node);
|
v_node_preds = NODE_PREDECESSORS (v_node);
|
sbitmap_a_and_b (tmp, v_node_preds, scc);
|
sbitmap_a_and_b (tmp, v_node_preds, scc);
|
|
|
/* Don't consider the already ordered predecessors again. */
|
/* Don't consider the already ordered predecessors again. */
|
sbitmap_difference (tmp, tmp, nodes_ordered);
|
sbitmap_difference (tmp, tmp, nodes_ordered);
|
sbitmap_a_or_b (workset, workset, tmp);
|
sbitmap_a_or_b (workset, workset, tmp);
|
RESET_BIT (workset, v);
|
RESET_BIT (workset, v);
|
SET_BIT (nodes_ordered, v);
|
SET_BIT (nodes_ordered, v);
|
}
|
}
|
dir = TOPDOWN;
|
dir = TOPDOWN;
|
sbitmap_zero (successors);
|
sbitmap_zero (successors);
|
find_successors (successors, g, nodes_ordered);
|
find_successors (successors, g, nodes_ordered);
|
sbitmap_a_and_b (workset, successors, scc);
|
sbitmap_a_and_b (workset, successors, scc);
|
}
|
}
|
}
|
}
|
sbitmap_free (tmp);
|
sbitmap_free (tmp);
|
sbitmap_free (workset);
|
sbitmap_free (workset);
|
sbitmap_free (zero_bitmap);
|
sbitmap_free (zero_bitmap);
|
sbitmap_free (predecessors);
|
sbitmap_free (predecessors);
|
sbitmap_free (successors);
|
sbitmap_free (successors);
|
return pos;
|
return pos;
|
}
|
}
|
|
|
|
|
/* This page contains functions for manipulating partial-schedules during
|
/* This page contains functions for manipulating partial-schedules during
|
modulo scheduling. */
|
modulo scheduling. */
|
|
|
/* Create a partial schedule and allocate a memory to hold II rows. */
|
/* Create a partial schedule and allocate a memory to hold II rows. */
|
|
|
static partial_schedule_ptr
|
static partial_schedule_ptr
|
create_partial_schedule (int ii, ddg_ptr g, int history)
|
create_partial_schedule (int ii, ddg_ptr g, int history)
|
{
|
{
|
partial_schedule_ptr ps = XNEW (struct partial_schedule);
|
partial_schedule_ptr ps = XNEW (struct partial_schedule);
|
ps->rows = (ps_insn_ptr *) xcalloc (ii, sizeof (ps_insn_ptr));
|
ps->rows = (ps_insn_ptr *) xcalloc (ii, sizeof (ps_insn_ptr));
|
ps->rows_length = (int *) xcalloc (ii, sizeof (int));
|
ps->rows_length = (int *) xcalloc (ii, sizeof (int));
|
ps->reg_moves = NULL;
|
ps->reg_moves = NULL;
|
ps->ii = ii;
|
ps->ii = ii;
|
ps->history = history;
|
ps->history = history;
|
ps->min_cycle = INT_MAX;
|
ps->min_cycle = INT_MAX;
|
ps->max_cycle = INT_MIN;
|
ps->max_cycle = INT_MIN;
|
ps->g = g;
|
ps->g = g;
|
|
|
return ps;
|
return ps;
|
}
|
}
|
|
|
/* Free the PS_INSNs in rows array of the given partial schedule.
|
/* Free the PS_INSNs in rows array of the given partial schedule.
|
??? Consider caching the PS_INSN's. */
|
??? Consider caching the PS_INSN's. */
|
static void
|
static void
|
free_ps_insns (partial_schedule_ptr ps)
|
free_ps_insns (partial_schedule_ptr ps)
|
{
|
{
|
int i;
|
int i;
|
|
|
for (i = 0; i < ps->ii; i++)
|
for (i = 0; i < ps->ii; i++)
|
{
|
{
|
while (ps->rows[i])
|
while (ps->rows[i])
|
{
|
{
|
ps_insn_ptr ps_insn = ps->rows[i]->next_in_row;
|
ps_insn_ptr ps_insn = ps->rows[i]->next_in_row;
|
|
|
free (ps->rows[i]);
|
free (ps->rows[i]);
|
ps->rows[i] = ps_insn;
|
ps->rows[i] = ps_insn;
|
}
|
}
|
ps->rows[i] = NULL;
|
ps->rows[i] = NULL;
|
}
|
}
|
}
|
}
|
|
|
/* Free all the memory allocated to the partial schedule. */
|
/* Free all the memory allocated to the partial schedule. */
|
|
|
static void
|
static void
|
free_partial_schedule (partial_schedule_ptr ps)
|
free_partial_schedule (partial_schedule_ptr ps)
|
{
|
{
|
ps_reg_move_info *move;
|
ps_reg_move_info *move;
|
unsigned int i;
|
unsigned int i;
|
|
|
if (!ps)
|
if (!ps)
|
return;
|
return;
|
|
|
FOR_EACH_VEC_ELT (ps_reg_move_info, ps->reg_moves, i, move)
|
FOR_EACH_VEC_ELT (ps_reg_move_info, ps->reg_moves, i, move)
|
sbitmap_free (move->uses);
|
sbitmap_free (move->uses);
|
VEC_free (ps_reg_move_info, heap, ps->reg_moves);
|
VEC_free (ps_reg_move_info, heap, ps->reg_moves);
|
|
|
free_ps_insns (ps);
|
free_ps_insns (ps);
|
free (ps->rows);
|
free (ps->rows);
|
free (ps->rows_length);
|
free (ps->rows_length);
|
free (ps);
|
free (ps);
|
}
|
}
|
|
|
/* Clear the rows array with its PS_INSNs, and create a new one with
|
/* Clear the rows array with its PS_INSNs, and create a new one with
|
NEW_II rows. */
|
NEW_II rows. */
|
|
|
static void
|
static void
|
reset_partial_schedule (partial_schedule_ptr ps, int new_ii)
|
reset_partial_schedule (partial_schedule_ptr ps, int new_ii)
|
{
|
{
|
if (!ps)
|
if (!ps)
|
return;
|
return;
|
free_ps_insns (ps);
|
free_ps_insns (ps);
|
if (new_ii == ps->ii)
|
if (new_ii == ps->ii)
|
return;
|
return;
|
ps->rows = (ps_insn_ptr *) xrealloc (ps->rows, new_ii
|
ps->rows = (ps_insn_ptr *) xrealloc (ps->rows, new_ii
|
* sizeof (ps_insn_ptr));
|
* sizeof (ps_insn_ptr));
|
memset (ps->rows, 0, new_ii * sizeof (ps_insn_ptr));
|
memset (ps->rows, 0, new_ii * sizeof (ps_insn_ptr));
|
ps->rows_length = (int *) xrealloc (ps->rows_length, new_ii * sizeof (int));
|
ps->rows_length = (int *) xrealloc (ps->rows_length, new_ii * sizeof (int));
|
memset (ps->rows_length, 0, new_ii * sizeof (int));
|
memset (ps->rows_length, 0, new_ii * sizeof (int));
|
ps->ii = new_ii;
|
ps->ii = new_ii;
|
ps->min_cycle = INT_MAX;
|
ps->min_cycle = INT_MAX;
|
ps->max_cycle = INT_MIN;
|
ps->max_cycle = INT_MIN;
|
}
|
}
|
|
|
/* Prints the partial schedule as an ii rows array, for each rows
|
/* Prints the partial schedule as an ii rows array, for each rows
|
print the ids of the insns in it. */
|
print the ids of the insns in it. */
|
void
|
void
|
print_partial_schedule (partial_schedule_ptr ps, FILE *dump)
|
print_partial_schedule (partial_schedule_ptr ps, FILE *dump)
|
{
|
{
|
int i;
|
int i;
|
|
|
for (i = 0; i < ps->ii; i++)
|
for (i = 0; i < ps->ii; i++)
|
{
|
{
|
ps_insn_ptr ps_i = ps->rows[i];
|
ps_insn_ptr ps_i = ps->rows[i];
|
|
|
fprintf (dump, "\n[ROW %d ]: ", i);
|
fprintf (dump, "\n[ROW %d ]: ", i);
|
while (ps_i)
|
while (ps_i)
|
{
|
{
|
rtx insn = ps_rtl_insn (ps, ps_i->id);
|
rtx insn = ps_rtl_insn (ps, ps_i->id);
|
|
|
if (JUMP_P (insn))
|
if (JUMP_P (insn))
|
fprintf (dump, "%d (branch), ", INSN_UID (insn));
|
fprintf (dump, "%d (branch), ", INSN_UID (insn));
|
else
|
else
|
fprintf (dump, "%d, ", INSN_UID (insn));
|
fprintf (dump, "%d, ", INSN_UID (insn));
|
|
|
ps_i = ps_i->next_in_row;
|
ps_i = ps_i->next_in_row;
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
/* Creates an object of PS_INSN and initializes it to the given parameters. */
|
/* Creates an object of PS_INSN and initializes it to the given parameters. */
|
static ps_insn_ptr
|
static ps_insn_ptr
|
create_ps_insn (int id, int cycle)
|
create_ps_insn (int id, int cycle)
|
{
|
{
|
ps_insn_ptr ps_i = XNEW (struct ps_insn);
|
ps_insn_ptr ps_i = XNEW (struct ps_insn);
|
|
|
ps_i->id = id;
|
ps_i->id = id;
|
ps_i->next_in_row = NULL;
|
ps_i->next_in_row = NULL;
|
ps_i->prev_in_row = NULL;
|
ps_i->prev_in_row = NULL;
|
ps_i->cycle = cycle;
|
ps_i->cycle = cycle;
|
|
|
return ps_i;
|
return ps_i;
|
}
|
}
|
|
|
|
|
/* Removes the given PS_INSN from the partial schedule. */
|
/* Removes the given PS_INSN from the partial schedule. */
|
static void
|
static void
|
remove_node_from_ps (partial_schedule_ptr ps, ps_insn_ptr ps_i)
|
remove_node_from_ps (partial_schedule_ptr ps, ps_insn_ptr ps_i)
|
{
|
{
|
int row;
|
int row;
|
|
|
gcc_assert (ps && ps_i);
|
gcc_assert (ps && ps_i);
|
|
|
row = SMODULO (ps_i->cycle, ps->ii);
|
row = SMODULO (ps_i->cycle, ps->ii);
|
if (! ps_i->prev_in_row)
|
if (! ps_i->prev_in_row)
|
{
|
{
|
gcc_assert (ps_i == ps->rows[row]);
|
gcc_assert (ps_i == ps->rows[row]);
|
ps->rows[row] = ps_i->next_in_row;
|
ps->rows[row] = ps_i->next_in_row;
|
if (ps->rows[row])
|
if (ps->rows[row])
|
ps->rows[row]->prev_in_row = NULL;
|
ps->rows[row]->prev_in_row = NULL;
|
}
|
}
|
else
|
else
|
{
|
{
|
ps_i->prev_in_row->next_in_row = ps_i->next_in_row;
|
ps_i->prev_in_row->next_in_row = ps_i->next_in_row;
|
if (ps_i->next_in_row)
|
if (ps_i->next_in_row)
|
ps_i->next_in_row->prev_in_row = ps_i->prev_in_row;
|
ps_i->next_in_row->prev_in_row = ps_i->prev_in_row;
|
}
|
}
|
|
|
ps->rows_length[row] -= 1;
|
ps->rows_length[row] -= 1;
|
free (ps_i);
|
free (ps_i);
|
return;
|
return;
|
}
|
}
|
|
|
/* Unlike what literature describes for modulo scheduling (which focuses
|
/* Unlike what literature describes for modulo scheduling (which focuses
|
on VLIW machines) the order of the instructions inside a cycle is
|
on VLIW machines) the order of the instructions inside a cycle is
|
important. Given the bitmaps MUST_FOLLOW and MUST_PRECEDE we know
|
important. Given the bitmaps MUST_FOLLOW and MUST_PRECEDE we know
|
where the current instruction should go relative to the already
|
where the current instruction should go relative to the already
|
scheduled instructions in the given cycle. Go over these
|
scheduled instructions in the given cycle. Go over these
|
instructions and find the first possible column to put it in. */
|
instructions and find the first possible column to put it in. */
|
static bool
|
static bool
|
ps_insn_find_column (partial_schedule_ptr ps, ps_insn_ptr ps_i,
|
ps_insn_find_column (partial_schedule_ptr ps, ps_insn_ptr ps_i,
|
sbitmap must_precede, sbitmap must_follow)
|
sbitmap must_precede, sbitmap must_follow)
|
{
|
{
|
ps_insn_ptr next_ps_i;
|
ps_insn_ptr next_ps_i;
|
ps_insn_ptr first_must_follow = NULL;
|
ps_insn_ptr first_must_follow = NULL;
|
ps_insn_ptr last_must_precede = NULL;
|
ps_insn_ptr last_must_precede = NULL;
|
ps_insn_ptr last_in_row = NULL;
|
ps_insn_ptr last_in_row = NULL;
|
int row;
|
int row;
|
|
|
if (! ps_i)
|
if (! ps_i)
|
return false;
|
return false;
|
|
|
row = SMODULO (ps_i->cycle, ps->ii);
|
row = SMODULO (ps_i->cycle, ps->ii);
|
|
|
/* Find the first must follow and the last must precede
|
/* Find the first must follow and the last must precede
|
and insert the node immediately after the must precede
|
and insert the node immediately after the must precede
|
but make sure that it there is no must follow after it. */
|
but make sure that it there is no must follow after it. */
|
for (next_ps_i = ps->rows[row];
|
for (next_ps_i = ps->rows[row];
|
next_ps_i;
|
next_ps_i;
|
next_ps_i = next_ps_i->next_in_row)
|
next_ps_i = next_ps_i->next_in_row)
|
{
|
{
|
if (must_follow
|
if (must_follow
|
&& TEST_BIT (must_follow, next_ps_i->id)
|
&& TEST_BIT (must_follow, next_ps_i->id)
|
&& ! first_must_follow)
|
&& ! first_must_follow)
|
first_must_follow = next_ps_i;
|
first_must_follow = next_ps_i;
|
if (must_precede && TEST_BIT (must_precede, next_ps_i->id))
|
if (must_precede && TEST_BIT (must_precede, next_ps_i->id))
|
{
|
{
|
/* If we have already met a node that must follow, then
|
/* If we have already met a node that must follow, then
|
there is no possible column. */
|
there is no possible column. */
|
if (first_must_follow)
|
if (first_must_follow)
|
return false;
|
return false;
|
else
|
else
|
last_must_precede = next_ps_i;
|
last_must_precede = next_ps_i;
|
}
|
}
|
/* The closing branch must be the last in the row. */
|
/* The closing branch must be the last in the row. */
|
if (must_precede
|
if (must_precede
|
&& TEST_BIT (must_precede, next_ps_i->id)
|
&& TEST_BIT (must_precede, next_ps_i->id)
|
&& JUMP_P (ps_rtl_insn (ps, next_ps_i->id)))
|
&& JUMP_P (ps_rtl_insn (ps, next_ps_i->id)))
|
return false;
|
return false;
|
|
|
last_in_row = next_ps_i;
|
last_in_row = next_ps_i;
|
}
|
}
|
|
|
/* The closing branch is scheduled as well. Make sure there is no
|
/* The closing branch is scheduled as well. Make sure there is no
|
dependent instruction after it as the branch should be the last
|
dependent instruction after it as the branch should be the last
|
instruction in the row. */
|
instruction in the row. */
|
if (JUMP_P (ps_rtl_insn (ps, ps_i->id)))
|
if (JUMP_P (ps_rtl_insn (ps, ps_i->id)))
|
{
|
{
|
if (first_must_follow)
|
if (first_must_follow)
|
return false;
|
return false;
|
if (last_in_row)
|
if (last_in_row)
|
{
|
{
|
/* Make the branch the last in the row. New instructions
|
/* Make the branch the last in the row. New instructions
|
will be inserted at the beginning of the row or after the
|
will be inserted at the beginning of the row or after the
|
last must_precede instruction thus the branch is guaranteed
|
last must_precede instruction thus the branch is guaranteed
|
to remain the last instruction in the row. */
|
to remain the last instruction in the row. */
|
last_in_row->next_in_row = ps_i;
|
last_in_row->next_in_row = ps_i;
|
ps_i->prev_in_row = last_in_row;
|
ps_i->prev_in_row = last_in_row;
|
ps_i->next_in_row = NULL;
|
ps_i->next_in_row = NULL;
|
}
|
}
|
else
|
else
|
ps->rows[row] = ps_i;
|
ps->rows[row] = ps_i;
|
return true;
|
return true;
|
}
|
}
|
|
|
/* Now insert the node after INSERT_AFTER_PSI. */
|
/* Now insert the node after INSERT_AFTER_PSI. */
|
|
|
if (! last_must_precede)
|
if (! last_must_precede)
|
{
|
{
|
ps_i->next_in_row = ps->rows[row];
|
ps_i->next_in_row = ps->rows[row];
|
ps_i->prev_in_row = NULL;
|
ps_i->prev_in_row = NULL;
|
if (ps_i->next_in_row)
|
if (ps_i->next_in_row)
|
ps_i->next_in_row->prev_in_row = ps_i;
|
ps_i->next_in_row->prev_in_row = ps_i;
|
ps->rows[row] = ps_i;
|
ps->rows[row] = ps_i;
|
}
|
}
|
else
|
else
|
{
|
{
|
ps_i->next_in_row = last_must_precede->next_in_row;
|
ps_i->next_in_row = last_must_precede->next_in_row;
|
last_must_precede->next_in_row = ps_i;
|
last_must_precede->next_in_row = ps_i;
|
ps_i->prev_in_row = last_must_precede;
|
ps_i->prev_in_row = last_must_precede;
|
if (ps_i->next_in_row)
|
if (ps_i->next_in_row)
|
ps_i->next_in_row->prev_in_row = ps_i;
|
ps_i->next_in_row->prev_in_row = ps_i;
|
}
|
}
|
|
|
return true;
|
return true;
|
}
|
}
|
|
|
/* Advances the PS_INSN one column in its current row; returns false
|
/* Advances the PS_INSN one column in its current row; returns false
|
in failure and true in success. Bit N is set in MUST_FOLLOW if
|
in failure and true in success. Bit N is set in MUST_FOLLOW if
|
the node with cuid N must be come after the node pointed to by
|
the node with cuid N must be come after the node pointed to by
|
PS_I when scheduled in the same cycle. */
|
PS_I when scheduled in the same cycle. */
|
static int
|
static int
|
ps_insn_advance_column (partial_schedule_ptr ps, ps_insn_ptr ps_i,
|
ps_insn_advance_column (partial_schedule_ptr ps, ps_insn_ptr ps_i,
|
sbitmap must_follow)
|
sbitmap must_follow)
|
{
|
{
|
ps_insn_ptr prev, next;
|
ps_insn_ptr prev, next;
|
int row;
|
int row;
|
|
|
if (!ps || !ps_i)
|
if (!ps || !ps_i)
|
return false;
|
return false;
|
|
|
row = SMODULO (ps_i->cycle, ps->ii);
|
row = SMODULO (ps_i->cycle, ps->ii);
|
|
|
if (! ps_i->next_in_row)
|
if (! ps_i->next_in_row)
|
return false;
|
return false;
|
|
|
/* Check if next_in_row is dependent on ps_i, both having same sched
|
/* Check if next_in_row is dependent on ps_i, both having same sched
|
times (typically ANTI_DEP). If so, ps_i cannot skip over it. */
|
times (typically ANTI_DEP). If so, ps_i cannot skip over it. */
|
if (must_follow && TEST_BIT (must_follow, ps_i->next_in_row->id))
|
if (must_follow && TEST_BIT (must_follow, ps_i->next_in_row->id))
|
return false;
|
return false;
|
|
|
/* Advance PS_I over its next_in_row in the doubly linked list. */
|
/* Advance PS_I over its next_in_row in the doubly linked list. */
|
prev = ps_i->prev_in_row;
|
prev = ps_i->prev_in_row;
|
next = ps_i->next_in_row;
|
next = ps_i->next_in_row;
|
|
|
if (ps_i == ps->rows[row])
|
if (ps_i == ps->rows[row])
|
ps->rows[row] = next;
|
ps->rows[row] = next;
|
|
|
ps_i->next_in_row = next->next_in_row;
|
ps_i->next_in_row = next->next_in_row;
|
|
|
if (next->next_in_row)
|
if (next->next_in_row)
|
next->next_in_row->prev_in_row = ps_i;
|
next->next_in_row->prev_in_row = ps_i;
|
|
|
next->next_in_row = ps_i;
|
next->next_in_row = ps_i;
|
ps_i->prev_in_row = next;
|
ps_i->prev_in_row = next;
|
|
|
next->prev_in_row = prev;
|
next->prev_in_row = prev;
|
if (prev)
|
if (prev)
|
prev->next_in_row = next;
|
prev->next_in_row = next;
|
|
|
return true;
|
return true;
|
}
|
}
|
|
|
/* Inserts a DDG_NODE to the given partial schedule at the given cycle.
|
/* Inserts a DDG_NODE to the given partial schedule at the given cycle.
|
Returns 0 if this is not possible and a PS_INSN otherwise. Bit N is
|
Returns 0 if this is not possible and a PS_INSN otherwise. Bit N is
|
set in MUST_PRECEDE/MUST_FOLLOW if the node with cuid N must be come
|
set in MUST_PRECEDE/MUST_FOLLOW if the node with cuid N must be come
|
before/after (respectively) the node pointed to by PS_I when scheduled
|
before/after (respectively) the node pointed to by PS_I when scheduled
|
in the same cycle. */
|
in the same cycle. */
|
static ps_insn_ptr
|
static ps_insn_ptr
|
add_node_to_ps (partial_schedule_ptr ps, int id, int cycle,
|
add_node_to_ps (partial_schedule_ptr ps, int id, int cycle,
|
sbitmap must_precede, sbitmap must_follow)
|
sbitmap must_precede, sbitmap must_follow)
|
{
|
{
|
ps_insn_ptr ps_i;
|
ps_insn_ptr ps_i;
|
int row = SMODULO (cycle, ps->ii);
|
int row = SMODULO (cycle, ps->ii);
|
|
|
if (ps->rows_length[row] >= issue_rate)
|
if (ps->rows_length[row] >= issue_rate)
|
return NULL;
|
return NULL;
|
|
|
ps_i = create_ps_insn (id, cycle);
|
ps_i = create_ps_insn (id, cycle);
|
|
|
/* Finds and inserts PS_I according to MUST_FOLLOW and
|
/* Finds and inserts PS_I according to MUST_FOLLOW and
|
MUST_PRECEDE. */
|
MUST_PRECEDE. */
|
if (! ps_insn_find_column (ps, ps_i, must_precede, must_follow))
|
if (! ps_insn_find_column (ps, ps_i, must_precede, must_follow))
|
{
|
{
|
free (ps_i);
|
free (ps_i);
|
return NULL;
|
return NULL;
|
}
|
}
|
|
|
ps->rows_length[row] += 1;
|
ps->rows_length[row] += 1;
|
return ps_i;
|
return ps_i;
|
}
|
}
|
|
|
/* Advance time one cycle. Assumes DFA is being used. */
|
/* Advance time one cycle. Assumes DFA is being used. */
|
static void
|
static void
|
advance_one_cycle (void)
|
advance_one_cycle (void)
|
{
|
{
|
if (targetm.sched.dfa_pre_cycle_insn)
|
if (targetm.sched.dfa_pre_cycle_insn)
|
state_transition (curr_state,
|
state_transition (curr_state,
|
targetm.sched.dfa_pre_cycle_insn ());
|
targetm.sched.dfa_pre_cycle_insn ());
|
|
|
state_transition (curr_state, NULL);
|
state_transition (curr_state, NULL);
|
|
|
if (targetm.sched.dfa_post_cycle_insn)
|
if (targetm.sched.dfa_post_cycle_insn)
|
state_transition (curr_state,
|
state_transition (curr_state,
|
targetm.sched.dfa_post_cycle_insn ());
|
targetm.sched.dfa_post_cycle_insn ());
|
}
|
}
|
|
|
|
|
|
|
/* Checks if PS has resource conflicts according to DFA, starting from
|
/* Checks if PS has resource conflicts according to DFA, starting from
|
FROM cycle to TO cycle; returns true if there are conflicts and false
|
FROM cycle to TO cycle; returns true if there are conflicts and false
|
if there are no conflicts. Assumes DFA is being used. */
|
if there are no conflicts. Assumes DFA is being used. */
|
static int
|
static int
|
ps_has_conflicts (partial_schedule_ptr ps, int from, int to)
|
ps_has_conflicts (partial_schedule_ptr ps, int from, int to)
|
{
|
{
|
int cycle;
|
int cycle;
|
|
|
state_reset (curr_state);
|
state_reset (curr_state);
|
|
|
for (cycle = from; cycle <= to; cycle++)
|
for (cycle = from; cycle <= to; cycle++)
|
{
|
{
|
ps_insn_ptr crr_insn;
|
ps_insn_ptr crr_insn;
|
/* Holds the remaining issue slots in the current row. */
|
/* Holds the remaining issue slots in the current row. */
|
int can_issue_more = issue_rate;
|
int can_issue_more = issue_rate;
|
|
|
/* Walk through the DFA for the current row. */
|
/* Walk through the DFA for the current row. */
|
for (crr_insn = ps->rows[SMODULO (cycle, ps->ii)];
|
for (crr_insn = ps->rows[SMODULO (cycle, ps->ii)];
|
crr_insn;
|
crr_insn;
|
crr_insn = crr_insn->next_in_row)
|
crr_insn = crr_insn->next_in_row)
|
{
|
{
|
rtx insn = ps_rtl_insn (ps, crr_insn->id);
|
rtx insn = ps_rtl_insn (ps, crr_insn->id);
|
|
|
if (!NONDEBUG_INSN_P (insn))
|
if (!NONDEBUG_INSN_P (insn))
|
continue;
|
continue;
|
|
|
/* Check if there is room for the current insn. */
|
/* Check if there is room for the current insn. */
|
if (!can_issue_more || state_dead_lock_p (curr_state))
|
if (!can_issue_more || state_dead_lock_p (curr_state))
|
return true;
|
return true;
|
|
|
/* Update the DFA state and return with failure if the DFA found
|
/* Update the DFA state and return with failure if the DFA found
|
resource conflicts. */
|
resource conflicts. */
|
if (state_transition (curr_state, insn) >= 0)
|
if (state_transition (curr_state, insn) >= 0)
|
return true;
|
return true;
|
|
|
if (targetm.sched.variable_issue)
|
if (targetm.sched.variable_issue)
|
can_issue_more =
|
can_issue_more =
|
targetm.sched.variable_issue (sched_dump, sched_verbose,
|
targetm.sched.variable_issue (sched_dump, sched_verbose,
|
insn, can_issue_more);
|
insn, can_issue_more);
|
/* A naked CLOBBER or USE generates no instruction, so don't
|
/* A naked CLOBBER or USE generates no instruction, so don't
|
let them consume issue slots. */
|
let them consume issue slots. */
|
else if (GET_CODE (PATTERN (insn)) != USE
|
else if (GET_CODE (PATTERN (insn)) != USE
|
&& GET_CODE (PATTERN (insn)) != CLOBBER)
|
&& GET_CODE (PATTERN (insn)) != CLOBBER)
|
can_issue_more--;
|
can_issue_more--;
|
}
|
}
|
|
|
/* Advance the DFA to the next cycle. */
|
/* Advance the DFA to the next cycle. */
|
advance_one_cycle ();
|
advance_one_cycle ();
|
}
|
}
|
return false;
|
return false;
|
}
|
}
|
|
|
/* Checks if the given node causes resource conflicts when added to PS at
|
/* Checks if the given node causes resource conflicts when added to PS at
|
cycle C. If not the node is added to PS and returned; otherwise zero
|
cycle C. If not the node is added to PS and returned; otherwise zero
|
is returned. Bit N is set in MUST_PRECEDE/MUST_FOLLOW if the node with
|
is returned. Bit N is set in MUST_PRECEDE/MUST_FOLLOW if the node with
|
cuid N must be come before/after (respectively) the node pointed to by
|
cuid N must be come before/after (respectively) the node pointed to by
|
PS_I when scheduled in the same cycle. */
|
PS_I when scheduled in the same cycle. */
|
ps_insn_ptr
|
ps_insn_ptr
|
ps_add_node_check_conflicts (partial_schedule_ptr ps, int n,
|
ps_add_node_check_conflicts (partial_schedule_ptr ps, int n,
|
int c, sbitmap must_precede,
|
int c, sbitmap must_precede,
|
sbitmap must_follow)
|
sbitmap must_follow)
|
{
|
{
|
int has_conflicts = 0;
|
int has_conflicts = 0;
|
ps_insn_ptr ps_i;
|
ps_insn_ptr ps_i;
|
|
|
/* First add the node to the PS, if this succeeds check for
|
/* First add the node to the PS, if this succeeds check for
|
conflicts, trying different issue slots in the same row. */
|
conflicts, trying different issue slots in the same row. */
|
if (! (ps_i = add_node_to_ps (ps, n, c, must_precede, must_follow)))
|
if (! (ps_i = add_node_to_ps (ps, n, c, must_precede, must_follow)))
|
return NULL; /* Failed to insert the node at the given cycle. */
|
return NULL; /* Failed to insert the node at the given cycle. */
|
|
|
has_conflicts = ps_has_conflicts (ps, c, c)
|
has_conflicts = ps_has_conflicts (ps, c, c)
|
|| (ps->history > 0
|
|| (ps->history > 0
|
&& ps_has_conflicts (ps,
|
&& ps_has_conflicts (ps,
|
c - ps->history,
|
c - ps->history,
|
c + ps->history));
|
c + ps->history));
|
|
|
/* Try different issue slots to find one that the given node can be
|
/* Try different issue slots to find one that the given node can be
|
scheduled in without conflicts. */
|
scheduled in without conflicts. */
|
while (has_conflicts)
|
while (has_conflicts)
|
{
|
{
|
if (! ps_insn_advance_column (ps, ps_i, must_follow))
|
if (! ps_insn_advance_column (ps, ps_i, must_follow))
|
break;
|
break;
|
has_conflicts = ps_has_conflicts (ps, c, c)
|
has_conflicts = ps_has_conflicts (ps, c, c)
|
|| (ps->history > 0
|
|| (ps->history > 0
|
&& ps_has_conflicts (ps,
|
&& ps_has_conflicts (ps,
|
c - ps->history,
|
c - ps->history,
|
c + ps->history));
|
c + ps->history));
|
}
|
}
|
|
|
if (has_conflicts)
|
if (has_conflicts)
|
{
|
{
|
remove_node_from_ps (ps, ps_i);
|
remove_node_from_ps (ps, ps_i);
|
return NULL;
|
return NULL;
|
}
|
}
|
|
|
ps->min_cycle = MIN (ps->min_cycle, c);
|
ps->min_cycle = MIN (ps->min_cycle, c);
|
ps->max_cycle = MAX (ps->max_cycle, c);
|
ps->max_cycle = MAX (ps->max_cycle, c);
|
return ps_i;
|
return ps_i;
|
}
|
}
|
|
|
/* Calculate the stage count of the partial schedule PS. The calculation
|
/* Calculate the stage count of the partial schedule PS. The calculation
|
takes into account the rotation amount passed in ROTATION_AMOUNT. */
|
takes into account the rotation amount passed in ROTATION_AMOUNT. */
|
int
|
int
|
calculate_stage_count (partial_schedule_ptr ps, int rotation_amount)
|
calculate_stage_count (partial_schedule_ptr ps, int rotation_amount)
|
{
|
{
|
int new_min_cycle = PS_MIN_CYCLE (ps) - rotation_amount;
|
int new_min_cycle = PS_MIN_CYCLE (ps) - rotation_amount;
|
int new_max_cycle = PS_MAX_CYCLE (ps) - rotation_amount;
|
int new_max_cycle = PS_MAX_CYCLE (ps) - rotation_amount;
|
int stage_count = CALC_STAGE_COUNT (-1, new_min_cycle, ps->ii);
|
int stage_count = CALC_STAGE_COUNT (-1, new_min_cycle, ps->ii);
|
|
|
/* The calculation of stage count is done adding the number of stages
|
/* The calculation of stage count is done adding the number of stages
|
before cycle zero and after cycle zero. */
|
before cycle zero and after cycle zero. */
|
stage_count += CALC_STAGE_COUNT (new_max_cycle, 0, ps->ii);
|
stage_count += CALC_STAGE_COUNT (new_max_cycle, 0, ps->ii);
|
|
|
return stage_count;
|
return stage_count;
|
}
|
}
|
|
|
/* Rotate the rows of PS such that insns scheduled at time
|
/* Rotate the rows of PS such that insns scheduled at time
|
START_CYCLE will appear in row 0. Updates max/min_cycles. */
|
START_CYCLE will appear in row 0. Updates max/min_cycles. */
|
void
|
void
|
rotate_partial_schedule (partial_schedule_ptr ps, int start_cycle)
|
rotate_partial_schedule (partial_schedule_ptr ps, int start_cycle)
|
{
|
{
|
int i, row, backward_rotates;
|
int i, row, backward_rotates;
|
int last_row = ps->ii - 1;
|
int last_row = ps->ii - 1;
|
|
|
if (start_cycle == 0)
|
if (start_cycle == 0)
|
return;
|
return;
|
|
|
backward_rotates = SMODULO (start_cycle, ps->ii);
|
backward_rotates = SMODULO (start_cycle, ps->ii);
|
|
|
/* Revisit later and optimize this into a single loop. */
|
/* Revisit later and optimize this into a single loop. */
|
for (i = 0; i < backward_rotates; i++)
|
for (i = 0; i < backward_rotates; i++)
|
{
|
{
|
ps_insn_ptr first_row = ps->rows[0];
|
ps_insn_ptr first_row = ps->rows[0];
|
int first_row_length = ps->rows_length[0];
|
int first_row_length = ps->rows_length[0];
|
|
|
for (row = 0; row < last_row; row++)
|
for (row = 0; row < last_row; row++)
|
{
|
{
|
ps->rows[row] = ps->rows[row + 1];
|
ps->rows[row] = ps->rows[row + 1];
|
ps->rows_length[row] = ps->rows_length[row + 1];
|
ps->rows_length[row] = ps->rows_length[row + 1];
|
}
|
}
|
|
|
ps->rows[last_row] = first_row;
|
ps->rows[last_row] = first_row;
|
ps->rows_length[last_row] = first_row_length;
|
ps->rows_length[last_row] = first_row_length;
|
}
|
}
|
|
|
ps->max_cycle -= start_cycle;
|
ps->max_cycle -= start_cycle;
|
ps->min_cycle -= start_cycle;
|
ps->min_cycle -= start_cycle;
|
}
|
}
|
|
|
#endif /* INSN_SCHEDULING */
|
#endif /* INSN_SCHEDULING */
|
|
|
static bool
|
static bool
|
gate_handle_sms (void)
|
gate_handle_sms (void)
|
{
|
{
|
return (optimize > 0 && flag_modulo_sched);
|
return (optimize > 0 && flag_modulo_sched);
|
}
|
}
|
|
|
|
|
/* Run instruction scheduler. */
|
/* Run instruction scheduler. */
|
/* Perform SMS module scheduling. */
|
/* Perform SMS module scheduling. */
|
static unsigned int
|
static unsigned int
|
rest_of_handle_sms (void)
|
rest_of_handle_sms (void)
|
{
|
{
|
#ifdef INSN_SCHEDULING
|
#ifdef INSN_SCHEDULING
|
basic_block bb;
|
basic_block bb;
|
|
|
/* Collect loop information to be used in SMS. */
|
/* Collect loop information to be used in SMS. */
|
cfg_layout_initialize (0);
|
cfg_layout_initialize (0);
|
sms_schedule ();
|
sms_schedule ();
|
|
|
/* Update the life information, because we add pseudos. */
|
/* Update the life information, because we add pseudos. */
|
max_regno = max_reg_num ();
|
max_regno = max_reg_num ();
|
|
|
/* Finalize layout changes. */
|
/* Finalize layout changes. */
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
if (bb->next_bb != EXIT_BLOCK_PTR)
|
if (bb->next_bb != EXIT_BLOCK_PTR)
|
bb->aux = bb->next_bb;
|
bb->aux = bb->next_bb;
|
free_dominance_info (CDI_DOMINATORS);
|
free_dominance_info (CDI_DOMINATORS);
|
cfg_layout_finalize ();
|
cfg_layout_finalize ();
|
#endif /* INSN_SCHEDULING */
|
#endif /* INSN_SCHEDULING */
|
return 0;
|
return 0;
|
}
|
}
|
|
|
struct rtl_opt_pass pass_sms =
|
struct rtl_opt_pass pass_sms =
|
{
|
{
|
{
|
{
|
RTL_PASS,
|
RTL_PASS,
|
"sms", /* name */
|
"sms", /* name */
|
gate_handle_sms, /* gate */
|
gate_handle_sms, /* gate */
|
rest_of_handle_sms, /* execute */
|
rest_of_handle_sms, /* execute */
|
NULL, /* sub */
|
NULL, /* sub */
|
NULL, /* next */
|
NULL, /* next */
|
0, /* static_pass_number */
|
0, /* static_pass_number */
|
TV_SMS, /* tv_id */
|
TV_SMS, /* tv_id */
|
0, /* properties_required */
|
0, /* properties_required */
|
0, /* properties_provided */
|
0, /* properties_provided */
|
0, /* properties_destroyed */
|
0, /* properties_destroyed */
|
0, /* todo_flags_start */
|
0, /* todo_flags_start */
|
TODO_df_finish
|
TODO_df_finish
|
| TODO_verify_flow
|
| TODO_verify_flow
|
| TODO_verify_rtl_sharing
|
| TODO_verify_rtl_sharing
|
| TODO_ggc_collect /* todo_flags_finish */
|
| TODO_ggc_collect /* todo_flags_finish */
|
}
|
}
|
};
|
};
|
|
|
