/* Instruction scheduling pass.
|
/* Instruction scheduling pass.
|
Copyright (C) 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
|
Copyright (C) 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
|
2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2010
|
2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2010
|
Free Software Foundation, Inc.
|
Free Software Foundation, Inc.
|
Contributed by Michael Tiemann (tiemann@cygnus.com) Enhanced by,
|
Contributed by Michael Tiemann (tiemann@cygnus.com) Enhanced by,
|
and currently maintained by, Jim Wilson (wilson@cygnus.com)
|
and currently maintained by, Jim Wilson (wilson@cygnus.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/>. */
|
|
|
/* This pass implements list scheduling within basic blocks. It is
|
/* This pass implements list scheduling within basic blocks. It is
|
run twice: (1) after flow analysis, but before register allocation,
|
run twice: (1) after flow analysis, but before register allocation,
|
and (2) after register allocation.
|
and (2) after register allocation.
|
|
|
The first run performs interblock scheduling, moving insns between
|
The first run performs interblock scheduling, moving insns between
|
different blocks in the same "region", and the second runs only
|
different blocks in the same "region", and the second runs only
|
basic block scheduling.
|
basic block scheduling.
|
|
|
Interblock motions performed are useful motions and speculative
|
Interblock motions performed are useful motions and speculative
|
motions, including speculative loads. Motions requiring code
|
motions, including speculative loads. Motions requiring code
|
duplication are not supported. The identification of motion type
|
duplication are not supported. The identification of motion type
|
and the check for validity of speculative motions requires
|
and the check for validity of speculative motions requires
|
construction and analysis of the function's control flow graph.
|
construction and analysis of the function's control flow graph.
|
|
|
The main entry point for this pass is schedule_insns(), called for
|
The main entry point for this pass is schedule_insns(), called for
|
each function. The work of the scheduler is organized in three
|
each function. The work of the scheduler is organized in three
|
levels: (1) function level: insns are subject to splitting,
|
levels: (1) function level: insns are subject to splitting,
|
control-flow-graph is constructed, regions are computed (after
|
control-flow-graph is constructed, regions are computed (after
|
reload, each region is of one block), (2) region level: control
|
reload, each region is of one block), (2) region level: control
|
flow graph attributes required for interblock scheduling are
|
flow graph attributes required for interblock scheduling are
|
computed (dominators, reachability, etc.), data dependences and
|
computed (dominators, reachability, etc.), data dependences and
|
priorities are computed, and (3) block level: insns in the block
|
priorities are computed, and (3) block level: insns in the block
|
are actually scheduled. */
|
are actually scheduled. */
|
|
|
#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 "toplev.h"
|
#include "toplev.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 "toplev.h"
|
#include "toplev.h"
|
#include "recog.h"
|
#include "recog.h"
|
#include "cfglayout.h"
|
#include "cfglayout.h"
|
#include "params.h"
|
#include "params.h"
|
#include "sched-int.h"
|
#include "sched-int.h"
|
#include "sel-sched.h"
|
#include "sel-sched.h"
|
#include "target.h"
|
#include "target.h"
|
#include "timevar.h"
|
#include "timevar.h"
|
#include "tree-pass.h"
|
#include "tree-pass.h"
|
#include "dbgcnt.h"
|
#include "dbgcnt.h"
|
|
|
#ifdef INSN_SCHEDULING
|
#ifdef INSN_SCHEDULING
|
|
|
/* Some accessor macros for h_i_d members only used within this file. */
|
/* Some accessor macros for h_i_d members only used within this file. */
|
#define FED_BY_SPEC_LOAD(INSN) (HID (INSN)->fed_by_spec_load)
|
#define FED_BY_SPEC_LOAD(INSN) (HID (INSN)->fed_by_spec_load)
|
#define IS_LOAD_INSN(INSN) (HID (insn)->is_load_insn)
|
#define IS_LOAD_INSN(INSN) (HID (insn)->is_load_insn)
|
|
|
/* nr_inter/spec counts interblock/speculative motion for the function. */
|
/* nr_inter/spec counts interblock/speculative motion for the function. */
|
static int nr_inter, nr_spec;
|
static int nr_inter, nr_spec;
|
|
|
static int is_cfg_nonregular (void);
|
static int is_cfg_nonregular (void);
|
|
|
/* Number of regions in the procedure. */
|
/* Number of regions in the procedure. */
|
int nr_regions = 0;
|
int nr_regions = 0;
|
|
|
/* Table of region descriptions. */
|
/* Table of region descriptions. */
|
region *rgn_table = NULL;
|
region *rgn_table = NULL;
|
|
|
/* Array of lists of regions' blocks. */
|
/* Array of lists of regions' blocks. */
|
int *rgn_bb_table = NULL;
|
int *rgn_bb_table = NULL;
|
|
|
/* Topological order of blocks in the region (if b2 is reachable from
|
/* Topological order of blocks in the region (if b2 is reachable from
|
b1, block_to_bb[b2] > block_to_bb[b1]). Note: A basic block is
|
b1, block_to_bb[b2] > block_to_bb[b1]). Note: A basic block is
|
always referred to by either block or b, while its topological
|
always referred to by either block or b, while its topological
|
order name (in the region) is referred to by bb. */
|
order name (in the region) is referred to by bb. */
|
int *block_to_bb = NULL;
|
int *block_to_bb = NULL;
|
|
|
/* The number of the region containing a block. */
|
/* The number of the region containing a block. */
|
int *containing_rgn = NULL;
|
int *containing_rgn = NULL;
|
|
|
/* ebb_head [i] - is index in rgn_bb_table of the head basic block of i'th ebb.
|
/* ebb_head [i] - is index in rgn_bb_table of the head basic block of i'th ebb.
|
Currently we can get a ebb only through splitting of currently
|
Currently we can get a ebb only through splitting of currently
|
scheduling block, therefore, we don't need ebb_head array for every region,
|
scheduling block, therefore, we don't need ebb_head array for every region,
|
hence, its sufficient to hold it for current one only. */
|
hence, its sufficient to hold it for current one only. */
|
int *ebb_head = NULL;
|
int *ebb_head = NULL;
|
|
|
/* The minimum probability of reaching a source block so that it will be
|
/* The minimum probability of reaching a source block so that it will be
|
considered for speculative scheduling. */
|
considered for speculative scheduling. */
|
static int min_spec_prob;
|
static int min_spec_prob;
|
|
|
static void find_single_block_region (bool);
|
static void find_single_block_region (bool);
|
static void find_rgns (void);
|
static void find_rgns (void);
|
static bool too_large (int, int *, int *);
|
static bool too_large (int, int *, int *);
|
|
|
/* Blocks of the current region being scheduled. */
|
/* Blocks of the current region being scheduled. */
|
int current_nr_blocks;
|
int current_nr_blocks;
|
int current_blocks;
|
int current_blocks;
|
|
|
/* A speculative motion requires checking live information on the path
|
/* A speculative motion requires checking live information on the path
|
from 'source' to 'target'. The split blocks are those to be checked.
|
from 'source' to 'target'. The split blocks are those to be checked.
|
After a speculative motion, live information should be modified in
|
After a speculative motion, live information should be modified in
|
the 'update' blocks.
|
the 'update' blocks.
|
|
|
Lists of split and update blocks for each candidate of the current
|
Lists of split and update blocks for each candidate of the current
|
target are in array bblst_table. */
|
target are in array bblst_table. */
|
static basic_block *bblst_table;
|
static basic_block *bblst_table;
|
static int bblst_size, bblst_last;
|
static int bblst_size, bblst_last;
|
|
|
/* Target info declarations.
|
/* Target info declarations.
|
|
|
The block currently being scheduled is referred to as the "target" block,
|
The block currently being scheduled is referred to as the "target" block,
|
while other blocks in the region from which insns can be moved to the
|
while other blocks in the region from which insns can be moved to the
|
target are called "source" blocks. The candidate structure holds info
|
target are called "source" blocks. The candidate structure holds info
|
about such sources: are they valid? Speculative? Etc. */
|
about such sources: are they valid? Speculative? Etc. */
|
typedef struct
|
typedef struct
|
{
|
{
|
basic_block *first_member;
|
basic_block *first_member;
|
int nr_members;
|
int nr_members;
|
}
|
}
|
bblst;
|
bblst;
|
|
|
typedef struct
|
typedef struct
|
{
|
{
|
char is_valid;
|
char is_valid;
|
char is_speculative;
|
char is_speculative;
|
int src_prob;
|
int src_prob;
|
bblst split_bbs;
|
bblst split_bbs;
|
bblst update_bbs;
|
bblst update_bbs;
|
}
|
}
|
candidate;
|
candidate;
|
|
|
static candidate *candidate_table;
|
static candidate *candidate_table;
|
#define IS_VALID(src) (candidate_table[src].is_valid)
|
#define IS_VALID(src) (candidate_table[src].is_valid)
|
#define IS_SPECULATIVE(src) (candidate_table[src].is_speculative)
|
#define IS_SPECULATIVE(src) (candidate_table[src].is_speculative)
|
#define IS_SPECULATIVE_INSN(INSN) \
|
#define IS_SPECULATIVE_INSN(INSN) \
|
(IS_SPECULATIVE (BLOCK_TO_BB (BLOCK_NUM (INSN))))
|
(IS_SPECULATIVE (BLOCK_TO_BB (BLOCK_NUM (INSN))))
|
#define SRC_PROB(src) ( candidate_table[src].src_prob )
|
#define SRC_PROB(src) ( candidate_table[src].src_prob )
|
|
|
/* The bb being currently scheduled. */
|
/* The bb being currently scheduled. */
|
int target_bb;
|
int target_bb;
|
|
|
/* List of edges. */
|
/* List of edges. */
|
typedef struct
|
typedef struct
|
{
|
{
|
edge *first_member;
|
edge *first_member;
|
int nr_members;
|
int nr_members;
|
}
|
}
|
edgelst;
|
edgelst;
|
|
|
static edge *edgelst_table;
|
static edge *edgelst_table;
|
static int edgelst_last;
|
static int edgelst_last;
|
|
|
static void extract_edgelst (sbitmap, edgelst *);
|
static void extract_edgelst (sbitmap, edgelst *);
|
|
|
/* Target info functions. */
|
/* Target info functions. */
|
static void split_edges (int, int, edgelst *);
|
static void split_edges (int, int, edgelst *);
|
static void compute_trg_info (int);
|
static void compute_trg_info (int);
|
void debug_candidate (int);
|
void debug_candidate (int);
|
void debug_candidates (int);
|
void debug_candidates (int);
|
|
|
/* Dominators array: dom[i] contains the sbitmap of dominators of
|
/* Dominators array: dom[i] contains the sbitmap of dominators of
|
bb i in the region. */
|
bb i in the region. */
|
static sbitmap *dom;
|
static sbitmap *dom;
|
|
|
/* bb 0 is the only region entry. */
|
/* bb 0 is the only region entry. */
|
#define IS_RGN_ENTRY(bb) (!bb)
|
#define IS_RGN_ENTRY(bb) (!bb)
|
|
|
/* Is bb_src dominated by bb_trg. */
|
/* Is bb_src dominated by bb_trg. */
|
#define IS_DOMINATED(bb_src, bb_trg) \
|
#define IS_DOMINATED(bb_src, bb_trg) \
|
( TEST_BIT (dom[bb_src], bb_trg) )
|
( TEST_BIT (dom[bb_src], bb_trg) )
|
|
|
/* Probability: Prob[i] is an int in [0, REG_BR_PROB_BASE] which is
|
/* Probability: Prob[i] is an int in [0, REG_BR_PROB_BASE] which is
|
the probability of bb i relative to the region entry. */
|
the probability of bb i relative to the region entry. */
|
static int *prob;
|
static int *prob;
|
|
|
/* Bit-set of edges, where bit i stands for edge i. */
|
/* Bit-set of edges, where bit i stands for edge i. */
|
typedef sbitmap edgeset;
|
typedef sbitmap edgeset;
|
|
|
/* Number of edges in the region. */
|
/* Number of edges in the region. */
|
static int rgn_nr_edges;
|
static int rgn_nr_edges;
|
|
|
/* Array of size rgn_nr_edges. */
|
/* Array of size rgn_nr_edges. */
|
static edge *rgn_edges;
|
static edge *rgn_edges;
|
|
|
/* Mapping from each edge in the graph to its number in the rgn. */
|
/* Mapping from each edge in the graph to its number in the rgn. */
|
#define EDGE_TO_BIT(edge) ((int)(size_t)(edge)->aux)
|
#define EDGE_TO_BIT(edge) ((int)(size_t)(edge)->aux)
|
#define SET_EDGE_TO_BIT(edge,nr) ((edge)->aux = (void *)(size_t)(nr))
|
#define SET_EDGE_TO_BIT(edge,nr) ((edge)->aux = (void *)(size_t)(nr))
|
|
|
/* The split edges of a source bb is different for each target
|
/* The split edges of a source bb is different for each target
|
bb. In order to compute this efficiently, the 'potential-split edges'
|
bb. In order to compute this efficiently, the 'potential-split edges'
|
are computed for each bb prior to scheduling a region. This is actually
|
are computed for each bb prior to scheduling a region. This is actually
|
the split edges of each bb relative to the region entry.
|
the split edges of each bb relative to the region entry.
|
|
|
pot_split[bb] is the set of potential split edges of bb. */
|
pot_split[bb] is the set of potential split edges of bb. */
|
static edgeset *pot_split;
|
static edgeset *pot_split;
|
|
|
/* For every bb, a set of its ancestor edges. */
|
/* For every bb, a set of its ancestor edges. */
|
static edgeset *ancestor_edges;
|
static edgeset *ancestor_edges;
|
|
|
#define INSN_PROBABILITY(INSN) (SRC_PROB (BLOCK_TO_BB (BLOCK_NUM (INSN))))
|
#define INSN_PROBABILITY(INSN) (SRC_PROB (BLOCK_TO_BB (BLOCK_NUM (INSN))))
|
|
|
/* Speculative scheduling functions. */
|
/* Speculative scheduling functions. */
|
static int check_live_1 (int, rtx);
|
static int check_live_1 (int, rtx);
|
static void update_live_1 (int, rtx);
|
static void update_live_1 (int, rtx);
|
static int is_pfree (rtx, int, int);
|
static int is_pfree (rtx, int, int);
|
static int find_conditional_protection (rtx, int);
|
static int find_conditional_protection (rtx, int);
|
static int is_conditionally_protected (rtx, int, int);
|
static int is_conditionally_protected (rtx, int, int);
|
static int is_prisky (rtx, int, int);
|
static int is_prisky (rtx, int, int);
|
static int is_exception_free (rtx, int, int);
|
static int is_exception_free (rtx, int, int);
|
|
|
static bool sets_likely_spilled (rtx);
|
static bool sets_likely_spilled (rtx);
|
static void sets_likely_spilled_1 (rtx, const_rtx, void *);
|
static void sets_likely_spilled_1 (rtx, const_rtx, void *);
|
static void add_branch_dependences (rtx, rtx);
|
static void add_branch_dependences (rtx, rtx);
|
static void compute_block_dependences (int);
|
static void compute_block_dependences (int);
|
|
|
static void schedule_region (int);
|
static void schedule_region (int);
|
static rtx concat_INSN_LIST (rtx, rtx);
|
static rtx concat_INSN_LIST (rtx, rtx);
|
static void concat_insn_mem_list (rtx, rtx, rtx *, rtx *);
|
static void concat_insn_mem_list (rtx, rtx, rtx *, rtx *);
|
static void propagate_deps (int, struct deps_desc *);
|
static void propagate_deps (int, struct deps_desc *);
|
static void free_pending_lists (void);
|
static void free_pending_lists (void);
|
|
|
/* Functions for construction of the control flow graph. */
|
/* Functions for construction of the control flow graph. */
|
|
|
/* Return 1 if control flow graph should not be constructed, 0 otherwise.
|
/* Return 1 if control flow graph should not be constructed, 0 otherwise.
|
|
|
We decide not to build the control flow graph if there is possibly more
|
We decide not to build the control flow graph if there is possibly more
|
than one entry to the function, if computed branches exist, if we
|
than one entry to the function, if computed branches exist, if we
|
have nonlocal gotos, or if we have an unreachable loop. */
|
have nonlocal gotos, or if we have an unreachable loop. */
|
|
|
static int
|
static int
|
is_cfg_nonregular (void)
|
is_cfg_nonregular (void)
|
{
|
{
|
basic_block b;
|
basic_block b;
|
rtx insn;
|
rtx insn;
|
|
|
/* If we have a label that could be the target of a nonlocal goto, then
|
/* If we have a label that could be the target of a nonlocal goto, then
|
the cfg is not well structured. */
|
the cfg is not well structured. */
|
if (nonlocal_goto_handler_labels)
|
if (nonlocal_goto_handler_labels)
|
return 1;
|
return 1;
|
|
|
/* If we have any forced labels, then the cfg is not well structured. */
|
/* If we have any forced labels, then the cfg is not well structured. */
|
if (forced_labels)
|
if (forced_labels)
|
return 1;
|
return 1;
|
|
|
/* If we have exception handlers, then we consider the cfg not well
|
/* If we have exception handlers, then we consider the cfg not well
|
structured. ?!? We should be able to handle this now that we
|
structured. ?!? We should be able to handle this now that we
|
compute an accurate cfg for EH. */
|
compute an accurate cfg for EH. */
|
if (current_function_has_exception_handlers ())
|
if (current_function_has_exception_handlers ())
|
return 1;
|
return 1;
|
|
|
/* If we have insns which refer to labels as non-jumped-to operands,
|
/* If we have insns which refer to labels as non-jumped-to operands,
|
then we consider the cfg not well structured. */
|
then we consider the cfg not well structured. */
|
FOR_EACH_BB (b)
|
FOR_EACH_BB (b)
|
FOR_BB_INSNS (b, insn)
|
FOR_BB_INSNS (b, insn)
|
{
|
{
|
rtx note, next, set, dest;
|
rtx note, next, set, dest;
|
|
|
/* If this function has a computed jump, then we consider the cfg
|
/* If this function has a computed jump, then we consider the cfg
|
not well structured. */
|
not well structured. */
|
if (JUMP_P (insn) && computed_jump_p (insn))
|
if (JUMP_P (insn) && computed_jump_p (insn))
|
return 1;
|
return 1;
|
|
|
if (!INSN_P (insn))
|
if (!INSN_P (insn))
|
continue;
|
continue;
|
|
|
note = find_reg_note (insn, REG_LABEL_OPERAND, NULL_RTX);
|
note = find_reg_note (insn, REG_LABEL_OPERAND, NULL_RTX);
|
if (note == NULL_RTX)
|
if (note == NULL_RTX)
|
continue;
|
continue;
|
|
|
/* For that label not to be seen as a referred-to label, this
|
/* For that label not to be seen as a referred-to label, this
|
must be a single-set which is feeding a jump *only*. This
|
must be a single-set which is feeding a jump *only*. This
|
could be a conditional jump with the label split off for
|
could be a conditional jump with the label split off for
|
machine-specific reasons or a casesi/tablejump. */
|
machine-specific reasons or a casesi/tablejump. */
|
next = next_nonnote_insn (insn);
|
next = next_nonnote_insn (insn);
|
if (next == NULL_RTX
|
if (next == NULL_RTX
|
|| !JUMP_P (next)
|
|| !JUMP_P (next)
|
|| (JUMP_LABEL (next) != XEXP (note, 0)
|
|| (JUMP_LABEL (next) != XEXP (note, 0)
|
&& find_reg_note (next, REG_LABEL_TARGET,
|
&& find_reg_note (next, REG_LABEL_TARGET,
|
XEXP (note, 0)) == NULL_RTX)
|
XEXP (note, 0)) == NULL_RTX)
|
|| BLOCK_FOR_INSN (insn) != BLOCK_FOR_INSN (next))
|
|| BLOCK_FOR_INSN (insn) != BLOCK_FOR_INSN (next))
|
return 1;
|
return 1;
|
|
|
set = single_set (insn);
|
set = single_set (insn);
|
if (set == NULL_RTX)
|
if (set == NULL_RTX)
|
return 1;
|
return 1;
|
|
|
dest = SET_DEST (set);
|
dest = SET_DEST (set);
|
if (!REG_P (dest) || !dead_or_set_p (next, dest))
|
if (!REG_P (dest) || !dead_or_set_p (next, dest))
|
return 1;
|
return 1;
|
}
|
}
|
|
|
/* Unreachable loops with more than one basic block are detected
|
/* Unreachable loops with more than one basic block are detected
|
during the DFS traversal in find_rgns.
|
during the DFS traversal in find_rgns.
|
|
|
Unreachable loops with a single block are detected here. This
|
Unreachable loops with a single block are detected here. This
|
test is redundant with the one in find_rgns, but it's much
|
test is redundant with the one in find_rgns, but it's much
|
cheaper to go ahead and catch the trivial case here. */
|
cheaper to go ahead and catch the trivial case here. */
|
FOR_EACH_BB (b)
|
FOR_EACH_BB (b)
|
{
|
{
|
if (EDGE_COUNT (b->preds) == 0
|
if (EDGE_COUNT (b->preds) == 0
|
|| (single_pred_p (b)
|
|| (single_pred_p (b)
|
&& single_pred (b) == b))
|
&& single_pred (b) == b))
|
return 1;
|
return 1;
|
}
|
}
|
|
|
/* All the tests passed. Consider the cfg well structured. */
|
/* All the tests passed. Consider the cfg well structured. */
|
return 0;
|
return 0;
|
}
|
}
|
|
|
/* Extract list of edges from a bitmap containing EDGE_TO_BIT bits. */
|
/* Extract list of edges from a bitmap containing EDGE_TO_BIT bits. */
|
|
|
static void
|
static void
|
extract_edgelst (sbitmap set, edgelst *el)
|
extract_edgelst (sbitmap set, edgelst *el)
|
{
|
{
|
unsigned int i = 0;
|
unsigned int i = 0;
|
sbitmap_iterator sbi;
|
sbitmap_iterator sbi;
|
|
|
/* edgelst table space is reused in each call to extract_edgelst. */
|
/* edgelst table space is reused in each call to extract_edgelst. */
|
edgelst_last = 0;
|
edgelst_last = 0;
|
|
|
el->first_member = &edgelst_table[edgelst_last];
|
el->first_member = &edgelst_table[edgelst_last];
|
el->nr_members = 0;
|
el->nr_members = 0;
|
|
|
/* Iterate over each word in the bitset. */
|
/* Iterate over each word in the bitset. */
|
EXECUTE_IF_SET_IN_SBITMAP (set, 0, i, sbi)
|
EXECUTE_IF_SET_IN_SBITMAP (set, 0, i, sbi)
|
{
|
{
|
edgelst_table[edgelst_last++] = rgn_edges[i];
|
edgelst_table[edgelst_last++] = rgn_edges[i];
|
el->nr_members++;
|
el->nr_members++;
|
}
|
}
|
}
|
}
|
|
|
/* Functions for the construction of regions. */
|
/* Functions for the construction of regions. */
|
|
|
/* Print the regions, for debugging purposes. Callable from debugger. */
|
/* Print the regions, for debugging purposes. Callable from debugger. */
|
|
|
void
|
void
|
debug_regions (void)
|
debug_regions (void)
|
{
|
{
|
int rgn, bb;
|
int rgn, bb;
|
|
|
fprintf (sched_dump, "\n;; ------------ REGIONS ----------\n\n");
|
fprintf (sched_dump, "\n;; ------------ REGIONS ----------\n\n");
|
for (rgn = 0; rgn < nr_regions; rgn++)
|
for (rgn = 0; rgn < nr_regions; rgn++)
|
{
|
{
|
fprintf (sched_dump, ";;\trgn %d nr_blocks %d:\n", rgn,
|
fprintf (sched_dump, ";;\trgn %d nr_blocks %d:\n", rgn,
|
rgn_table[rgn].rgn_nr_blocks);
|
rgn_table[rgn].rgn_nr_blocks);
|
fprintf (sched_dump, ";;\tbb/block: ");
|
fprintf (sched_dump, ";;\tbb/block: ");
|
|
|
/* We don't have ebb_head initialized yet, so we can't use
|
/* We don't have ebb_head initialized yet, so we can't use
|
BB_TO_BLOCK (). */
|
BB_TO_BLOCK (). */
|
current_blocks = RGN_BLOCKS (rgn);
|
current_blocks = RGN_BLOCKS (rgn);
|
|
|
for (bb = 0; bb < rgn_table[rgn].rgn_nr_blocks; bb++)
|
for (bb = 0; bb < rgn_table[rgn].rgn_nr_blocks; bb++)
|
fprintf (sched_dump, " %d/%d ", bb, rgn_bb_table[current_blocks + bb]);
|
fprintf (sched_dump, " %d/%d ", bb, rgn_bb_table[current_blocks + bb]);
|
|
|
fprintf (sched_dump, "\n\n");
|
fprintf (sched_dump, "\n\n");
|
}
|
}
|
}
|
}
|
|
|
/* Print the region's basic blocks. */
|
/* Print the region's basic blocks. */
|
|
|
void
|
void
|
debug_region (int rgn)
|
debug_region (int rgn)
|
{
|
{
|
int bb;
|
int bb;
|
|
|
fprintf (stderr, "\n;; ------------ REGION %d ----------\n\n", rgn);
|
fprintf (stderr, "\n;; ------------ REGION %d ----------\n\n", rgn);
|
fprintf (stderr, ";;\trgn %d nr_blocks %d:\n", rgn,
|
fprintf (stderr, ";;\trgn %d nr_blocks %d:\n", rgn,
|
rgn_table[rgn].rgn_nr_blocks);
|
rgn_table[rgn].rgn_nr_blocks);
|
fprintf (stderr, ";;\tbb/block: ");
|
fprintf (stderr, ";;\tbb/block: ");
|
|
|
/* We don't have ebb_head initialized yet, so we can't use
|
/* We don't have ebb_head initialized yet, so we can't use
|
BB_TO_BLOCK (). */
|
BB_TO_BLOCK (). */
|
current_blocks = RGN_BLOCKS (rgn);
|
current_blocks = RGN_BLOCKS (rgn);
|
|
|
for (bb = 0; bb < rgn_table[rgn].rgn_nr_blocks; bb++)
|
for (bb = 0; bb < rgn_table[rgn].rgn_nr_blocks; bb++)
|
fprintf (stderr, " %d/%d ", bb, rgn_bb_table[current_blocks + bb]);
|
fprintf (stderr, " %d/%d ", bb, rgn_bb_table[current_blocks + bb]);
|
|
|
fprintf (stderr, "\n\n");
|
fprintf (stderr, "\n\n");
|
|
|
for (bb = 0; bb < rgn_table[rgn].rgn_nr_blocks; bb++)
|
for (bb = 0; bb < rgn_table[rgn].rgn_nr_blocks; bb++)
|
{
|
{
|
debug_bb_n_slim (rgn_bb_table[current_blocks + bb]);
|
debug_bb_n_slim (rgn_bb_table[current_blocks + bb]);
|
fprintf (stderr, "\n");
|
fprintf (stderr, "\n");
|
}
|
}
|
|
|
fprintf (stderr, "\n");
|
fprintf (stderr, "\n");
|
|
|
}
|
}
|
|
|
/* True when a bb with index BB_INDEX contained in region RGN. */
|
/* True when a bb with index BB_INDEX contained in region RGN. */
|
static bool
|
static bool
|
bb_in_region_p (int bb_index, int rgn)
|
bb_in_region_p (int bb_index, int rgn)
|
{
|
{
|
int i;
|
int i;
|
|
|
for (i = 0; i < rgn_table[rgn].rgn_nr_blocks; i++)
|
for (i = 0; i < rgn_table[rgn].rgn_nr_blocks; i++)
|
if (rgn_bb_table[current_blocks + i] == bb_index)
|
if (rgn_bb_table[current_blocks + i] == bb_index)
|
return true;
|
return true;
|
|
|
return false;
|
return false;
|
}
|
}
|
|
|
/* Dump region RGN to file F using dot syntax. */
|
/* Dump region RGN to file F using dot syntax. */
|
void
|
void
|
dump_region_dot (FILE *f, int rgn)
|
dump_region_dot (FILE *f, int rgn)
|
{
|
{
|
int i;
|
int i;
|
|
|
fprintf (f, "digraph Region_%d {\n", rgn);
|
fprintf (f, "digraph Region_%d {\n", rgn);
|
|
|
/* We don't have ebb_head initialized yet, so we can't use
|
/* We don't have ebb_head initialized yet, so we can't use
|
BB_TO_BLOCK (). */
|
BB_TO_BLOCK (). */
|
current_blocks = RGN_BLOCKS (rgn);
|
current_blocks = RGN_BLOCKS (rgn);
|
|
|
for (i = 0; i < rgn_table[rgn].rgn_nr_blocks; i++)
|
for (i = 0; i < rgn_table[rgn].rgn_nr_blocks; i++)
|
{
|
{
|
edge e;
|
edge e;
|
edge_iterator ei;
|
edge_iterator ei;
|
int src_bb_num = rgn_bb_table[current_blocks + i];
|
int src_bb_num = rgn_bb_table[current_blocks + i];
|
struct basic_block_def *bb = BASIC_BLOCK (src_bb_num);
|
struct basic_block_def *bb = BASIC_BLOCK (src_bb_num);
|
|
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
if (bb_in_region_p (e->dest->index, rgn))
|
if (bb_in_region_p (e->dest->index, rgn))
|
fprintf (f, "\t%d -> %d\n", src_bb_num, e->dest->index);
|
fprintf (f, "\t%d -> %d\n", src_bb_num, e->dest->index);
|
}
|
}
|
fprintf (f, "}\n");
|
fprintf (f, "}\n");
|
}
|
}
|
|
|
/* The same, but first open a file specified by FNAME. */
|
/* The same, but first open a file specified by FNAME. */
|
void
|
void
|
dump_region_dot_file (const char *fname, int rgn)
|
dump_region_dot_file (const char *fname, int rgn)
|
{
|
{
|
FILE *f = fopen (fname, "wt");
|
FILE *f = fopen (fname, "wt");
|
dump_region_dot (f, rgn);
|
dump_region_dot (f, rgn);
|
fclose (f);
|
fclose (f);
|
}
|
}
|
|
|
/* Build a single block region for each basic block in the function.
|
/* Build a single block region for each basic block in the function.
|
This allows for using the same code for interblock and basic block
|
This allows for using the same code for interblock and basic block
|
scheduling. */
|
scheduling. */
|
|
|
static void
|
static void
|
find_single_block_region (bool ebbs_p)
|
find_single_block_region (bool ebbs_p)
|
{
|
{
|
basic_block bb, ebb_start;
|
basic_block bb, ebb_start;
|
int i = 0;
|
int i = 0;
|
|
|
nr_regions = 0;
|
nr_regions = 0;
|
|
|
if (ebbs_p) {
|
if (ebbs_p) {
|
int probability_cutoff;
|
int probability_cutoff;
|
if (profile_info && flag_branch_probabilities)
|
if (profile_info && flag_branch_probabilities)
|
probability_cutoff = PARAM_VALUE (TRACER_MIN_BRANCH_PROBABILITY_FEEDBACK);
|
probability_cutoff = PARAM_VALUE (TRACER_MIN_BRANCH_PROBABILITY_FEEDBACK);
|
else
|
else
|
probability_cutoff = PARAM_VALUE (TRACER_MIN_BRANCH_PROBABILITY);
|
probability_cutoff = PARAM_VALUE (TRACER_MIN_BRANCH_PROBABILITY);
|
probability_cutoff = REG_BR_PROB_BASE / 100 * probability_cutoff;
|
probability_cutoff = REG_BR_PROB_BASE / 100 * probability_cutoff;
|
|
|
FOR_EACH_BB (ebb_start)
|
FOR_EACH_BB (ebb_start)
|
{
|
{
|
RGN_NR_BLOCKS (nr_regions) = 0;
|
RGN_NR_BLOCKS (nr_regions) = 0;
|
RGN_BLOCKS (nr_regions) = i;
|
RGN_BLOCKS (nr_regions) = i;
|
RGN_DONT_CALC_DEPS (nr_regions) = 0;
|
RGN_DONT_CALC_DEPS (nr_regions) = 0;
|
RGN_HAS_REAL_EBB (nr_regions) = 0;
|
RGN_HAS_REAL_EBB (nr_regions) = 0;
|
|
|
for (bb = ebb_start; ; bb = bb->next_bb)
|
for (bb = ebb_start; ; bb = bb->next_bb)
|
{
|
{
|
edge e;
|
edge e;
|
edge_iterator ei;
|
edge_iterator ei;
|
|
|
rgn_bb_table[i] = bb->index;
|
rgn_bb_table[i] = bb->index;
|
RGN_NR_BLOCKS (nr_regions)++;
|
RGN_NR_BLOCKS (nr_regions)++;
|
CONTAINING_RGN (bb->index) = nr_regions;
|
CONTAINING_RGN (bb->index) = nr_regions;
|
BLOCK_TO_BB (bb->index) = i - RGN_BLOCKS (nr_regions);
|
BLOCK_TO_BB (bb->index) = i - RGN_BLOCKS (nr_regions);
|
i++;
|
i++;
|
|
|
if (bb->next_bb == EXIT_BLOCK_PTR
|
if (bb->next_bb == EXIT_BLOCK_PTR
|
|| LABEL_P (BB_HEAD (bb->next_bb)))
|
|| LABEL_P (BB_HEAD (bb->next_bb)))
|
break;
|
break;
|
|
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
if ((e->flags & EDGE_FALLTHRU) != 0)
|
if ((e->flags & EDGE_FALLTHRU) != 0)
|
break;
|
break;
|
if (! e)
|
if (! e)
|
break;
|
break;
|
if (e->probability <= probability_cutoff)
|
if (e->probability <= probability_cutoff)
|
break;
|
break;
|
}
|
}
|
|
|
ebb_start = bb;
|
ebb_start = bb;
|
nr_regions++;
|
nr_regions++;
|
}
|
}
|
}
|
}
|
else
|
else
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
{
|
{
|
rgn_bb_table[nr_regions] = bb->index;
|
rgn_bb_table[nr_regions] = bb->index;
|
RGN_NR_BLOCKS (nr_regions) = 1;
|
RGN_NR_BLOCKS (nr_regions) = 1;
|
RGN_BLOCKS (nr_regions) = nr_regions;
|
RGN_BLOCKS (nr_regions) = nr_regions;
|
RGN_DONT_CALC_DEPS (nr_regions) = 0;
|
RGN_DONT_CALC_DEPS (nr_regions) = 0;
|
RGN_HAS_REAL_EBB (nr_regions) = 0;
|
RGN_HAS_REAL_EBB (nr_regions) = 0;
|
|
|
CONTAINING_RGN (bb->index) = nr_regions;
|
CONTAINING_RGN (bb->index) = nr_regions;
|
BLOCK_TO_BB (bb->index) = 0;
|
BLOCK_TO_BB (bb->index) = 0;
|
nr_regions++;
|
nr_regions++;
|
}
|
}
|
}
|
}
|
|
|
/* Estimate number of the insns in the BB. */
|
/* Estimate number of the insns in the BB. */
|
static int
|
static int
|
rgn_estimate_number_of_insns (basic_block bb)
|
rgn_estimate_number_of_insns (basic_block bb)
|
{
|
{
|
int count;
|
int count;
|
|
|
count = INSN_LUID (BB_END (bb)) - INSN_LUID (BB_HEAD (bb));
|
count = INSN_LUID (BB_END (bb)) - INSN_LUID (BB_HEAD (bb));
|
|
|
if (MAY_HAVE_DEBUG_INSNS)
|
if (MAY_HAVE_DEBUG_INSNS)
|
{
|
{
|
rtx insn;
|
rtx insn;
|
|
|
FOR_BB_INSNS (bb, insn)
|
FOR_BB_INSNS (bb, insn)
|
if (DEBUG_INSN_P (insn))
|
if (DEBUG_INSN_P (insn))
|
count--;
|
count--;
|
}
|
}
|
|
|
return count;
|
return count;
|
}
|
}
|
|
|
/* Update number of blocks and the estimate for number of insns
|
/* Update number of blocks and the estimate for number of insns
|
in the region. Return true if the region is "too large" for interblock
|
in the region. Return true if the region is "too large" for interblock
|
scheduling (compile time considerations). */
|
scheduling (compile time considerations). */
|
|
|
static bool
|
static bool
|
too_large (int block, int *num_bbs, int *num_insns)
|
too_large (int block, int *num_bbs, int *num_insns)
|
{
|
{
|
(*num_bbs)++;
|
(*num_bbs)++;
|
(*num_insns) += (common_sched_info->estimate_number_of_insns
|
(*num_insns) += (common_sched_info->estimate_number_of_insns
|
(BASIC_BLOCK (block)));
|
(BASIC_BLOCK (block)));
|
|
|
return ((*num_bbs > PARAM_VALUE (PARAM_MAX_SCHED_REGION_BLOCKS))
|
return ((*num_bbs > PARAM_VALUE (PARAM_MAX_SCHED_REGION_BLOCKS))
|
|| (*num_insns > PARAM_VALUE (PARAM_MAX_SCHED_REGION_INSNS)));
|
|| (*num_insns > PARAM_VALUE (PARAM_MAX_SCHED_REGION_INSNS)));
|
}
|
}
|
|
|
/* Update_loop_relations(blk, hdr): Check if the loop headed by max_hdr[blk]
|
/* Update_loop_relations(blk, hdr): Check if the loop headed by max_hdr[blk]
|
is still an inner loop. Put in max_hdr[blk] the header of the most inner
|
is still an inner loop. Put in max_hdr[blk] the header of the most inner
|
loop containing blk. */
|
loop containing blk. */
|
#define UPDATE_LOOP_RELATIONS(blk, hdr) \
|
#define UPDATE_LOOP_RELATIONS(blk, hdr) \
|
{ \
|
{ \
|
if (max_hdr[blk] == -1) \
|
if (max_hdr[blk] == -1) \
|
max_hdr[blk] = hdr; \
|
max_hdr[blk] = hdr; \
|
else if (dfs_nr[max_hdr[blk]] > dfs_nr[hdr]) \
|
else if (dfs_nr[max_hdr[blk]] > dfs_nr[hdr]) \
|
RESET_BIT (inner, hdr); \
|
RESET_BIT (inner, hdr); \
|
else if (dfs_nr[max_hdr[blk]] < dfs_nr[hdr]) \
|
else if (dfs_nr[max_hdr[blk]] < dfs_nr[hdr]) \
|
{ \
|
{ \
|
RESET_BIT (inner,max_hdr[blk]); \
|
RESET_BIT (inner,max_hdr[blk]); \
|
max_hdr[blk] = hdr; \
|
max_hdr[blk] = hdr; \
|
} \
|
} \
|
}
|
}
|
|
|
/* Find regions for interblock scheduling.
|
/* Find regions for interblock scheduling.
|
|
|
A region for scheduling can be:
|
A region for scheduling can be:
|
|
|
* A loop-free procedure, or
|
* A loop-free procedure, or
|
|
|
* A reducible inner loop, or
|
* A reducible inner loop, or
|
|
|
* A basic block not contained in any other region.
|
* A basic block not contained in any other region.
|
|
|
?!? In theory we could build other regions based on extended basic
|
?!? In theory we could build other regions based on extended basic
|
blocks or reverse extended basic blocks. Is it worth the trouble?
|
blocks or reverse extended basic blocks. Is it worth the trouble?
|
|
|
Loop blocks that form a region are put into the region's block list
|
Loop blocks that form a region are put into the region's block list
|
in topological order.
|
in topological order.
|
|
|
This procedure stores its results into the following global (ick) variables
|
This procedure stores its results into the following global (ick) variables
|
|
|
* rgn_nr
|
* rgn_nr
|
* rgn_table
|
* rgn_table
|
* rgn_bb_table
|
* rgn_bb_table
|
* block_to_bb
|
* block_to_bb
|
* containing region
|
* containing region
|
|
|
We use dominator relationships to avoid making regions out of non-reducible
|
We use dominator relationships to avoid making regions out of non-reducible
|
loops.
|
loops.
|
|
|
This procedure needs to be converted to work on pred/succ lists instead
|
This procedure needs to be converted to work on pred/succ lists instead
|
of edge tables. That would simplify it somewhat. */
|
of edge tables. That would simplify it somewhat. */
|
|
|
static void
|
static void
|
haifa_find_rgns (void)
|
haifa_find_rgns (void)
|
{
|
{
|
int *max_hdr, *dfs_nr, *degree;
|
int *max_hdr, *dfs_nr, *degree;
|
char no_loops = 1;
|
char no_loops = 1;
|
int node, child, loop_head, i, head, tail;
|
int node, child, loop_head, i, head, tail;
|
int count = 0, sp, idx = 0;
|
int count = 0, sp, idx = 0;
|
edge_iterator current_edge;
|
edge_iterator current_edge;
|
edge_iterator *stack;
|
edge_iterator *stack;
|
int num_bbs, num_insns, unreachable;
|
int num_bbs, num_insns, unreachable;
|
int too_large_failure;
|
int too_large_failure;
|
basic_block bb;
|
basic_block bb;
|
|
|
/* Note if a block is a natural loop header. */
|
/* Note if a block is a natural loop header. */
|
sbitmap header;
|
sbitmap header;
|
|
|
/* Note if a block is a natural inner loop header. */
|
/* Note if a block is a natural inner loop header. */
|
sbitmap inner;
|
sbitmap inner;
|
|
|
/* Note if a block is in the block queue. */
|
/* Note if a block is in the block queue. */
|
sbitmap in_queue;
|
sbitmap in_queue;
|
|
|
/* Note if a block is in the block queue. */
|
/* Note if a block is in the block queue. */
|
sbitmap in_stack;
|
sbitmap in_stack;
|
|
|
/* Perform a DFS traversal of the cfg. Identify loop headers, inner loops
|
/* Perform a DFS traversal of the cfg. Identify loop headers, inner loops
|
and a mapping from block to its loop header (if the block is contained
|
and a mapping from block to its loop header (if the block is contained
|
in a loop, else -1).
|
in a loop, else -1).
|
|
|
Store results in HEADER, INNER, and MAX_HDR respectively, these will
|
Store results in HEADER, INNER, and MAX_HDR respectively, these will
|
be used as inputs to the second traversal.
|
be used as inputs to the second traversal.
|
|
|
STACK, SP and DFS_NR are only used during the first traversal. */
|
STACK, SP and DFS_NR are only used during the first traversal. */
|
|
|
/* Allocate and initialize variables for the first traversal. */
|
/* Allocate and initialize variables for the first traversal. */
|
max_hdr = XNEWVEC (int, last_basic_block);
|
max_hdr = XNEWVEC (int, last_basic_block);
|
dfs_nr = XCNEWVEC (int, last_basic_block);
|
dfs_nr = XCNEWVEC (int, last_basic_block);
|
stack = XNEWVEC (edge_iterator, n_edges);
|
stack = XNEWVEC (edge_iterator, n_edges);
|
|
|
inner = sbitmap_alloc (last_basic_block);
|
inner = sbitmap_alloc (last_basic_block);
|
sbitmap_ones (inner);
|
sbitmap_ones (inner);
|
|
|
header = sbitmap_alloc (last_basic_block);
|
header = sbitmap_alloc (last_basic_block);
|
sbitmap_zero (header);
|
sbitmap_zero (header);
|
|
|
in_queue = sbitmap_alloc (last_basic_block);
|
in_queue = sbitmap_alloc (last_basic_block);
|
sbitmap_zero (in_queue);
|
sbitmap_zero (in_queue);
|
|
|
in_stack = sbitmap_alloc (last_basic_block);
|
in_stack = sbitmap_alloc (last_basic_block);
|
sbitmap_zero (in_stack);
|
sbitmap_zero (in_stack);
|
|
|
for (i = 0; i < last_basic_block; i++)
|
for (i = 0; i < last_basic_block; i++)
|
max_hdr[i] = -1;
|
max_hdr[i] = -1;
|
|
|
#define EDGE_PASSED(E) (ei_end_p ((E)) || ei_edge ((E))->aux)
|
#define EDGE_PASSED(E) (ei_end_p ((E)) || ei_edge ((E))->aux)
|
#define SET_EDGE_PASSED(E) (ei_edge ((E))->aux = ei_edge ((E)))
|
#define SET_EDGE_PASSED(E) (ei_edge ((E))->aux = ei_edge ((E)))
|
|
|
/* DFS traversal to find inner loops in the cfg. */
|
/* DFS traversal to find inner loops in the cfg. */
|
|
|
current_edge = ei_start (single_succ (ENTRY_BLOCK_PTR)->succs);
|
current_edge = ei_start (single_succ (ENTRY_BLOCK_PTR)->succs);
|
sp = -1;
|
sp = -1;
|
|
|
while (1)
|
while (1)
|
{
|
{
|
if (EDGE_PASSED (current_edge))
|
if (EDGE_PASSED (current_edge))
|
{
|
{
|
/* We have reached a leaf node or a node that was already
|
/* We have reached a leaf node or a node that was already
|
processed. Pop edges off the stack until we find
|
processed. Pop edges off the stack until we find
|
an edge that has not yet been processed. */
|
an edge that has not yet been processed. */
|
while (sp >= 0 && EDGE_PASSED (current_edge))
|
while (sp >= 0 && EDGE_PASSED (current_edge))
|
{
|
{
|
/* Pop entry off the stack. */
|
/* Pop entry off the stack. */
|
current_edge = stack[sp--];
|
current_edge = stack[sp--];
|
node = ei_edge (current_edge)->src->index;
|
node = ei_edge (current_edge)->src->index;
|
gcc_assert (node != ENTRY_BLOCK);
|
gcc_assert (node != ENTRY_BLOCK);
|
child = ei_edge (current_edge)->dest->index;
|
child = ei_edge (current_edge)->dest->index;
|
gcc_assert (child != EXIT_BLOCK);
|
gcc_assert (child != EXIT_BLOCK);
|
RESET_BIT (in_stack, child);
|
RESET_BIT (in_stack, child);
|
if (max_hdr[child] >= 0 && TEST_BIT (in_stack, max_hdr[child]))
|
if (max_hdr[child] >= 0 && TEST_BIT (in_stack, max_hdr[child]))
|
UPDATE_LOOP_RELATIONS (node, max_hdr[child]);
|
UPDATE_LOOP_RELATIONS (node, max_hdr[child]);
|
ei_next (¤t_edge);
|
ei_next (¤t_edge);
|
}
|
}
|
|
|
/* See if have finished the DFS tree traversal. */
|
/* See if have finished the DFS tree traversal. */
|
if (sp < 0 && EDGE_PASSED (current_edge))
|
if (sp < 0 && EDGE_PASSED (current_edge))
|
break;
|
break;
|
|
|
/* Nope, continue the traversal with the popped node. */
|
/* Nope, continue the traversal with the popped node. */
|
continue;
|
continue;
|
}
|
}
|
|
|
/* Process a node. */
|
/* Process a node. */
|
node = ei_edge (current_edge)->src->index;
|
node = ei_edge (current_edge)->src->index;
|
gcc_assert (node != ENTRY_BLOCK);
|
gcc_assert (node != ENTRY_BLOCK);
|
SET_BIT (in_stack, node);
|
SET_BIT (in_stack, node);
|
dfs_nr[node] = ++count;
|
dfs_nr[node] = ++count;
|
|
|
/* We don't traverse to the exit block. */
|
/* We don't traverse to the exit block. */
|
child = ei_edge (current_edge)->dest->index;
|
child = ei_edge (current_edge)->dest->index;
|
if (child == EXIT_BLOCK)
|
if (child == EXIT_BLOCK)
|
{
|
{
|
SET_EDGE_PASSED (current_edge);
|
SET_EDGE_PASSED (current_edge);
|
ei_next (¤t_edge);
|
ei_next (¤t_edge);
|
continue;
|
continue;
|
}
|
}
|
|
|
/* If the successor is in the stack, then we've found a loop.
|
/* If the successor is in the stack, then we've found a loop.
|
Mark the loop, if it is not a natural loop, then it will
|
Mark the loop, if it is not a natural loop, then it will
|
be rejected during the second traversal. */
|
be rejected during the second traversal. */
|
if (TEST_BIT (in_stack, child))
|
if (TEST_BIT (in_stack, child))
|
{
|
{
|
no_loops = 0;
|
no_loops = 0;
|
SET_BIT (header, child);
|
SET_BIT (header, child);
|
UPDATE_LOOP_RELATIONS (node, child);
|
UPDATE_LOOP_RELATIONS (node, child);
|
SET_EDGE_PASSED (current_edge);
|
SET_EDGE_PASSED (current_edge);
|
ei_next (¤t_edge);
|
ei_next (¤t_edge);
|
continue;
|
continue;
|
}
|
}
|
|
|
/* If the child was already visited, then there is no need to visit
|
/* If the child was already visited, then there is no need to visit
|
it again. Just update the loop relationships and restart
|
it again. Just update the loop relationships and restart
|
with a new edge. */
|
with a new edge. */
|
if (dfs_nr[child])
|
if (dfs_nr[child])
|
{
|
{
|
if (max_hdr[child] >= 0 && TEST_BIT (in_stack, max_hdr[child]))
|
if (max_hdr[child] >= 0 && TEST_BIT (in_stack, max_hdr[child]))
|
UPDATE_LOOP_RELATIONS (node, max_hdr[child]);
|
UPDATE_LOOP_RELATIONS (node, max_hdr[child]);
|
SET_EDGE_PASSED (current_edge);
|
SET_EDGE_PASSED (current_edge);
|
ei_next (¤t_edge);
|
ei_next (¤t_edge);
|
continue;
|
continue;
|
}
|
}
|
|
|
/* Push an entry on the stack and continue DFS traversal. */
|
/* Push an entry on the stack and continue DFS traversal. */
|
stack[++sp] = current_edge;
|
stack[++sp] = current_edge;
|
SET_EDGE_PASSED (current_edge);
|
SET_EDGE_PASSED (current_edge);
|
current_edge = ei_start (ei_edge (current_edge)->dest->succs);
|
current_edge = ei_start (ei_edge (current_edge)->dest->succs);
|
}
|
}
|
|
|
/* Reset ->aux field used by EDGE_PASSED. */
|
/* Reset ->aux field used by EDGE_PASSED. */
|
FOR_ALL_BB (bb)
|
FOR_ALL_BB (bb)
|
{
|
{
|
edge_iterator ei;
|
edge_iterator ei;
|
edge e;
|
edge e;
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
e->aux = NULL;
|
e->aux = NULL;
|
}
|
}
|
|
|
|
|
/* Another check for unreachable blocks. The earlier test in
|
/* Another check for unreachable blocks. The earlier test in
|
is_cfg_nonregular only finds unreachable blocks that do not
|
is_cfg_nonregular only finds unreachable blocks that do not
|
form a loop.
|
form a loop.
|
|
|
The DFS traversal will mark every block that is reachable from
|
The DFS traversal will mark every block that is reachable from
|
the entry node by placing a nonzero value in dfs_nr. Thus if
|
the entry node by placing a nonzero value in dfs_nr. Thus if
|
dfs_nr is zero for any block, then it must be unreachable. */
|
dfs_nr is zero for any block, then it must be unreachable. */
|
unreachable = 0;
|
unreachable = 0;
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
if (dfs_nr[bb->index] == 0)
|
if (dfs_nr[bb->index] == 0)
|
{
|
{
|
unreachable = 1;
|
unreachable = 1;
|
break;
|
break;
|
}
|
}
|
|
|
/* Gross. To avoid wasting memory, the second pass uses the dfs_nr array
|
/* Gross. To avoid wasting memory, the second pass uses the dfs_nr array
|
to hold degree counts. */
|
to hold degree counts. */
|
degree = dfs_nr;
|
degree = dfs_nr;
|
|
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
degree[bb->index] = EDGE_COUNT (bb->preds);
|
degree[bb->index] = EDGE_COUNT (bb->preds);
|
|
|
/* Do not perform region scheduling if there are any unreachable
|
/* Do not perform region scheduling if there are any unreachable
|
blocks. */
|
blocks. */
|
if (!unreachable)
|
if (!unreachable)
|
{
|
{
|
int *queue, *degree1 = NULL;
|
int *queue, *degree1 = NULL;
|
/* We use EXTENDED_RGN_HEADER as an addition to HEADER and put
|
/* We use EXTENDED_RGN_HEADER as an addition to HEADER and put
|
there basic blocks, which are forced to be region heads.
|
there basic blocks, which are forced to be region heads.
|
This is done to try to assemble few smaller regions
|
This is done to try to assemble few smaller regions
|
from a too_large region. */
|
from a too_large region. */
|
sbitmap extended_rgn_header = NULL;
|
sbitmap extended_rgn_header = NULL;
|
bool extend_regions_p;
|
bool extend_regions_p;
|
|
|
if (no_loops)
|
if (no_loops)
|
SET_BIT (header, 0);
|
SET_BIT (header, 0);
|
|
|
/* Second traversal:find reducible inner loops and topologically sort
|
/* Second traversal:find reducible inner loops and topologically sort
|
block of each region. */
|
block of each region. */
|
|
|
queue = XNEWVEC (int, n_basic_blocks);
|
queue = XNEWVEC (int, n_basic_blocks);
|
|
|
extend_regions_p = PARAM_VALUE (PARAM_MAX_SCHED_EXTEND_REGIONS_ITERS) > 0;
|
extend_regions_p = PARAM_VALUE (PARAM_MAX_SCHED_EXTEND_REGIONS_ITERS) > 0;
|
if (extend_regions_p)
|
if (extend_regions_p)
|
{
|
{
|
degree1 = XNEWVEC (int, last_basic_block);
|
degree1 = XNEWVEC (int, last_basic_block);
|
extended_rgn_header = sbitmap_alloc (last_basic_block);
|
extended_rgn_header = sbitmap_alloc (last_basic_block);
|
sbitmap_zero (extended_rgn_header);
|
sbitmap_zero (extended_rgn_header);
|
}
|
}
|
|
|
/* Find blocks which are inner loop headers. We still have non-reducible
|
/* Find blocks which are inner loop headers. We still have non-reducible
|
loops to consider at this point. */
|
loops to consider at this point. */
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
{
|
{
|
if (TEST_BIT (header, bb->index) && TEST_BIT (inner, bb->index))
|
if (TEST_BIT (header, bb->index) && TEST_BIT (inner, bb->index))
|
{
|
{
|
edge e;
|
edge e;
|
edge_iterator ei;
|
edge_iterator ei;
|
basic_block jbb;
|
basic_block jbb;
|
|
|
/* Now check that the loop is reducible. We do this separate
|
/* Now check that the loop is reducible. We do this separate
|
from finding inner loops so that we do not find a reducible
|
from finding inner loops so that we do not find a reducible
|
loop which contains an inner non-reducible loop.
|
loop which contains an inner non-reducible loop.
|
|
|
A simple way to find reducible/natural loops is to verify
|
A simple way to find reducible/natural loops is to verify
|
that each block in the loop is dominated by the loop
|
that each block in the loop is dominated by the loop
|
header.
|
header.
|
|
|
If there exists a block that is not dominated by the loop
|
If there exists a block that is not dominated by the loop
|
header, then the block is reachable from outside the loop
|
header, then the block is reachable from outside the loop
|
and thus the loop is not a natural loop. */
|
and thus the loop is not a natural loop. */
|
FOR_EACH_BB (jbb)
|
FOR_EACH_BB (jbb)
|
{
|
{
|
/* First identify blocks in the loop, except for the loop
|
/* First identify blocks in the loop, except for the loop
|
entry block. */
|
entry block. */
|
if (bb->index == max_hdr[jbb->index] && bb != jbb)
|
if (bb->index == max_hdr[jbb->index] && bb != jbb)
|
{
|
{
|
/* Now verify that the block is dominated by the loop
|
/* Now verify that the block is dominated by the loop
|
header. */
|
header. */
|
if (!dominated_by_p (CDI_DOMINATORS, jbb, bb))
|
if (!dominated_by_p (CDI_DOMINATORS, jbb, bb))
|
break;
|
break;
|
}
|
}
|
}
|
}
|
|
|
/* If we exited the loop early, then I is the header of
|
/* If we exited the loop early, then I is the header of
|
a non-reducible loop and we should quit processing it
|
a non-reducible loop and we should quit processing it
|
now. */
|
now. */
|
if (jbb != EXIT_BLOCK_PTR)
|
if (jbb != EXIT_BLOCK_PTR)
|
continue;
|
continue;
|
|
|
/* I is a header of an inner loop, or block 0 in a subroutine
|
/* I is a header of an inner loop, or block 0 in a subroutine
|
with no loops at all. */
|
with no loops at all. */
|
head = tail = -1;
|
head = tail = -1;
|
too_large_failure = 0;
|
too_large_failure = 0;
|
loop_head = max_hdr[bb->index];
|
loop_head = max_hdr[bb->index];
|
|
|
if (extend_regions_p)
|
if (extend_regions_p)
|
/* We save degree in case when we meet a too_large region
|
/* We save degree in case when we meet a too_large region
|
and cancel it. We need a correct degree later when
|
and cancel it. We need a correct degree later when
|
calling extend_rgns. */
|
calling extend_rgns. */
|
memcpy (degree1, degree, last_basic_block * sizeof (int));
|
memcpy (degree1, degree, last_basic_block * sizeof (int));
|
|
|
/* Decrease degree of all I's successors for topological
|
/* Decrease degree of all I's successors for topological
|
ordering. */
|
ordering. */
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
if (e->dest != EXIT_BLOCK_PTR)
|
if (e->dest != EXIT_BLOCK_PTR)
|
--degree[e->dest->index];
|
--degree[e->dest->index];
|
|
|
/* Estimate # insns, and count # blocks in the region. */
|
/* Estimate # insns, and count # blocks in the region. */
|
num_bbs = 1;
|
num_bbs = 1;
|
num_insns = common_sched_info->estimate_number_of_insns (bb);
|
num_insns = common_sched_info->estimate_number_of_insns (bb);
|
|
|
/* Find all loop latches (blocks with back edges to the loop
|
/* Find all loop latches (blocks with back edges to the loop
|
header) or all the leaf blocks in the cfg has no loops.
|
header) or all the leaf blocks in the cfg has no loops.
|
|
|
Place those blocks into the queue. */
|
Place those blocks into the queue. */
|
if (no_loops)
|
if (no_loops)
|
{
|
{
|
FOR_EACH_BB (jbb)
|
FOR_EACH_BB (jbb)
|
/* Leaf nodes have only a single successor which must
|
/* Leaf nodes have only a single successor which must
|
be EXIT_BLOCK. */
|
be EXIT_BLOCK. */
|
if (single_succ_p (jbb)
|
if (single_succ_p (jbb)
|
&& single_succ (jbb) == EXIT_BLOCK_PTR)
|
&& single_succ (jbb) == EXIT_BLOCK_PTR)
|
{
|
{
|
queue[++tail] = jbb->index;
|
queue[++tail] = jbb->index;
|
SET_BIT (in_queue, jbb->index);
|
SET_BIT (in_queue, jbb->index);
|
|
|
if (too_large (jbb->index, &num_bbs, &num_insns))
|
if (too_large (jbb->index, &num_bbs, &num_insns))
|
{
|
{
|
too_large_failure = 1;
|
too_large_failure = 1;
|
break;
|
break;
|
}
|
}
|
}
|
}
|
}
|
}
|
else
|
else
|
{
|
{
|
edge e;
|
edge e;
|
|
|
FOR_EACH_EDGE (e, ei, bb->preds)
|
FOR_EACH_EDGE (e, ei, bb->preds)
|
{
|
{
|
if (e->src == ENTRY_BLOCK_PTR)
|
if (e->src == ENTRY_BLOCK_PTR)
|
continue;
|
continue;
|
|
|
node = e->src->index;
|
node = e->src->index;
|
|
|
if (max_hdr[node] == loop_head && node != bb->index)
|
if (max_hdr[node] == loop_head && node != bb->index)
|
{
|
{
|
/* This is a loop latch. */
|
/* This is a loop latch. */
|
queue[++tail] = node;
|
queue[++tail] = node;
|
SET_BIT (in_queue, node);
|
SET_BIT (in_queue, node);
|
|
|
if (too_large (node, &num_bbs, &num_insns))
|
if (too_large (node, &num_bbs, &num_insns))
|
{
|
{
|
too_large_failure = 1;
|
too_large_failure = 1;
|
break;
|
break;
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
/* Now add all the blocks in the loop to the queue.
|
/* Now add all the blocks in the loop to the queue.
|
|
|
We know the loop is a natural loop; however the algorithm
|
We know the loop is a natural loop; however the algorithm
|
above will not always mark certain blocks as being in the
|
above will not always mark certain blocks as being in the
|
loop. Consider:
|
loop. Consider:
|
node children
|
node children
|
a b,c
|
a b,c
|
b c
|
b c
|
c a,d
|
c a,d
|
d b
|
d b
|
|
|
The algorithm in the DFS traversal may not mark B & D as part
|
The algorithm in the DFS traversal may not mark B & D as part
|
of the loop (i.e. they will not have max_hdr set to A).
|
of the loop (i.e. they will not have max_hdr set to A).
|
|
|
We know they can not be loop latches (else they would have
|
We know they can not be loop latches (else they would have
|
had max_hdr set since they'd have a backedge to a dominator
|
had max_hdr set since they'd have a backedge to a dominator
|
block). So we don't need them on the initial queue.
|
block). So we don't need them on the initial queue.
|
|
|
We know they are part of the loop because they are dominated
|
We know they are part of the loop because they are dominated
|
by the loop header and can be reached by a backwards walk of
|
by the loop header and can be reached by a backwards walk of
|
the edges starting with nodes on the initial queue.
|
the edges starting with nodes on the initial queue.
|
|
|
It is safe and desirable to include those nodes in the
|
It is safe and desirable to include those nodes in the
|
loop/scheduling region. To do so we would need to decrease
|
loop/scheduling region. To do so we would need to decrease
|
the degree of a node if it is the target of a backedge
|
the degree of a node if it is the target of a backedge
|
within the loop itself as the node is placed in the queue.
|
within the loop itself as the node is placed in the queue.
|
|
|
We do not do this because I'm not sure that the actual
|
We do not do this because I'm not sure that the actual
|
scheduling code will properly handle this case. ?!? */
|
scheduling code will properly handle this case. ?!? */
|
|
|
while (head < tail && !too_large_failure)
|
while (head < tail && !too_large_failure)
|
{
|
{
|
edge e;
|
edge e;
|
child = queue[++head];
|
child = queue[++head];
|
|
|
FOR_EACH_EDGE (e, ei, BASIC_BLOCK (child)->preds)
|
FOR_EACH_EDGE (e, ei, BASIC_BLOCK (child)->preds)
|
{
|
{
|
node = e->src->index;
|
node = e->src->index;
|
|
|
/* See discussion above about nodes not marked as in
|
/* See discussion above about nodes not marked as in
|
this loop during the initial DFS traversal. */
|
this loop during the initial DFS traversal. */
|
if (e->src == ENTRY_BLOCK_PTR
|
if (e->src == ENTRY_BLOCK_PTR
|
|| max_hdr[node] != loop_head)
|
|| max_hdr[node] != loop_head)
|
{
|
{
|
tail = -1;
|
tail = -1;
|
break;
|
break;
|
}
|
}
|
else if (!TEST_BIT (in_queue, node) && node != bb->index)
|
else if (!TEST_BIT (in_queue, node) && node != bb->index)
|
{
|
{
|
queue[++tail] = node;
|
queue[++tail] = node;
|
SET_BIT (in_queue, node);
|
SET_BIT (in_queue, node);
|
|
|
if (too_large (node, &num_bbs, &num_insns))
|
if (too_large (node, &num_bbs, &num_insns))
|
{
|
{
|
too_large_failure = 1;
|
too_large_failure = 1;
|
break;
|
break;
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
if (tail >= 0 && !too_large_failure)
|
if (tail >= 0 && !too_large_failure)
|
{
|
{
|
/* Place the loop header into list of region blocks. */
|
/* Place the loop header into list of region blocks. */
|
degree[bb->index] = -1;
|
degree[bb->index] = -1;
|
rgn_bb_table[idx] = bb->index;
|
rgn_bb_table[idx] = bb->index;
|
RGN_NR_BLOCKS (nr_regions) = num_bbs;
|
RGN_NR_BLOCKS (nr_regions) = num_bbs;
|
RGN_BLOCKS (nr_regions) = idx++;
|
RGN_BLOCKS (nr_regions) = idx++;
|
RGN_DONT_CALC_DEPS (nr_regions) = 0;
|
RGN_DONT_CALC_DEPS (nr_regions) = 0;
|
RGN_HAS_REAL_EBB (nr_regions) = 0;
|
RGN_HAS_REAL_EBB (nr_regions) = 0;
|
CONTAINING_RGN (bb->index) = nr_regions;
|
CONTAINING_RGN (bb->index) = nr_regions;
|
BLOCK_TO_BB (bb->index) = count = 0;
|
BLOCK_TO_BB (bb->index) = count = 0;
|
|
|
/* Remove blocks from queue[] when their in degree
|
/* Remove blocks from queue[] when their in degree
|
becomes zero. Repeat until no blocks are left on the
|
becomes zero. Repeat until no blocks are left on the
|
list. This produces a topological list of blocks in
|
list. This produces a topological list of blocks in
|
the region. */
|
the region. */
|
while (tail >= 0)
|
while (tail >= 0)
|
{
|
{
|
if (head < 0)
|
if (head < 0)
|
head = tail;
|
head = tail;
|
child = queue[head];
|
child = queue[head];
|
if (degree[child] == 0)
|
if (degree[child] == 0)
|
{
|
{
|
edge e;
|
edge e;
|
|
|
degree[child] = -1;
|
degree[child] = -1;
|
rgn_bb_table[idx++] = child;
|
rgn_bb_table[idx++] = child;
|
BLOCK_TO_BB (child) = ++count;
|
BLOCK_TO_BB (child) = ++count;
|
CONTAINING_RGN (child) = nr_regions;
|
CONTAINING_RGN (child) = nr_regions;
|
queue[head] = queue[tail--];
|
queue[head] = queue[tail--];
|
|
|
FOR_EACH_EDGE (e, ei, BASIC_BLOCK (child)->succs)
|
FOR_EACH_EDGE (e, ei, BASIC_BLOCK (child)->succs)
|
if (e->dest != EXIT_BLOCK_PTR)
|
if (e->dest != EXIT_BLOCK_PTR)
|
--degree[e->dest->index];
|
--degree[e->dest->index];
|
}
|
}
|
else
|
else
|
--head;
|
--head;
|
}
|
}
|
++nr_regions;
|
++nr_regions;
|
}
|
}
|
else if (extend_regions_p)
|
else if (extend_regions_p)
|
{
|
{
|
/* Restore DEGREE. */
|
/* Restore DEGREE. */
|
int *t = degree;
|
int *t = degree;
|
|
|
degree = degree1;
|
degree = degree1;
|
degree1 = t;
|
degree1 = t;
|
|
|
/* And force successors of BB to be region heads.
|
/* And force successors of BB to be region heads.
|
This may provide several smaller regions instead
|
This may provide several smaller regions instead
|
of one too_large region. */
|
of one too_large region. */
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
if (e->dest != EXIT_BLOCK_PTR)
|
if (e->dest != EXIT_BLOCK_PTR)
|
SET_BIT (extended_rgn_header, e->dest->index);
|
SET_BIT (extended_rgn_header, e->dest->index);
|
}
|
}
|
}
|
}
|
}
|
}
|
free (queue);
|
free (queue);
|
|
|
if (extend_regions_p)
|
if (extend_regions_p)
|
{
|
{
|
free (degree1);
|
free (degree1);
|
|
|
sbitmap_a_or_b (header, header, extended_rgn_header);
|
sbitmap_a_or_b (header, header, extended_rgn_header);
|
sbitmap_free (extended_rgn_header);
|
sbitmap_free (extended_rgn_header);
|
|
|
extend_rgns (degree, &idx, header, max_hdr);
|
extend_rgns (degree, &idx, header, max_hdr);
|
}
|
}
|
}
|
}
|
|
|
/* Any block that did not end up in a region is placed into a region
|
/* Any block that did not end up in a region is placed into a region
|
by itself. */
|
by itself. */
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
if (degree[bb->index] >= 0)
|
if (degree[bb->index] >= 0)
|
{
|
{
|
rgn_bb_table[idx] = bb->index;
|
rgn_bb_table[idx] = bb->index;
|
RGN_NR_BLOCKS (nr_regions) = 1;
|
RGN_NR_BLOCKS (nr_regions) = 1;
|
RGN_BLOCKS (nr_regions) = idx++;
|
RGN_BLOCKS (nr_regions) = idx++;
|
RGN_DONT_CALC_DEPS (nr_regions) = 0;
|
RGN_DONT_CALC_DEPS (nr_regions) = 0;
|
RGN_HAS_REAL_EBB (nr_regions) = 0;
|
RGN_HAS_REAL_EBB (nr_regions) = 0;
|
CONTAINING_RGN (bb->index) = nr_regions++;
|
CONTAINING_RGN (bb->index) = nr_regions++;
|
BLOCK_TO_BB (bb->index) = 0;
|
BLOCK_TO_BB (bb->index) = 0;
|
}
|
}
|
|
|
free (max_hdr);
|
free (max_hdr);
|
free (degree);
|
free (degree);
|
free (stack);
|
free (stack);
|
sbitmap_free (header);
|
sbitmap_free (header);
|
sbitmap_free (inner);
|
sbitmap_free (inner);
|
sbitmap_free (in_queue);
|
sbitmap_free (in_queue);
|
sbitmap_free (in_stack);
|
sbitmap_free (in_stack);
|
}
|
}
|
|
|
|
|
/* Wrapper function.
|
/* Wrapper function.
|
If FLAG_SEL_SCHED_PIPELINING is set, then use custom function to form
|
If FLAG_SEL_SCHED_PIPELINING is set, then use custom function to form
|
regions. Otherwise just call find_rgns_haifa. */
|
regions. Otherwise just call find_rgns_haifa. */
|
static void
|
static void
|
find_rgns (void)
|
find_rgns (void)
|
{
|
{
|
if (sel_sched_p () && flag_sel_sched_pipelining)
|
if (sel_sched_p () && flag_sel_sched_pipelining)
|
sel_find_rgns ();
|
sel_find_rgns ();
|
else
|
else
|
haifa_find_rgns ();
|
haifa_find_rgns ();
|
}
|
}
|
|
|
static int gather_region_statistics (int **);
|
static int gather_region_statistics (int **);
|
static void print_region_statistics (int *, int, int *, int);
|
static void print_region_statistics (int *, int, int *, int);
|
|
|
/* Calculate the histogram that shows the number of regions having the
|
/* Calculate the histogram that shows the number of regions having the
|
given number of basic blocks, and store it in the RSP array. Return
|
given number of basic blocks, and store it in the RSP array. Return
|
the size of this array. */
|
the size of this array. */
|
static int
|
static int
|
gather_region_statistics (int **rsp)
|
gather_region_statistics (int **rsp)
|
{
|
{
|
int i, *a = 0, a_sz = 0;
|
int i, *a = 0, a_sz = 0;
|
|
|
/* a[i] is the number of regions that have (i + 1) basic blocks. */
|
/* a[i] is the number of regions that have (i + 1) basic blocks. */
|
for (i = 0; i < nr_regions; i++)
|
for (i = 0; i < nr_regions; i++)
|
{
|
{
|
int nr_blocks = RGN_NR_BLOCKS (i);
|
int nr_blocks = RGN_NR_BLOCKS (i);
|
|
|
gcc_assert (nr_blocks >= 1);
|
gcc_assert (nr_blocks >= 1);
|
|
|
if (nr_blocks > a_sz)
|
if (nr_blocks > a_sz)
|
{
|
{
|
a = XRESIZEVEC (int, a, nr_blocks);
|
a = XRESIZEVEC (int, a, nr_blocks);
|
do
|
do
|
a[a_sz++] = 0;
|
a[a_sz++] = 0;
|
while (a_sz != nr_blocks);
|
while (a_sz != nr_blocks);
|
}
|
}
|
|
|
a[nr_blocks - 1]++;
|
a[nr_blocks - 1]++;
|
}
|
}
|
|
|
*rsp = a;
|
*rsp = a;
|
return a_sz;
|
return a_sz;
|
}
|
}
|
|
|
/* Print regions statistics. S1 and S2 denote the data before and after
|
/* Print regions statistics. S1 and S2 denote the data before and after
|
calling extend_rgns, respectively. */
|
calling extend_rgns, respectively. */
|
static void
|
static void
|
print_region_statistics (int *s1, int s1_sz, int *s2, int s2_sz)
|
print_region_statistics (int *s1, int s1_sz, int *s2, int s2_sz)
|
{
|
{
|
int i;
|
int i;
|
|
|
/* We iterate until s2_sz because extend_rgns does not decrease
|
/* We iterate until s2_sz because extend_rgns does not decrease
|
the maximal region size. */
|
the maximal region size. */
|
for (i = 1; i < s2_sz; i++)
|
for (i = 1; i < s2_sz; i++)
|
{
|
{
|
int n1, n2;
|
int n1, n2;
|
|
|
n2 = s2[i];
|
n2 = s2[i];
|
|
|
if (n2 == 0)
|
if (n2 == 0)
|
continue;
|
continue;
|
|
|
if (i >= s1_sz)
|
if (i >= s1_sz)
|
n1 = 0;
|
n1 = 0;
|
else
|
else
|
n1 = s1[i];
|
n1 = s1[i];
|
|
|
fprintf (sched_dump, ";; Region extension statistics: size %d: " \
|
fprintf (sched_dump, ";; Region extension statistics: size %d: " \
|
"was %d + %d more\n", i + 1, n1, n2 - n1);
|
"was %d + %d more\n", i + 1, n1, n2 - n1);
|
}
|
}
|
}
|
}
|
|
|
/* Extend regions.
|
/* Extend regions.
|
DEGREE - Array of incoming edge count, considering only
|
DEGREE - Array of incoming edge count, considering only
|
the edges, that don't have their sources in formed regions yet.
|
the edges, that don't have their sources in formed regions yet.
|
IDXP - pointer to the next available index in rgn_bb_table.
|
IDXP - pointer to the next available index in rgn_bb_table.
|
HEADER - set of all region heads.
|
HEADER - set of all region heads.
|
LOOP_HDR - mapping from block to the containing loop
|
LOOP_HDR - mapping from block to the containing loop
|
(two blocks can reside within one region if they have
|
(two blocks can reside within one region if they have
|
the same loop header). */
|
the same loop header). */
|
void
|
void
|
extend_rgns (int *degree, int *idxp, sbitmap header, int *loop_hdr)
|
extend_rgns (int *degree, int *idxp, sbitmap header, int *loop_hdr)
|
{
|
{
|
int *order, i, rescan = 0, idx = *idxp, iter = 0, max_iter, *max_hdr;
|
int *order, i, rescan = 0, idx = *idxp, iter = 0, max_iter, *max_hdr;
|
int nblocks = n_basic_blocks - NUM_FIXED_BLOCKS;
|
int nblocks = n_basic_blocks - NUM_FIXED_BLOCKS;
|
|
|
max_iter = PARAM_VALUE (PARAM_MAX_SCHED_EXTEND_REGIONS_ITERS);
|
max_iter = PARAM_VALUE (PARAM_MAX_SCHED_EXTEND_REGIONS_ITERS);
|
|
|
max_hdr = XNEWVEC (int, last_basic_block);
|
max_hdr = XNEWVEC (int, last_basic_block);
|
|
|
order = XNEWVEC (int, last_basic_block);
|
order = XNEWVEC (int, last_basic_block);
|
post_order_compute (order, false, false);
|
post_order_compute (order, false, false);
|
|
|
for (i = nblocks - 1; i >= 0; i--)
|
for (i = nblocks - 1; i >= 0; i--)
|
{
|
{
|
int bbn = order[i];
|
int bbn = order[i];
|
if (degree[bbn] >= 0)
|
if (degree[bbn] >= 0)
|
{
|
{
|
max_hdr[bbn] = bbn;
|
max_hdr[bbn] = bbn;
|
rescan = 1;
|
rescan = 1;
|
}
|
}
|
else
|
else
|
/* This block already was processed in find_rgns. */
|
/* This block already was processed in find_rgns. */
|
max_hdr[bbn] = -1;
|
max_hdr[bbn] = -1;
|
}
|
}
|
|
|
/* The idea is to topologically walk through CFG in top-down order.
|
/* The idea is to topologically walk through CFG in top-down order.
|
During the traversal, if all the predecessors of a node are
|
During the traversal, if all the predecessors of a node are
|
marked to be in the same region (they all have the same max_hdr),
|
marked to be in the same region (they all have the same max_hdr),
|
then current node is also marked to be a part of that region.
|
then current node is also marked to be a part of that region.
|
Otherwise the node starts its own region.
|
Otherwise the node starts its own region.
|
CFG should be traversed until no further changes are made. On each
|
CFG should be traversed until no further changes are made. On each
|
iteration the set of the region heads is extended (the set of those
|
iteration the set of the region heads is extended (the set of those
|
blocks that have max_hdr[bbi] == bbi). This set is upper bounded by the
|
blocks that have max_hdr[bbi] == bbi). This set is upper bounded by the
|
set of all basic blocks, thus the algorithm is guaranteed to
|
set of all basic blocks, thus the algorithm is guaranteed to
|
terminate. */
|
terminate. */
|
|
|
while (rescan && iter < max_iter)
|
while (rescan && iter < max_iter)
|
{
|
{
|
rescan = 0;
|
rescan = 0;
|
|
|
for (i = nblocks - 1; i >= 0; i--)
|
for (i = nblocks - 1; i >= 0; i--)
|
{
|
{
|
edge e;
|
edge e;
|
edge_iterator ei;
|
edge_iterator ei;
|
int bbn = order[i];
|
int bbn = order[i];
|
|
|
if (max_hdr[bbn] != -1 && !TEST_BIT (header, bbn))
|
if (max_hdr[bbn] != -1 && !TEST_BIT (header, bbn))
|
{
|
{
|
int hdr = -1;
|
int hdr = -1;
|
|
|
FOR_EACH_EDGE (e, ei, BASIC_BLOCK (bbn)->preds)
|
FOR_EACH_EDGE (e, ei, BASIC_BLOCK (bbn)->preds)
|
{
|
{
|
int predn = e->src->index;
|
int predn = e->src->index;
|
|
|
if (predn != ENTRY_BLOCK
|
if (predn != ENTRY_BLOCK
|
/* If pred wasn't processed in find_rgns. */
|
/* If pred wasn't processed in find_rgns. */
|
&& max_hdr[predn] != -1
|
&& max_hdr[predn] != -1
|
/* And pred and bb reside in the same loop.
|
/* And pred and bb reside in the same loop.
|
(Or out of any loop). */
|
(Or out of any loop). */
|
&& loop_hdr[bbn] == loop_hdr[predn])
|
&& loop_hdr[bbn] == loop_hdr[predn])
|
{
|
{
|
if (hdr == -1)
|
if (hdr == -1)
|
/* Then bb extends the containing region of pred. */
|
/* Then bb extends the containing region of pred. */
|
hdr = max_hdr[predn];
|
hdr = max_hdr[predn];
|
else if (hdr != max_hdr[predn])
|
else if (hdr != max_hdr[predn])
|
/* Too bad, there are at least two predecessors
|
/* Too bad, there are at least two predecessors
|
that reside in different regions. Thus, BB should
|
that reside in different regions. Thus, BB should
|
begin its own region. */
|
begin its own region. */
|
{
|
{
|
hdr = bbn;
|
hdr = bbn;
|
break;
|
break;
|
}
|
}
|
}
|
}
|
else
|
else
|
/* BB starts its own region. */
|
/* BB starts its own region. */
|
{
|
{
|
hdr = bbn;
|
hdr = bbn;
|
break;
|
break;
|
}
|
}
|
}
|
}
|
|
|
if (hdr == bbn)
|
if (hdr == bbn)
|
{
|
{
|
/* If BB start its own region,
|
/* If BB start its own region,
|
update set of headers with BB. */
|
update set of headers with BB. */
|
SET_BIT (header, bbn);
|
SET_BIT (header, bbn);
|
rescan = 1;
|
rescan = 1;
|
}
|
}
|
else
|
else
|
gcc_assert (hdr != -1);
|
gcc_assert (hdr != -1);
|
|
|
max_hdr[bbn] = hdr;
|
max_hdr[bbn] = hdr;
|
}
|
}
|
}
|
}
|
|
|
iter++;
|
iter++;
|
}
|
}
|
|
|
/* Statistics were gathered on the SPEC2000 package of tests with
|
/* Statistics were gathered on the SPEC2000 package of tests with
|
mainline weekly snapshot gcc-4.1-20051015 on ia64.
|
mainline weekly snapshot gcc-4.1-20051015 on ia64.
|
|
|
Statistics for SPECint:
|
Statistics for SPECint:
|
1 iteration : 1751 cases (38.7%)
|
1 iteration : 1751 cases (38.7%)
|
2 iterations: 2770 cases (61.3%)
|
2 iterations: 2770 cases (61.3%)
|
Blocks wrapped in regions by find_rgns without extension: 18295 blocks
|
Blocks wrapped in regions by find_rgns without extension: 18295 blocks
|
Blocks wrapped in regions by 2 iterations in extend_rgns: 23821 blocks
|
Blocks wrapped in regions by 2 iterations in extend_rgns: 23821 blocks
|
(We don't count single block regions here).
|
(We don't count single block regions here).
|
|
|
Statistics for SPECfp:
|
Statistics for SPECfp:
|
1 iteration : 621 cases (35.9%)
|
1 iteration : 621 cases (35.9%)
|
2 iterations: 1110 cases (64.1%)
|
2 iterations: 1110 cases (64.1%)
|
Blocks wrapped in regions by find_rgns without extension: 6476 blocks
|
Blocks wrapped in regions by find_rgns without extension: 6476 blocks
|
Blocks wrapped in regions by 2 iterations in extend_rgns: 11155 blocks
|
Blocks wrapped in regions by 2 iterations in extend_rgns: 11155 blocks
|
(We don't count single block regions here).
|
(We don't count single block regions here).
|
|
|
By default we do at most 2 iterations.
|
By default we do at most 2 iterations.
|
This can be overridden with max-sched-extend-regions-iters parameter:
|
This can be overridden with max-sched-extend-regions-iters parameter:
|
0 - disable region extension,
|
0 - disable region extension,
|
N > 0 - do at most N iterations. */
|
N > 0 - do at most N iterations. */
|
|
|
if (sched_verbose && iter != 0)
|
if (sched_verbose && iter != 0)
|
fprintf (sched_dump, ";; Region extension iterations: %d%s\n", iter,
|
fprintf (sched_dump, ";; Region extension iterations: %d%s\n", iter,
|
rescan ? "... failed" : "");
|
rescan ? "... failed" : "");
|
|
|
if (!rescan && iter != 0)
|
if (!rescan && iter != 0)
|
{
|
{
|
int *s1 = NULL, s1_sz = 0;
|
int *s1 = NULL, s1_sz = 0;
|
|
|
/* Save the old statistics for later printout. */
|
/* Save the old statistics for later printout. */
|
if (sched_verbose >= 6)
|
if (sched_verbose >= 6)
|
s1_sz = gather_region_statistics (&s1);
|
s1_sz = gather_region_statistics (&s1);
|
|
|
/* We have succeeded. Now assemble the regions. */
|
/* We have succeeded. Now assemble the regions. */
|
for (i = nblocks - 1; i >= 0; i--)
|
for (i = nblocks - 1; i >= 0; i--)
|
{
|
{
|
int bbn = order[i];
|
int bbn = order[i];
|
|
|
if (max_hdr[bbn] == bbn)
|
if (max_hdr[bbn] == bbn)
|
/* BBN is a region head. */
|
/* BBN is a region head. */
|
{
|
{
|
edge e;
|
edge e;
|
edge_iterator ei;
|
edge_iterator ei;
|
int num_bbs = 0, j, num_insns = 0, large;
|
int num_bbs = 0, j, num_insns = 0, large;
|
|
|
large = too_large (bbn, &num_bbs, &num_insns);
|
large = too_large (bbn, &num_bbs, &num_insns);
|
|
|
degree[bbn] = -1;
|
degree[bbn] = -1;
|
rgn_bb_table[idx] = bbn;
|
rgn_bb_table[idx] = bbn;
|
RGN_BLOCKS (nr_regions) = idx++;
|
RGN_BLOCKS (nr_regions) = idx++;
|
RGN_DONT_CALC_DEPS (nr_regions) = 0;
|
RGN_DONT_CALC_DEPS (nr_regions) = 0;
|
RGN_HAS_REAL_EBB (nr_regions) = 0;
|
RGN_HAS_REAL_EBB (nr_regions) = 0;
|
CONTAINING_RGN (bbn) = nr_regions;
|
CONTAINING_RGN (bbn) = nr_regions;
|
BLOCK_TO_BB (bbn) = 0;
|
BLOCK_TO_BB (bbn) = 0;
|
|
|
FOR_EACH_EDGE (e, ei, BASIC_BLOCK (bbn)->succs)
|
FOR_EACH_EDGE (e, ei, BASIC_BLOCK (bbn)->succs)
|
if (e->dest != EXIT_BLOCK_PTR)
|
if (e->dest != EXIT_BLOCK_PTR)
|
degree[e->dest->index]--;
|
degree[e->dest->index]--;
|
|
|
if (!large)
|
if (!large)
|
/* Here we check whether the region is too_large. */
|
/* Here we check whether the region is too_large. */
|
for (j = i - 1; j >= 0; j--)
|
for (j = i - 1; j >= 0; j--)
|
{
|
{
|
int succn = order[j];
|
int succn = order[j];
|
if (max_hdr[succn] == bbn)
|
if (max_hdr[succn] == bbn)
|
{
|
{
|
if ((large = too_large (succn, &num_bbs, &num_insns)))
|
if ((large = too_large (succn, &num_bbs, &num_insns)))
|
break;
|
break;
|
}
|
}
|
}
|
}
|
|
|
if (large)
|
if (large)
|
/* If the region is too_large, then wrap every block of
|
/* If the region is too_large, then wrap every block of
|
the region into single block region.
|
the region into single block region.
|
Here we wrap region head only. Other blocks are
|
Here we wrap region head only. Other blocks are
|
processed in the below cycle. */
|
processed in the below cycle. */
|
{
|
{
|
RGN_NR_BLOCKS (nr_regions) = 1;
|
RGN_NR_BLOCKS (nr_regions) = 1;
|
nr_regions++;
|
nr_regions++;
|
}
|
}
|
|
|
num_bbs = 1;
|
num_bbs = 1;
|
|
|
for (j = i - 1; j >= 0; j--)
|
for (j = i - 1; j >= 0; j--)
|
{
|
{
|
int succn = order[j];
|
int succn = order[j];
|
|
|
if (max_hdr[succn] == bbn)
|
if (max_hdr[succn] == bbn)
|
/* This cycle iterates over all basic blocks, that
|
/* This cycle iterates over all basic blocks, that
|
are supposed to be in the region with head BBN,
|
are supposed to be in the region with head BBN,
|
and wraps them into that region (or in single
|
and wraps them into that region (or in single
|
block region). */
|
block region). */
|
{
|
{
|
gcc_assert (degree[succn] == 0);
|
gcc_assert (degree[succn] == 0);
|
|
|
degree[succn] = -1;
|
degree[succn] = -1;
|
rgn_bb_table[idx] = succn;
|
rgn_bb_table[idx] = succn;
|
BLOCK_TO_BB (succn) = large ? 0 : num_bbs++;
|
BLOCK_TO_BB (succn) = large ? 0 : num_bbs++;
|
CONTAINING_RGN (succn) = nr_regions;
|
CONTAINING_RGN (succn) = nr_regions;
|
|
|
if (large)
|
if (large)
|
/* Wrap SUCCN into single block region. */
|
/* Wrap SUCCN into single block region. */
|
{
|
{
|
RGN_BLOCKS (nr_regions) = idx;
|
RGN_BLOCKS (nr_regions) = idx;
|
RGN_NR_BLOCKS (nr_regions) = 1;
|
RGN_NR_BLOCKS (nr_regions) = 1;
|
RGN_DONT_CALC_DEPS (nr_regions) = 0;
|
RGN_DONT_CALC_DEPS (nr_regions) = 0;
|
RGN_HAS_REAL_EBB (nr_regions) = 0;
|
RGN_HAS_REAL_EBB (nr_regions) = 0;
|
nr_regions++;
|
nr_regions++;
|
}
|
}
|
|
|
idx++;
|
idx++;
|
|
|
FOR_EACH_EDGE (e, ei, BASIC_BLOCK (succn)->succs)
|
FOR_EACH_EDGE (e, ei, BASIC_BLOCK (succn)->succs)
|
if (e->dest != EXIT_BLOCK_PTR)
|
if (e->dest != EXIT_BLOCK_PTR)
|
degree[e->dest->index]--;
|
degree[e->dest->index]--;
|
}
|
}
|
}
|
}
|
|
|
if (!large)
|
if (!large)
|
{
|
{
|
RGN_NR_BLOCKS (nr_regions) = num_bbs;
|
RGN_NR_BLOCKS (nr_regions) = num_bbs;
|
nr_regions++;
|
nr_regions++;
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
if (sched_verbose >= 6)
|
if (sched_verbose >= 6)
|
{
|
{
|
int *s2, s2_sz;
|
int *s2, s2_sz;
|
|
|
/* Get the new statistics and print the comparison with the
|
/* Get the new statistics and print the comparison with the
|
one before calling this function. */
|
one before calling this function. */
|
s2_sz = gather_region_statistics (&s2);
|
s2_sz = gather_region_statistics (&s2);
|
print_region_statistics (s1, s1_sz, s2, s2_sz);
|
print_region_statistics (s1, s1_sz, s2, s2_sz);
|
free (s1);
|
free (s1);
|
free (s2);
|
free (s2);
|
}
|
}
|
}
|
}
|
|
|
free (order);
|
free (order);
|
free (max_hdr);
|
free (max_hdr);
|
|
|
*idxp = idx;
|
*idxp = idx;
|
}
|
}
|
|
|
/* Functions for regions scheduling information. */
|
/* Functions for regions scheduling information. */
|
|
|
/* Compute dominators, probability, and potential-split-edges of bb.
|
/* Compute dominators, probability, and potential-split-edges of bb.
|
Assume that these values were already computed for bb's predecessors. */
|
Assume that these values were already computed for bb's predecessors. */
|
|
|
static void
|
static void
|
compute_dom_prob_ps (int bb)
|
compute_dom_prob_ps (int bb)
|
{
|
{
|
edge_iterator in_ei;
|
edge_iterator in_ei;
|
edge in_edge;
|
edge in_edge;
|
|
|
/* We shouldn't have any real ebbs yet. */
|
/* We shouldn't have any real ebbs yet. */
|
gcc_assert (ebb_head [bb] == bb + current_blocks);
|
gcc_assert (ebb_head [bb] == bb + current_blocks);
|
|
|
if (IS_RGN_ENTRY (bb))
|
if (IS_RGN_ENTRY (bb))
|
{
|
{
|
SET_BIT (dom[bb], 0);
|
SET_BIT (dom[bb], 0);
|
prob[bb] = REG_BR_PROB_BASE;
|
prob[bb] = REG_BR_PROB_BASE;
|
return;
|
return;
|
}
|
}
|
|
|
prob[bb] = 0;
|
prob[bb] = 0;
|
|
|
/* Initialize dom[bb] to '111..1'. */
|
/* Initialize dom[bb] to '111..1'. */
|
sbitmap_ones (dom[bb]);
|
sbitmap_ones (dom[bb]);
|
|
|
FOR_EACH_EDGE (in_edge, in_ei, BASIC_BLOCK (BB_TO_BLOCK (bb))->preds)
|
FOR_EACH_EDGE (in_edge, in_ei, BASIC_BLOCK (BB_TO_BLOCK (bb))->preds)
|
{
|
{
|
int pred_bb;
|
int pred_bb;
|
edge out_edge;
|
edge out_edge;
|
edge_iterator out_ei;
|
edge_iterator out_ei;
|
|
|
if (in_edge->src == ENTRY_BLOCK_PTR)
|
if (in_edge->src == ENTRY_BLOCK_PTR)
|
continue;
|
continue;
|
|
|
pred_bb = BLOCK_TO_BB (in_edge->src->index);
|
pred_bb = BLOCK_TO_BB (in_edge->src->index);
|
sbitmap_a_and_b (dom[bb], dom[bb], dom[pred_bb]);
|
sbitmap_a_and_b (dom[bb], dom[bb], dom[pred_bb]);
|
sbitmap_a_or_b (ancestor_edges[bb],
|
sbitmap_a_or_b (ancestor_edges[bb],
|
ancestor_edges[bb], ancestor_edges[pred_bb]);
|
ancestor_edges[bb], ancestor_edges[pred_bb]);
|
|
|
SET_BIT (ancestor_edges[bb], EDGE_TO_BIT (in_edge));
|
SET_BIT (ancestor_edges[bb], EDGE_TO_BIT (in_edge));
|
|
|
sbitmap_a_or_b (pot_split[bb], pot_split[bb], pot_split[pred_bb]);
|
sbitmap_a_or_b (pot_split[bb], pot_split[bb], pot_split[pred_bb]);
|
|
|
FOR_EACH_EDGE (out_edge, out_ei, in_edge->src->succs)
|
FOR_EACH_EDGE (out_edge, out_ei, in_edge->src->succs)
|
SET_BIT (pot_split[bb], EDGE_TO_BIT (out_edge));
|
SET_BIT (pot_split[bb], EDGE_TO_BIT (out_edge));
|
|
|
prob[bb] += ((prob[pred_bb] * in_edge->probability) / REG_BR_PROB_BASE);
|
prob[bb] += ((prob[pred_bb] * in_edge->probability) / REG_BR_PROB_BASE);
|
}
|
}
|
|
|
SET_BIT (dom[bb], bb);
|
SET_BIT (dom[bb], bb);
|
sbitmap_difference (pot_split[bb], pot_split[bb], ancestor_edges[bb]);
|
sbitmap_difference (pot_split[bb], pot_split[bb], ancestor_edges[bb]);
|
|
|
if (sched_verbose >= 2)
|
if (sched_verbose >= 2)
|
fprintf (sched_dump, ";; bb_prob(%d, %d) = %3d\n", bb, BB_TO_BLOCK (bb),
|
fprintf (sched_dump, ";; bb_prob(%d, %d) = %3d\n", bb, BB_TO_BLOCK (bb),
|
(100 * prob[bb]) / REG_BR_PROB_BASE);
|
(100 * prob[bb]) / REG_BR_PROB_BASE);
|
}
|
}
|
|
|
/* Functions for target info. */
|
/* Functions for target info. */
|
|
|
/* Compute in BL the list of split-edges of bb_src relatively to bb_trg.
|
/* Compute in BL the list of split-edges of bb_src relatively to bb_trg.
|
Note that bb_trg dominates bb_src. */
|
Note that bb_trg dominates bb_src. */
|
|
|
static void
|
static void
|
split_edges (int bb_src, int bb_trg, edgelst *bl)
|
split_edges (int bb_src, int bb_trg, edgelst *bl)
|
{
|
{
|
sbitmap src = sbitmap_alloc (pot_split[bb_src]->n_bits);
|
sbitmap src = sbitmap_alloc (pot_split[bb_src]->n_bits);
|
sbitmap_copy (src, pot_split[bb_src]);
|
sbitmap_copy (src, pot_split[bb_src]);
|
|
|
sbitmap_difference (src, src, pot_split[bb_trg]);
|
sbitmap_difference (src, src, pot_split[bb_trg]);
|
extract_edgelst (src, bl);
|
extract_edgelst (src, bl);
|
sbitmap_free (src);
|
sbitmap_free (src);
|
}
|
}
|
|
|
/* Find the valid candidate-source-blocks for the target block TRG, compute
|
/* Find the valid candidate-source-blocks for the target block TRG, compute
|
their probability, and check if they are speculative or not.
|
their probability, and check if they are speculative or not.
|
For speculative sources, compute their update-blocks and split-blocks. */
|
For speculative sources, compute their update-blocks and split-blocks. */
|
|
|
static void
|
static void
|
compute_trg_info (int trg)
|
compute_trg_info (int trg)
|
{
|
{
|
candidate *sp;
|
candidate *sp;
|
edgelst el = { NULL, 0 };
|
edgelst el = { NULL, 0 };
|
int i, j, k, update_idx;
|
int i, j, k, update_idx;
|
basic_block block;
|
basic_block block;
|
sbitmap visited;
|
sbitmap visited;
|
edge_iterator ei;
|
edge_iterator ei;
|
edge e;
|
edge e;
|
|
|
candidate_table = XNEWVEC (candidate, current_nr_blocks);
|
candidate_table = XNEWVEC (candidate, current_nr_blocks);
|
|
|
bblst_last = 0;
|
bblst_last = 0;
|
/* bblst_table holds split blocks and update blocks for each block after
|
/* bblst_table holds split blocks and update blocks for each block after
|
the current one in the region. split blocks and update blocks are
|
the current one in the region. split blocks and update blocks are
|
the TO blocks of region edges, so there can be at most rgn_nr_edges
|
the TO blocks of region edges, so there can be at most rgn_nr_edges
|
of them. */
|
of them. */
|
bblst_size = (current_nr_blocks - target_bb) * rgn_nr_edges;
|
bblst_size = (current_nr_blocks - target_bb) * rgn_nr_edges;
|
bblst_table = XNEWVEC (basic_block, bblst_size);
|
bblst_table = XNEWVEC (basic_block, bblst_size);
|
|
|
edgelst_last = 0;
|
edgelst_last = 0;
|
edgelst_table = XNEWVEC (edge, rgn_nr_edges);
|
edgelst_table = XNEWVEC (edge, rgn_nr_edges);
|
|
|
/* Define some of the fields for the target bb as well. */
|
/* Define some of the fields for the target bb as well. */
|
sp = candidate_table + trg;
|
sp = candidate_table + trg;
|
sp->is_valid = 1;
|
sp->is_valid = 1;
|
sp->is_speculative = 0;
|
sp->is_speculative = 0;
|
sp->src_prob = REG_BR_PROB_BASE;
|
sp->src_prob = REG_BR_PROB_BASE;
|
|
|
visited = sbitmap_alloc (last_basic_block);
|
visited = sbitmap_alloc (last_basic_block);
|
|
|
for (i = trg + 1; i < current_nr_blocks; i++)
|
for (i = trg + 1; i < current_nr_blocks; i++)
|
{
|
{
|
sp = candidate_table + i;
|
sp = candidate_table + i;
|
|
|
sp->is_valid = IS_DOMINATED (i, trg);
|
sp->is_valid = IS_DOMINATED (i, trg);
|
if (sp->is_valid)
|
if (sp->is_valid)
|
{
|
{
|
int tf = prob[trg], cf = prob[i];
|
int tf = prob[trg], cf = prob[i];
|
|
|
/* In CFGs with low probability edges TF can possibly be zero. */
|
/* In CFGs with low probability edges TF can possibly be zero. */
|
sp->src_prob = (tf ? ((cf * REG_BR_PROB_BASE) / tf) : 0);
|
sp->src_prob = (tf ? ((cf * REG_BR_PROB_BASE) / tf) : 0);
|
sp->is_valid = (sp->src_prob >= min_spec_prob);
|
sp->is_valid = (sp->src_prob >= min_spec_prob);
|
}
|
}
|
|
|
if (sp->is_valid)
|
if (sp->is_valid)
|
{
|
{
|
split_edges (i, trg, &el);
|
split_edges (i, trg, &el);
|
sp->is_speculative = (el.nr_members) ? 1 : 0;
|
sp->is_speculative = (el.nr_members) ? 1 : 0;
|
if (sp->is_speculative && !flag_schedule_speculative)
|
if (sp->is_speculative && !flag_schedule_speculative)
|
sp->is_valid = 0;
|
sp->is_valid = 0;
|
}
|
}
|
|
|
if (sp->is_valid)
|
if (sp->is_valid)
|
{
|
{
|
/* Compute split blocks and store them in bblst_table.
|
/* Compute split blocks and store them in bblst_table.
|
The TO block of every split edge is a split block. */
|
The TO block of every split edge is a split block. */
|
sp->split_bbs.first_member = &bblst_table[bblst_last];
|
sp->split_bbs.first_member = &bblst_table[bblst_last];
|
sp->split_bbs.nr_members = el.nr_members;
|
sp->split_bbs.nr_members = el.nr_members;
|
for (j = 0; j < el.nr_members; bblst_last++, j++)
|
for (j = 0; j < el.nr_members; bblst_last++, j++)
|
bblst_table[bblst_last] = el.first_member[j]->dest;
|
bblst_table[bblst_last] = el.first_member[j]->dest;
|
sp->update_bbs.first_member = &bblst_table[bblst_last];
|
sp->update_bbs.first_member = &bblst_table[bblst_last];
|
|
|
/* Compute update blocks and store them in bblst_table.
|
/* Compute update blocks and store them in bblst_table.
|
For every split edge, look at the FROM block, and check
|
For every split edge, look at the FROM block, and check
|
all out edges. For each out edge that is not a split edge,
|
all out edges. For each out edge that is not a split edge,
|
add the TO block to the update block list. This list can end
|
add the TO block to the update block list. This list can end
|
up with a lot of duplicates. We need to weed them out to avoid
|
up with a lot of duplicates. We need to weed them out to avoid
|
overrunning the end of the bblst_table. */
|
overrunning the end of the bblst_table. */
|
|
|
update_idx = 0;
|
update_idx = 0;
|
sbitmap_zero (visited);
|
sbitmap_zero (visited);
|
for (j = 0; j < el.nr_members; j++)
|
for (j = 0; j < el.nr_members; j++)
|
{
|
{
|
block = el.first_member[j]->src;
|
block = el.first_member[j]->src;
|
FOR_EACH_EDGE (e, ei, block->succs)
|
FOR_EACH_EDGE (e, ei, block->succs)
|
{
|
{
|
if (!TEST_BIT (visited, e->dest->index))
|
if (!TEST_BIT (visited, e->dest->index))
|
{
|
{
|
for (k = 0; k < el.nr_members; k++)
|
for (k = 0; k < el.nr_members; k++)
|
if (e == el.first_member[k])
|
if (e == el.first_member[k])
|
break;
|
break;
|
|
|
if (k >= el.nr_members)
|
if (k >= el.nr_members)
|
{
|
{
|
bblst_table[bblst_last++] = e->dest;
|
bblst_table[bblst_last++] = e->dest;
|
SET_BIT (visited, e->dest->index);
|
SET_BIT (visited, e->dest->index);
|
update_idx++;
|
update_idx++;
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
sp->update_bbs.nr_members = update_idx;
|
sp->update_bbs.nr_members = update_idx;
|
|
|
/* Make sure we didn't overrun the end of bblst_table. */
|
/* Make sure we didn't overrun the end of bblst_table. */
|
gcc_assert (bblst_last <= bblst_size);
|
gcc_assert (bblst_last <= bblst_size);
|
}
|
}
|
else
|
else
|
{
|
{
|
sp->split_bbs.nr_members = sp->update_bbs.nr_members = 0;
|
sp->split_bbs.nr_members = sp->update_bbs.nr_members = 0;
|
|
|
sp->is_speculative = 0;
|
sp->is_speculative = 0;
|
sp->src_prob = 0;
|
sp->src_prob = 0;
|
}
|
}
|
}
|
}
|
|
|
sbitmap_free (visited);
|
sbitmap_free (visited);
|
}
|
}
|
|
|
/* Free the computed target info. */
|
/* Free the computed target info. */
|
static void
|
static void
|
free_trg_info (void)
|
free_trg_info (void)
|
{
|
{
|
free (candidate_table);
|
free (candidate_table);
|
free (bblst_table);
|
free (bblst_table);
|
free (edgelst_table);
|
free (edgelst_table);
|
}
|
}
|
|
|
/* Print candidates info, for debugging purposes. Callable from debugger. */
|
/* Print candidates info, for debugging purposes. Callable from debugger. */
|
|
|
void
|
void
|
debug_candidate (int i)
|
debug_candidate (int i)
|
{
|
{
|
if (!candidate_table[i].is_valid)
|
if (!candidate_table[i].is_valid)
|
return;
|
return;
|
|
|
if (candidate_table[i].is_speculative)
|
if (candidate_table[i].is_speculative)
|
{
|
{
|
int j;
|
int j;
|
fprintf (sched_dump, "src b %d bb %d speculative \n", BB_TO_BLOCK (i), i);
|
fprintf (sched_dump, "src b %d bb %d speculative \n", BB_TO_BLOCK (i), i);
|
|
|
fprintf (sched_dump, "split path: ");
|
fprintf (sched_dump, "split path: ");
|
for (j = 0; j < candidate_table[i].split_bbs.nr_members; j++)
|
for (j = 0; j < candidate_table[i].split_bbs.nr_members; j++)
|
{
|
{
|
int b = candidate_table[i].split_bbs.first_member[j]->index;
|
int b = candidate_table[i].split_bbs.first_member[j]->index;
|
|
|
fprintf (sched_dump, " %d ", b);
|
fprintf (sched_dump, " %d ", b);
|
}
|
}
|
fprintf (sched_dump, "\n");
|
fprintf (sched_dump, "\n");
|
|
|
fprintf (sched_dump, "update path: ");
|
fprintf (sched_dump, "update path: ");
|
for (j = 0; j < candidate_table[i].update_bbs.nr_members; j++)
|
for (j = 0; j < candidate_table[i].update_bbs.nr_members; j++)
|
{
|
{
|
int b = candidate_table[i].update_bbs.first_member[j]->index;
|
int b = candidate_table[i].update_bbs.first_member[j]->index;
|
|
|
fprintf (sched_dump, " %d ", b);
|
fprintf (sched_dump, " %d ", b);
|
}
|
}
|
fprintf (sched_dump, "\n");
|
fprintf (sched_dump, "\n");
|
}
|
}
|
else
|
else
|
{
|
{
|
fprintf (sched_dump, " src %d equivalent\n", BB_TO_BLOCK (i));
|
fprintf (sched_dump, " src %d equivalent\n", BB_TO_BLOCK (i));
|
}
|
}
|
}
|
}
|
|
|
/* Print candidates info, for debugging purposes. Callable from debugger. */
|
/* Print candidates info, for debugging purposes. Callable from debugger. */
|
|
|
void
|
void
|
debug_candidates (int trg)
|
debug_candidates (int trg)
|
{
|
{
|
int i;
|
int i;
|
|
|
fprintf (sched_dump, "----------- candidate table: target: b=%d bb=%d ---\n",
|
fprintf (sched_dump, "----------- candidate table: target: b=%d bb=%d ---\n",
|
BB_TO_BLOCK (trg), trg);
|
BB_TO_BLOCK (trg), trg);
|
for (i = trg + 1; i < current_nr_blocks; i++)
|
for (i = trg + 1; i < current_nr_blocks; i++)
|
debug_candidate (i);
|
debug_candidate (i);
|
}
|
}
|
|
|
/* Functions for speculative scheduling. */
|
/* Functions for speculative scheduling. */
|
|
|
static bitmap_head not_in_df;
|
static bitmap_head not_in_df;
|
|
|
/* Return 0 if x is a set of a register alive in the beginning of one
|
/* Return 0 if x is a set of a register alive in the beginning of one
|
of the split-blocks of src, otherwise return 1. */
|
of the split-blocks of src, otherwise return 1. */
|
|
|
static int
|
static int
|
check_live_1 (int src, rtx x)
|
check_live_1 (int src, rtx x)
|
{
|
{
|
int i;
|
int i;
|
int regno;
|
int regno;
|
rtx reg = SET_DEST (x);
|
rtx reg = SET_DEST (x);
|
|
|
if (reg == 0)
|
if (reg == 0)
|
return 1;
|
return 1;
|
|
|
while (GET_CODE (reg) == SUBREG
|
while (GET_CODE (reg) == SUBREG
|
|| GET_CODE (reg) == ZERO_EXTRACT
|
|| GET_CODE (reg) == ZERO_EXTRACT
|
|| GET_CODE (reg) == STRICT_LOW_PART)
|
|| GET_CODE (reg) == STRICT_LOW_PART)
|
reg = XEXP (reg, 0);
|
reg = XEXP (reg, 0);
|
|
|
if (GET_CODE (reg) == PARALLEL)
|
if (GET_CODE (reg) == PARALLEL)
|
{
|
{
|
int i;
|
int i;
|
|
|
for (i = XVECLEN (reg, 0) - 1; i >= 0; i--)
|
for (i = XVECLEN (reg, 0) - 1; i >= 0; i--)
|
if (XEXP (XVECEXP (reg, 0, i), 0) != 0)
|
if (XEXP (XVECEXP (reg, 0, i), 0) != 0)
|
if (check_live_1 (src, XEXP (XVECEXP (reg, 0, i), 0)))
|
if (check_live_1 (src, XEXP (XVECEXP (reg, 0, i), 0)))
|
return 1;
|
return 1;
|
|
|
return 0;
|
return 0;
|
}
|
}
|
|
|
if (!REG_P (reg))
|
if (!REG_P (reg))
|
return 1;
|
return 1;
|
|
|
regno = REGNO (reg);
|
regno = REGNO (reg);
|
|
|
if (regno < FIRST_PSEUDO_REGISTER && global_regs[regno])
|
if (regno < FIRST_PSEUDO_REGISTER && global_regs[regno])
|
{
|
{
|
/* Global registers are assumed live. */
|
/* Global registers are assumed live. */
|
return 0;
|
return 0;
|
}
|
}
|
else
|
else
|
{
|
{
|
if (regno < FIRST_PSEUDO_REGISTER)
|
if (regno < FIRST_PSEUDO_REGISTER)
|
{
|
{
|
/* Check for hard registers. */
|
/* Check for hard registers. */
|
int j = hard_regno_nregs[regno][GET_MODE (reg)];
|
int j = hard_regno_nregs[regno][GET_MODE (reg)];
|
while (--j >= 0)
|
while (--j >= 0)
|
{
|
{
|
for (i = 0; i < candidate_table[src].split_bbs.nr_members; i++)
|
for (i = 0; i < candidate_table[src].split_bbs.nr_members; i++)
|
{
|
{
|
basic_block b = candidate_table[src].split_bbs.first_member[i];
|
basic_block b = candidate_table[src].split_bbs.first_member[i];
|
int t = bitmap_bit_p (¬_in_df, b->index);
|
int t = bitmap_bit_p (¬_in_df, b->index);
|
|
|
/* We can have split blocks, that were recently generated.
|
/* We can have split blocks, that were recently generated.
|
Such blocks are always outside current region. */
|
Such blocks are always outside current region. */
|
gcc_assert (!t || (CONTAINING_RGN (b->index)
|
gcc_assert (!t || (CONTAINING_RGN (b->index)
|
!= CONTAINING_RGN (BB_TO_BLOCK (src))));
|
!= CONTAINING_RGN (BB_TO_BLOCK (src))));
|
|
|
if (t || REGNO_REG_SET_P (df_get_live_in (b), regno + j))
|
if (t || REGNO_REG_SET_P (df_get_live_in (b), regno + j))
|
return 0;
|
return 0;
|
}
|
}
|
}
|
}
|
}
|
}
|
else
|
else
|
{
|
{
|
/* Check for pseudo registers. */
|
/* Check for pseudo registers. */
|
for (i = 0; i < candidate_table[src].split_bbs.nr_members; i++)
|
for (i = 0; i < candidate_table[src].split_bbs.nr_members; i++)
|
{
|
{
|
basic_block b = candidate_table[src].split_bbs.first_member[i];
|
basic_block b = candidate_table[src].split_bbs.first_member[i];
|
int t = bitmap_bit_p (¬_in_df, b->index);
|
int t = bitmap_bit_p (¬_in_df, b->index);
|
|
|
gcc_assert (!t || (CONTAINING_RGN (b->index)
|
gcc_assert (!t || (CONTAINING_RGN (b->index)
|
!= CONTAINING_RGN (BB_TO_BLOCK (src))));
|
!= CONTAINING_RGN (BB_TO_BLOCK (src))));
|
|
|
if (t || REGNO_REG_SET_P (df_get_live_in (b), regno))
|
if (t || REGNO_REG_SET_P (df_get_live_in (b), regno))
|
return 0;
|
return 0;
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
return 1;
|
return 1;
|
}
|
}
|
|
|
/* If x is a set of a register R, mark that R is alive in the beginning
|
/* If x is a set of a register R, mark that R is alive in the beginning
|
of every update-block of src. */
|
of every update-block of src. */
|
|
|
static void
|
static void
|
update_live_1 (int src, rtx x)
|
update_live_1 (int src, rtx x)
|
{
|
{
|
int i;
|
int i;
|
int regno;
|
int regno;
|
rtx reg = SET_DEST (x);
|
rtx reg = SET_DEST (x);
|
|
|
if (reg == 0)
|
if (reg == 0)
|
return;
|
return;
|
|
|
while (GET_CODE (reg) == SUBREG
|
while (GET_CODE (reg) == SUBREG
|
|| GET_CODE (reg) == ZERO_EXTRACT
|
|| GET_CODE (reg) == ZERO_EXTRACT
|
|| GET_CODE (reg) == STRICT_LOW_PART)
|
|| GET_CODE (reg) == STRICT_LOW_PART)
|
reg = XEXP (reg, 0);
|
reg = XEXP (reg, 0);
|
|
|
if (GET_CODE (reg) == PARALLEL)
|
if (GET_CODE (reg) == PARALLEL)
|
{
|
{
|
int i;
|
int i;
|
|
|
for (i = XVECLEN (reg, 0) - 1; i >= 0; i--)
|
for (i = XVECLEN (reg, 0) - 1; i >= 0; i--)
|
if (XEXP (XVECEXP (reg, 0, i), 0) != 0)
|
if (XEXP (XVECEXP (reg, 0, i), 0) != 0)
|
update_live_1 (src, XEXP (XVECEXP (reg, 0, i), 0));
|
update_live_1 (src, XEXP (XVECEXP (reg, 0, i), 0));
|
|
|
return;
|
return;
|
}
|
}
|
|
|
if (!REG_P (reg))
|
if (!REG_P (reg))
|
return;
|
return;
|
|
|
/* Global registers are always live, so the code below does not apply
|
/* Global registers are always live, so the code below does not apply
|
to them. */
|
to them. */
|
|
|
regno = REGNO (reg);
|
regno = REGNO (reg);
|
|
|
if (regno >= FIRST_PSEUDO_REGISTER || !global_regs[regno])
|
if (regno >= FIRST_PSEUDO_REGISTER || !global_regs[regno])
|
{
|
{
|
if (regno < FIRST_PSEUDO_REGISTER)
|
if (regno < FIRST_PSEUDO_REGISTER)
|
{
|
{
|
int j = hard_regno_nregs[regno][GET_MODE (reg)];
|
int j = hard_regno_nregs[regno][GET_MODE (reg)];
|
while (--j >= 0)
|
while (--j >= 0)
|
{
|
{
|
for (i = 0; i < candidate_table[src].update_bbs.nr_members; i++)
|
for (i = 0; i < candidate_table[src].update_bbs.nr_members; i++)
|
{
|
{
|
basic_block b = candidate_table[src].update_bbs.first_member[i];
|
basic_block b = candidate_table[src].update_bbs.first_member[i];
|
|
|
SET_REGNO_REG_SET (df_get_live_in (b), regno + j);
|
SET_REGNO_REG_SET (df_get_live_in (b), regno + j);
|
}
|
}
|
}
|
}
|
}
|
}
|
else
|
else
|
{
|
{
|
for (i = 0; i < candidate_table[src].update_bbs.nr_members; i++)
|
for (i = 0; i < candidate_table[src].update_bbs.nr_members; i++)
|
{
|
{
|
basic_block b = candidate_table[src].update_bbs.first_member[i];
|
basic_block b = candidate_table[src].update_bbs.first_member[i];
|
|
|
SET_REGNO_REG_SET (df_get_live_in (b), regno);
|
SET_REGNO_REG_SET (df_get_live_in (b), regno);
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
/* Return 1 if insn can be speculatively moved from block src to trg,
|
/* Return 1 if insn can be speculatively moved from block src to trg,
|
otherwise return 0. Called before first insertion of insn to
|
otherwise return 0. Called before first insertion of insn to
|
ready-list or before the scheduling. */
|
ready-list or before the scheduling. */
|
|
|
static int
|
static int
|
check_live (rtx insn, int src)
|
check_live (rtx insn, int src)
|
{
|
{
|
/* Find the registers set by instruction. */
|
/* Find the registers set by instruction. */
|
if (GET_CODE (PATTERN (insn)) == SET
|
if (GET_CODE (PATTERN (insn)) == SET
|
|| GET_CODE (PATTERN (insn)) == CLOBBER)
|
|| GET_CODE (PATTERN (insn)) == CLOBBER)
|
return check_live_1 (src, PATTERN (insn));
|
return check_live_1 (src, PATTERN (insn));
|
else if (GET_CODE (PATTERN (insn)) == PARALLEL)
|
else if (GET_CODE (PATTERN (insn)) == PARALLEL)
|
{
|
{
|
int j;
|
int j;
|
for (j = XVECLEN (PATTERN (insn), 0) - 1; j >= 0; j--)
|
for (j = XVECLEN (PATTERN (insn), 0) - 1; j >= 0; j--)
|
if ((GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == SET
|
if ((GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == SET
|
|| GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == CLOBBER)
|
|| GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == CLOBBER)
|
&& !check_live_1 (src, XVECEXP (PATTERN (insn), 0, j)))
|
&& !check_live_1 (src, XVECEXP (PATTERN (insn), 0, j)))
|
return 0;
|
return 0;
|
|
|
return 1;
|
return 1;
|
}
|
}
|
|
|
return 1;
|
return 1;
|
}
|
}
|
|
|
/* Update the live registers info after insn was moved speculatively from
|
/* Update the live registers info after insn was moved speculatively from
|
block src to trg. */
|
block src to trg. */
|
|
|
static void
|
static void
|
update_live (rtx insn, int src)
|
update_live (rtx insn, int src)
|
{
|
{
|
/* Find the registers set by instruction. */
|
/* Find the registers set by instruction. */
|
if (GET_CODE (PATTERN (insn)) == SET
|
if (GET_CODE (PATTERN (insn)) == SET
|
|| GET_CODE (PATTERN (insn)) == CLOBBER)
|
|| GET_CODE (PATTERN (insn)) == CLOBBER)
|
update_live_1 (src, PATTERN (insn));
|
update_live_1 (src, PATTERN (insn));
|
else if (GET_CODE (PATTERN (insn)) == PARALLEL)
|
else if (GET_CODE (PATTERN (insn)) == PARALLEL)
|
{
|
{
|
int j;
|
int j;
|
for (j = XVECLEN (PATTERN (insn), 0) - 1; j >= 0; j--)
|
for (j = XVECLEN (PATTERN (insn), 0) - 1; j >= 0; j--)
|
if (GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == SET
|
if (GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == SET
|
|| GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == CLOBBER)
|
|| GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == CLOBBER)
|
update_live_1 (src, XVECEXP (PATTERN (insn), 0, j));
|
update_live_1 (src, XVECEXP (PATTERN (insn), 0, j));
|
}
|
}
|
}
|
}
|
|
|
/* Nonzero if block bb_to is equal to, or reachable from block bb_from. */
|
/* Nonzero if block bb_to is equal to, or reachable from block bb_from. */
|
#define IS_REACHABLE(bb_from, bb_to) \
|
#define IS_REACHABLE(bb_from, bb_to) \
|
(bb_from == bb_to \
|
(bb_from == bb_to \
|
|| IS_RGN_ENTRY (bb_from) \
|
|| IS_RGN_ENTRY (bb_from) \
|
|| (TEST_BIT (ancestor_edges[bb_to], \
|
|| (TEST_BIT (ancestor_edges[bb_to], \
|
EDGE_TO_BIT (single_pred_edge (BASIC_BLOCK (BB_TO_BLOCK (bb_from)))))))
|
EDGE_TO_BIT (single_pred_edge (BASIC_BLOCK (BB_TO_BLOCK (bb_from)))))))
|
|
|
/* Turns on the fed_by_spec_load flag for insns fed by load_insn. */
|
/* Turns on the fed_by_spec_load flag for insns fed by load_insn. */
|
|
|
static void
|
static void
|
set_spec_fed (rtx load_insn)
|
set_spec_fed (rtx load_insn)
|
{
|
{
|
sd_iterator_def sd_it;
|
sd_iterator_def sd_it;
|
dep_t dep;
|
dep_t dep;
|
|
|
FOR_EACH_DEP (load_insn, SD_LIST_FORW, sd_it, dep)
|
FOR_EACH_DEP (load_insn, SD_LIST_FORW, sd_it, dep)
|
if (DEP_TYPE (dep) == REG_DEP_TRUE)
|
if (DEP_TYPE (dep) == REG_DEP_TRUE)
|
FED_BY_SPEC_LOAD (DEP_CON (dep)) = 1;
|
FED_BY_SPEC_LOAD (DEP_CON (dep)) = 1;
|
}
|
}
|
|
|
/* On the path from the insn to load_insn_bb, find a conditional
|
/* On the path from the insn to load_insn_bb, find a conditional
|
branch depending on insn, that guards the speculative load. */
|
branch depending on insn, that guards the speculative load. */
|
|
|
static int
|
static int
|
find_conditional_protection (rtx insn, int load_insn_bb)
|
find_conditional_protection (rtx insn, int load_insn_bb)
|
{
|
{
|
sd_iterator_def sd_it;
|
sd_iterator_def sd_it;
|
dep_t dep;
|
dep_t dep;
|
|
|
/* Iterate through DEF-USE forward dependences. */
|
/* Iterate through DEF-USE forward dependences. */
|
FOR_EACH_DEP (insn, SD_LIST_FORW, sd_it, dep)
|
FOR_EACH_DEP (insn, SD_LIST_FORW, sd_it, dep)
|
{
|
{
|
rtx next = DEP_CON (dep);
|
rtx next = DEP_CON (dep);
|
|
|
if ((CONTAINING_RGN (BLOCK_NUM (next)) ==
|
if ((CONTAINING_RGN (BLOCK_NUM (next)) ==
|
CONTAINING_RGN (BB_TO_BLOCK (load_insn_bb)))
|
CONTAINING_RGN (BB_TO_BLOCK (load_insn_bb)))
|
&& IS_REACHABLE (INSN_BB (next), load_insn_bb)
|
&& IS_REACHABLE (INSN_BB (next), load_insn_bb)
|
&& load_insn_bb != INSN_BB (next)
|
&& load_insn_bb != INSN_BB (next)
|
&& DEP_TYPE (dep) == REG_DEP_TRUE
|
&& DEP_TYPE (dep) == REG_DEP_TRUE
|
&& (JUMP_P (next)
|
&& (JUMP_P (next)
|
|| find_conditional_protection (next, load_insn_bb)))
|
|| find_conditional_protection (next, load_insn_bb)))
|
return 1;
|
return 1;
|
}
|
}
|
return 0;
|
return 0;
|
} /* find_conditional_protection */
|
} /* find_conditional_protection */
|
|
|
/* Returns 1 if the same insn1 that participates in the computation
|
/* Returns 1 if the same insn1 that participates in the computation
|
of load_insn's address is feeding a conditional branch that is
|
of load_insn's address is feeding a conditional branch that is
|
guarding on load_insn. This is true if we find two DEF-USE
|
guarding on load_insn. This is true if we find two DEF-USE
|
chains:
|
chains:
|
insn1 -> ... -> conditional-branch
|
insn1 -> ... -> conditional-branch
|
insn1 -> ... -> load_insn,
|
insn1 -> ... -> load_insn,
|
and if a flow path exists:
|
and if a flow path exists:
|
insn1 -> ... -> conditional-branch -> ... -> load_insn,
|
insn1 -> ... -> conditional-branch -> ... -> load_insn,
|
and if insn1 is on the path
|
and if insn1 is on the path
|
region-entry -> ... -> bb_trg -> ... load_insn.
|
region-entry -> ... -> bb_trg -> ... load_insn.
|
|
|
Locate insn1 by climbing on INSN_BACK_DEPS from load_insn.
|
Locate insn1 by climbing on INSN_BACK_DEPS from load_insn.
|
Locate the branch by following INSN_FORW_DEPS from insn1. */
|
Locate the branch by following INSN_FORW_DEPS from insn1. */
|
|
|
static int
|
static int
|
is_conditionally_protected (rtx load_insn, int bb_src, int bb_trg)
|
is_conditionally_protected (rtx load_insn, int bb_src, int bb_trg)
|
{
|
{
|
sd_iterator_def sd_it;
|
sd_iterator_def sd_it;
|
dep_t dep;
|
dep_t dep;
|
|
|
FOR_EACH_DEP (load_insn, SD_LIST_BACK, sd_it, dep)
|
FOR_EACH_DEP (load_insn, SD_LIST_BACK, sd_it, dep)
|
{
|
{
|
rtx insn1 = DEP_PRO (dep);
|
rtx insn1 = DEP_PRO (dep);
|
|
|
/* Must be a DEF-USE dependence upon non-branch. */
|
/* Must be a DEF-USE dependence upon non-branch. */
|
if (DEP_TYPE (dep) != REG_DEP_TRUE
|
if (DEP_TYPE (dep) != REG_DEP_TRUE
|
|| JUMP_P (insn1))
|
|| JUMP_P (insn1))
|
continue;
|
continue;
|
|
|
/* Must exist a path: region-entry -> ... -> bb_trg -> ... load_insn. */
|
/* Must exist a path: region-entry -> ... -> bb_trg -> ... load_insn. */
|
if (INSN_BB (insn1) == bb_src
|
if (INSN_BB (insn1) == bb_src
|
|| (CONTAINING_RGN (BLOCK_NUM (insn1))
|
|| (CONTAINING_RGN (BLOCK_NUM (insn1))
|
!= CONTAINING_RGN (BB_TO_BLOCK (bb_src)))
|
!= CONTAINING_RGN (BB_TO_BLOCK (bb_src)))
|
|| (!IS_REACHABLE (bb_trg, INSN_BB (insn1))
|
|| (!IS_REACHABLE (bb_trg, INSN_BB (insn1))
|
&& !IS_REACHABLE (INSN_BB (insn1), bb_trg)))
|
&& !IS_REACHABLE (INSN_BB (insn1), bb_trg)))
|
continue;
|
continue;
|
|
|
/* Now search for the conditional-branch. */
|
/* Now search for the conditional-branch. */
|
if (find_conditional_protection (insn1, bb_src))
|
if (find_conditional_protection (insn1, bb_src))
|
return 1;
|
return 1;
|
|
|
/* Recursive step: search another insn1, "above" current insn1. */
|
/* Recursive step: search another insn1, "above" current insn1. */
|
return is_conditionally_protected (insn1, bb_src, bb_trg);
|
return is_conditionally_protected (insn1, bb_src, bb_trg);
|
}
|
}
|
|
|
/* The chain does not exist. */
|
/* The chain does not exist. */
|
return 0;
|
return 0;
|
} /* is_conditionally_protected */
|
} /* is_conditionally_protected */
|
|
|
/* Returns 1 if a clue for "similar load" 'insn2' is found, and hence
|
/* Returns 1 if a clue for "similar load" 'insn2' is found, and hence
|
load_insn can move speculatively from bb_src to bb_trg. All the
|
load_insn can move speculatively from bb_src to bb_trg. All the
|
following must hold:
|
following must hold:
|
|
|
(1) both loads have 1 base register (PFREE_CANDIDATEs).
|
(1) both loads have 1 base register (PFREE_CANDIDATEs).
|
(2) load_insn and load1 have a def-use dependence upon
|
(2) load_insn and load1 have a def-use dependence upon
|
the same insn 'insn1'.
|
the same insn 'insn1'.
|
(3) either load2 is in bb_trg, or:
|
(3) either load2 is in bb_trg, or:
|
- there's only one split-block, and
|
- there's only one split-block, and
|
- load1 is on the escape path, and
|
- load1 is on the escape path, and
|
|
|
From all these we can conclude that the two loads access memory
|
From all these we can conclude that the two loads access memory
|
addresses that differ at most by a constant, and hence if moving
|
addresses that differ at most by a constant, and hence if moving
|
load_insn would cause an exception, it would have been caused by
|
load_insn would cause an exception, it would have been caused by
|
load2 anyhow. */
|
load2 anyhow. */
|
|
|
static int
|
static int
|
is_pfree (rtx load_insn, int bb_src, int bb_trg)
|
is_pfree (rtx load_insn, int bb_src, int bb_trg)
|
{
|
{
|
sd_iterator_def back_sd_it;
|
sd_iterator_def back_sd_it;
|
dep_t back_dep;
|
dep_t back_dep;
|
candidate *candp = candidate_table + bb_src;
|
candidate *candp = candidate_table + bb_src;
|
|
|
if (candp->split_bbs.nr_members != 1)
|
if (candp->split_bbs.nr_members != 1)
|
/* Must have exactly one escape block. */
|
/* Must have exactly one escape block. */
|
return 0;
|
return 0;
|
|
|
FOR_EACH_DEP (load_insn, SD_LIST_BACK, back_sd_it, back_dep)
|
FOR_EACH_DEP (load_insn, SD_LIST_BACK, back_sd_it, back_dep)
|
{
|
{
|
rtx insn1 = DEP_PRO (back_dep);
|
rtx insn1 = DEP_PRO (back_dep);
|
|
|
if (DEP_TYPE (back_dep) == REG_DEP_TRUE)
|
if (DEP_TYPE (back_dep) == REG_DEP_TRUE)
|
/* Found a DEF-USE dependence (insn1, load_insn). */
|
/* Found a DEF-USE dependence (insn1, load_insn). */
|
{
|
{
|
sd_iterator_def fore_sd_it;
|
sd_iterator_def fore_sd_it;
|
dep_t fore_dep;
|
dep_t fore_dep;
|
|
|
FOR_EACH_DEP (insn1, SD_LIST_FORW, fore_sd_it, fore_dep)
|
FOR_EACH_DEP (insn1, SD_LIST_FORW, fore_sd_it, fore_dep)
|
{
|
{
|
rtx insn2 = DEP_CON (fore_dep);
|
rtx insn2 = DEP_CON (fore_dep);
|
|
|
if (DEP_TYPE (fore_dep) == REG_DEP_TRUE)
|
if (DEP_TYPE (fore_dep) == REG_DEP_TRUE)
|
{
|
{
|
/* Found a DEF-USE dependence (insn1, insn2). */
|
/* Found a DEF-USE dependence (insn1, insn2). */
|
if (haifa_classify_insn (insn2) != PFREE_CANDIDATE)
|
if (haifa_classify_insn (insn2) != PFREE_CANDIDATE)
|
/* insn2 not guaranteed to be a 1 base reg load. */
|
/* insn2 not guaranteed to be a 1 base reg load. */
|
continue;
|
continue;
|
|
|
if (INSN_BB (insn2) == bb_trg)
|
if (INSN_BB (insn2) == bb_trg)
|
/* insn2 is the similar load, in the target block. */
|
/* insn2 is the similar load, in the target block. */
|
return 1;
|
return 1;
|
|
|
if (*(candp->split_bbs.first_member) == BLOCK_FOR_INSN (insn2))
|
if (*(candp->split_bbs.first_member) == BLOCK_FOR_INSN (insn2))
|
/* insn2 is a similar load, in a split-block. */
|
/* insn2 is a similar load, in a split-block. */
|
return 1;
|
return 1;
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
/* Couldn't find a similar load. */
|
/* Couldn't find a similar load. */
|
return 0;
|
return 0;
|
} /* is_pfree */
|
} /* is_pfree */
|
|
|
/* Return 1 if load_insn is prisky (i.e. if load_insn is fed by
|
/* Return 1 if load_insn is prisky (i.e. if load_insn is fed by
|
a load moved speculatively, or if load_insn is protected by
|
a load moved speculatively, or if load_insn is protected by
|
a compare on load_insn's address). */
|
a compare on load_insn's address). */
|
|
|
static int
|
static int
|
is_prisky (rtx load_insn, int bb_src, int bb_trg)
|
is_prisky (rtx load_insn, int bb_src, int bb_trg)
|
{
|
{
|
if (FED_BY_SPEC_LOAD (load_insn))
|
if (FED_BY_SPEC_LOAD (load_insn))
|
return 1;
|
return 1;
|
|
|
if (sd_lists_empty_p (load_insn, SD_LIST_BACK))
|
if (sd_lists_empty_p (load_insn, SD_LIST_BACK))
|
/* Dependence may 'hide' out of the region. */
|
/* Dependence may 'hide' out of the region. */
|
return 1;
|
return 1;
|
|
|
if (is_conditionally_protected (load_insn, bb_src, bb_trg))
|
if (is_conditionally_protected (load_insn, bb_src, bb_trg))
|
return 1;
|
return 1;
|
|
|
return 0;
|
return 0;
|
}
|
}
|
|
|
/* Insn is a candidate to be moved speculatively from bb_src to bb_trg.
|
/* Insn is a candidate to be moved speculatively from bb_src to bb_trg.
|
Return 1 if insn is exception-free (and the motion is valid)
|
Return 1 if insn is exception-free (and the motion is valid)
|
and 0 otherwise. */
|
and 0 otherwise. */
|
|
|
static int
|
static int
|
is_exception_free (rtx insn, int bb_src, int bb_trg)
|
is_exception_free (rtx insn, int bb_src, int bb_trg)
|
{
|
{
|
int insn_class = haifa_classify_insn (insn);
|
int insn_class = haifa_classify_insn (insn);
|
|
|
/* Handle non-load insns. */
|
/* Handle non-load insns. */
|
switch (insn_class)
|
switch (insn_class)
|
{
|
{
|
case TRAP_FREE:
|
case TRAP_FREE:
|
return 1;
|
return 1;
|
case TRAP_RISKY:
|
case TRAP_RISKY:
|
return 0;
|
return 0;
|
default:;
|
default:;
|
}
|
}
|
|
|
/* Handle loads. */
|
/* Handle loads. */
|
if (!flag_schedule_speculative_load)
|
if (!flag_schedule_speculative_load)
|
return 0;
|
return 0;
|
IS_LOAD_INSN (insn) = 1;
|
IS_LOAD_INSN (insn) = 1;
|
switch (insn_class)
|
switch (insn_class)
|
{
|
{
|
case IFREE:
|
case IFREE:
|
return (1);
|
return (1);
|
case IRISKY:
|
case IRISKY:
|
return 0;
|
return 0;
|
case PFREE_CANDIDATE:
|
case PFREE_CANDIDATE:
|
if (is_pfree (insn, bb_src, bb_trg))
|
if (is_pfree (insn, bb_src, bb_trg))
|
return 1;
|
return 1;
|
/* Don't 'break' here: PFREE-candidate is also PRISKY-candidate. */
|
/* Don't 'break' here: PFREE-candidate is also PRISKY-candidate. */
|
case PRISKY_CANDIDATE:
|
case PRISKY_CANDIDATE:
|
if (!flag_schedule_speculative_load_dangerous
|
if (!flag_schedule_speculative_load_dangerous
|
|| is_prisky (insn, bb_src, bb_trg))
|
|| is_prisky (insn, bb_src, bb_trg))
|
return 0;
|
return 0;
|
break;
|
break;
|
default:;
|
default:;
|
}
|
}
|
|
|
return flag_schedule_speculative_load_dangerous;
|
return flag_schedule_speculative_load_dangerous;
|
}
|
}
|
|
|
/* The number of insns from the current block scheduled so far. */
|
/* The number of insns from the current block scheduled so far. */
|
static int sched_target_n_insns;
|
static int sched_target_n_insns;
|
/* The number of insns from the current block to be scheduled in total. */
|
/* The number of insns from the current block to be scheduled in total. */
|
static int target_n_insns;
|
static int target_n_insns;
|
/* The number of insns from the entire region scheduled so far. */
|
/* The number of insns from the entire region scheduled so far. */
|
static int sched_n_insns;
|
static int sched_n_insns;
|
|
|
/* Implementations of the sched_info functions for region scheduling. */
|
/* Implementations of the sched_info functions for region scheduling. */
|
static void init_ready_list (void);
|
static void init_ready_list (void);
|
static int can_schedule_ready_p (rtx);
|
static int can_schedule_ready_p (rtx);
|
static void begin_schedule_ready (rtx, rtx);
|
static void begin_schedule_ready (rtx, rtx);
|
static ds_t new_ready (rtx, ds_t);
|
static ds_t new_ready (rtx, ds_t);
|
static int schedule_more_p (void);
|
static int schedule_more_p (void);
|
static const char *rgn_print_insn (const_rtx, int);
|
static const char *rgn_print_insn (const_rtx, int);
|
static int rgn_rank (rtx, rtx);
|
static int rgn_rank (rtx, rtx);
|
static void compute_jump_reg_dependencies (rtx, regset, regset, regset);
|
static void compute_jump_reg_dependencies (rtx, regset, regset, regset);
|
|
|
/* Functions for speculative scheduling. */
|
/* Functions for speculative scheduling. */
|
static void rgn_add_remove_insn (rtx, int);
|
static void rgn_add_remove_insn (rtx, int);
|
static void rgn_add_block (basic_block, basic_block);
|
static void rgn_add_block (basic_block, basic_block);
|
static void rgn_fix_recovery_cfg (int, int, int);
|
static void rgn_fix_recovery_cfg (int, int, int);
|
static basic_block advance_target_bb (basic_block, rtx);
|
static basic_block advance_target_bb (basic_block, rtx);
|
|
|
/* Return nonzero if there are more insns that should be scheduled. */
|
/* Return nonzero if there are more insns that should be scheduled. */
|
|
|
static int
|
static int
|
schedule_more_p (void)
|
schedule_more_p (void)
|
{
|
{
|
return sched_target_n_insns < target_n_insns;
|
return sched_target_n_insns < target_n_insns;
|
}
|
}
|
|
|
/* Add all insns that are initially ready to the ready list READY. Called
|
/* Add all insns that are initially ready to the ready list READY. Called
|
once before scheduling a set of insns. */
|
once before scheduling a set of insns. */
|
|
|
static void
|
static void
|
init_ready_list (void)
|
init_ready_list (void)
|
{
|
{
|
rtx prev_head = current_sched_info->prev_head;
|
rtx prev_head = current_sched_info->prev_head;
|
rtx next_tail = current_sched_info->next_tail;
|
rtx next_tail = current_sched_info->next_tail;
|
int bb_src;
|
int bb_src;
|
rtx insn;
|
rtx insn;
|
|
|
target_n_insns = 0;
|
target_n_insns = 0;
|
sched_target_n_insns = 0;
|
sched_target_n_insns = 0;
|
sched_n_insns = 0;
|
sched_n_insns = 0;
|
|
|
/* Print debugging information. */
|
/* Print debugging information. */
|
if (sched_verbose >= 5)
|
if (sched_verbose >= 5)
|
debug_rgn_dependencies (target_bb);
|
debug_rgn_dependencies (target_bb);
|
|
|
/* Prepare current target block info. */
|
/* Prepare current target block info. */
|
if (current_nr_blocks > 1)
|
if (current_nr_blocks > 1)
|
compute_trg_info (target_bb);
|
compute_trg_info (target_bb);
|
|
|
/* Initialize ready list with all 'ready' insns in target block.
|
/* Initialize ready list with all 'ready' insns in target block.
|
Count number of insns in the target block being scheduled. */
|
Count number of insns in the target block being scheduled. */
|
for (insn = NEXT_INSN (prev_head); insn != next_tail; insn = NEXT_INSN (insn))
|
for (insn = NEXT_INSN (prev_head); insn != next_tail; insn = NEXT_INSN (insn))
|
{
|
{
|
try_ready (insn);
|
try_ready (insn);
|
target_n_insns++;
|
target_n_insns++;
|
|
|
gcc_assert (!(TODO_SPEC (insn) & BEGIN_CONTROL));
|
gcc_assert (!(TODO_SPEC (insn) & BEGIN_CONTROL));
|
}
|
}
|
|
|
/* Add to ready list all 'ready' insns in valid source blocks.
|
/* Add to ready list all 'ready' insns in valid source blocks.
|
For speculative insns, check-live, exception-free, and
|
For speculative insns, check-live, exception-free, and
|
issue-delay. */
|
issue-delay. */
|
for (bb_src = target_bb + 1; bb_src < current_nr_blocks; bb_src++)
|
for (bb_src = target_bb + 1; bb_src < current_nr_blocks; bb_src++)
|
if (IS_VALID (bb_src))
|
if (IS_VALID (bb_src))
|
{
|
{
|
rtx src_head;
|
rtx src_head;
|
rtx src_next_tail;
|
rtx src_next_tail;
|
rtx tail, head;
|
rtx tail, head;
|
|
|
get_ebb_head_tail (EBB_FIRST_BB (bb_src), EBB_LAST_BB (bb_src),
|
get_ebb_head_tail (EBB_FIRST_BB (bb_src), EBB_LAST_BB (bb_src),
|
&head, &tail);
|
&head, &tail);
|
src_next_tail = NEXT_INSN (tail);
|
src_next_tail = NEXT_INSN (tail);
|
src_head = head;
|
src_head = head;
|
|
|
for (insn = src_head; insn != src_next_tail; insn = NEXT_INSN (insn))
|
for (insn = src_head; insn != src_next_tail; insn = NEXT_INSN (insn))
|
if (INSN_P (insn) && !BOUNDARY_DEBUG_INSN_P (insn))
|
if (INSN_P (insn) && !BOUNDARY_DEBUG_INSN_P (insn))
|
try_ready (insn);
|
try_ready (insn);
|
}
|
}
|
}
|
}
|
|
|
/* Called after taking INSN from the ready list. Returns nonzero if this
|
/* Called after taking INSN from the ready list. Returns nonzero if this
|
insn can be scheduled, nonzero if we should silently discard it. */
|
insn can be scheduled, nonzero if we should silently discard it. */
|
|
|
static int
|
static int
|
can_schedule_ready_p (rtx insn)
|
can_schedule_ready_p (rtx insn)
|
{
|
{
|
/* An interblock motion? */
|
/* An interblock motion? */
|
if (INSN_BB (insn) != target_bb
|
if (INSN_BB (insn) != target_bb
|
&& IS_SPECULATIVE_INSN (insn)
|
&& IS_SPECULATIVE_INSN (insn)
|
&& !check_live (insn, INSN_BB (insn)))
|
&& !check_live (insn, INSN_BB (insn)))
|
return 0;
|
return 0;
|
else
|
else
|
return 1;
|
return 1;
|
}
|
}
|
|
|
/* Updates counter and other information. Split from can_schedule_ready_p ()
|
/* Updates counter and other information. Split from can_schedule_ready_p ()
|
because when we schedule insn speculatively then insn passed to
|
because when we schedule insn speculatively then insn passed to
|
can_schedule_ready_p () differs from the one passed to
|
can_schedule_ready_p () differs from the one passed to
|
begin_schedule_ready (). */
|
begin_schedule_ready (). */
|
static void
|
static void
|
begin_schedule_ready (rtx insn, rtx last ATTRIBUTE_UNUSED)
|
begin_schedule_ready (rtx insn, rtx last ATTRIBUTE_UNUSED)
|
{
|
{
|
/* An interblock motion? */
|
/* An interblock motion? */
|
if (INSN_BB (insn) != target_bb)
|
if (INSN_BB (insn) != target_bb)
|
{
|
{
|
if (IS_SPECULATIVE_INSN (insn))
|
if (IS_SPECULATIVE_INSN (insn))
|
{
|
{
|
gcc_assert (check_live (insn, INSN_BB (insn)));
|
gcc_assert (check_live (insn, INSN_BB (insn)));
|
|
|
update_live (insn, INSN_BB (insn));
|
update_live (insn, INSN_BB (insn));
|
|
|
/* For speculative load, mark insns fed by it. */
|
/* For speculative load, mark insns fed by it. */
|
if (IS_LOAD_INSN (insn) || FED_BY_SPEC_LOAD (insn))
|
if (IS_LOAD_INSN (insn) || FED_BY_SPEC_LOAD (insn))
|
set_spec_fed (insn);
|
set_spec_fed (insn);
|
|
|
nr_spec++;
|
nr_spec++;
|
}
|
}
|
nr_inter++;
|
nr_inter++;
|
}
|
}
|
else
|
else
|
{
|
{
|
/* In block motion. */
|
/* In block motion. */
|
sched_target_n_insns++;
|
sched_target_n_insns++;
|
}
|
}
|
sched_n_insns++;
|
sched_n_insns++;
|
}
|
}
|
|
|
/* Called after INSN has all its hard dependencies resolved and the speculation
|
/* Called after INSN has all its hard dependencies resolved and the speculation
|
of type TS is enough to overcome them all.
|
of type TS is enough to overcome them all.
|
Return nonzero if it should be moved to the ready list or the queue, or zero
|
Return nonzero if it should be moved to the ready list or the queue, or zero
|
if we should silently discard it. */
|
if we should silently discard it. */
|
static ds_t
|
static ds_t
|
new_ready (rtx next, ds_t ts)
|
new_ready (rtx next, ds_t ts)
|
{
|
{
|
if (INSN_BB (next) != target_bb)
|
if (INSN_BB (next) != target_bb)
|
{
|
{
|
int not_ex_free = 0;
|
int not_ex_free = 0;
|
|
|
/* For speculative insns, before inserting to ready/queue,
|
/* For speculative insns, before inserting to ready/queue,
|
check live, exception-free, and issue-delay. */
|
check live, exception-free, and issue-delay. */
|
if (!IS_VALID (INSN_BB (next))
|
if (!IS_VALID (INSN_BB (next))
|
|| CANT_MOVE (next)
|
|| CANT_MOVE (next)
|
|| (IS_SPECULATIVE_INSN (next)
|
|| (IS_SPECULATIVE_INSN (next)
|
&& ((recog_memoized (next) >= 0
|
&& ((recog_memoized (next) >= 0
|
&& min_insn_conflict_delay (curr_state, next, next)
|
&& min_insn_conflict_delay (curr_state, next, next)
|
> PARAM_VALUE (PARAM_MAX_SCHED_INSN_CONFLICT_DELAY))
|
> PARAM_VALUE (PARAM_MAX_SCHED_INSN_CONFLICT_DELAY))
|
|| IS_SPECULATION_CHECK_P (next)
|
|| IS_SPECULATION_CHECK_P (next)
|
|| !check_live (next, INSN_BB (next))
|
|| !check_live (next, INSN_BB (next))
|
|| (not_ex_free = !is_exception_free (next, INSN_BB (next),
|
|| (not_ex_free = !is_exception_free (next, INSN_BB (next),
|
target_bb)))))
|
target_bb)))))
|
{
|
{
|
if (not_ex_free
|
if (not_ex_free
|
/* We are here because is_exception_free () == false.
|
/* We are here because is_exception_free () == false.
|
But we possibly can handle that with control speculation. */
|
But we possibly can handle that with control speculation. */
|
&& sched_deps_info->generate_spec_deps
|
&& sched_deps_info->generate_spec_deps
|
&& spec_info->mask & BEGIN_CONTROL)
|
&& spec_info->mask & BEGIN_CONTROL)
|
{
|
{
|
ds_t new_ds;
|
ds_t new_ds;
|
|
|
/* Add control speculation to NEXT's dependency type. */
|
/* Add control speculation to NEXT's dependency type. */
|
new_ds = set_dep_weak (ts, BEGIN_CONTROL, MAX_DEP_WEAK);
|
new_ds = set_dep_weak (ts, BEGIN_CONTROL, MAX_DEP_WEAK);
|
|
|
/* Check if NEXT can be speculated with new dependency type. */
|
/* Check if NEXT can be speculated with new dependency type. */
|
if (sched_insn_is_legitimate_for_speculation_p (next, new_ds))
|
if (sched_insn_is_legitimate_for_speculation_p (next, new_ds))
|
/* Here we got new control-speculative instruction. */
|
/* Here we got new control-speculative instruction. */
|
ts = new_ds;
|
ts = new_ds;
|
else
|
else
|
/* NEXT isn't ready yet. */
|
/* NEXT isn't ready yet. */
|
ts = (ts & ~SPECULATIVE) | HARD_DEP;
|
ts = (ts & ~SPECULATIVE) | HARD_DEP;
|
}
|
}
|
else
|
else
|
/* NEXT isn't ready yet. */
|
/* NEXT isn't ready yet. */
|
ts = (ts & ~SPECULATIVE) | HARD_DEP;
|
ts = (ts & ~SPECULATIVE) | HARD_DEP;
|
}
|
}
|
}
|
}
|
|
|
return ts;
|
return ts;
|
}
|
}
|
|
|
/* Return a string that contains the insn uid and optionally anything else
|
/* Return a string that contains the insn uid and optionally anything else
|
necessary to identify this insn in an output. It's valid to use a
|
necessary to identify this insn in an output. It's valid to use a
|
static buffer for this. The ALIGNED parameter should cause the string
|
static buffer for this. The ALIGNED parameter should cause the string
|
to be formatted so that multiple output lines will line up nicely. */
|
to be formatted so that multiple output lines will line up nicely. */
|
|
|
static const char *
|
static const char *
|
rgn_print_insn (const_rtx insn, int aligned)
|
rgn_print_insn (const_rtx insn, int aligned)
|
{
|
{
|
static char tmp[80];
|
static char tmp[80];
|
|
|
if (aligned)
|
if (aligned)
|
sprintf (tmp, "b%3d: i%4d", INSN_BB (insn), INSN_UID (insn));
|
sprintf (tmp, "b%3d: i%4d", INSN_BB (insn), INSN_UID (insn));
|
else
|
else
|
{
|
{
|
if (current_nr_blocks > 1 && INSN_BB (insn) != target_bb)
|
if (current_nr_blocks > 1 && INSN_BB (insn) != target_bb)
|
sprintf (tmp, "%d/b%d", INSN_UID (insn), INSN_BB (insn));
|
sprintf (tmp, "%d/b%d", INSN_UID (insn), INSN_BB (insn));
|
else
|
else
|
sprintf (tmp, "%d", INSN_UID (insn));
|
sprintf (tmp, "%d", INSN_UID (insn));
|
}
|
}
|
return tmp;
|
return tmp;
|
}
|
}
|
|
|
/* Compare priority of two insns. Return a positive number if the second
|
/* Compare priority of two insns. Return a positive number if the second
|
insn is to be preferred for scheduling, and a negative one if the first
|
insn is to be preferred for scheduling, and a negative one if the first
|
is to be preferred. Zero if they are equally good. */
|
is to be preferred. Zero if they are equally good. */
|
|
|
static int
|
static int
|
rgn_rank (rtx insn1, rtx insn2)
|
rgn_rank (rtx insn1, rtx insn2)
|
{
|
{
|
/* Some comparison make sense in interblock scheduling only. */
|
/* Some comparison make sense in interblock scheduling only. */
|
if (INSN_BB (insn1) != INSN_BB (insn2))
|
if (INSN_BB (insn1) != INSN_BB (insn2))
|
{
|
{
|
int spec_val, prob_val;
|
int spec_val, prob_val;
|
|
|
/* Prefer an inblock motion on an interblock motion. */
|
/* Prefer an inblock motion on an interblock motion. */
|
if ((INSN_BB (insn2) == target_bb) && (INSN_BB (insn1) != target_bb))
|
if ((INSN_BB (insn2) == target_bb) && (INSN_BB (insn1) != target_bb))
|
return 1;
|
return 1;
|
if ((INSN_BB (insn1) == target_bb) && (INSN_BB (insn2) != target_bb))
|
if ((INSN_BB (insn1) == target_bb) && (INSN_BB (insn2) != target_bb))
|
return -1;
|
return -1;
|
|
|
/* Prefer a useful motion on a speculative one. */
|
/* Prefer a useful motion on a speculative one. */
|
spec_val = IS_SPECULATIVE_INSN (insn1) - IS_SPECULATIVE_INSN (insn2);
|
spec_val = IS_SPECULATIVE_INSN (insn1) - IS_SPECULATIVE_INSN (insn2);
|
if (spec_val)
|
if (spec_val)
|
return spec_val;
|
return spec_val;
|
|
|
/* Prefer a more probable (speculative) insn. */
|
/* Prefer a more probable (speculative) insn. */
|
prob_val = INSN_PROBABILITY (insn2) - INSN_PROBABILITY (insn1);
|
prob_val = INSN_PROBABILITY (insn2) - INSN_PROBABILITY (insn1);
|
if (prob_val)
|
if (prob_val)
|
return prob_val;
|
return prob_val;
|
}
|
}
|
return 0;
|
return 0;
|
}
|
}
|
|
|
/* NEXT is an instruction that depends on INSN (a backward dependence);
|
/* NEXT is an instruction that depends on INSN (a backward dependence);
|
return nonzero if we should include this dependence in priority
|
return nonzero if we should include this dependence in priority
|
calculations. */
|
calculations. */
|
|
|
int
|
int
|
contributes_to_priority (rtx next, rtx insn)
|
contributes_to_priority (rtx next, rtx insn)
|
{
|
{
|
/* NEXT and INSN reside in one ebb. */
|
/* NEXT and INSN reside in one ebb. */
|
return BLOCK_TO_BB (BLOCK_NUM (next)) == BLOCK_TO_BB (BLOCK_NUM (insn));
|
return BLOCK_TO_BB (BLOCK_NUM (next)) == BLOCK_TO_BB (BLOCK_NUM (insn));
|
}
|
}
|
|
|
/* INSN is a JUMP_INSN, COND_SET is the set of registers that are
|
/* INSN is a JUMP_INSN, COND_SET is the set of registers that are
|
conditionally set before INSN. Store the set of registers that
|
conditionally set before INSN. Store the set of registers that
|
must be considered as used by this jump in USED and that of
|
must be considered as used by this jump in USED and that of
|
registers that must be considered as set in SET. */
|
registers that must be considered as set in SET. */
|
|
|
static void
|
static void
|
compute_jump_reg_dependencies (rtx insn ATTRIBUTE_UNUSED,
|
compute_jump_reg_dependencies (rtx insn ATTRIBUTE_UNUSED,
|
regset cond_exec ATTRIBUTE_UNUSED,
|
regset cond_exec ATTRIBUTE_UNUSED,
|
regset used ATTRIBUTE_UNUSED,
|
regset used ATTRIBUTE_UNUSED,
|
regset set ATTRIBUTE_UNUSED)
|
regset set ATTRIBUTE_UNUSED)
|
{
|
{
|
/* Nothing to do here, since we postprocess jumps in
|
/* Nothing to do here, since we postprocess jumps in
|
add_branch_dependences. */
|
add_branch_dependences. */
|
}
|
}
|
|
|
/* This variable holds common_sched_info hooks and data relevant to
|
/* This variable holds common_sched_info hooks and data relevant to
|
the interblock scheduler. */
|
the interblock scheduler. */
|
static struct common_sched_info_def rgn_common_sched_info;
|
static struct common_sched_info_def rgn_common_sched_info;
|
|
|
|
|
/* This holds data for the dependence analysis relevant to
|
/* This holds data for the dependence analysis relevant to
|
the interblock scheduler. */
|
the interblock scheduler. */
|
static struct sched_deps_info_def rgn_sched_deps_info;
|
static struct sched_deps_info_def rgn_sched_deps_info;
|
|
|
/* This holds constant data used for initializing the above structure
|
/* This holds constant data used for initializing the above structure
|
for the Haifa scheduler. */
|
for the Haifa scheduler. */
|
static const struct sched_deps_info_def rgn_const_sched_deps_info =
|
static const struct sched_deps_info_def rgn_const_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
|
};
|
};
|
|
|
/* Same as above, but for the selective scheduler. */
|
/* Same as above, but for the selective scheduler. */
|
static const struct sched_deps_info_def rgn_const_sel_sched_deps_info =
|
static const struct sched_deps_info_def rgn_const_sel_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
|
};
|
};
|
|
|
/* Return true if scheduling INSN will trigger finish of scheduling
|
/* Return true if scheduling INSN will trigger finish of scheduling
|
current block. */
|
current block. */
|
static bool
|
static bool
|
rgn_insn_finishes_block_p (rtx insn)
|
rgn_insn_finishes_block_p (rtx insn)
|
{
|
{
|
if (INSN_BB (insn) == target_bb
|
if (INSN_BB (insn) == target_bb
|
&& sched_target_n_insns + 1 == target_n_insns)
|
&& sched_target_n_insns + 1 == target_n_insns)
|
/* INSN is the last not-scheduled instruction in the current block. */
|
/* INSN is the last not-scheduled instruction in the current block. */
|
return true;
|
return true;
|
|
|
return false;
|
return false;
|
}
|
}
|
|
|
/* Used in schedule_insns to initialize current_sched_info for scheduling
|
/* Used in schedule_insns to initialize current_sched_info for scheduling
|
regions (or single basic blocks). */
|
regions (or single basic blocks). */
|
|
|
static const struct haifa_sched_info rgn_const_sched_info =
|
static const struct haifa_sched_info rgn_const_sched_info =
|
{
|
{
|
init_ready_list,
|
init_ready_list,
|
can_schedule_ready_p,
|
can_schedule_ready_p,
|
schedule_more_p,
|
schedule_more_p,
|
new_ready,
|
new_ready,
|
rgn_rank,
|
rgn_rank,
|
rgn_print_insn,
|
rgn_print_insn,
|
contributes_to_priority,
|
contributes_to_priority,
|
rgn_insn_finishes_block_p,
|
rgn_insn_finishes_block_p,
|
|
|
NULL, NULL,
|
NULL, NULL,
|
NULL, NULL,
|
NULL, NULL,
|
0, 0,
|
0, 0,
|
|
|
rgn_add_remove_insn,
|
rgn_add_remove_insn,
|
begin_schedule_ready,
|
begin_schedule_ready,
|
advance_target_bb,
|
advance_target_bb,
|
SCHED_RGN
|
SCHED_RGN
|
};
|
};
|
|
|
/* This variable holds the data and hooks needed to the Haifa scheduler backend
|
/* This variable holds the data and hooks needed to the Haifa scheduler backend
|
for the interblock scheduler frontend. */
|
for the interblock scheduler frontend. */
|
static struct haifa_sched_info rgn_sched_info;
|
static struct haifa_sched_info rgn_sched_info;
|
|
|
/* Returns maximum priority that an insn was assigned to. */
|
/* Returns maximum priority that an insn was assigned to. */
|
|
|
int
|
int
|
get_rgn_sched_max_insns_priority (void)
|
get_rgn_sched_max_insns_priority (void)
|
{
|
{
|
return rgn_sched_info.sched_max_insns_priority;
|
return rgn_sched_info.sched_max_insns_priority;
|
}
|
}
|
|
|
/* Determine if PAT sets a CLASS_LIKELY_SPILLED_P register. */
|
/* Determine if PAT sets a CLASS_LIKELY_SPILLED_P register. */
|
|
|
static bool
|
static bool
|
sets_likely_spilled (rtx pat)
|
sets_likely_spilled (rtx pat)
|
{
|
{
|
bool ret = false;
|
bool ret = false;
|
note_stores (pat, sets_likely_spilled_1, &ret);
|
note_stores (pat, sets_likely_spilled_1, &ret);
|
return ret;
|
return ret;
|
}
|
}
|
|
|
static void
|
static void
|
sets_likely_spilled_1 (rtx x, const_rtx pat, void *data)
|
sets_likely_spilled_1 (rtx x, const_rtx pat, void *data)
|
{
|
{
|
bool *ret = (bool *) data;
|
bool *ret = (bool *) data;
|
|
|
if (GET_CODE (pat) == SET
|
if (GET_CODE (pat) == SET
|
&& REG_P (x)
|
&& REG_P (x)
|
&& REGNO (x) < FIRST_PSEUDO_REGISTER
|
&& REGNO (x) < FIRST_PSEUDO_REGISTER
|
&& CLASS_LIKELY_SPILLED_P (REGNO_REG_CLASS (REGNO (x))))
|
&& CLASS_LIKELY_SPILLED_P (REGNO_REG_CLASS (REGNO (x))))
|
*ret = true;
|
*ret = true;
|
}
|
}
|
|
|
/* A bitmap to note insns that participate in any dependency. Used in
|
/* A bitmap to note insns that participate in any dependency. Used in
|
add_branch_dependences. */
|
add_branch_dependences. */
|
static sbitmap insn_referenced;
|
static sbitmap insn_referenced;
|
|
|
/* Add dependences so that branches are scheduled to run last in their
|
/* Add dependences so that branches are scheduled to run last in their
|
block. */
|
block. */
|
static void
|
static void
|
add_branch_dependences (rtx head, rtx tail)
|
add_branch_dependences (rtx head, rtx tail)
|
{
|
{
|
rtx insn, last;
|
rtx insn, last;
|
|
|
/* For all branches, calls, uses, clobbers, cc0 setters, and instructions
|
/* For all branches, calls, uses, clobbers, cc0 setters, and instructions
|
that can throw exceptions, force them to remain in order at the end of
|
that can throw exceptions, force them to remain in order at the end of
|
the block by adding dependencies and giving the last a high priority.
|
the block by adding dependencies and giving the last a high priority.
|
There may be notes present, and prev_head may also be a note.
|
There may be notes present, and prev_head may also be a note.
|
|
|
Branches must obviously remain at the end. Calls should remain at the
|
Branches must obviously remain at the end. Calls should remain at the
|
end since moving them results in worse register allocation. Uses remain
|
end since moving them results in worse register allocation. Uses remain
|
at the end to ensure proper register allocation.
|
at the end to ensure proper register allocation.
|
|
|
cc0 setters remain at the end because they can't be moved away from
|
cc0 setters remain at the end because they can't be moved away from
|
their cc0 user.
|
their cc0 user.
|
|
|
COND_EXEC insns cannot be moved past a branch (see e.g. PR17808).
|
COND_EXEC insns cannot be moved past a branch (see e.g. PR17808).
|
|
|
Insns setting CLASS_LIKELY_SPILLED_P registers (usually return values)
|
Insns setting CLASS_LIKELY_SPILLED_P registers (usually return values)
|
are not moved before reload because we can wind up with register
|
are not moved before reload because we can wind up with register
|
allocation failures. */
|
allocation failures. */
|
|
|
while (tail != head && DEBUG_INSN_P (tail))
|
while (tail != head && DEBUG_INSN_P (tail))
|
tail = PREV_INSN (tail);
|
tail = PREV_INSN (tail);
|
|
|
insn = tail;
|
insn = tail;
|
last = 0;
|
last = 0;
|
while (CALL_P (insn)
|
while (CALL_P (insn)
|
|| JUMP_P (insn)
|
|| JUMP_P (insn)
|
|| (NONJUMP_INSN_P (insn)
|
|| (NONJUMP_INSN_P (insn)
|
&& (GET_CODE (PATTERN (insn)) == USE
|
&& (GET_CODE (PATTERN (insn)) == USE
|
|| GET_CODE (PATTERN (insn)) == CLOBBER
|
|| GET_CODE (PATTERN (insn)) == CLOBBER
|
|| can_throw_internal (insn)
|
|| can_throw_internal (insn)
|
#ifdef HAVE_cc0
|
#ifdef HAVE_cc0
|
|| sets_cc0_p (PATTERN (insn))
|
|| sets_cc0_p (PATTERN (insn))
|
#endif
|
#endif
|
|| (!reload_completed
|
|| (!reload_completed
|
&& sets_likely_spilled (PATTERN (insn)))))
|
&& sets_likely_spilled (PATTERN (insn)))))
|
|| NOTE_P (insn))
|
|| NOTE_P (insn))
|
{
|
{
|
if (!NOTE_P (insn))
|
if (!NOTE_P (insn))
|
{
|
{
|
if (last != 0
|
if (last != 0
|
&& sd_find_dep_between (insn, last, false) == NULL)
|
&& sd_find_dep_between (insn, last, false) == NULL)
|
{
|
{
|
if (! sched_insns_conditions_mutex_p (last, insn))
|
if (! sched_insns_conditions_mutex_p (last, insn))
|
add_dependence (last, insn, REG_DEP_ANTI);
|
add_dependence (last, insn, REG_DEP_ANTI);
|
SET_BIT (insn_referenced, INSN_LUID (insn));
|
SET_BIT (insn_referenced, INSN_LUID (insn));
|
}
|
}
|
|
|
CANT_MOVE (insn) = 1;
|
CANT_MOVE (insn) = 1;
|
|
|
last = insn;
|
last = insn;
|
}
|
}
|
|
|
/* Don't overrun the bounds of the basic block. */
|
/* Don't overrun the bounds of the basic block. */
|
if (insn == head)
|
if (insn == head)
|
break;
|
break;
|
|
|
do
|
do
|
insn = PREV_INSN (insn);
|
insn = PREV_INSN (insn);
|
while (insn != head && DEBUG_INSN_P (insn));
|
while (insn != head && DEBUG_INSN_P (insn));
|
}
|
}
|
|
|
/* Make sure these insns are scheduled last in their block. */
|
/* Make sure these insns are scheduled last in their block. */
|
insn = last;
|
insn = last;
|
if (insn != 0)
|
if (insn != 0)
|
while (insn != head)
|
while (insn != head)
|
{
|
{
|
insn = prev_nonnote_insn (insn);
|
insn = prev_nonnote_insn (insn);
|
|
|
if (TEST_BIT (insn_referenced, INSN_LUID (insn))
|
if (TEST_BIT (insn_referenced, INSN_LUID (insn))
|
|| DEBUG_INSN_P (insn))
|
|| DEBUG_INSN_P (insn))
|
continue;
|
continue;
|
|
|
if (! sched_insns_conditions_mutex_p (last, insn))
|
if (! sched_insns_conditions_mutex_p (last, insn))
|
add_dependence (last, insn, REG_DEP_ANTI);
|
add_dependence (last, insn, REG_DEP_ANTI);
|
}
|
}
|
|
|
if (!targetm.have_conditional_execution ())
|
if (!targetm.have_conditional_execution ())
|
return;
|
return;
|
|
|
/* Finally, if the block ends in a jump, and we are doing intra-block
|
/* Finally, if the block ends in a jump, and we are doing intra-block
|
scheduling, make sure that the branch depends on any COND_EXEC insns
|
scheduling, make sure that the branch depends on any COND_EXEC insns
|
inside the block to avoid moving the COND_EXECs past the branch insn.
|
inside the block to avoid moving the COND_EXECs past the branch insn.
|
|
|
We only have to do this after reload, because (1) before reload there
|
We only have to do this after reload, because (1) before reload there
|
are no COND_EXEC insns, and (2) the region scheduler is an intra-block
|
are no COND_EXEC insns, and (2) the region scheduler is an intra-block
|
scheduler after reload.
|
scheduler after reload.
|
|
|
FIXME: We could in some cases move COND_EXEC insns past the branch if
|
FIXME: We could in some cases move COND_EXEC insns past the branch if
|
this scheduler would be a little smarter. Consider this code:
|
this scheduler would be a little smarter. Consider this code:
|
|
|
T = [addr]
|
T = [addr]
|
C ? addr += 4
|
C ? addr += 4
|
!C ? X += 12
|
!C ? X += 12
|
C ? T += 1
|
C ? T += 1
|
C ? jump foo
|
C ? jump foo
|
|
|
On a target with a one cycle stall on a memory access the optimal
|
On a target with a one cycle stall on a memory access the optimal
|
sequence would be:
|
sequence would be:
|
|
|
T = [addr]
|
T = [addr]
|
C ? addr += 4
|
C ? addr += 4
|
C ? T += 1
|
C ? T += 1
|
C ? jump foo
|
C ? jump foo
|
!C ? X += 12
|
!C ? X += 12
|
|
|
We don't want to put the 'X += 12' before the branch because it just
|
We don't want to put the 'X += 12' before the branch because it just
|
wastes a cycle of execution time when the branch is taken.
|
wastes a cycle of execution time when the branch is taken.
|
|
|
Note that in the example "!C" will always be true. That is another
|
Note that in the example "!C" will always be true. That is another
|
possible improvement for handling COND_EXECs in this scheduler: it
|
possible improvement for handling COND_EXECs in this scheduler: it
|
could remove always-true predicates. */
|
could remove always-true predicates. */
|
|
|
if (!reload_completed || ! JUMP_P (tail))
|
if (!reload_completed || ! JUMP_P (tail))
|
return;
|
return;
|
|
|
insn = tail;
|
insn = tail;
|
while (insn != head)
|
while (insn != head)
|
{
|
{
|
insn = PREV_INSN (insn);
|
insn = PREV_INSN (insn);
|
|
|
/* Note that we want to add this dependency even when
|
/* Note that we want to add this dependency even when
|
sched_insns_conditions_mutex_p returns true. The whole point
|
sched_insns_conditions_mutex_p returns true. The whole point
|
is that we _want_ this dependency, even if these insns really
|
is that we _want_ this dependency, even if these insns really
|
are independent. */
|
are independent. */
|
if (INSN_P (insn) && GET_CODE (PATTERN (insn)) == COND_EXEC)
|
if (INSN_P (insn) && GET_CODE (PATTERN (insn)) == COND_EXEC)
|
add_dependence (tail, insn, REG_DEP_ANTI);
|
add_dependence (tail, insn, REG_DEP_ANTI);
|
}
|
}
|
}
|
}
|
|
|
/* Data structures for the computation of data dependences in a regions. We
|
/* Data structures for the computation of data dependences in a regions. We
|
keep one `deps' structure for every basic block. Before analyzing the
|
keep one `deps' structure for every basic block. Before analyzing the
|
data dependences for a bb, its variables are initialized as a function of
|
data dependences for a bb, its variables are initialized as a function of
|
the variables of its predecessors. When the analysis for a bb completes,
|
the variables of its predecessors. When the analysis for a bb completes,
|
we save the contents to the corresponding bb_deps[bb] variable. */
|
we save the contents to the corresponding bb_deps[bb] variable. */
|
|
|
static struct deps_desc *bb_deps;
|
static struct deps_desc *bb_deps;
|
|
|
/* Duplicate the INSN_LIST elements of COPY and prepend them to OLD. */
|
/* Duplicate the INSN_LIST elements of COPY and prepend them to OLD. */
|
|
|
static rtx
|
static rtx
|
concat_INSN_LIST (rtx copy, rtx old)
|
concat_INSN_LIST (rtx copy, rtx old)
|
{
|
{
|
rtx new_rtx = old;
|
rtx new_rtx = old;
|
for (; copy ; copy = XEXP (copy, 1))
|
for (; copy ; copy = XEXP (copy, 1))
|
new_rtx = alloc_INSN_LIST (XEXP (copy, 0), new_rtx);
|
new_rtx = alloc_INSN_LIST (XEXP (copy, 0), new_rtx);
|
return new_rtx;
|
return new_rtx;
|
}
|
}
|
|
|
static void
|
static void
|
concat_insn_mem_list (rtx copy_insns, rtx copy_mems, rtx *old_insns_p,
|
concat_insn_mem_list (rtx copy_insns, rtx copy_mems, rtx *old_insns_p,
|
rtx *old_mems_p)
|
rtx *old_mems_p)
|
{
|
{
|
rtx new_insns = *old_insns_p;
|
rtx new_insns = *old_insns_p;
|
rtx new_mems = *old_mems_p;
|
rtx new_mems = *old_mems_p;
|
|
|
while (copy_insns)
|
while (copy_insns)
|
{
|
{
|
new_insns = alloc_INSN_LIST (XEXP (copy_insns, 0), new_insns);
|
new_insns = alloc_INSN_LIST (XEXP (copy_insns, 0), new_insns);
|
new_mems = alloc_EXPR_LIST (VOIDmode, XEXP (copy_mems, 0), new_mems);
|
new_mems = alloc_EXPR_LIST (VOIDmode, XEXP (copy_mems, 0), new_mems);
|
copy_insns = XEXP (copy_insns, 1);
|
copy_insns = XEXP (copy_insns, 1);
|
copy_mems = XEXP (copy_mems, 1);
|
copy_mems = XEXP (copy_mems, 1);
|
}
|
}
|
|
|
*old_insns_p = new_insns;
|
*old_insns_p = new_insns;
|
*old_mems_p = new_mems;
|
*old_mems_p = new_mems;
|
}
|
}
|
|
|
/* Join PRED_DEPS to the SUCC_DEPS. */
|
/* Join PRED_DEPS to the SUCC_DEPS. */
|
void
|
void
|
deps_join (struct deps_desc *succ_deps, struct deps_desc *pred_deps)
|
deps_join (struct deps_desc *succ_deps, struct deps_desc *pred_deps)
|
{
|
{
|
unsigned reg;
|
unsigned reg;
|
reg_set_iterator rsi;
|
reg_set_iterator rsi;
|
|
|
/* The reg_last lists are inherited by successor. */
|
/* The reg_last lists are inherited by successor. */
|
EXECUTE_IF_SET_IN_REG_SET (&pred_deps->reg_last_in_use, 0, reg, rsi)
|
EXECUTE_IF_SET_IN_REG_SET (&pred_deps->reg_last_in_use, 0, reg, rsi)
|
{
|
{
|
struct deps_reg *pred_rl = &pred_deps->reg_last[reg];
|
struct deps_reg *pred_rl = &pred_deps->reg_last[reg];
|
struct deps_reg *succ_rl = &succ_deps->reg_last[reg];
|
struct deps_reg *succ_rl = &succ_deps->reg_last[reg];
|
|
|
succ_rl->uses = concat_INSN_LIST (pred_rl->uses, succ_rl->uses);
|
succ_rl->uses = concat_INSN_LIST (pred_rl->uses, succ_rl->uses);
|
succ_rl->sets = concat_INSN_LIST (pred_rl->sets, succ_rl->sets);
|
succ_rl->sets = concat_INSN_LIST (pred_rl->sets, succ_rl->sets);
|
succ_rl->implicit_sets
|
succ_rl->implicit_sets
|
= concat_INSN_LIST (pred_rl->implicit_sets, succ_rl->implicit_sets);
|
= concat_INSN_LIST (pred_rl->implicit_sets, succ_rl->implicit_sets);
|
succ_rl->clobbers = concat_INSN_LIST (pred_rl->clobbers,
|
succ_rl->clobbers = concat_INSN_LIST (pred_rl->clobbers,
|
succ_rl->clobbers);
|
succ_rl->clobbers);
|
succ_rl->uses_length += pred_rl->uses_length;
|
succ_rl->uses_length += pred_rl->uses_length;
|
succ_rl->clobbers_length += pred_rl->clobbers_length;
|
succ_rl->clobbers_length += pred_rl->clobbers_length;
|
}
|
}
|
IOR_REG_SET (&succ_deps->reg_last_in_use, &pred_deps->reg_last_in_use);
|
IOR_REG_SET (&succ_deps->reg_last_in_use, &pred_deps->reg_last_in_use);
|
|
|
/* Mem read/write lists are inherited by successor. */
|
/* Mem read/write lists are inherited by successor. */
|
concat_insn_mem_list (pred_deps->pending_read_insns,
|
concat_insn_mem_list (pred_deps->pending_read_insns,
|
pred_deps->pending_read_mems,
|
pred_deps->pending_read_mems,
|
&succ_deps->pending_read_insns,
|
&succ_deps->pending_read_insns,
|
&succ_deps->pending_read_mems);
|
&succ_deps->pending_read_mems);
|
concat_insn_mem_list (pred_deps->pending_write_insns,
|
concat_insn_mem_list (pred_deps->pending_write_insns,
|
pred_deps->pending_write_mems,
|
pred_deps->pending_write_mems,
|
&succ_deps->pending_write_insns,
|
&succ_deps->pending_write_insns,
|
&succ_deps->pending_write_mems);
|
&succ_deps->pending_write_mems);
|
|
|
succ_deps->last_pending_memory_flush
|
succ_deps->last_pending_memory_flush
|
= concat_INSN_LIST (pred_deps->last_pending_memory_flush,
|
= concat_INSN_LIST (pred_deps->last_pending_memory_flush,
|
succ_deps->last_pending_memory_flush);
|
succ_deps->last_pending_memory_flush);
|
|
|
succ_deps->pending_read_list_length += pred_deps->pending_read_list_length;
|
succ_deps->pending_read_list_length += pred_deps->pending_read_list_length;
|
succ_deps->pending_write_list_length += pred_deps->pending_write_list_length;
|
succ_deps->pending_write_list_length += pred_deps->pending_write_list_length;
|
succ_deps->pending_flush_length += pred_deps->pending_flush_length;
|
succ_deps->pending_flush_length += pred_deps->pending_flush_length;
|
|
|
/* last_function_call is inherited by successor. */
|
/* last_function_call is inherited by successor. */
|
succ_deps->last_function_call
|
succ_deps->last_function_call
|
= concat_INSN_LIST (pred_deps->last_function_call,
|
= concat_INSN_LIST (pred_deps->last_function_call,
|
succ_deps->last_function_call);
|
succ_deps->last_function_call);
|
|
|
/* last_function_call_may_noreturn is inherited by successor. */
|
/* last_function_call_may_noreturn is inherited by successor. */
|
succ_deps->last_function_call_may_noreturn
|
succ_deps->last_function_call_may_noreturn
|
= concat_INSN_LIST (pred_deps->last_function_call_may_noreturn,
|
= concat_INSN_LIST (pred_deps->last_function_call_may_noreturn,
|
succ_deps->last_function_call_may_noreturn);
|
succ_deps->last_function_call_may_noreturn);
|
|
|
/* sched_before_next_call is inherited by successor. */
|
/* sched_before_next_call is inherited by successor. */
|
succ_deps->sched_before_next_call
|
succ_deps->sched_before_next_call
|
= concat_INSN_LIST (pred_deps->sched_before_next_call,
|
= concat_INSN_LIST (pred_deps->sched_before_next_call,
|
succ_deps->sched_before_next_call);
|
succ_deps->sched_before_next_call);
|
}
|
}
|
|
|
/* After computing the dependencies for block BB, propagate the dependencies
|
/* After computing the dependencies for block BB, propagate the dependencies
|
found in TMP_DEPS to the successors of the block. */
|
found in TMP_DEPS to the successors of the block. */
|
static void
|
static void
|
propagate_deps (int bb, struct deps_desc *pred_deps)
|
propagate_deps (int bb, struct deps_desc *pred_deps)
|
{
|
{
|
basic_block block = BASIC_BLOCK (BB_TO_BLOCK (bb));
|
basic_block block = BASIC_BLOCK (BB_TO_BLOCK (bb));
|
edge_iterator ei;
|
edge_iterator ei;
|
edge e;
|
edge e;
|
|
|
/* bb's structures are inherited by its successors. */
|
/* bb's structures are inherited by its successors. */
|
FOR_EACH_EDGE (e, ei, block->succs)
|
FOR_EACH_EDGE (e, ei, block->succs)
|
{
|
{
|
/* Only bbs "below" bb, in the same region, are interesting. */
|
/* Only bbs "below" bb, in the same region, are interesting. */
|
if (e->dest == EXIT_BLOCK_PTR
|
if (e->dest == EXIT_BLOCK_PTR
|
|| CONTAINING_RGN (block->index) != CONTAINING_RGN (e->dest->index)
|
|| CONTAINING_RGN (block->index) != CONTAINING_RGN (e->dest->index)
|
|| BLOCK_TO_BB (e->dest->index) <= bb)
|
|| BLOCK_TO_BB (e->dest->index) <= bb)
|
continue;
|
continue;
|
|
|
deps_join (bb_deps + BLOCK_TO_BB (e->dest->index), pred_deps);
|
deps_join (bb_deps + BLOCK_TO_BB (e->dest->index), pred_deps);
|
}
|
}
|
|
|
/* These lists should point to the right place, for correct
|
/* These lists should point to the right place, for correct
|
freeing later. */
|
freeing later. */
|
bb_deps[bb].pending_read_insns = pred_deps->pending_read_insns;
|
bb_deps[bb].pending_read_insns = pred_deps->pending_read_insns;
|
bb_deps[bb].pending_read_mems = pred_deps->pending_read_mems;
|
bb_deps[bb].pending_read_mems = pred_deps->pending_read_mems;
|
bb_deps[bb].pending_write_insns = pred_deps->pending_write_insns;
|
bb_deps[bb].pending_write_insns = pred_deps->pending_write_insns;
|
bb_deps[bb].pending_write_mems = pred_deps->pending_write_mems;
|
bb_deps[bb].pending_write_mems = pred_deps->pending_write_mems;
|
|
|
/* Can't allow these to be freed twice. */
|
/* Can't allow these to be freed twice. */
|
pred_deps->pending_read_insns = 0;
|
pred_deps->pending_read_insns = 0;
|
pred_deps->pending_read_mems = 0;
|
pred_deps->pending_read_mems = 0;
|
pred_deps->pending_write_insns = 0;
|
pred_deps->pending_write_insns = 0;
|
pred_deps->pending_write_mems = 0;
|
pred_deps->pending_write_mems = 0;
|
}
|
}
|
|
|
/* Compute dependences inside bb. In a multiple blocks region:
|
/* Compute dependences inside bb. In a multiple blocks region:
|
(1) a bb is analyzed after its predecessors, and (2) the lists in
|
(1) a bb is analyzed after its predecessors, and (2) the lists in
|
effect at the end of bb (after analyzing for bb) are inherited by
|
effect at the end of bb (after analyzing for bb) are inherited by
|
bb's successors.
|
bb's successors.
|
|
|
Specifically for reg-reg data dependences, the block insns are
|
Specifically for reg-reg data dependences, the block insns are
|
scanned by sched_analyze () top-to-bottom. Three lists are
|
scanned by sched_analyze () top-to-bottom. Three lists are
|
maintained by sched_analyze (): reg_last[].sets for register DEFs,
|
maintained by sched_analyze (): reg_last[].sets for register DEFs,
|
reg_last[].implicit_sets for implicit hard register DEFs, and
|
reg_last[].implicit_sets for implicit hard register DEFs, and
|
reg_last[].uses for register USEs.
|
reg_last[].uses for register USEs.
|
|
|
When analysis is completed for bb, we update for its successors:
|
When analysis is completed for bb, we update for its successors:
|
; - DEFS[succ] = Union (DEFS [succ], DEFS [bb])
|
; - DEFS[succ] = Union (DEFS [succ], DEFS [bb])
|
; - IMPLICIT_DEFS[succ] = Union (IMPLICIT_DEFS [succ], IMPLICIT_DEFS [bb])
|
; - IMPLICIT_DEFS[succ] = Union (IMPLICIT_DEFS [succ], IMPLICIT_DEFS [bb])
|
; - USES[succ] = Union (USES [succ], DEFS [bb])
|
; - USES[succ] = Union (USES [succ], DEFS [bb])
|
|
|
The mechanism for computing mem-mem data dependence is very
|
The mechanism for computing mem-mem data dependence is very
|
similar, and the result is interblock dependences in the region. */
|
similar, and the result is interblock dependences in the region. */
|
|
|
static void
|
static void
|
compute_block_dependences (int bb)
|
compute_block_dependences (int bb)
|
{
|
{
|
rtx head, tail;
|
rtx head, tail;
|
struct deps_desc tmp_deps;
|
struct deps_desc tmp_deps;
|
|
|
tmp_deps = bb_deps[bb];
|
tmp_deps = bb_deps[bb];
|
|
|
/* Do the analysis for this block. */
|
/* Do the analysis for this block. */
|
gcc_assert (EBB_FIRST_BB (bb) == EBB_LAST_BB (bb));
|
gcc_assert (EBB_FIRST_BB (bb) == EBB_LAST_BB (bb));
|
get_ebb_head_tail (EBB_FIRST_BB (bb), EBB_LAST_BB (bb), &head, &tail);
|
get_ebb_head_tail (EBB_FIRST_BB (bb), EBB_LAST_BB (bb), &head, &tail);
|
|
|
sched_analyze (&tmp_deps, head, tail);
|
sched_analyze (&tmp_deps, head, tail);
|
|
|
/* Selective scheduling handles control dependencies by itself. */
|
/* Selective scheduling handles control dependencies by itself. */
|
if (!sel_sched_p ())
|
if (!sel_sched_p ())
|
add_branch_dependences (head, tail);
|
add_branch_dependences (head, tail);
|
|
|
if (current_nr_blocks > 1)
|
if (current_nr_blocks > 1)
|
propagate_deps (bb, &tmp_deps);
|
propagate_deps (bb, &tmp_deps);
|
|
|
/* Free up the INSN_LISTs. */
|
/* Free up the INSN_LISTs. */
|
free_deps (&tmp_deps);
|
free_deps (&tmp_deps);
|
|
|
if (targetm.sched.dependencies_evaluation_hook)
|
if (targetm.sched.dependencies_evaluation_hook)
|
targetm.sched.dependencies_evaluation_hook (head, tail);
|
targetm.sched.dependencies_evaluation_hook (head, tail);
|
}
|
}
|
|
|
/* Free dependencies of instructions inside BB. */
|
/* Free dependencies of instructions inside BB. */
|
static void
|
static void
|
free_block_dependencies (int bb)
|
free_block_dependencies (int bb)
|
{
|
{
|
rtx head;
|
rtx head;
|
rtx tail;
|
rtx tail;
|
|
|
get_ebb_head_tail (EBB_FIRST_BB (bb), EBB_LAST_BB (bb), &head, &tail);
|
get_ebb_head_tail (EBB_FIRST_BB (bb), EBB_LAST_BB (bb), &head, &tail);
|
|
|
if (no_real_insns_p (head, tail))
|
if (no_real_insns_p (head, tail))
|
return;
|
return;
|
|
|
sched_free_deps (head, tail, true);
|
sched_free_deps (head, tail, true);
|
}
|
}
|
|
|
/* Remove all INSN_LISTs and EXPR_LISTs from the pending lists and add
|
/* Remove all INSN_LISTs and EXPR_LISTs from the pending lists and add
|
them to the unused_*_list variables, so that they can be reused. */
|
them to the unused_*_list variables, so that they can be reused. */
|
|
|
static void
|
static void
|
free_pending_lists (void)
|
free_pending_lists (void)
|
{
|
{
|
int bb;
|
int bb;
|
|
|
for (bb = 0; bb < current_nr_blocks; bb++)
|
for (bb = 0; bb < current_nr_blocks; bb++)
|
{
|
{
|
free_INSN_LIST_list (&bb_deps[bb].pending_read_insns);
|
free_INSN_LIST_list (&bb_deps[bb].pending_read_insns);
|
free_INSN_LIST_list (&bb_deps[bb].pending_write_insns);
|
free_INSN_LIST_list (&bb_deps[bb].pending_write_insns);
|
free_EXPR_LIST_list (&bb_deps[bb].pending_read_mems);
|
free_EXPR_LIST_list (&bb_deps[bb].pending_read_mems);
|
free_EXPR_LIST_list (&bb_deps[bb].pending_write_mems);
|
free_EXPR_LIST_list (&bb_deps[bb].pending_write_mems);
|
}
|
}
|
}
|
}
|
|
|
/* Print dependences for debugging starting from FROM_BB.
|
/* Print dependences for debugging starting from FROM_BB.
|
Callable from debugger. */
|
Callable from debugger. */
|
/* Print dependences for debugging starting from FROM_BB.
|
/* Print dependences for debugging starting from FROM_BB.
|
Callable from debugger. */
|
Callable from debugger. */
|
void
|
void
|
debug_rgn_dependencies (int from_bb)
|
debug_rgn_dependencies (int from_bb)
|
{
|
{
|
int bb;
|
int bb;
|
|
|
fprintf (sched_dump,
|
fprintf (sched_dump,
|
";; --------------- forward dependences: ------------ \n");
|
";; --------------- forward dependences: ------------ \n");
|
|
|
for (bb = from_bb; bb < current_nr_blocks; bb++)
|
for (bb = from_bb; bb < current_nr_blocks; bb++)
|
{
|
{
|
rtx head, tail;
|
rtx head, tail;
|
|
|
get_ebb_head_tail (EBB_FIRST_BB (bb), EBB_LAST_BB (bb), &head, &tail);
|
get_ebb_head_tail (EBB_FIRST_BB (bb), EBB_LAST_BB (bb), &head, &tail);
|
fprintf (sched_dump, "\n;; --- Region Dependences --- b %d bb %d \n",
|
fprintf (sched_dump, "\n;; --- Region Dependences --- b %d bb %d \n",
|
BB_TO_BLOCK (bb), bb);
|
BB_TO_BLOCK (bb), bb);
|
|
|
debug_dependencies (head, tail);
|
debug_dependencies (head, tail);
|
}
|
}
|
}
|
}
|
|
|
/* Print dependencies information for instructions between HEAD and TAIL.
|
/* Print dependencies information for instructions between HEAD and TAIL.
|
??? This function would probably fit best in haifa-sched.c. */
|
??? This function would probably fit best in haifa-sched.c. */
|
void debug_dependencies (rtx head, rtx tail)
|
void debug_dependencies (rtx head, rtx tail)
|
{
|
{
|
rtx insn;
|
rtx insn;
|
rtx next_tail = NEXT_INSN (tail);
|
rtx next_tail = NEXT_INSN (tail);
|
|
|
fprintf (sched_dump, ";; %7s%6s%6s%6s%6s%6s%14s\n",
|
fprintf (sched_dump, ";; %7s%6s%6s%6s%6s%6s%14s\n",
|
"insn", "code", "bb", "dep", "prio", "cost",
|
"insn", "code", "bb", "dep", "prio", "cost",
|
"reservation");
|
"reservation");
|
fprintf (sched_dump, ";; %7s%6s%6s%6s%6s%6s%14s\n",
|
fprintf (sched_dump, ";; %7s%6s%6s%6s%6s%6s%14s\n",
|
"----", "----", "--", "---", "----", "----",
|
"----", "----", "--", "---", "----", "----",
|
"-----------");
|
"-----------");
|
|
|
for (insn = head; insn != next_tail; insn = NEXT_INSN (insn))
|
for (insn = head; insn != next_tail; insn = NEXT_INSN (insn))
|
{
|
{
|
if (! INSN_P (insn))
|
if (! INSN_P (insn))
|
{
|
{
|
int n;
|
int n;
|
fprintf (sched_dump, ";; %6d ", INSN_UID (insn));
|
fprintf (sched_dump, ";; %6d ", INSN_UID (insn));
|
if (NOTE_P (insn))
|
if (NOTE_P (insn))
|
{
|
{
|
n = NOTE_KIND (insn);
|
n = NOTE_KIND (insn);
|
fprintf (sched_dump, "%s\n", GET_NOTE_INSN_NAME (n));
|
fprintf (sched_dump, "%s\n", GET_NOTE_INSN_NAME (n));
|
}
|
}
|
else
|
else
|
fprintf (sched_dump, " {%s}\n", GET_RTX_NAME (GET_CODE (insn)));
|
fprintf (sched_dump, " {%s}\n", GET_RTX_NAME (GET_CODE (insn)));
|
continue;
|
continue;
|
}
|
}
|
|
|
fprintf (sched_dump,
|
fprintf (sched_dump,
|
";; %s%5d%6d%6d%6d%6d%6d ",
|
";; %s%5d%6d%6d%6d%6d%6d ",
|
(SCHED_GROUP_P (insn) ? "+" : " "),
|
(SCHED_GROUP_P (insn) ? "+" : " "),
|
INSN_UID (insn),
|
INSN_UID (insn),
|
INSN_CODE (insn),
|
INSN_CODE (insn),
|
BLOCK_NUM (insn),
|
BLOCK_NUM (insn),
|
sched_emulate_haifa_p ? -1 : sd_lists_size (insn, SD_LIST_BACK),
|
sched_emulate_haifa_p ? -1 : sd_lists_size (insn, SD_LIST_BACK),
|
(sel_sched_p () ? (sched_emulate_haifa_p ? -1
|
(sel_sched_p () ? (sched_emulate_haifa_p ? -1
|
: INSN_PRIORITY (insn))
|
: INSN_PRIORITY (insn))
|
: INSN_PRIORITY (insn)),
|
: INSN_PRIORITY (insn)),
|
(sel_sched_p () ? (sched_emulate_haifa_p ? -1
|
(sel_sched_p () ? (sched_emulate_haifa_p ? -1
|
: insn_cost (insn))
|
: insn_cost (insn))
|
: insn_cost (insn)));
|
: insn_cost (insn)));
|
|
|
if (recog_memoized (insn) < 0)
|
if (recog_memoized (insn) < 0)
|
fprintf (sched_dump, "nothing");
|
fprintf (sched_dump, "nothing");
|
else
|
else
|
print_reservation (sched_dump, insn);
|
print_reservation (sched_dump, insn);
|
|
|
fprintf (sched_dump, "\t: ");
|
fprintf (sched_dump, "\t: ");
|
{
|
{
|
sd_iterator_def sd_it;
|
sd_iterator_def sd_it;
|
dep_t dep;
|
dep_t dep;
|
|
|
FOR_EACH_DEP (insn, SD_LIST_FORW, sd_it, dep)
|
FOR_EACH_DEP (insn, SD_LIST_FORW, sd_it, dep)
|
fprintf (sched_dump, "%d ", INSN_UID (DEP_CON (dep)));
|
fprintf (sched_dump, "%d ", INSN_UID (DEP_CON (dep)));
|
}
|
}
|
fprintf (sched_dump, "\n");
|
fprintf (sched_dump, "\n");
|
}
|
}
|
|
|
fprintf (sched_dump, "\n");
|
fprintf (sched_dump, "\n");
|
}
|
}
|
|
|
/* Returns true if all the basic blocks of the current region have
|
/* Returns true if all the basic blocks of the current region have
|
NOTE_DISABLE_SCHED_OF_BLOCK which means not to schedule that region. */
|
NOTE_DISABLE_SCHED_OF_BLOCK which means not to schedule that region. */
|
bool
|
bool
|
sched_is_disabled_for_current_region_p (void)
|
sched_is_disabled_for_current_region_p (void)
|
{
|
{
|
int bb;
|
int bb;
|
|
|
for (bb = 0; bb < current_nr_blocks; bb++)
|
for (bb = 0; bb < current_nr_blocks; bb++)
|
if (!(BASIC_BLOCK (BB_TO_BLOCK (bb))->flags & BB_DISABLE_SCHEDULE))
|
if (!(BASIC_BLOCK (BB_TO_BLOCK (bb))->flags & BB_DISABLE_SCHEDULE))
|
return false;
|
return false;
|
|
|
return true;
|
return true;
|
}
|
}
|
|
|
/* Free all region dependencies saved in INSN_BACK_DEPS and
|
/* Free all region dependencies saved in INSN_BACK_DEPS and
|
INSN_RESOLVED_BACK_DEPS. The Haifa scheduler does this on the fly
|
INSN_RESOLVED_BACK_DEPS. The Haifa scheduler does this on the fly
|
when scheduling, so this function is supposed to be called from
|
when scheduling, so this function is supposed to be called from
|
the selective scheduling only. */
|
the selective scheduling only. */
|
void
|
void
|
free_rgn_deps (void)
|
free_rgn_deps (void)
|
{
|
{
|
int bb;
|
int bb;
|
|
|
for (bb = 0; bb < current_nr_blocks; bb++)
|
for (bb = 0; bb < current_nr_blocks; bb++)
|
{
|
{
|
rtx head, tail;
|
rtx head, tail;
|
|
|
gcc_assert (EBB_FIRST_BB (bb) == EBB_LAST_BB (bb));
|
gcc_assert (EBB_FIRST_BB (bb) == EBB_LAST_BB (bb));
|
get_ebb_head_tail (EBB_FIRST_BB (bb), EBB_LAST_BB (bb), &head, &tail);
|
get_ebb_head_tail (EBB_FIRST_BB (bb), EBB_LAST_BB (bb), &head, &tail);
|
|
|
sched_free_deps (head, tail, false);
|
sched_free_deps (head, tail, false);
|
}
|
}
|
}
|
}
|
|
|
static int rgn_n_insns;
|
static int rgn_n_insns;
|
|
|
/* Compute insn priority for a current region. */
|
/* Compute insn priority for a current region. */
|
void
|
void
|
compute_priorities (void)
|
compute_priorities (void)
|
{
|
{
|
int bb;
|
int bb;
|
|
|
current_sched_info->sched_max_insns_priority = 0;
|
current_sched_info->sched_max_insns_priority = 0;
|
for (bb = 0; bb < current_nr_blocks; bb++)
|
for (bb = 0; bb < current_nr_blocks; bb++)
|
{
|
{
|
rtx head, tail;
|
rtx head, tail;
|
|
|
gcc_assert (EBB_FIRST_BB (bb) == EBB_LAST_BB (bb));
|
gcc_assert (EBB_FIRST_BB (bb) == EBB_LAST_BB (bb));
|
get_ebb_head_tail (EBB_FIRST_BB (bb), EBB_LAST_BB (bb), &head, &tail);
|
get_ebb_head_tail (EBB_FIRST_BB (bb), EBB_LAST_BB (bb), &head, &tail);
|
|
|
if (no_real_insns_p (head, tail))
|
if (no_real_insns_p (head, tail))
|
continue;
|
continue;
|
|
|
rgn_n_insns += set_priorities (head, tail);
|
rgn_n_insns += set_priorities (head, tail);
|
}
|
}
|
current_sched_info->sched_max_insns_priority++;
|
current_sched_info->sched_max_insns_priority++;
|
}
|
}
|
|
|
/* Schedule a region. A region is either an inner loop, a loop-free
|
/* Schedule a region. A region is either an inner loop, a loop-free
|
subroutine, or a single basic block. Each bb in the region is
|
subroutine, or a single basic block. Each bb in the region is
|
scheduled after its flow predecessors. */
|
scheduled after its flow predecessors. */
|
|
|
static void
|
static void
|
schedule_region (int rgn)
|
schedule_region (int rgn)
|
{
|
{
|
int bb;
|
int bb;
|
int sched_rgn_n_insns = 0;
|
int sched_rgn_n_insns = 0;
|
|
|
rgn_n_insns = 0;
|
rgn_n_insns = 0;
|
|
|
rgn_setup_region (rgn);
|
rgn_setup_region (rgn);
|
|
|
/* Don't schedule region that is marked by
|
/* Don't schedule region that is marked by
|
NOTE_DISABLE_SCHED_OF_BLOCK. */
|
NOTE_DISABLE_SCHED_OF_BLOCK. */
|
if (sched_is_disabled_for_current_region_p ())
|
if (sched_is_disabled_for_current_region_p ())
|
return;
|
return;
|
|
|
sched_rgn_compute_dependencies (rgn);
|
sched_rgn_compute_dependencies (rgn);
|
|
|
sched_rgn_local_init (rgn);
|
sched_rgn_local_init (rgn);
|
|
|
/* Set priorities. */
|
/* Set priorities. */
|
compute_priorities ();
|
compute_priorities ();
|
|
|
sched_extend_ready_list (rgn_n_insns);
|
sched_extend_ready_list (rgn_n_insns);
|
|
|
if (sched_pressure_p)
|
if (sched_pressure_p)
|
{
|
{
|
sched_init_region_reg_pressure_info ();
|
sched_init_region_reg_pressure_info ();
|
for (bb = 0; bb < current_nr_blocks; bb++)
|
for (bb = 0; bb < current_nr_blocks; bb++)
|
{
|
{
|
basic_block first_bb, last_bb;
|
basic_block first_bb, last_bb;
|
rtx head, tail;
|
rtx head, tail;
|
|
|
first_bb = EBB_FIRST_BB (bb);
|
first_bb = EBB_FIRST_BB (bb);
|
last_bb = EBB_LAST_BB (bb);
|
last_bb = EBB_LAST_BB (bb);
|
|
|
get_ebb_head_tail (first_bb, last_bb, &head, &tail);
|
get_ebb_head_tail (first_bb, last_bb, &head, &tail);
|
|
|
if (no_real_insns_p (head, tail))
|
if (no_real_insns_p (head, tail))
|
{
|
{
|
gcc_assert (first_bb == last_bb);
|
gcc_assert (first_bb == last_bb);
|
continue;
|
continue;
|
}
|
}
|
sched_setup_bb_reg_pressure_info (first_bb, PREV_INSN (head));
|
sched_setup_bb_reg_pressure_info (first_bb, PREV_INSN (head));
|
}
|
}
|
}
|
}
|
|
|
/* Now we can schedule all blocks. */
|
/* Now we can schedule all blocks. */
|
for (bb = 0; bb < current_nr_blocks; bb++)
|
for (bb = 0; bb < current_nr_blocks; bb++)
|
{
|
{
|
basic_block first_bb, last_bb, curr_bb;
|
basic_block first_bb, last_bb, curr_bb;
|
rtx head, tail;
|
rtx head, tail;
|
|
|
first_bb = EBB_FIRST_BB (bb);
|
first_bb = EBB_FIRST_BB (bb);
|
last_bb = EBB_LAST_BB (bb);
|
last_bb = EBB_LAST_BB (bb);
|
|
|
get_ebb_head_tail (first_bb, last_bb, &head, &tail);
|
get_ebb_head_tail (first_bb, last_bb, &head, &tail);
|
|
|
if (no_real_insns_p (head, tail))
|
if (no_real_insns_p (head, tail))
|
{
|
{
|
gcc_assert (first_bb == last_bb);
|
gcc_assert (first_bb == last_bb);
|
continue;
|
continue;
|
}
|
}
|
|
|
current_sched_info->prev_head = PREV_INSN (head);
|
current_sched_info->prev_head = PREV_INSN (head);
|
current_sched_info->next_tail = NEXT_INSN (tail);
|
current_sched_info->next_tail = NEXT_INSN (tail);
|
|
|
remove_notes (head, tail);
|
remove_notes (head, tail);
|
|
|
unlink_bb_notes (first_bb, last_bb);
|
unlink_bb_notes (first_bb, last_bb);
|
|
|
target_bb = bb;
|
target_bb = bb;
|
|
|
gcc_assert (flag_schedule_interblock || current_nr_blocks == 1);
|
gcc_assert (flag_schedule_interblock || current_nr_blocks == 1);
|
current_sched_info->queue_must_finish_empty = current_nr_blocks == 1;
|
current_sched_info->queue_must_finish_empty = current_nr_blocks == 1;
|
|
|
curr_bb = first_bb;
|
curr_bb = first_bb;
|
if (dbg_cnt (sched_block))
|
if (dbg_cnt (sched_block))
|
{
|
{
|
schedule_block (&curr_bb);
|
schedule_block (&curr_bb);
|
gcc_assert (EBB_FIRST_BB (bb) == first_bb);
|
gcc_assert (EBB_FIRST_BB (bb) == first_bb);
|
sched_rgn_n_insns += sched_n_insns;
|
sched_rgn_n_insns += sched_n_insns;
|
}
|
}
|
else
|
else
|
{
|
{
|
sched_rgn_n_insns += rgn_n_insns;
|
sched_rgn_n_insns += rgn_n_insns;
|
}
|
}
|
|
|
/* Clean up. */
|
/* Clean up. */
|
if (current_nr_blocks > 1)
|
if (current_nr_blocks > 1)
|
free_trg_info ();
|
free_trg_info ();
|
}
|
}
|
|
|
/* Sanity check: verify that all region insns were scheduled. */
|
/* Sanity check: verify that all region insns were scheduled. */
|
gcc_assert (sched_rgn_n_insns == rgn_n_insns);
|
gcc_assert (sched_rgn_n_insns == rgn_n_insns);
|
|
|
sched_finish_ready_list ();
|
sched_finish_ready_list ();
|
|
|
/* Done with this region. */
|
/* Done with this region. */
|
sched_rgn_local_finish ();
|
sched_rgn_local_finish ();
|
|
|
/* Free dependencies. */
|
/* Free dependencies. */
|
for (bb = 0; bb < current_nr_blocks; ++bb)
|
for (bb = 0; bb < current_nr_blocks; ++bb)
|
free_block_dependencies (bb);
|
free_block_dependencies (bb);
|
|
|
gcc_assert (haifa_recovery_bb_ever_added_p
|
gcc_assert (haifa_recovery_bb_ever_added_p
|
|| deps_pools_are_empty_p ());
|
|| deps_pools_are_empty_p ());
|
}
|
}
|
|
|
/* Initialize data structures for region scheduling. */
|
/* Initialize data structures for region scheduling. */
|
|
|
void
|
void
|
sched_rgn_init (bool single_blocks_p)
|
sched_rgn_init (bool single_blocks_p)
|
{
|
{
|
min_spec_prob = ((PARAM_VALUE (PARAM_MIN_SPEC_PROB) * REG_BR_PROB_BASE)
|
min_spec_prob = ((PARAM_VALUE (PARAM_MIN_SPEC_PROB) * REG_BR_PROB_BASE)
|
/ 100);
|
/ 100);
|
|
|
nr_inter = 0;
|
nr_inter = 0;
|
nr_spec = 0;
|
nr_spec = 0;
|
|
|
extend_regions ();
|
extend_regions ();
|
|
|
CONTAINING_RGN (ENTRY_BLOCK) = -1;
|
CONTAINING_RGN (ENTRY_BLOCK) = -1;
|
CONTAINING_RGN (EXIT_BLOCK) = -1;
|
CONTAINING_RGN (EXIT_BLOCK) = -1;
|
|
|
/* Compute regions for scheduling. */
|
/* Compute regions for scheduling. */
|
if (single_blocks_p
|
if (single_blocks_p
|
|| n_basic_blocks == NUM_FIXED_BLOCKS + 1
|
|| n_basic_blocks == NUM_FIXED_BLOCKS + 1
|
|| !flag_schedule_interblock
|
|| !flag_schedule_interblock
|
|| is_cfg_nonregular ())
|
|| is_cfg_nonregular ())
|
{
|
{
|
find_single_block_region (sel_sched_p ());
|
find_single_block_region (sel_sched_p ());
|
}
|
}
|
else
|
else
|
{
|
{
|
/* Compute the dominators and post dominators. */
|
/* Compute the dominators and post dominators. */
|
if (!sel_sched_p ())
|
if (!sel_sched_p ())
|
calculate_dominance_info (CDI_DOMINATORS);
|
calculate_dominance_info (CDI_DOMINATORS);
|
|
|
/* Find regions. */
|
/* Find regions. */
|
find_rgns ();
|
find_rgns ();
|
|
|
if (sched_verbose >= 3)
|
if (sched_verbose >= 3)
|
debug_regions ();
|
debug_regions ();
|
|
|
/* For now. This will move as more and more of haifa is converted
|
/* For now. This will move as more and more of haifa is converted
|
to using the cfg code. */
|
to using the cfg code. */
|
if (!sel_sched_p ())
|
if (!sel_sched_p ())
|
free_dominance_info (CDI_DOMINATORS);
|
free_dominance_info (CDI_DOMINATORS);
|
}
|
}
|
|
|
gcc_assert (0 < nr_regions && nr_regions <= n_basic_blocks);
|
gcc_assert (0 < nr_regions && nr_regions <= n_basic_blocks);
|
|
|
RGN_BLOCKS (nr_regions) = (RGN_BLOCKS (nr_regions - 1) +
|
RGN_BLOCKS (nr_regions) = (RGN_BLOCKS (nr_regions - 1) +
|
RGN_NR_BLOCKS (nr_regions - 1));
|
RGN_NR_BLOCKS (nr_regions - 1));
|
}
|
}
|
|
|
/* Free data structures for region scheduling. */
|
/* Free data structures for region scheduling. */
|
void
|
void
|
sched_rgn_finish (void)
|
sched_rgn_finish (void)
|
{
|
{
|
/* Reposition the prologue and epilogue notes in case we moved the
|
/* Reposition the prologue and epilogue notes in case we moved the
|
prologue/epilogue insns. */
|
prologue/epilogue insns. */
|
if (reload_completed)
|
if (reload_completed)
|
reposition_prologue_and_epilogue_notes ();
|
reposition_prologue_and_epilogue_notes ();
|
|
|
if (sched_verbose)
|
if (sched_verbose)
|
{
|
{
|
if (reload_completed == 0
|
if (reload_completed == 0
|
&& flag_schedule_interblock)
|
&& flag_schedule_interblock)
|
{
|
{
|
fprintf (sched_dump,
|
fprintf (sched_dump,
|
"\n;; Procedure interblock/speculative motions == %d/%d \n",
|
"\n;; Procedure interblock/speculative motions == %d/%d \n",
|
nr_inter, nr_spec);
|
nr_inter, nr_spec);
|
}
|
}
|
else
|
else
|
gcc_assert (nr_inter <= 0);
|
gcc_assert (nr_inter <= 0);
|
fprintf (sched_dump, "\n\n");
|
fprintf (sched_dump, "\n\n");
|
}
|
}
|
|
|
nr_regions = 0;
|
nr_regions = 0;
|
|
|
free (rgn_table);
|
free (rgn_table);
|
rgn_table = NULL;
|
rgn_table = NULL;
|
|
|
free (rgn_bb_table);
|
free (rgn_bb_table);
|
rgn_bb_table = NULL;
|
rgn_bb_table = NULL;
|
|
|
free (block_to_bb);
|
free (block_to_bb);
|
block_to_bb = NULL;
|
block_to_bb = NULL;
|
|
|
free (containing_rgn);
|
free (containing_rgn);
|
containing_rgn = NULL;
|
containing_rgn = NULL;
|
|
|
free (ebb_head);
|
free (ebb_head);
|
ebb_head = NULL;
|
ebb_head = NULL;
|
}
|
}
|
|
|
/* Setup global variables like CURRENT_BLOCKS and CURRENT_NR_BLOCK to
|
/* Setup global variables like CURRENT_BLOCKS and CURRENT_NR_BLOCK to
|
point to the region RGN. */
|
point to the region RGN. */
|
void
|
void
|
rgn_setup_region (int rgn)
|
rgn_setup_region (int rgn)
|
{
|
{
|
int bb;
|
int bb;
|
|
|
/* Set variables for the current region. */
|
/* Set variables for the current region. */
|
current_nr_blocks = RGN_NR_BLOCKS (rgn);
|
current_nr_blocks = RGN_NR_BLOCKS (rgn);
|
current_blocks = RGN_BLOCKS (rgn);
|
current_blocks = RGN_BLOCKS (rgn);
|
|
|
/* EBB_HEAD is a region-scope structure. But we realloc it for
|
/* EBB_HEAD is a region-scope structure. But we realloc it for
|
each region to save time/memory/something else.
|
each region to save time/memory/something else.
|
See comments in add_block1, for what reasons we allocate +1 element. */
|
See comments in add_block1, for what reasons we allocate +1 element. */
|
ebb_head = XRESIZEVEC (int, ebb_head, current_nr_blocks + 1);
|
ebb_head = XRESIZEVEC (int, ebb_head, current_nr_blocks + 1);
|
for (bb = 0; bb <= current_nr_blocks; bb++)
|
for (bb = 0; bb <= current_nr_blocks; bb++)
|
ebb_head[bb] = current_blocks + bb;
|
ebb_head[bb] = current_blocks + bb;
|
}
|
}
|
|
|
/* Compute instruction dependencies in region RGN. */
|
/* Compute instruction dependencies in region RGN. */
|
void
|
void
|
sched_rgn_compute_dependencies (int rgn)
|
sched_rgn_compute_dependencies (int rgn)
|
{
|
{
|
if (!RGN_DONT_CALC_DEPS (rgn))
|
if (!RGN_DONT_CALC_DEPS (rgn))
|
{
|
{
|
int bb;
|
int bb;
|
|
|
if (sel_sched_p ())
|
if (sel_sched_p ())
|
sched_emulate_haifa_p = 1;
|
sched_emulate_haifa_p = 1;
|
|
|
init_deps_global ();
|
init_deps_global ();
|
|
|
/* Initializations for region data dependence analysis. */
|
/* Initializations for region data dependence analysis. */
|
bb_deps = XNEWVEC (struct deps_desc, current_nr_blocks);
|
bb_deps = XNEWVEC (struct deps_desc, current_nr_blocks);
|
for (bb = 0; bb < current_nr_blocks; bb++)
|
for (bb = 0; bb < current_nr_blocks; bb++)
|
init_deps (bb_deps + bb, false);
|
init_deps (bb_deps + bb, false);
|
|
|
/* Initialize bitmap used in add_branch_dependences. */
|
/* Initialize bitmap used in add_branch_dependences. */
|
insn_referenced = sbitmap_alloc (sched_max_luid);
|
insn_referenced = sbitmap_alloc (sched_max_luid);
|
sbitmap_zero (insn_referenced);
|
sbitmap_zero (insn_referenced);
|
|
|
/* Compute backward dependencies. */
|
/* Compute backward dependencies. */
|
for (bb = 0; bb < current_nr_blocks; bb++)
|
for (bb = 0; bb < current_nr_blocks; bb++)
|
compute_block_dependences (bb);
|
compute_block_dependences (bb);
|
|
|
sbitmap_free (insn_referenced);
|
sbitmap_free (insn_referenced);
|
free_pending_lists ();
|
free_pending_lists ();
|
finish_deps_global ();
|
finish_deps_global ();
|
free (bb_deps);
|
free (bb_deps);
|
|
|
/* We don't want to recalculate this twice. */
|
/* We don't want to recalculate this twice. */
|
RGN_DONT_CALC_DEPS (rgn) = 1;
|
RGN_DONT_CALC_DEPS (rgn) = 1;
|
|
|
if (sel_sched_p ())
|
if (sel_sched_p ())
|
sched_emulate_haifa_p = 0;
|
sched_emulate_haifa_p = 0;
|
}
|
}
|
else
|
else
|
/* (This is a recovery block. It is always a single block region.)
|
/* (This is a recovery block. It is always a single block region.)
|
OR (We use selective scheduling.) */
|
OR (We use selective scheduling.) */
|
gcc_assert (current_nr_blocks == 1 || sel_sched_p ());
|
gcc_assert (current_nr_blocks == 1 || sel_sched_p ());
|
}
|
}
|
|
|
/* Init region data structures. Returns true if this region should
|
/* Init region data structures. Returns true if this region should
|
not be scheduled. */
|
not be scheduled. */
|
void
|
void
|
sched_rgn_local_init (int rgn)
|
sched_rgn_local_init (int rgn)
|
{
|
{
|
int bb;
|
int bb;
|
|
|
/* Compute interblock info: probabilities, split-edges, dominators, etc. */
|
/* Compute interblock info: probabilities, split-edges, dominators, etc. */
|
if (current_nr_blocks > 1)
|
if (current_nr_blocks > 1)
|
{
|
{
|
basic_block block;
|
basic_block block;
|
edge e;
|
edge e;
|
edge_iterator ei;
|
edge_iterator ei;
|
|
|
prob = XNEWVEC (int, current_nr_blocks);
|
prob = XNEWVEC (int, current_nr_blocks);
|
|
|
dom = sbitmap_vector_alloc (current_nr_blocks, current_nr_blocks);
|
dom = sbitmap_vector_alloc (current_nr_blocks, current_nr_blocks);
|
sbitmap_vector_zero (dom, current_nr_blocks);
|
sbitmap_vector_zero (dom, current_nr_blocks);
|
|
|
/* Use ->aux to implement EDGE_TO_BIT mapping. */
|
/* Use ->aux to implement EDGE_TO_BIT mapping. */
|
rgn_nr_edges = 0;
|
rgn_nr_edges = 0;
|
FOR_EACH_BB (block)
|
FOR_EACH_BB (block)
|
{
|
{
|
if (CONTAINING_RGN (block->index) != rgn)
|
if (CONTAINING_RGN (block->index) != rgn)
|
continue;
|
continue;
|
FOR_EACH_EDGE (e, ei, block->succs)
|
FOR_EACH_EDGE (e, ei, block->succs)
|
SET_EDGE_TO_BIT (e, rgn_nr_edges++);
|
SET_EDGE_TO_BIT (e, rgn_nr_edges++);
|
}
|
}
|
|
|
rgn_edges = XNEWVEC (edge, rgn_nr_edges);
|
rgn_edges = XNEWVEC (edge, rgn_nr_edges);
|
rgn_nr_edges = 0;
|
rgn_nr_edges = 0;
|
FOR_EACH_BB (block)
|
FOR_EACH_BB (block)
|
{
|
{
|
if (CONTAINING_RGN (block->index) != rgn)
|
if (CONTAINING_RGN (block->index) != rgn)
|
continue;
|
continue;
|
FOR_EACH_EDGE (e, ei, block->succs)
|
FOR_EACH_EDGE (e, ei, block->succs)
|
rgn_edges[rgn_nr_edges++] = e;
|
rgn_edges[rgn_nr_edges++] = e;
|
}
|
}
|
|
|
/* Split edges. */
|
/* Split edges. */
|
pot_split = sbitmap_vector_alloc (current_nr_blocks, rgn_nr_edges);
|
pot_split = sbitmap_vector_alloc (current_nr_blocks, rgn_nr_edges);
|
sbitmap_vector_zero (pot_split, current_nr_blocks);
|
sbitmap_vector_zero (pot_split, current_nr_blocks);
|
ancestor_edges = sbitmap_vector_alloc (current_nr_blocks, rgn_nr_edges);
|
ancestor_edges = sbitmap_vector_alloc (current_nr_blocks, rgn_nr_edges);
|
sbitmap_vector_zero (ancestor_edges, current_nr_blocks);
|
sbitmap_vector_zero (ancestor_edges, current_nr_blocks);
|
|
|
/* Compute probabilities, dominators, split_edges. */
|
/* Compute probabilities, dominators, split_edges. */
|
for (bb = 0; bb < current_nr_blocks; bb++)
|
for (bb = 0; bb < current_nr_blocks; bb++)
|
compute_dom_prob_ps (bb);
|
compute_dom_prob_ps (bb);
|
|
|
/* Cleanup ->aux used for EDGE_TO_BIT mapping. */
|
/* Cleanup ->aux used for EDGE_TO_BIT mapping. */
|
/* We don't need them anymore. But we want to avoid duplication of
|
/* We don't need them anymore. But we want to avoid duplication of
|
aux fields in the newly created edges. */
|
aux fields in the newly created edges. */
|
FOR_EACH_BB (block)
|
FOR_EACH_BB (block)
|
{
|
{
|
if (CONTAINING_RGN (block->index) != rgn)
|
if (CONTAINING_RGN (block->index) != rgn)
|
continue;
|
continue;
|
FOR_EACH_EDGE (e, ei, block->succs)
|
FOR_EACH_EDGE (e, ei, block->succs)
|
e->aux = NULL;
|
e->aux = NULL;
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
/* Free data computed for the finished region. */
|
/* Free data computed for the finished region. */
|
void
|
void
|
sched_rgn_local_free (void)
|
sched_rgn_local_free (void)
|
{
|
{
|
free (prob);
|
free (prob);
|
sbitmap_vector_free (dom);
|
sbitmap_vector_free (dom);
|
sbitmap_vector_free (pot_split);
|
sbitmap_vector_free (pot_split);
|
sbitmap_vector_free (ancestor_edges);
|
sbitmap_vector_free (ancestor_edges);
|
free (rgn_edges);
|
free (rgn_edges);
|
}
|
}
|
|
|
/* Free data computed for the finished region. */
|
/* Free data computed for the finished region. */
|
void
|
void
|
sched_rgn_local_finish (void)
|
sched_rgn_local_finish (void)
|
{
|
{
|
if (current_nr_blocks > 1 && !sel_sched_p ())
|
if (current_nr_blocks > 1 && !sel_sched_p ())
|
{
|
{
|
sched_rgn_local_free ();
|
sched_rgn_local_free ();
|
}
|
}
|
}
|
}
|
|
|
/* Setup scheduler infos. */
|
/* Setup scheduler infos. */
|
void
|
void
|
rgn_setup_common_sched_info (void)
|
rgn_setup_common_sched_info (void)
|
{
|
{
|
memcpy (&rgn_common_sched_info, &haifa_common_sched_info,
|
memcpy (&rgn_common_sched_info, &haifa_common_sched_info,
|
sizeof (rgn_common_sched_info));
|
sizeof (rgn_common_sched_info));
|
|
|
rgn_common_sched_info.fix_recovery_cfg = rgn_fix_recovery_cfg;
|
rgn_common_sched_info.fix_recovery_cfg = rgn_fix_recovery_cfg;
|
rgn_common_sched_info.add_block = rgn_add_block;
|
rgn_common_sched_info.add_block = rgn_add_block;
|
rgn_common_sched_info.estimate_number_of_insns
|
rgn_common_sched_info.estimate_number_of_insns
|
= rgn_estimate_number_of_insns;
|
= rgn_estimate_number_of_insns;
|
rgn_common_sched_info.sched_pass_id = SCHED_RGN_PASS;
|
rgn_common_sched_info.sched_pass_id = SCHED_RGN_PASS;
|
|
|
common_sched_info = &rgn_common_sched_info;
|
common_sched_info = &rgn_common_sched_info;
|
}
|
}
|
|
|
/* Setup all *_sched_info structures (for the Haifa frontend
|
/* Setup all *_sched_info structures (for the Haifa frontend
|
and for the dependence analysis) in the interblock scheduler. */
|
and for the dependence analysis) in the interblock scheduler. */
|
void
|
void
|
rgn_setup_sched_infos (void)
|
rgn_setup_sched_infos (void)
|
{
|
{
|
if (!sel_sched_p ())
|
if (!sel_sched_p ())
|
memcpy (&rgn_sched_deps_info, &rgn_const_sched_deps_info,
|
memcpy (&rgn_sched_deps_info, &rgn_const_sched_deps_info,
|
sizeof (rgn_sched_deps_info));
|
sizeof (rgn_sched_deps_info));
|
else
|
else
|
memcpy (&rgn_sched_deps_info, &rgn_const_sel_sched_deps_info,
|
memcpy (&rgn_sched_deps_info, &rgn_const_sel_sched_deps_info,
|
sizeof (rgn_sched_deps_info));
|
sizeof (rgn_sched_deps_info));
|
|
|
sched_deps_info = &rgn_sched_deps_info;
|
sched_deps_info = &rgn_sched_deps_info;
|
|
|
memcpy (&rgn_sched_info, &rgn_const_sched_info, sizeof (rgn_sched_info));
|
memcpy (&rgn_sched_info, &rgn_const_sched_info, sizeof (rgn_sched_info));
|
current_sched_info = &rgn_sched_info;
|
current_sched_info = &rgn_sched_info;
|
}
|
}
|
|
|
/* The one entry point in this file. */
|
/* The one entry point in this file. */
|
void
|
void
|
schedule_insns (void)
|
schedule_insns (void)
|
{
|
{
|
int rgn;
|
int rgn;
|
|
|
/* Taking care of this degenerate case makes the rest of
|
/* Taking care of this degenerate case makes the rest of
|
this code simpler. */
|
this code simpler. */
|
if (n_basic_blocks == NUM_FIXED_BLOCKS)
|
if (n_basic_blocks == NUM_FIXED_BLOCKS)
|
return;
|
return;
|
|
|
rgn_setup_common_sched_info ();
|
rgn_setup_common_sched_info ();
|
rgn_setup_sched_infos ();
|
rgn_setup_sched_infos ();
|
|
|
haifa_sched_init ();
|
haifa_sched_init ();
|
sched_rgn_init (reload_completed);
|
sched_rgn_init (reload_completed);
|
|
|
bitmap_initialize (¬_in_df, 0);
|
bitmap_initialize (¬_in_df, 0);
|
bitmap_clear (¬_in_df);
|
bitmap_clear (¬_in_df);
|
|
|
/* Schedule every region in the subroutine. */
|
/* Schedule every region in the subroutine. */
|
for (rgn = 0; rgn < nr_regions; rgn++)
|
for (rgn = 0; rgn < nr_regions; rgn++)
|
if (dbg_cnt (sched_region))
|
if (dbg_cnt (sched_region))
|
schedule_region (rgn);
|
schedule_region (rgn);
|
|
|
/* Clean up. */
|
/* Clean up. */
|
sched_rgn_finish ();
|
sched_rgn_finish ();
|
bitmap_clear (¬_in_df);
|
bitmap_clear (¬_in_df);
|
|
|
haifa_sched_finish ();
|
haifa_sched_finish ();
|
}
|
}
|
|
|
/* INSN has been added to/removed from current region. */
|
/* INSN has been added to/removed from current region. */
|
static void
|
static void
|
rgn_add_remove_insn (rtx insn, int remove_p)
|
rgn_add_remove_insn (rtx insn, int remove_p)
|
{
|
{
|
if (!remove_p)
|
if (!remove_p)
|
rgn_n_insns++;
|
rgn_n_insns++;
|
else
|
else
|
rgn_n_insns--;
|
rgn_n_insns--;
|
|
|
if (INSN_BB (insn) == target_bb)
|
if (INSN_BB (insn) == target_bb)
|
{
|
{
|
if (!remove_p)
|
if (!remove_p)
|
target_n_insns++;
|
target_n_insns++;
|
else
|
else
|
target_n_insns--;
|
target_n_insns--;
|
}
|
}
|
}
|
}
|
|
|
/* Extend internal data structures. */
|
/* Extend internal data structures. */
|
void
|
void
|
extend_regions (void)
|
extend_regions (void)
|
{
|
{
|
rgn_table = XRESIZEVEC (region, rgn_table, n_basic_blocks);
|
rgn_table = XRESIZEVEC (region, rgn_table, n_basic_blocks);
|
rgn_bb_table = XRESIZEVEC (int, rgn_bb_table, n_basic_blocks);
|
rgn_bb_table = XRESIZEVEC (int, rgn_bb_table, n_basic_blocks);
|
block_to_bb = XRESIZEVEC (int, block_to_bb, last_basic_block);
|
block_to_bb = XRESIZEVEC (int, block_to_bb, last_basic_block);
|
containing_rgn = XRESIZEVEC (int, containing_rgn, last_basic_block);
|
containing_rgn = XRESIZEVEC (int, containing_rgn, last_basic_block);
|
}
|
}
|
|
|
void
|
void
|
rgn_make_new_region_out_of_new_block (basic_block bb)
|
rgn_make_new_region_out_of_new_block (basic_block bb)
|
{
|
{
|
int i;
|
int i;
|
|
|
i = RGN_BLOCKS (nr_regions);
|
i = RGN_BLOCKS (nr_regions);
|
/* I - first free position in rgn_bb_table. */
|
/* I - first free position in rgn_bb_table. */
|
|
|
rgn_bb_table[i] = bb->index;
|
rgn_bb_table[i] = bb->index;
|
RGN_NR_BLOCKS (nr_regions) = 1;
|
RGN_NR_BLOCKS (nr_regions) = 1;
|
RGN_HAS_REAL_EBB (nr_regions) = 0;
|
RGN_HAS_REAL_EBB (nr_regions) = 0;
|
RGN_DONT_CALC_DEPS (nr_regions) = 0;
|
RGN_DONT_CALC_DEPS (nr_regions) = 0;
|
CONTAINING_RGN (bb->index) = nr_regions;
|
CONTAINING_RGN (bb->index) = nr_regions;
|
BLOCK_TO_BB (bb->index) = 0;
|
BLOCK_TO_BB (bb->index) = 0;
|
|
|
nr_regions++;
|
nr_regions++;
|
|
|
RGN_BLOCKS (nr_regions) = i + 1;
|
RGN_BLOCKS (nr_regions) = i + 1;
|
}
|
}
|
|
|
/* BB was added to ebb after AFTER. */
|
/* BB was added to ebb after AFTER. */
|
static void
|
static void
|
rgn_add_block (basic_block bb, basic_block after)
|
rgn_add_block (basic_block bb, basic_block after)
|
{
|
{
|
extend_regions ();
|
extend_regions ();
|
bitmap_set_bit (¬_in_df, bb->index);
|
bitmap_set_bit (¬_in_df, bb->index);
|
|
|
if (after == 0 || after == EXIT_BLOCK_PTR)
|
if (after == 0 || after == EXIT_BLOCK_PTR)
|
{
|
{
|
rgn_make_new_region_out_of_new_block (bb);
|
rgn_make_new_region_out_of_new_block (bb);
|
RGN_DONT_CALC_DEPS (nr_regions - 1) = (after == EXIT_BLOCK_PTR);
|
RGN_DONT_CALC_DEPS (nr_regions - 1) = (after == EXIT_BLOCK_PTR);
|
}
|
}
|
else
|
else
|
{
|
{
|
int i, pos;
|
int i, pos;
|
|
|
/* We need to fix rgn_table, block_to_bb, containing_rgn
|
/* We need to fix rgn_table, block_to_bb, containing_rgn
|
and ebb_head. */
|
and ebb_head. */
|
|
|
BLOCK_TO_BB (bb->index) = BLOCK_TO_BB (after->index);
|
BLOCK_TO_BB (bb->index) = BLOCK_TO_BB (after->index);
|
|
|
/* We extend ebb_head to one more position to
|
/* We extend ebb_head to one more position to
|
easily find the last position of the last ebb in
|
easily find the last position of the last ebb in
|
the current region. Thus, ebb_head[BLOCK_TO_BB (after) + 1]
|
the current region. Thus, ebb_head[BLOCK_TO_BB (after) + 1]
|
is _always_ valid for access. */
|
is _always_ valid for access. */
|
|
|
i = BLOCK_TO_BB (after->index) + 1;
|
i = BLOCK_TO_BB (after->index) + 1;
|
pos = ebb_head[i] - 1;
|
pos = ebb_head[i] - 1;
|
/* Now POS is the index of the last block in the region. */
|
/* Now POS is the index of the last block in the region. */
|
|
|
/* Find index of basic block AFTER. */
|
/* Find index of basic block AFTER. */
|
for (; rgn_bb_table[pos] != after->index; pos--);
|
for (; rgn_bb_table[pos] != after->index; pos--);
|
|
|
pos++;
|
pos++;
|
gcc_assert (pos > ebb_head[i - 1]);
|
gcc_assert (pos > ebb_head[i - 1]);
|
|
|
/* i - ebb right after "AFTER". */
|
/* i - ebb right after "AFTER". */
|
/* ebb_head[i] - VALID. */
|
/* ebb_head[i] - VALID. */
|
|
|
/* Source position: ebb_head[i]
|
/* Source position: ebb_head[i]
|
Destination position: ebb_head[i] + 1
|
Destination position: ebb_head[i] + 1
|
Last position:
|
Last position:
|
RGN_BLOCKS (nr_regions) - 1
|
RGN_BLOCKS (nr_regions) - 1
|
Number of elements to copy: (last_position) - (source_position) + 1
|
Number of elements to copy: (last_position) - (source_position) + 1
|
*/
|
*/
|
|
|
memmove (rgn_bb_table + pos + 1,
|
memmove (rgn_bb_table + pos + 1,
|
rgn_bb_table + pos,
|
rgn_bb_table + pos,
|
((RGN_BLOCKS (nr_regions) - 1) - (pos) + 1)
|
((RGN_BLOCKS (nr_regions) - 1) - (pos) + 1)
|
* sizeof (*rgn_bb_table));
|
* sizeof (*rgn_bb_table));
|
|
|
rgn_bb_table[pos] = bb->index;
|
rgn_bb_table[pos] = bb->index;
|
|
|
for (; i <= current_nr_blocks; i++)
|
for (; i <= current_nr_blocks; i++)
|
ebb_head [i]++;
|
ebb_head [i]++;
|
|
|
i = CONTAINING_RGN (after->index);
|
i = CONTAINING_RGN (after->index);
|
CONTAINING_RGN (bb->index) = i;
|
CONTAINING_RGN (bb->index) = i;
|
|
|
RGN_HAS_REAL_EBB (i) = 1;
|
RGN_HAS_REAL_EBB (i) = 1;
|
|
|
for (++i; i <= nr_regions; i++)
|
for (++i; i <= nr_regions; i++)
|
RGN_BLOCKS (i)++;
|
RGN_BLOCKS (i)++;
|
}
|
}
|
}
|
}
|
|
|
/* Fix internal data after interblock movement of jump instruction.
|
/* Fix internal data after interblock movement of jump instruction.
|
For parameter meaning please refer to
|
For parameter meaning please refer to
|
sched-int.h: struct sched_info: fix_recovery_cfg. */
|
sched-int.h: struct sched_info: fix_recovery_cfg. */
|
static void
|
static void
|
rgn_fix_recovery_cfg (int bbi, int check_bbi, int check_bb_nexti)
|
rgn_fix_recovery_cfg (int bbi, int check_bbi, int check_bb_nexti)
|
{
|
{
|
int old_pos, new_pos, i;
|
int old_pos, new_pos, i;
|
|
|
BLOCK_TO_BB (check_bb_nexti) = BLOCK_TO_BB (bbi);
|
BLOCK_TO_BB (check_bb_nexti) = BLOCK_TO_BB (bbi);
|
|
|
for (old_pos = ebb_head[BLOCK_TO_BB (check_bbi) + 1] - 1;
|
for (old_pos = ebb_head[BLOCK_TO_BB (check_bbi) + 1] - 1;
|
rgn_bb_table[old_pos] != check_bb_nexti;
|
rgn_bb_table[old_pos] != check_bb_nexti;
|
old_pos--);
|
old_pos--);
|
gcc_assert (old_pos > ebb_head[BLOCK_TO_BB (check_bbi)]);
|
gcc_assert (old_pos > ebb_head[BLOCK_TO_BB (check_bbi)]);
|
|
|
for (new_pos = ebb_head[BLOCK_TO_BB (bbi) + 1] - 1;
|
for (new_pos = ebb_head[BLOCK_TO_BB (bbi) + 1] - 1;
|
rgn_bb_table[new_pos] != bbi;
|
rgn_bb_table[new_pos] != bbi;
|
new_pos--);
|
new_pos--);
|
new_pos++;
|
new_pos++;
|
gcc_assert (new_pos > ebb_head[BLOCK_TO_BB (bbi)]);
|
gcc_assert (new_pos > ebb_head[BLOCK_TO_BB (bbi)]);
|
|
|
gcc_assert (new_pos < old_pos);
|
gcc_assert (new_pos < old_pos);
|
|
|
memmove (rgn_bb_table + new_pos + 1,
|
memmove (rgn_bb_table + new_pos + 1,
|
rgn_bb_table + new_pos,
|
rgn_bb_table + new_pos,
|
(old_pos - new_pos) * sizeof (*rgn_bb_table));
|
(old_pos - new_pos) * sizeof (*rgn_bb_table));
|
|
|
rgn_bb_table[new_pos] = check_bb_nexti;
|
rgn_bb_table[new_pos] = check_bb_nexti;
|
|
|
for (i = BLOCK_TO_BB (bbi) + 1; i <= BLOCK_TO_BB (check_bbi); i++)
|
for (i = BLOCK_TO_BB (bbi) + 1; i <= BLOCK_TO_BB (check_bbi); i++)
|
ebb_head[i]++;
|
ebb_head[i]++;
|
}
|
}
|
|
|
/* Return next block in ebb chain. For parameter meaning please refer to
|
/* Return next block in ebb chain. For parameter meaning please refer to
|
sched-int.h: struct sched_info: advance_target_bb. */
|
sched-int.h: struct sched_info: advance_target_bb. */
|
static basic_block
|
static basic_block
|
advance_target_bb (basic_block bb, rtx insn)
|
advance_target_bb (basic_block bb, rtx insn)
|
{
|
{
|
if (insn)
|
if (insn)
|
return 0;
|
return 0;
|
|
|
gcc_assert (BLOCK_TO_BB (bb->index) == target_bb
|
gcc_assert (BLOCK_TO_BB (bb->index) == target_bb
|
&& BLOCK_TO_BB (bb->next_bb->index) == target_bb);
|
&& BLOCK_TO_BB (bb->next_bb->index) == target_bb);
|
return bb->next_bb;
|
return bb->next_bb;
|
}
|
}
|
|
|
#endif
|
#endif
|
|
|
static bool
|
static bool
|
gate_handle_sched (void)
|
gate_handle_sched (void)
|
{
|
{
|
#ifdef INSN_SCHEDULING
|
#ifdef INSN_SCHEDULING
|
return flag_schedule_insns && dbg_cnt (sched_func);
|
return flag_schedule_insns && dbg_cnt (sched_func);
|
#else
|
#else
|
return 0;
|
return 0;
|
#endif
|
#endif
|
}
|
}
|
|
|
/* Run instruction scheduler. */
|
/* Run instruction scheduler. */
|
static unsigned int
|
static unsigned int
|
rest_of_handle_sched (void)
|
rest_of_handle_sched (void)
|
{
|
{
|
#ifdef INSN_SCHEDULING
|
#ifdef INSN_SCHEDULING
|
if (flag_selective_scheduling
|
if (flag_selective_scheduling
|
&& ! maybe_skip_selective_scheduling ())
|
&& ! maybe_skip_selective_scheduling ())
|
run_selective_scheduling ();
|
run_selective_scheduling ();
|
else
|
else
|
schedule_insns ();
|
schedule_insns ();
|
#endif
|
#endif
|
return 0;
|
return 0;
|
}
|
}
|
|
|
static bool
|
static bool
|
gate_handle_sched2 (void)
|
gate_handle_sched2 (void)
|
{
|
{
|
#ifdef INSN_SCHEDULING
|
#ifdef INSN_SCHEDULING
|
return optimize > 0 && flag_schedule_insns_after_reload
|
return optimize > 0 && flag_schedule_insns_after_reload
|
&& dbg_cnt (sched2_func);
|
&& dbg_cnt (sched2_func);
|
#else
|
#else
|
return 0;
|
return 0;
|
#endif
|
#endif
|
}
|
}
|
|
|
/* Run second scheduling pass after reload. */
|
/* Run second scheduling pass after reload. */
|
static unsigned int
|
static unsigned int
|
rest_of_handle_sched2 (void)
|
rest_of_handle_sched2 (void)
|
{
|
{
|
#ifdef INSN_SCHEDULING
|
#ifdef INSN_SCHEDULING
|
if (flag_selective_scheduling2
|
if (flag_selective_scheduling2
|
&& ! maybe_skip_selective_scheduling ())
|
&& ! maybe_skip_selective_scheduling ())
|
run_selective_scheduling ();
|
run_selective_scheduling ();
|
else
|
else
|
{
|
{
|
/* Do control and data sched analysis again,
|
/* Do control and data sched analysis again,
|
and write some more of the results to dump file. */
|
and write some more of the results to dump file. */
|
if (flag_sched2_use_superblocks)
|
if (flag_sched2_use_superblocks)
|
schedule_ebbs ();
|
schedule_ebbs ();
|
else
|
else
|
schedule_insns ();
|
schedule_insns ();
|
}
|
}
|
#endif
|
#endif
|
return 0;
|
return 0;
|
}
|
}
|
|
|
struct rtl_opt_pass pass_sched =
|
struct rtl_opt_pass pass_sched =
|
{
|
{
|
{
|
{
|
RTL_PASS,
|
RTL_PASS,
|
"sched1", /* name */
|
"sched1", /* name */
|
gate_handle_sched, /* gate */
|
gate_handle_sched, /* gate */
|
rest_of_handle_sched, /* execute */
|
rest_of_handle_sched, /* execute */
|
NULL, /* sub */
|
NULL, /* sub */
|
NULL, /* next */
|
NULL, /* next */
|
0, /* static_pass_number */
|
0, /* static_pass_number */
|
TV_SCHED, /* tv_id */
|
TV_SCHED, /* 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_verify_rtl_sharing |
|
TODO_df_finish | TODO_verify_rtl_sharing |
|
TODO_dump_func |
|
TODO_dump_func |
|
TODO_verify_flow |
|
TODO_verify_flow |
|
TODO_ggc_collect /* todo_flags_finish */
|
TODO_ggc_collect /* todo_flags_finish */
|
}
|
}
|
};
|
};
|
|
|
struct rtl_opt_pass pass_sched2 =
|
struct rtl_opt_pass pass_sched2 =
|
{
|
{
|
{
|
{
|
RTL_PASS,
|
RTL_PASS,
|
"sched2", /* name */
|
"sched2", /* name */
|
gate_handle_sched2, /* gate */
|
gate_handle_sched2, /* gate */
|
rest_of_handle_sched2, /* execute */
|
rest_of_handle_sched2, /* execute */
|
NULL, /* sub */
|
NULL, /* sub */
|
NULL, /* next */
|
NULL, /* next */
|
0, /* static_pass_number */
|
0, /* static_pass_number */
|
TV_SCHED2, /* tv_id */
|
TV_SCHED2, /* 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_verify_rtl_sharing |
|
TODO_df_finish | TODO_verify_rtl_sharing |
|
TODO_dump_func |
|
TODO_dump_func |
|
TODO_verify_flow |
|
TODO_verify_flow |
|
TODO_ggc_collect /* todo_flags_finish */
|
TODO_ggc_collect /* todo_flags_finish */
|
}
|
}
|
};
|
};
|
|
|
