/* Loop unrolling and peeling.
|
/* Loop unrolling and peeling.
|
Copyright (C) 2002, 2003, 2004, 2005, 2007 Free Software Foundation, Inc.
|
Copyright (C) 2002, 2003, 2004, 2005, 2007 Free Software Foundation, Inc.
|
|
|
This file is part of GCC.
|
This file is part of GCC.
|
|
|
GCC is free software; you can redistribute it and/or modify it under
|
GCC is free software; you can redistribute it and/or modify it under
|
the terms of the GNU General Public License as published by the Free
|
the terms of the GNU General Public License as published by the Free
|
Software Foundation; either version 3, or (at your option) any later
|
Software Foundation; either version 3, or (at your option) any later
|
version.
|
version.
|
|
|
GCC is distributed in the hope that it will be useful, but WITHOUT ANY
|
GCC is distributed in the hope that it will be useful, but WITHOUT ANY
|
WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
for more details.
|
for more details.
|
|
|
You should have received a copy of the GNU General Public License
|
You should have received a copy of the GNU General Public License
|
along with GCC; see the file COPYING3. If not see
|
along with GCC; see the file COPYING3. If not see
|
<http://www.gnu.org/licenses/>. */
|
<http://www.gnu.org/licenses/>. */
|
|
|
#include "config.h"
|
#include "config.h"
|
#include "system.h"
|
#include "system.h"
|
#include "coretypes.h"
|
#include "coretypes.h"
|
#include "tm.h"
|
#include "tm.h"
|
#include "rtl.h"
|
#include "rtl.h"
|
#include "hard-reg-set.h"
|
#include "hard-reg-set.h"
|
#include "obstack.h"
|
#include "obstack.h"
|
#include "basic-block.h"
|
#include "basic-block.h"
|
#include "cfgloop.h"
|
#include "cfgloop.h"
|
#include "cfglayout.h"
|
#include "cfglayout.h"
|
#include "params.h"
|
#include "params.h"
|
#include "output.h"
|
#include "output.h"
|
#include "expr.h"
|
#include "expr.h"
|
#include "hashtab.h"
|
#include "hashtab.h"
|
#include "recog.h"
|
#include "recog.h"
|
|
|
/* This pass performs loop unrolling and peeling. We only perform these
|
/* This pass performs loop unrolling and peeling. We only perform these
|
optimizations on innermost loops (with single exception) because
|
optimizations on innermost loops (with single exception) because
|
the impact on performance is greatest here, and we want to avoid
|
the impact on performance is greatest here, and we want to avoid
|
unnecessary code size growth. The gain is caused by greater sequentiality
|
unnecessary code size growth. The gain is caused by greater sequentiality
|
of code, better code to optimize for further passes and in some cases
|
of code, better code to optimize for further passes and in some cases
|
by fewer testings of exit conditions. The main problem is code growth,
|
by fewer testings of exit conditions. The main problem is code growth,
|
that impacts performance negatively due to effect of caches.
|
that impacts performance negatively due to effect of caches.
|
|
|
What we do:
|
What we do:
|
|
|
-- complete peeling of once-rolling loops; this is the above mentioned
|
-- complete peeling of once-rolling loops; this is the above mentioned
|
exception, as this causes loop to be cancelled completely and
|
exception, as this causes loop to be cancelled completely and
|
does not cause code growth
|
does not cause code growth
|
-- complete peeling of loops that roll (small) constant times.
|
-- complete peeling of loops that roll (small) constant times.
|
-- simple peeling of first iterations of loops that do not roll much
|
-- simple peeling of first iterations of loops that do not roll much
|
(according to profile feedback)
|
(according to profile feedback)
|
-- unrolling of loops that roll constant times; this is almost always
|
-- unrolling of loops that roll constant times; this is almost always
|
win, as we get rid of exit condition tests.
|
win, as we get rid of exit condition tests.
|
-- unrolling of loops that roll number of times that we can compute
|
-- unrolling of loops that roll number of times that we can compute
|
in runtime; we also get rid of exit condition tests here, but there
|
in runtime; we also get rid of exit condition tests here, but there
|
is the extra expense for calculating the number of iterations
|
is the extra expense for calculating the number of iterations
|
-- simple unrolling of remaining loops; this is performed only if we
|
-- simple unrolling of remaining loops; this is performed only if we
|
are asked to, as the gain is questionable in this case and often
|
are asked to, as the gain is questionable in this case and often
|
it may even slow down the code
|
it may even slow down the code
|
For more detailed descriptions of each of those, see comments at
|
For more detailed descriptions of each of those, see comments at
|
appropriate function below.
|
appropriate function below.
|
|
|
There is a lot of parameters (defined and described in params.def) that
|
There is a lot of parameters (defined and described in params.def) that
|
control how much we unroll/peel.
|
control how much we unroll/peel.
|
|
|
??? A great problem is that we don't have a good way how to determine
|
??? A great problem is that we don't have a good way how to determine
|
how many times we should unroll the loop; the experiments I have made
|
how many times we should unroll the loop; the experiments I have made
|
showed that this choice may affect performance in order of several %.
|
showed that this choice may affect performance in order of several %.
|
*/
|
*/
|
|
|
/* Information about induction variables to split. */
|
/* Information about induction variables to split. */
|
|
|
struct iv_to_split
|
struct iv_to_split
|
{
|
{
|
rtx insn; /* The insn in that the induction variable occurs. */
|
rtx insn; /* The insn in that the induction variable occurs. */
|
rtx base_var; /* The variable on that the values in the further
|
rtx base_var; /* The variable on that the values in the further
|
iterations are based. */
|
iterations are based. */
|
rtx step; /* Step of the induction variable. */
|
rtx step; /* Step of the induction variable. */
|
unsigned n_loc;
|
unsigned n_loc;
|
unsigned loc[3]; /* Location where the definition of the induction
|
unsigned loc[3]; /* Location where the definition of the induction
|
variable occurs in the insn. For example if
|
variable occurs in the insn. For example if
|
N_LOC is 2, the expression is located at
|
N_LOC is 2, the expression is located at
|
XEXP (XEXP (single_set, loc[0]), loc[1]). */
|
XEXP (XEXP (single_set, loc[0]), loc[1]). */
|
};
|
};
|
|
|
/* Information about accumulators to expand. */
|
/* Information about accumulators to expand. */
|
|
|
struct var_to_expand
|
struct var_to_expand
|
{
|
{
|
rtx insn; /* The insn in that the variable expansion occurs. */
|
rtx insn; /* The insn in that the variable expansion occurs. */
|
rtx reg; /* The accumulator which is expanded. */
|
rtx reg; /* The accumulator which is expanded. */
|
VEC(rtx,heap) *var_expansions; /* The copies of the accumulator which is expanded. */
|
VEC(rtx,heap) *var_expansions; /* The copies of the accumulator which is expanded. */
|
enum rtx_code op; /* The type of the accumulation - addition, subtraction
|
enum rtx_code op; /* The type of the accumulation - addition, subtraction
|
or multiplication. */
|
or multiplication. */
|
int expansion_count; /* Count the number of expansions generated so far. */
|
int expansion_count; /* Count the number of expansions generated so far. */
|
int reuse_expansion; /* The expansion we intend to reuse to expand
|
int reuse_expansion; /* The expansion we intend to reuse to expand
|
the accumulator. If REUSE_EXPANSION is 0 reuse
|
the accumulator. If REUSE_EXPANSION is 0 reuse
|
the original accumulator. Else use
|
the original accumulator. Else use
|
var_expansions[REUSE_EXPANSION - 1]. */
|
var_expansions[REUSE_EXPANSION - 1]. */
|
};
|
};
|
|
|
/* Information about optimization applied in
|
/* Information about optimization applied in
|
the unrolled loop. */
|
the unrolled loop. */
|
|
|
struct opt_info
|
struct opt_info
|
{
|
{
|
htab_t insns_to_split; /* A hashtable of insns to split. */
|
htab_t insns_to_split; /* A hashtable of insns to split. */
|
htab_t insns_with_var_to_expand; /* A hashtable of insns with accumulators
|
htab_t insns_with_var_to_expand; /* A hashtable of insns with accumulators
|
to expand. */
|
to expand. */
|
unsigned first_new_block; /* The first basic block that was
|
unsigned first_new_block; /* The first basic block that was
|
duplicated. */
|
duplicated. */
|
basic_block loop_exit; /* The loop exit basic block. */
|
basic_block loop_exit; /* The loop exit basic block. */
|
basic_block loop_preheader; /* The loop preheader basic block. */
|
basic_block loop_preheader; /* The loop preheader basic block. */
|
};
|
};
|
|
|
static void decide_unrolling_and_peeling (struct loops *, int);
|
static void decide_unrolling_and_peeling (struct loops *, int);
|
static void peel_loops_completely (struct loops *, int);
|
static void peel_loops_completely (struct loops *, int);
|
static void decide_peel_simple (struct loop *, int);
|
static void decide_peel_simple (struct loop *, int);
|
static void decide_peel_once_rolling (struct loop *, int);
|
static void decide_peel_once_rolling (struct loop *, int);
|
static void decide_peel_completely (struct loop *, int);
|
static void decide_peel_completely (struct loop *, int);
|
static void decide_unroll_stupid (struct loop *, int);
|
static void decide_unroll_stupid (struct loop *, int);
|
static void decide_unroll_constant_iterations (struct loop *, int);
|
static void decide_unroll_constant_iterations (struct loop *, int);
|
static void decide_unroll_runtime_iterations (struct loop *, int);
|
static void decide_unroll_runtime_iterations (struct loop *, int);
|
static void peel_loop_simple (struct loops *, struct loop *);
|
static void peel_loop_simple (struct loops *, struct loop *);
|
static void peel_loop_completely (struct loops *, struct loop *);
|
static void peel_loop_completely (struct loops *, struct loop *);
|
static void unroll_loop_stupid (struct loops *, struct loop *);
|
static void unroll_loop_stupid (struct loops *, struct loop *);
|
static void unroll_loop_constant_iterations (struct loops *, struct loop *);
|
static void unroll_loop_constant_iterations (struct loops *, struct loop *);
|
static void unroll_loop_runtime_iterations (struct loops *, struct loop *);
|
static void unroll_loop_runtime_iterations (struct loops *, struct loop *);
|
static struct opt_info *analyze_insns_in_loop (struct loop *);
|
static struct opt_info *analyze_insns_in_loop (struct loop *);
|
static void opt_info_start_duplication (struct opt_info *);
|
static void opt_info_start_duplication (struct opt_info *);
|
static void apply_opt_in_copies (struct opt_info *, unsigned, bool, bool);
|
static void apply_opt_in_copies (struct opt_info *, unsigned, bool, bool);
|
static void free_opt_info (struct opt_info *);
|
static void free_opt_info (struct opt_info *);
|
static struct var_to_expand *analyze_insn_to_expand_var (struct loop*, rtx);
|
static struct var_to_expand *analyze_insn_to_expand_var (struct loop*, rtx);
|
static bool referenced_in_one_insn_in_loop_p (struct loop *, rtx);
|
static bool referenced_in_one_insn_in_loop_p (struct loop *, rtx);
|
static struct iv_to_split *analyze_iv_to_split_insn (rtx);
|
static struct iv_to_split *analyze_iv_to_split_insn (rtx);
|
static void expand_var_during_unrolling (struct var_to_expand *, rtx);
|
static void expand_var_during_unrolling (struct var_to_expand *, rtx);
|
static int insert_var_expansion_initialization (void **, void *);
|
static int insert_var_expansion_initialization (void **, void *);
|
static int combine_var_copies_in_loop_exit (void **, void *);
|
static int combine_var_copies_in_loop_exit (void **, void *);
|
static int release_var_copies (void **, void *);
|
static int release_var_copies (void **, void *);
|
static rtx get_expansion (struct var_to_expand *);
|
static rtx get_expansion (struct var_to_expand *);
|
|
|
/* Unroll and/or peel (depending on FLAGS) LOOPS. */
|
/* Unroll and/or peel (depending on FLAGS) LOOPS. */
|
void
|
void
|
unroll_and_peel_loops (struct loops *loops, int flags)
|
unroll_and_peel_loops (struct loops *loops, int flags)
|
{
|
{
|
struct loop *loop, *next;
|
struct loop *loop, *next;
|
bool check;
|
bool check;
|
|
|
/* First perform complete loop peeling (it is almost surely a win,
|
/* First perform complete loop peeling (it is almost surely a win,
|
and affects parameters for further decision a lot). */
|
and affects parameters for further decision a lot). */
|
peel_loops_completely (loops, flags);
|
peel_loops_completely (loops, flags);
|
|
|
/* Now decide rest of unrolling and peeling. */
|
/* Now decide rest of unrolling and peeling. */
|
decide_unrolling_and_peeling (loops, flags);
|
decide_unrolling_and_peeling (loops, flags);
|
|
|
loop = loops->tree_root;
|
loop = loops->tree_root;
|
while (loop->inner)
|
while (loop->inner)
|
loop = loop->inner;
|
loop = loop->inner;
|
|
|
/* Scan the loops, inner ones first. */
|
/* Scan the loops, inner ones first. */
|
while (loop != loops->tree_root)
|
while (loop != loops->tree_root)
|
{
|
{
|
if (loop->next)
|
if (loop->next)
|
{
|
{
|
next = loop->next;
|
next = loop->next;
|
while (next->inner)
|
while (next->inner)
|
next = next->inner;
|
next = next->inner;
|
}
|
}
|
else
|
else
|
next = loop->outer;
|
next = loop->outer;
|
|
|
check = true;
|
check = true;
|
/* And perform the appropriate transformations. */
|
/* And perform the appropriate transformations. */
|
switch (loop->lpt_decision.decision)
|
switch (loop->lpt_decision.decision)
|
{
|
{
|
case LPT_PEEL_COMPLETELY:
|
case LPT_PEEL_COMPLETELY:
|
/* Already done. */
|
/* Already done. */
|
gcc_unreachable ();
|
gcc_unreachable ();
|
case LPT_PEEL_SIMPLE:
|
case LPT_PEEL_SIMPLE:
|
peel_loop_simple (loops, loop);
|
peel_loop_simple (loops, loop);
|
break;
|
break;
|
case LPT_UNROLL_CONSTANT:
|
case LPT_UNROLL_CONSTANT:
|
unroll_loop_constant_iterations (loops, loop);
|
unroll_loop_constant_iterations (loops, loop);
|
break;
|
break;
|
case LPT_UNROLL_RUNTIME:
|
case LPT_UNROLL_RUNTIME:
|
unroll_loop_runtime_iterations (loops, loop);
|
unroll_loop_runtime_iterations (loops, loop);
|
break;
|
break;
|
case LPT_UNROLL_STUPID:
|
case LPT_UNROLL_STUPID:
|
unroll_loop_stupid (loops, loop);
|
unroll_loop_stupid (loops, loop);
|
break;
|
break;
|
case LPT_NONE:
|
case LPT_NONE:
|
check = false;
|
check = false;
|
break;
|
break;
|
default:
|
default:
|
gcc_unreachable ();
|
gcc_unreachable ();
|
}
|
}
|
if (check)
|
if (check)
|
{
|
{
|
#ifdef ENABLE_CHECKING
|
#ifdef ENABLE_CHECKING
|
verify_dominators (CDI_DOMINATORS);
|
verify_dominators (CDI_DOMINATORS);
|
verify_loop_structure (loops);
|
verify_loop_structure (loops);
|
#endif
|
#endif
|
}
|
}
|
loop = next;
|
loop = next;
|
}
|
}
|
|
|
iv_analysis_done ();
|
iv_analysis_done ();
|
}
|
}
|
|
|
/* Check whether exit of the LOOP is at the end of loop body. */
|
/* Check whether exit of the LOOP is at the end of loop body. */
|
|
|
static bool
|
static bool
|
loop_exit_at_end_p (struct loop *loop)
|
loop_exit_at_end_p (struct loop *loop)
|
{
|
{
|
struct niter_desc *desc = get_simple_loop_desc (loop);
|
struct niter_desc *desc = get_simple_loop_desc (loop);
|
rtx insn;
|
rtx insn;
|
|
|
if (desc->in_edge->dest != loop->latch)
|
if (desc->in_edge->dest != loop->latch)
|
return false;
|
return false;
|
|
|
/* Check that the latch is empty. */
|
/* Check that the latch is empty. */
|
FOR_BB_INSNS (loop->latch, insn)
|
FOR_BB_INSNS (loop->latch, insn)
|
{
|
{
|
if (INSN_P (insn))
|
if (INSN_P (insn))
|
return false;
|
return false;
|
}
|
}
|
|
|
return true;
|
return true;
|
}
|
}
|
|
|
/* Check whether to peel LOOPS (depending on FLAGS) completely and do so. */
|
/* Check whether to peel LOOPS (depending on FLAGS) completely and do so. */
|
static void
|
static void
|
peel_loops_completely (struct loops *loops, int flags)
|
peel_loops_completely (struct loops *loops, int flags)
|
{
|
{
|
struct loop *loop;
|
struct loop *loop;
|
unsigned i;
|
unsigned i;
|
|
|
/* Scan the loops, the inner ones first. */
|
/* Scan the loops, the inner ones first. */
|
for (i = loops->num - 1; i > 0; i--)
|
for (i = loops->num - 1; i > 0; i--)
|
{
|
{
|
loop = loops->parray[i];
|
loop = loops->parray[i];
|
if (!loop)
|
if (!loop)
|
continue;
|
continue;
|
|
|
loop->lpt_decision.decision = LPT_NONE;
|
loop->lpt_decision.decision = LPT_NONE;
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
"\n;; *** Considering loop %d for complete peeling ***\n",
|
"\n;; *** Considering loop %d for complete peeling ***\n",
|
loop->num);
|
loop->num);
|
|
|
loop->ninsns = num_loop_insns (loop);
|
loop->ninsns = num_loop_insns (loop);
|
|
|
decide_peel_once_rolling (loop, flags);
|
decide_peel_once_rolling (loop, flags);
|
if (loop->lpt_decision.decision == LPT_NONE)
|
if (loop->lpt_decision.decision == LPT_NONE)
|
decide_peel_completely (loop, flags);
|
decide_peel_completely (loop, flags);
|
|
|
if (loop->lpt_decision.decision == LPT_PEEL_COMPLETELY)
|
if (loop->lpt_decision.decision == LPT_PEEL_COMPLETELY)
|
{
|
{
|
peel_loop_completely (loops, loop);
|
peel_loop_completely (loops, loop);
|
#ifdef ENABLE_CHECKING
|
#ifdef ENABLE_CHECKING
|
verify_dominators (CDI_DOMINATORS);
|
verify_dominators (CDI_DOMINATORS);
|
verify_loop_structure (loops);
|
verify_loop_structure (loops);
|
#endif
|
#endif
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
/* Decide whether unroll or peel LOOPS (depending on FLAGS) and how much. */
|
/* Decide whether unroll or peel LOOPS (depending on FLAGS) and how much. */
|
static void
|
static void
|
decide_unrolling_and_peeling (struct loops *loops, int flags)
|
decide_unrolling_and_peeling (struct loops *loops, int flags)
|
{
|
{
|
struct loop *loop = loops->tree_root, *next;
|
struct loop *loop = loops->tree_root, *next;
|
|
|
while (loop->inner)
|
while (loop->inner)
|
loop = loop->inner;
|
loop = loop->inner;
|
|
|
/* Scan the loops, inner ones first. */
|
/* Scan the loops, inner ones first. */
|
while (loop != loops->tree_root)
|
while (loop != loops->tree_root)
|
{
|
{
|
if (loop->next)
|
if (loop->next)
|
{
|
{
|
next = loop->next;
|
next = loop->next;
|
while (next->inner)
|
while (next->inner)
|
next = next->inner;
|
next = next->inner;
|
}
|
}
|
else
|
else
|
next = loop->outer;
|
next = loop->outer;
|
|
|
loop->lpt_decision.decision = LPT_NONE;
|
loop->lpt_decision.decision = LPT_NONE;
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "\n;; *** Considering loop %d ***\n", loop->num);
|
fprintf (dump_file, "\n;; *** Considering loop %d ***\n", loop->num);
|
|
|
/* Do not peel cold areas. */
|
/* Do not peel cold areas. */
|
if (!maybe_hot_bb_p (loop->header))
|
if (!maybe_hot_bb_p (loop->header))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Not considering loop, cold area\n");
|
fprintf (dump_file, ";; Not considering loop, cold area\n");
|
loop = next;
|
loop = next;
|
continue;
|
continue;
|
}
|
}
|
|
|
/* Can the loop be manipulated? */
|
/* Can the loop be manipulated? */
|
if (!can_duplicate_loop_p (loop))
|
if (!can_duplicate_loop_p (loop))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
";; Not considering loop, cannot duplicate\n");
|
";; Not considering loop, cannot duplicate\n");
|
loop = next;
|
loop = next;
|
continue;
|
continue;
|
}
|
}
|
|
|
/* Skip non-innermost loops. */
|
/* Skip non-innermost loops. */
|
if (loop->inner)
|
if (loop->inner)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Not considering loop, is not innermost\n");
|
fprintf (dump_file, ";; Not considering loop, is not innermost\n");
|
loop = next;
|
loop = next;
|
continue;
|
continue;
|
}
|
}
|
|
|
loop->ninsns = num_loop_insns (loop);
|
loop->ninsns = num_loop_insns (loop);
|
loop->av_ninsns = average_num_loop_insns (loop);
|
loop->av_ninsns = average_num_loop_insns (loop);
|
|
|
/* Try transformations one by one in decreasing order of
|
/* Try transformations one by one in decreasing order of
|
priority. */
|
priority. */
|
|
|
decide_unroll_constant_iterations (loop, flags);
|
decide_unroll_constant_iterations (loop, flags);
|
if (loop->lpt_decision.decision == LPT_NONE)
|
if (loop->lpt_decision.decision == LPT_NONE)
|
decide_unroll_runtime_iterations (loop, flags);
|
decide_unroll_runtime_iterations (loop, flags);
|
if (loop->lpt_decision.decision == LPT_NONE)
|
if (loop->lpt_decision.decision == LPT_NONE)
|
decide_unroll_stupid (loop, flags);
|
decide_unroll_stupid (loop, flags);
|
if (loop->lpt_decision.decision == LPT_NONE)
|
if (loop->lpt_decision.decision == LPT_NONE)
|
decide_peel_simple (loop, flags);
|
decide_peel_simple (loop, flags);
|
|
|
loop = next;
|
loop = next;
|
}
|
}
|
}
|
}
|
|
|
/* Decide whether the LOOP is once rolling and suitable for complete
|
/* Decide whether the LOOP is once rolling and suitable for complete
|
peeling. */
|
peeling. */
|
static void
|
static void
|
decide_peel_once_rolling (struct loop *loop, int flags ATTRIBUTE_UNUSED)
|
decide_peel_once_rolling (struct loop *loop, int flags ATTRIBUTE_UNUSED)
|
{
|
{
|
struct niter_desc *desc;
|
struct niter_desc *desc;
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "\n;; Considering peeling once rolling loop\n");
|
fprintf (dump_file, "\n;; Considering peeling once rolling loop\n");
|
|
|
/* Is the loop small enough? */
|
/* Is the loop small enough? */
|
if ((unsigned) PARAM_VALUE (PARAM_MAX_ONCE_PEELED_INSNS) < loop->ninsns)
|
if ((unsigned) PARAM_VALUE (PARAM_MAX_ONCE_PEELED_INSNS) < loop->ninsns)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Not considering loop, is too big\n");
|
fprintf (dump_file, ";; Not considering loop, is too big\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* Check for simple loops. */
|
/* Check for simple loops. */
|
desc = get_simple_loop_desc (loop);
|
desc = get_simple_loop_desc (loop);
|
|
|
/* Check number of iterations. */
|
/* Check number of iterations. */
|
if (!desc->simple_p
|
if (!desc->simple_p
|
|| desc->assumptions
|
|| desc->assumptions
|
|| desc->infinite
|
|| desc->infinite
|
|| !desc->const_iter
|
|| !desc->const_iter
|
|| desc->niter != 0)
|
|| desc->niter != 0)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
";; Unable to prove that the loop rolls exactly once\n");
|
";; Unable to prove that the loop rolls exactly once\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* Success. */
|
/* Success. */
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Decided to peel exactly once rolling loop\n");
|
fprintf (dump_file, ";; Decided to peel exactly once rolling loop\n");
|
loop->lpt_decision.decision = LPT_PEEL_COMPLETELY;
|
loop->lpt_decision.decision = LPT_PEEL_COMPLETELY;
|
}
|
}
|
|
|
/* Decide whether the LOOP is suitable for complete peeling. */
|
/* Decide whether the LOOP is suitable for complete peeling. */
|
static void
|
static void
|
decide_peel_completely (struct loop *loop, int flags ATTRIBUTE_UNUSED)
|
decide_peel_completely (struct loop *loop, int flags ATTRIBUTE_UNUSED)
|
{
|
{
|
unsigned npeel;
|
unsigned npeel;
|
struct niter_desc *desc;
|
struct niter_desc *desc;
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "\n;; Considering peeling completely\n");
|
fprintf (dump_file, "\n;; Considering peeling completely\n");
|
|
|
/* Skip non-innermost loops. */
|
/* Skip non-innermost loops. */
|
if (loop->inner)
|
if (loop->inner)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Not considering loop, is not innermost\n");
|
fprintf (dump_file, ";; Not considering loop, is not innermost\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* Do not peel cold areas. */
|
/* Do not peel cold areas. */
|
if (!maybe_hot_bb_p (loop->header))
|
if (!maybe_hot_bb_p (loop->header))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Not considering loop, cold area\n");
|
fprintf (dump_file, ";; Not considering loop, cold area\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* Can the loop be manipulated? */
|
/* Can the loop be manipulated? */
|
if (!can_duplicate_loop_p (loop))
|
if (!can_duplicate_loop_p (loop))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
";; Not considering loop, cannot duplicate\n");
|
";; Not considering loop, cannot duplicate\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* npeel = number of iterations to peel. */
|
/* npeel = number of iterations to peel. */
|
npeel = PARAM_VALUE (PARAM_MAX_COMPLETELY_PEELED_INSNS) / loop->ninsns;
|
npeel = PARAM_VALUE (PARAM_MAX_COMPLETELY_PEELED_INSNS) / loop->ninsns;
|
if (npeel > (unsigned) PARAM_VALUE (PARAM_MAX_COMPLETELY_PEEL_TIMES))
|
if (npeel > (unsigned) PARAM_VALUE (PARAM_MAX_COMPLETELY_PEEL_TIMES))
|
npeel = PARAM_VALUE (PARAM_MAX_COMPLETELY_PEEL_TIMES);
|
npeel = PARAM_VALUE (PARAM_MAX_COMPLETELY_PEEL_TIMES);
|
|
|
/* Is the loop small enough? */
|
/* Is the loop small enough? */
|
if (!npeel)
|
if (!npeel)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Not considering loop, is too big\n");
|
fprintf (dump_file, ";; Not considering loop, is too big\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* Check for simple loops. */
|
/* Check for simple loops. */
|
desc = get_simple_loop_desc (loop);
|
desc = get_simple_loop_desc (loop);
|
|
|
/* Check number of iterations. */
|
/* Check number of iterations. */
|
if (!desc->simple_p
|
if (!desc->simple_p
|
|| desc->assumptions
|
|| desc->assumptions
|
|| !desc->const_iter
|
|| !desc->const_iter
|
|| desc->infinite)
|
|| desc->infinite)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
";; Unable to prove that the loop iterates constant times\n");
|
";; Unable to prove that the loop iterates constant times\n");
|
return;
|
return;
|
}
|
}
|
|
|
if (desc->niter > npeel - 1)
|
if (desc->niter > npeel - 1)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file,
|
fprintf (dump_file,
|
";; Not peeling loop completely, rolls too much (");
|
";; Not peeling loop completely, rolls too much (");
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC, desc->niter);
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC, desc->niter);
|
fprintf (dump_file, " iterations > %d [maximum peelings])\n", npeel);
|
fprintf (dump_file, " iterations > %d [maximum peelings])\n", npeel);
|
}
|
}
|
return;
|
return;
|
}
|
}
|
|
|
/* Success. */
|
/* Success. */
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Decided to peel loop completely\n");
|
fprintf (dump_file, ";; Decided to peel loop completely\n");
|
loop->lpt_decision.decision = LPT_PEEL_COMPLETELY;
|
loop->lpt_decision.decision = LPT_PEEL_COMPLETELY;
|
}
|
}
|
|
|
/* Peel all iterations of LOOP, remove exit edges and cancel the loop
|
/* Peel all iterations of LOOP, remove exit edges and cancel the loop
|
completely. The transformation done:
|
completely. The transformation done:
|
|
|
for (i = 0; i < 4; i++)
|
for (i = 0; i < 4; i++)
|
body;
|
body;
|
|
|
==>
|
==>
|
|
|
i = 0;
|
i = 0;
|
body; i++;
|
body; i++;
|
body; i++;
|
body; i++;
|
body; i++;
|
body; i++;
|
body; i++;
|
body; i++;
|
*/
|
*/
|
static void
|
static void
|
peel_loop_completely (struct loops *loops, struct loop *loop)
|
peel_loop_completely (struct loops *loops, struct loop *loop)
|
{
|
{
|
sbitmap wont_exit;
|
sbitmap wont_exit;
|
unsigned HOST_WIDE_INT npeel;
|
unsigned HOST_WIDE_INT npeel;
|
unsigned n_remove_edges, i;
|
unsigned n_remove_edges, i;
|
edge *remove_edges, ein;
|
edge *remove_edges, ein;
|
struct niter_desc *desc = get_simple_loop_desc (loop);
|
struct niter_desc *desc = get_simple_loop_desc (loop);
|
struct opt_info *opt_info = NULL;
|
struct opt_info *opt_info = NULL;
|
|
|
npeel = desc->niter;
|
npeel = desc->niter;
|
|
|
if (npeel)
|
if (npeel)
|
{
|
{
|
bool ok;
|
bool ok;
|
|
|
wont_exit = sbitmap_alloc (npeel + 1);
|
wont_exit = sbitmap_alloc (npeel + 1);
|
sbitmap_ones (wont_exit);
|
sbitmap_ones (wont_exit);
|
RESET_BIT (wont_exit, 0);
|
RESET_BIT (wont_exit, 0);
|
if (desc->noloop_assumptions)
|
if (desc->noloop_assumptions)
|
RESET_BIT (wont_exit, 1);
|
RESET_BIT (wont_exit, 1);
|
|
|
remove_edges = XCNEWVEC (edge, npeel);
|
remove_edges = XCNEWVEC (edge, npeel);
|
n_remove_edges = 0;
|
n_remove_edges = 0;
|
|
|
if (flag_split_ivs_in_unroller)
|
if (flag_split_ivs_in_unroller)
|
opt_info = analyze_insns_in_loop (loop);
|
opt_info = analyze_insns_in_loop (loop);
|
|
|
opt_info_start_duplication (opt_info);
|
opt_info_start_duplication (opt_info);
|
ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
|
ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
|
loops, npeel,
|
loops, npeel,
|
wont_exit, desc->out_edge,
|
wont_exit, desc->out_edge,
|
remove_edges, &n_remove_edges,
|
remove_edges, &n_remove_edges,
|
DLTHE_FLAG_UPDATE_FREQ
|
DLTHE_FLAG_UPDATE_FREQ
|
| DLTHE_FLAG_COMPLETTE_PEEL
|
| DLTHE_FLAG_COMPLETTE_PEEL
|
| (opt_info
|
| (opt_info
|
? DLTHE_RECORD_COPY_NUMBER : 0));
|
? DLTHE_RECORD_COPY_NUMBER : 0));
|
gcc_assert (ok);
|
gcc_assert (ok);
|
|
|
free (wont_exit);
|
free (wont_exit);
|
|
|
if (opt_info)
|
if (opt_info)
|
{
|
{
|
apply_opt_in_copies (opt_info, npeel, false, true);
|
apply_opt_in_copies (opt_info, npeel, false, true);
|
free_opt_info (opt_info);
|
free_opt_info (opt_info);
|
}
|
}
|
|
|
/* Remove the exit edges. */
|
/* Remove the exit edges. */
|
for (i = 0; i < n_remove_edges; i++)
|
for (i = 0; i < n_remove_edges; i++)
|
remove_path (loops, remove_edges[i]);
|
remove_path (loops, remove_edges[i]);
|
free (remove_edges);
|
free (remove_edges);
|
}
|
}
|
|
|
ein = desc->in_edge;
|
ein = desc->in_edge;
|
free_simple_loop_desc (loop);
|
free_simple_loop_desc (loop);
|
|
|
/* Now remove the unreachable part of the last iteration and cancel
|
/* Now remove the unreachable part of the last iteration and cancel
|
the loop. */
|
the loop. */
|
remove_path (loops, ein);
|
remove_path (loops, ein);
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Peeled loop completely, %d times\n", (int) npeel);
|
fprintf (dump_file, ";; Peeled loop completely, %d times\n", (int) npeel);
|
}
|
}
|
|
|
/* Decide whether to unroll LOOP iterating constant number of times
|
/* Decide whether to unroll LOOP iterating constant number of times
|
and how much. */
|
and how much. */
|
|
|
static void
|
static void
|
decide_unroll_constant_iterations (struct loop *loop, int flags)
|
decide_unroll_constant_iterations (struct loop *loop, int flags)
|
{
|
{
|
unsigned nunroll, nunroll_by_av, best_copies, best_unroll = 0, n_copies, i;
|
unsigned nunroll, nunroll_by_av, best_copies, best_unroll = 0, n_copies, i;
|
struct niter_desc *desc;
|
struct niter_desc *desc;
|
|
|
if (!(flags & UAP_UNROLL))
|
if (!(flags & UAP_UNROLL))
|
{
|
{
|
/* We were not asked to, just return back silently. */
|
/* We were not asked to, just return back silently. */
|
return;
|
return;
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
"\n;; Considering unrolling loop with constant "
|
"\n;; Considering unrolling loop with constant "
|
"number of iterations\n");
|
"number of iterations\n");
|
|
|
/* nunroll = total number of copies of the original loop body in
|
/* nunroll = total number of copies of the original loop body in
|
unrolled loop (i.e. if it is 2, we have to duplicate loop body once. */
|
unrolled loop (i.e. if it is 2, we have to duplicate loop body once. */
|
nunroll = PARAM_VALUE (PARAM_MAX_UNROLLED_INSNS) / loop->ninsns;
|
nunroll = PARAM_VALUE (PARAM_MAX_UNROLLED_INSNS) / loop->ninsns;
|
nunroll_by_av
|
nunroll_by_av
|
= PARAM_VALUE (PARAM_MAX_AVERAGE_UNROLLED_INSNS) / loop->av_ninsns;
|
= PARAM_VALUE (PARAM_MAX_AVERAGE_UNROLLED_INSNS) / loop->av_ninsns;
|
if (nunroll > nunroll_by_av)
|
if (nunroll > nunroll_by_av)
|
nunroll = nunroll_by_av;
|
nunroll = nunroll_by_av;
|
if (nunroll > (unsigned) PARAM_VALUE (PARAM_MAX_UNROLL_TIMES))
|
if (nunroll > (unsigned) PARAM_VALUE (PARAM_MAX_UNROLL_TIMES))
|
nunroll = PARAM_VALUE (PARAM_MAX_UNROLL_TIMES);
|
nunroll = PARAM_VALUE (PARAM_MAX_UNROLL_TIMES);
|
|
|
/* Skip big loops. */
|
/* Skip big loops. */
|
if (nunroll <= 1)
|
if (nunroll <= 1)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Not considering loop, is too big\n");
|
fprintf (dump_file, ";; Not considering loop, is too big\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* Check for simple loops. */
|
/* Check for simple loops. */
|
desc = get_simple_loop_desc (loop);
|
desc = get_simple_loop_desc (loop);
|
|
|
/* Check number of iterations. */
|
/* Check number of iterations. */
|
if (!desc->simple_p || !desc->const_iter || desc->assumptions)
|
if (!desc->simple_p || !desc->const_iter || desc->assumptions)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
";; Unable to prove that the loop iterates constant times\n");
|
";; Unable to prove that the loop iterates constant times\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* Check whether the loop rolls enough to consider. */
|
/* Check whether the loop rolls enough to consider. */
|
if (desc->niter < 2 * nunroll)
|
if (desc->niter < 2 * nunroll)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Not unrolling loop, doesn't roll\n");
|
fprintf (dump_file, ";; Not unrolling loop, doesn't roll\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* Success; now compute number of iterations to unroll. We alter
|
/* Success; now compute number of iterations to unroll. We alter
|
nunroll so that as few as possible copies of loop body are
|
nunroll so that as few as possible copies of loop body are
|
necessary, while still not decreasing the number of unrollings
|
necessary, while still not decreasing the number of unrollings
|
too much (at most by 1). */
|
too much (at most by 1). */
|
best_copies = 2 * nunroll + 10;
|
best_copies = 2 * nunroll + 10;
|
|
|
i = 2 * nunroll + 2;
|
i = 2 * nunroll + 2;
|
if (i - 1 >= desc->niter)
|
if (i - 1 >= desc->niter)
|
i = desc->niter - 2;
|
i = desc->niter - 2;
|
|
|
for (; i >= nunroll - 1; i--)
|
for (; i >= nunroll - 1; i--)
|
{
|
{
|
unsigned exit_mod = desc->niter % (i + 1);
|
unsigned exit_mod = desc->niter % (i + 1);
|
|
|
if (!loop_exit_at_end_p (loop))
|
if (!loop_exit_at_end_p (loop))
|
n_copies = exit_mod + i + 1;
|
n_copies = exit_mod + i + 1;
|
else if (exit_mod != (unsigned) i
|
else if (exit_mod != (unsigned) i
|
|| desc->noloop_assumptions != NULL_RTX)
|
|| desc->noloop_assumptions != NULL_RTX)
|
n_copies = exit_mod + i + 2;
|
n_copies = exit_mod + i + 2;
|
else
|
else
|
n_copies = i + 1;
|
n_copies = i + 1;
|
|
|
if (n_copies < best_copies)
|
if (n_copies < best_copies)
|
{
|
{
|
best_copies = n_copies;
|
best_copies = n_copies;
|
best_unroll = i;
|
best_unroll = i;
|
}
|
}
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; max_unroll %d (%d copies, initial %d).\n",
|
fprintf (dump_file, ";; max_unroll %d (%d copies, initial %d).\n",
|
best_unroll + 1, best_copies, nunroll);
|
best_unroll + 1, best_copies, nunroll);
|
|
|
loop->lpt_decision.decision = LPT_UNROLL_CONSTANT;
|
loop->lpt_decision.decision = LPT_UNROLL_CONSTANT;
|
loop->lpt_decision.times = best_unroll;
|
loop->lpt_decision.times = best_unroll;
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
";; Decided to unroll the constant times rolling loop, %d times.\n",
|
";; Decided to unroll the constant times rolling loop, %d times.\n",
|
loop->lpt_decision.times);
|
loop->lpt_decision.times);
|
}
|
}
|
|
|
/* Unroll LOOP with constant number of iterations LOOP->LPT_DECISION.TIMES + 1
|
/* Unroll LOOP with constant number of iterations LOOP->LPT_DECISION.TIMES + 1
|
times. The transformation does this:
|
times. The transformation does this:
|
|
|
for (i = 0; i < 102; i++)
|
for (i = 0; i < 102; i++)
|
body;
|
body;
|
|
|
==>
|
==>
|
|
|
i = 0;
|
i = 0;
|
body; i++;
|
body; i++;
|
body; i++;
|
body; i++;
|
while (i < 102)
|
while (i < 102)
|
{
|
{
|
body; i++;
|
body; i++;
|
body; i++;
|
body; i++;
|
body; i++;
|
body; i++;
|
body; i++;
|
body; i++;
|
}
|
}
|
*/
|
*/
|
static void
|
static void
|
unroll_loop_constant_iterations (struct loops *loops, struct loop *loop)
|
unroll_loop_constant_iterations (struct loops *loops, struct loop *loop)
|
{
|
{
|
unsigned HOST_WIDE_INT niter;
|
unsigned HOST_WIDE_INT niter;
|
unsigned exit_mod;
|
unsigned exit_mod;
|
sbitmap wont_exit;
|
sbitmap wont_exit;
|
unsigned n_remove_edges, i;
|
unsigned n_remove_edges, i;
|
edge *remove_edges;
|
edge *remove_edges;
|
unsigned max_unroll = loop->lpt_decision.times;
|
unsigned max_unroll = loop->lpt_decision.times;
|
struct niter_desc *desc = get_simple_loop_desc (loop);
|
struct niter_desc *desc = get_simple_loop_desc (loop);
|
bool exit_at_end = loop_exit_at_end_p (loop);
|
bool exit_at_end = loop_exit_at_end_p (loop);
|
struct opt_info *opt_info = NULL;
|
struct opt_info *opt_info = NULL;
|
bool ok;
|
bool ok;
|
|
|
niter = desc->niter;
|
niter = desc->niter;
|
|
|
/* Should not get here (such loop should be peeled instead). */
|
/* Should not get here (such loop should be peeled instead). */
|
gcc_assert (niter > max_unroll + 1);
|
gcc_assert (niter > max_unroll + 1);
|
|
|
exit_mod = niter % (max_unroll + 1);
|
exit_mod = niter % (max_unroll + 1);
|
|
|
wont_exit = sbitmap_alloc (max_unroll + 1);
|
wont_exit = sbitmap_alloc (max_unroll + 1);
|
sbitmap_ones (wont_exit);
|
sbitmap_ones (wont_exit);
|
|
|
remove_edges = XCNEWVEC (edge, max_unroll + exit_mod + 1);
|
remove_edges = XCNEWVEC (edge, max_unroll + exit_mod + 1);
|
n_remove_edges = 0;
|
n_remove_edges = 0;
|
if (flag_split_ivs_in_unroller
|
if (flag_split_ivs_in_unroller
|
|| flag_variable_expansion_in_unroller)
|
|| flag_variable_expansion_in_unroller)
|
opt_info = analyze_insns_in_loop (loop);
|
opt_info = analyze_insns_in_loop (loop);
|
|
|
if (!exit_at_end)
|
if (!exit_at_end)
|
{
|
{
|
/* The exit is not at the end of the loop; leave exit test
|
/* The exit is not at the end of the loop; leave exit test
|
in the first copy, so that the loops that start with test
|
in the first copy, so that the loops that start with test
|
of exit condition have continuous body after unrolling. */
|
of exit condition have continuous body after unrolling. */
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Condition on beginning of loop.\n");
|
fprintf (dump_file, ";; Condition on beginning of loop.\n");
|
|
|
/* Peel exit_mod iterations. */
|
/* Peel exit_mod iterations. */
|
RESET_BIT (wont_exit, 0);
|
RESET_BIT (wont_exit, 0);
|
if (desc->noloop_assumptions)
|
if (desc->noloop_assumptions)
|
RESET_BIT (wont_exit, 1);
|
RESET_BIT (wont_exit, 1);
|
|
|
if (exit_mod)
|
if (exit_mod)
|
{
|
{
|
opt_info_start_duplication (opt_info);
|
opt_info_start_duplication (opt_info);
|
ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
|
ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
|
loops, exit_mod,
|
loops, exit_mod,
|
wont_exit, desc->out_edge,
|
wont_exit, desc->out_edge,
|
remove_edges, &n_remove_edges,
|
remove_edges, &n_remove_edges,
|
DLTHE_FLAG_UPDATE_FREQ
|
DLTHE_FLAG_UPDATE_FREQ
|
| (opt_info && exit_mod > 1
|
| (opt_info && exit_mod > 1
|
? DLTHE_RECORD_COPY_NUMBER
|
? DLTHE_RECORD_COPY_NUMBER
|
: 0));
|
: 0));
|
gcc_assert (ok);
|
gcc_assert (ok);
|
|
|
if (opt_info && exit_mod > 1)
|
if (opt_info && exit_mod > 1)
|
apply_opt_in_copies (opt_info, exit_mod, false, false);
|
apply_opt_in_copies (opt_info, exit_mod, false, false);
|
|
|
desc->noloop_assumptions = NULL_RTX;
|
desc->noloop_assumptions = NULL_RTX;
|
desc->niter -= exit_mod;
|
desc->niter -= exit_mod;
|
desc->niter_max -= exit_mod;
|
desc->niter_max -= exit_mod;
|
}
|
}
|
|
|
SET_BIT (wont_exit, 1);
|
SET_BIT (wont_exit, 1);
|
}
|
}
|
else
|
else
|
{
|
{
|
/* Leave exit test in last copy, for the same reason as above if
|
/* Leave exit test in last copy, for the same reason as above if
|
the loop tests the condition at the end of loop body. */
|
the loop tests the condition at the end of loop body. */
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Condition on end of loop.\n");
|
fprintf (dump_file, ";; Condition on end of loop.\n");
|
|
|
/* We know that niter >= max_unroll + 2; so we do not need to care of
|
/* We know that niter >= max_unroll + 2; so we do not need to care of
|
case when we would exit before reaching the loop. So just peel
|
case when we would exit before reaching the loop. So just peel
|
exit_mod + 1 iterations. */
|
exit_mod + 1 iterations. */
|
if (exit_mod != max_unroll
|
if (exit_mod != max_unroll
|
|| desc->noloop_assumptions)
|
|| desc->noloop_assumptions)
|
{
|
{
|
RESET_BIT (wont_exit, 0);
|
RESET_BIT (wont_exit, 0);
|
if (desc->noloop_assumptions)
|
if (desc->noloop_assumptions)
|
RESET_BIT (wont_exit, 1);
|
RESET_BIT (wont_exit, 1);
|
|
|
opt_info_start_duplication (opt_info);
|
opt_info_start_duplication (opt_info);
|
ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
|
ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
|
loops, exit_mod + 1,
|
loops, exit_mod + 1,
|
wont_exit, desc->out_edge,
|
wont_exit, desc->out_edge,
|
remove_edges, &n_remove_edges,
|
remove_edges, &n_remove_edges,
|
DLTHE_FLAG_UPDATE_FREQ
|
DLTHE_FLAG_UPDATE_FREQ
|
| (opt_info && exit_mod > 0
|
| (opt_info && exit_mod > 0
|
? DLTHE_RECORD_COPY_NUMBER
|
? DLTHE_RECORD_COPY_NUMBER
|
: 0));
|
: 0));
|
gcc_assert (ok);
|
gcc_assert (ok);
|
|
|
if (opt_info && exit_mod > 0)
|
if (opt_info && exit_mod > 0)
|
apply_opt_in_copies (opt_info, exit_mod + 1, false, false);
|
apply_opt_in_copies (opt_info, exit_mod + 1, false, false);
|
|
|
desc->niter -= exit_mod + 1;
|
desc->niter -= exit_mod + 1;
|
desc->niter_max -= exit_mod + 1;
|
desc->niter_max -= exit_mod + 1;
|
desc->noloop_assumptions = NULL_RTX;
|
desc->noloop_assumptions = NULL_RTX;
|
|
|
SET_BIT (wont_exit, 0);
|
SET_BIT (wont_exit, 0);
|
SET_BIT (wont_exit, 1);
|
SET_BIT (wont_exit, 1);
|
}
|
}
|
|
|
RESET_BIT (wont_exit, max_unroll);
|
RESET_BIT (wont_exit, max_unroll);
|
}
|
}
|
|
|
/* Now unroll the loop. */
|
/* Now unroll the loop. */
|
|
|
opt_info_start_duplication (opt_info);
|
opt_info_start_duplication (opt_info);
|
ok = duplicate_loop_to_header_edge (loop, loop_latch_edge (loop),
|
ok = duplicate_loop_to_header_edge (loop, loop_latch_edge (loop),
|
loops, max_unroll,
|
loops, max_unroll,
|
wont_exit, desc->out_edge,
|
wont_exit, desc->out_edge,
|
remove_edges, &n_remove_edges,
|
remove_edges, &n_remove_edges,
|
DLTHE_FLAG_UPDATE_FREQ
|
DLTHE_FLAG_UPDATE_FREQ
|
| (opt_info
|
| (opt_info
|
? DLTHE_RECORD_COPY_NUMBER
|
? DLTHE_RECORD_COPY_NUMBER
|
: 0));
|
: 0));
|
gcc_assert (ok);
|
gcc_assert (ok);
|
|
|
if (opt_info)
|
if (opt_info)
|
{
|
{
|
apply_opt_in_copies (opt_info, max_unroll, true, true);
|
apply_opt_in_copies (opt_info, max_unroll, true, true);
|
free_opt_info (opt_info);
|
free_opt_info (opt_info);
|
}
|
}
|
|
|
free (wont_exit);
|
free (wont_exit);
|
|
|
if (exit_at_end)
|
if (exit_at_end)
|
{
|
{
|
basic_block exit_block = get_bb_copy (desc->in_edge->src);
|
basic_block exit_block = get_bb_copy (desc->in_edge->src);
|
/* Find a new in and out edge; they are in the last copy we have made. */
|
/* Find a new in and out edge; they are in the last copy we have made. */
|
|
|
if (EDGE_SUCC (exit_block, 0)->dest == desc->out_edge->dest)
|
if (EDGE_SUCC (exit_block, 0)->dest == desc->out_edge->dest)
|
{
|
{
|
desc->out_edge = EDGE_SUCC (exit_block, 0);
|
desc->out_edge = EDGE_SUCC (exit_block, 0);
|
desc->in_edge = EDGE_SUCC (exit_block, 1);
|
desc->in_edge = EDGE_SUCC (exit_block, 1);
|
}
|
}
|
else
|
else
|
{
|
{
|
desc->out_edge = EDGE_SUCC (exit_block, 1);
|
desc->out_edge = EDGE_SUCC (exit_block, 1);
|
desc->in_edge = EDGE_SUCC (exit_block, 0);
|
desc->in_edge = EDGE_SUCC (exit_block, 0);
|
}
|
}
|
}
|
}
|
|
|
desc->niter /= max_unroll + 1;
|
desc->niter /= max_unroll + 1;
|
desc->niter_max /= max_unroll + 1;
|
desc->niter_max /= max_unroll + 1;
|
desc->niter_expr = GEN_INT (desc->niter);
|
desc->niter_expr = GEN_INT (desc->niter);
|
|
|
/* Remove the edges. */
|
/* Remove the edges. */
|
for (i = 0; i < n_remove_edges; i++)
|
for (i = 0; i < n_remove_edges; i++)
|
remove_path (loops, remove_edges[i]);
|
remove_path (loops, remove_edges[i]);
|
free (remove_edges);
|
free (remove_edges);
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
";; Unrolled loop %d times, constant # of iterations %i insns\n",
|
";; Unrolled loop %d times, constant # of iterations %i insns\n",
|
max_unroll, num_loop_insns (loop));
|
max_unroll, num_loop_insns (loop));
|
}
|
}
|
|
|
/* Decide whether to unroll LOOP iterating runtime computable number of times
|
/* Decide whether to unroll LOOP iterating runtime computable number of times
|
and how much. */
|
and how much. */
|
static void
|
static void
|
decide_unroll_runtime_iterations (struct loop *loop, int flags)
|
decide_unroll_runtime_iterations (struct loop *loop, int flags)
|
{
|
{
|
unsigned nunroll, nunroll_by_av, i;
|
unsigned nunroll, nunroll_by_av, i;
|
struct niter_desc *desc;
|
struct niter_desc *desc;
|
|
|
if (!(flags & UAP_UNROLL))
|
if (!(flags & UAP_UNROLL))
|
{
|
{
|
/* We were not asked to, just return back silently. */
|
/* We were not asked to, just return back silently. */
|
return;
|
return;
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
"\n;; Considering unrolling loop with runtime "
|
"\n;; Considering unrolling loop with runtime "
|
"computable number of iterations\n");
|
"computable number of iterations\n");
|
|
|
/* nunroll = total number of copies of the original loop body in
|
/* nunroll = total number of copies of the original loop body in
|
unrolled loop (i.e. if it is 2, we have to duplicate loop body once. */
|
unrolled loop (i.e. if it is 2, we have to duplicate loop body once. */
|
nunroll = PARAM_VALUE (PARAM_MAX_UNROLLED_INSNS) / loop->ninsns;
|
nunroll = PARAM_VALUE (PARAM_MAX_UNROLLED_INSNS) / loop->ninsns;
|
nunroll_by_av = PARAM_VALUE (PARAM_MAX_AVERAGE_UNROLLED_INSNS) / loop->av_ninsns;
|
nunroll_by_av = PARAM_VALUE (PARAM_MAX_AVERAGE_UNROLLED_INSNS) / loop->av_ninsns;
|
if (nunroll > nunroll_by_av)
|
if (nunroll > nunroll_by_av)
|
nunroll = nunroll_by_av;
|
nunroll = nunroll_by_av;
|
if (nunroll > (unsigned) PARAM_VALUE (PARAM_MAX_UNROLL_TIMES))
|
if (nunroll > (unsigned) PARAM_VALUE (PARAM_MAX_UNROLL_TIMES))
|
nunroll = PARAM_VALUE (PARAM_MAX_UNROLL_TIMES);
|
nunroll = PARAM_VALUE (PARAM_MAX_UNROLL_TIMES);
|
|
|
/* Skip big loops. */
|
/* Skip big loops. */
|
if (nunroll <= 1)
|
if (nunroll <= 1)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Not considering loop, is too big\n");
|
fprintf (dump_file, ";; Not considering loop, is too big\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* Check for simple loops. */
|
/* Check for simple loops. */
|
desc = get_simple_loop_desc (loop);
|
desc = get_simple_loop_desc (loop);
|
|
|
/* Check simpleness. */
|
/* Check simpleness. */
|
if (!desc->simple_p || desc->assumptions)
|
if (!desc->simple_p || desc->assumptions)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
";; Unable to prove that the number of iterations "
|
";; Unable to prove that the number of iterations "
|
"can be counted in runtime\n");
|
"can be counted in runtime\n");
|
return;
|
return;
|
}
|
}
|
|
|
if (desc->const_iter)
|
if (desc->const_iter)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Loop iterates constant times\n");
|
fprintf (dump_file, ";; Loop iterates constant times\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* If we have profile feedback, check whether the loop rolls. */
|
/* If we have profile feedback, check whether the loop rolls. */
|
if (loop->header->count && expected_loop_iterations (loop) < 2 * nunroll)
|
if (loop->header->count && expected_loop_iterations (loop) < 2 * nunroll)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Not unrolling loop, doesn't roll\n");
|
fprintf (dump_file, ";; Not unrolling loop, doesn't roll\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* Success; now force nunroll to be power of 2, as we are unable to
|
/* Success; now force nunroll to be power of 2, as we are unable to
|
cope with overflows in computation of number of iterations. */
|
cope with overflows in computation of number of iterations. */
|
for (i = 1; 2 * i <= nunroll; i *= 2)
|
for (i = 1; 2 * i <= nunroll; i *= 2)
|
continue;
|
continue;
|
|
|
loop->lpt_decision.decision = LPT_UNROLL_RUNTIME;
|
loop->lpt_decision.decision = LPT_UNROLL_RUNTIME;
|
loop->lpt_decision.times = i - 1;
|
loop->lpt_decision.times = i - 1;
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
";; Decided to unroll the runtime computable "
|
";; Decided to unroll the runtime computable "
|
"times rolling loop, %d times.\n",
|
"times rolling loop, %d times.\n",
|
loop->lpt_decision.times);
|
loop->lpt_decision.times);
|
}
|
}
|
|
|
/* Unroll LOOP for that we are able to count number of iterations in runtime
|
/* Unroll LOOP for that we are able to count number of iterations in runtime
|
LOOP->LPT_DECISION.TIMES + 1 times. The transformation does this (with some
|
LOOP->LPT_DECISION.TIMES + 1 times. The transformation does this (with some
|
extra care for case n < 0):
|
extra care for case n < 0):
|
|
|
for (i = 0; i < n; i++)
|
for (i = 0; i < n; i++)
|
body;
|
body;
|
|
|
==>
|
==>
|
|
|
i = 0;
|
i = 0;
|
mod = n % 4;
|
mod = n % 4;
|
|
|
switch (mod)
|
switch (mod)
|
{
|
{
|
case 3:
|
case 3:
|
body; i++;
|
body; i++;
|
case 2:
|
case 2:
|
body; i++;
|
body; i++;
|
case 1:
|
case 1:
|
body; i++;
|
body; i++;
|
case 0: ;
|
case 0: ;
|
}
|
}
|
|
|
while (i < n)
|
while (i < n)
|
{
|
{
|
body; i++;
|
body; i++;
|
body; i++;
|
body; i++;
|
body; i++;
|
body; i++;
|
body; i++;
|
body; i++;
|
}
|
}
|
*/
|
*/
|
static void
|
static void
|
unroll_loop_runtime_iterations (struct loops *loops, struct loop *loop)
|
unroll_loop_runtime_iterations (struct loops *loops, struct loop *loop)
|
{
|
{
|
rtx old_niter, niter, init_code, branch_code, tmp;
|
rtx old_niter, niter, init_code, branch_code, tmp;
|
unsigned i, j, p;
|
unsigned i, j, p;
|
basic_block preheader, *body, *dom_bbs, swtch, ezc_swtch;
|
basic_block preheader, *body, *dom_bbs, swtch, ezc_swtch;
|
unsigned n_dom_bbs;
|
unsigned n_dom_bbs;
|
sbitmap wont_exit;
|
sbitmap wont_exit;
|
int may_exit_copy;
|
int may_exit_copy;
|
unsigned n_peel, n_remove_edges;
|
unsigned n_peel, n_remove_edges;
|
edge *remove_edges, e;
|
edge *remove_edges, e;
|
bool extra_zero_check, last_may_exit;
|
bool extra_zero_check, last_may_exit;
|
unsigned max_unroll = loop->lpt_decision.times;
|
unsigned max_unroll = loop->lpt_decision.times;
|
struct niter_desc *desc = get_simple_loop_desc (loop);
|
struct niter_desc *desc = get_simple_loop_desc (loop);
|
bool exit_at_end = loop_exit_at_end_p (loop);
|
bool exit_at_end = loop_exit_at_end_p (loop);
|
struct opt_info *opt_info = NULL;
|
struct opt_info *opt_info = NULL;
|
bool ok;
|
bool ok;
|
|
|
if (flag_split_ivs_in_unroller
|
if (flag_split_ivs_in_unroller
|
|| flag_variable_expansion_in_unroller)
|
|| flag_variable_expansion_in_unroller)
|
opt_info = analyze_insns_in_loop (loop);
|
opt_info = analyze_insns_in_loop (loop);
|
|
|
/* Remember blocks whose dominators will have to be updated. */
|
/* Remember blocks whose dominators will have to be updated. */
|
dom_bbs = XCNEWVEC (basic_block, n_basic_blocks);
|
dom_bbs = XCNEWVEC (basic_block, n_basic_blocks);
|
n_dom_bbs = 0;
|
n_dom_bbs = 0;
|
|
|
body = get_loop_body (loop);
|
body = get_loop_body (loop);
|
for (i = 0; i < loop->num_nodes; i++)
|
for (i = 0; i < loop->num_nodes; i++)
|
{
|
{
|
unsigned nldom;
|
unsigned nldom;
|
basic_block *ldom;
|
basic_block *ldom;
|
|
|
nldom = get_dominated_by (CDI_DOMINATORS, body[i], &ldom);
|
nldom = get_dominated_by (CDI_DOMINATORS, body[i], &ldom);
|
for (j = 0; j < nldom; j++)
|
for (j = 0; j < nldom; j++)
|
if (!flow_bb_inside_loop_p (loop, ldom[j]))
|
if (!flow_bb_inside_loop_p (loop, ldom[j]))
|
dom_bbs[n_dom_bbs++] = ldom[j];
|
dom_bbs[n_dom_bbs++] = ldom[j];
|
|
|
free (ldom);
|
free (ldom);
|
}
|
}
|
free (body);
|
free (body);
|
|
|
if (!exit_at_end)
|
if (!exit_at_end)
|
{
|
{
|
/* Leave exit in first copy (for explanation why see comment in
|
/* Leave exit in first copy (for explanation why see comment in
|
unroll_loop_constant_iterations). */
|
unroll_loop_constant_iterations). */
|
may_exit_copy = 0;
|
may_exit_copy = 0;
|
n_peel = max_unroll - 1;
|
n_peel = max_unroll - 1;
|
extra_zero_check = true;
|
extra_zero_check = true;
|
last_may_exit = false;
|
last_may_exit = false;
|
}
|
}
|
else
|
else
|
{
|
{
|
/* Leave exit in last copy (for explanation why see comment in
|
/* Leave exit in last copy (for explanation why see comment in
|
unroll_loop_constant_iterations). */
|
unroll_loop_constant_iterations). */
|
may_exit_copy = max_unroll;
|
may_exit_copy = max_unroll;
|
n_peel = max_unroll;
|
n_peel = max_unroll;
|
extra_zero_check = false;
|
extra_zero_check = false;
|
last_may_exit = true;
|
last_may_exit = true;
|
}
|
}
|
|
|
/* Get expression for number of iterations. */
|
/* Get expression for number of iterations. */
|
start_sequence ();
|
start_sequence ();
|
old_niter = niter = gen_reg_rtx (desc->mode);
|
old_niter = niter = gen_reg_rtx (desc->mode);
|
tmp = force_operand (copy_rtx (desc->niter_expr), niter);
|
tmp = force_operand (copy_rtx (desc->niter_expr), niter);
|
if (tmp != niter)
|
if (tmp != niter)
|
emit_move_insn (niter, tmp);
|
emit_move_insn (niter, tmp);
|
|
|
/* Count modulo by ANDing it with max_unroll; we use the fact that
|
/* Count modulo by ANDing it with max_unroll; we use the fact that
|
the number of unrollings is a power of two, and thus this is correct
|
the number of unrollings is a power of two, and thus this is correct
|
even if there is overflow in the computation. */
|
even if there is overflow in the computation. */
|
niter = expand_simple_binop (desc->mode, AND,
|
niter = expand_simple_binop (desc->mode, AND,
|
niter,
|
niter,
|
GEN_INT (max_unroll),
|
GEN_INT (max_unroll),
|
NULL_RTX, 0, OPTAB_LIB_WIDEN);
|
NULL_RTX, 0, OPTAB_LIB_WIDEN);
|
|
|
init_code = get_insns ();
|
init_code = get_insns ();
|
end_sequence ();
|
end_sequence ();
|
|
|
/* Precondition the loop. */
|
/* Precondition the loop. */
|
loop_split_edge_with (loop_preheader_edge (loop), init_code);
|
loop_split_edge_with (loop_preheader_edge (loop), init_code);
|
|
|
remove_edges = XCNEWVEC (edge, max_unroll + n_peel + 1);
|
remove_edges = XCNEWVEC (edge, max_unroll + n_peel + 1);
|
n_remove_edges = 0;
|
n_remove_edges = 0;
|
|
|
wont_exit = sbitmap_alloc (max_unroll + 2);
|
wont_exit = sbitmap_alloc (max_unroll + 2);
|
|
|
/* Peel the first copy of loop body (almost always we must leave exit test
|
/* Peel the first copy of loop body (almost always we must leave exit test
|
here; the only exception is when we have extra zero check and the number
|
here; the only exception is when we have extra zero check and the number
|
of iterations is reliable. Also record the place of (possible) extra
|
of iterations is reliable. Also record the place of (possible) extra
|
zero check. */
|
zero check. */
|
sbitmap_zero (wont_exit);
|
sbitmap_zero (wont_exit);
|
if (extra_zero_check
|
if (extra_zero_check
|
&& !desc->noloop_assumptions)
|
&& !desc->noloop_assumptions)
|
SET_BIT (wont_exit, 1);
|
SET_BIT (wont_exit, 1);
|
ezc_swtch = loop_preheader_edge (loop)->src;
|
ezc_swtch = loop_preheader_edge (loop)->src;
|
ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
|
ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
|
loops, 1,
|
loops, 1,
|
wont_exit, desc->out_edge,
|
wont_exit, desc->out_edge,
|
remove_edges, &n_remove_edges,
|
remove_edges, &n_remove_edges,
|
DLTHE_FLAG_UPDATE_FREQ);
|
DLTHE_FLAG_UPDATE_FREQ);
|
gcc_assert (ok);
|
gcc_assert (ok);
|
|
|
/* Record the place where switch will be built for preconditioning. */
|
/* Record the place where switch will be built for preconditioning. */
|
swtch = loop_split_edge_with (loop_preheader_edge (loop),
|
swtch = loop_split_edge_with (loop_preheader_edge (loop),
|
NULL_RTX);
|
NULL_RTX);
|
|
|
for (i = 0; i < n_peel; i++)
|
for (i = 0; i < n_peel; i++)
|
{
|
{
|
/* Peel the copy. */
|
/* Peel the copy. */
|
sbitmap_zero (wont_exit);
|
sbitmap_zero (wont_exit);
|
if (i != n_peel - 1 || !last_may_exit)
|
if (i != n_peel - 1 || !last_may_exit)
|
SET_BIT (wont_exit, 1);
|
SET_BIT (wont_exit, 1);
|
ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
|
ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
|
loops, 1,
|
loops, 1,
|
wont_exit, desc->out_edge,
|
wont_exit, desc->out_edge,
|
remove_edges, &n_remove_edges,
|
remove_edges, &n_remove_edges,
|
DLTHE_FLAG_UPDATE_FREQ);
|
DLTHE_FLAG_UPDATE_FREQ);
|
gcc_assert (ok);
|
gcc_assert (ok);
|
|
|
/* Create item for switch. */
|
/* Create item for switch. */
|
j = n_peel - i - (extra_zero_check ? 0 : 1);
|
j = n_peel - i - (extra_zero_check ? 0 : 1);
|
p = REG_BR_PROB_BASE / (i + 2);
|
p = REG_BR_PROB_BASE / (i + 2);
|
|
|
preheader = loop_split_edge_with (loop_preheader_edge (loop), NULL_RTX);
|
preheader = loop_split_edge_with (loop_preheader_edge (loop), NULL_RTX);
|
branch_code = compare_and_jump_seq (copy_rtx (niter), GEN_INT (j), EQ,
|
branch_code = compare_and_jump_seq (copy_rtx (niter), GEN_INT (j), EQ,
|
block_label (preheader), p,
|
block_label (preheader), p,
|
NULL_RTX);
|
NULL_RTX);
|
|
|
swtch = loop_split_edge_with (single_pred_edge (swtch), branch_code);
|
swtch = loop_split_edge_with (single_pred_edge (swtch), branch_code);
|
set_immediate_dominator (CDI_DOMINATORS, preheader, swtch);
|
set_immediate_dominator (CDI_DOMINATORS, preheader, swtch);
|
single_pred_edge (swtch)->probability = REG_BR_PROB_BASE - p;
|
single_pred_edge (swtch)->probability = REG_BR_PROB_BASE - p;
|
e = make_edge (swtch, preheader,
|
e = make_edge (swtch, preheader,
|
single_succ_edge (swtch)->flags & EDGE_IRREDUCIBLE_LOOP);
|
single_succ_edge (swtch)->flags & EDGE_IRREDUCIBLE_LOOP);
|
e->probability = p;
|
e->probability = p;
|
}
|
}
|
|
|
if (extra_zero_check)
|
if (extra_zero_check)
|
{
|
{
|
/* Add branch for zero iterations. */
|
/* Add branch for zero iterations. */
|
p = REG_BR_PROB_BASE / (max_unroll + 1);
|
p = REG_BR_PROB_BASE / (max_unroll + 1);
|
swtch = ezc_swtch;
|
swtch = ezc_swtch;
|
preheader = loop_split_edge_with (loop_preheader_edge (loop), NULL_RTX);
|
preheader = loop_split_edge_with (loop_preheader_edge (loop), NULL_RTX);
|
branch_code = compare_and_jump_seq (copy_rtx (niter), const0_rtx, EQ,
|
branch_code = compare_and_jump_seq (copy_rtx (niter), const0_rtx, EQ,
|
block_label (preheader), p,
|
block_label (preheader), p,
|
NULL_RTX);
|
NULL_RTX);
|
|
|
swtch = loop_split_edge_with (single_succ_edge (swtch), branch_code);
|
swtch = loop_split_edge_with (single_succ_edge (swtch), branch_code);
|
set_immediate_dominator (CDI_DOMINATORS, preheader, swtch);
|
set_immediate_dominator (CDI_DOMINATORS, preheader, swtch);
|
single_succ_edge (swtch)->probability = REG_BR_PROB_BASE - p;
|
single_succ_edge (swtch)->probability = REG_BR_PROB_BASE - p;
|
e = make_edge (swtch, preheader,
|
e = make_edge (swtch, preheader,
|
single_succ_edge (swtch)->flags & EDGE_IRREDUCIBLE_LOOP);
|
single_succ_edge (swtch)->flags & EDGE_IRREDUCIBLE_LOOP);
|
e->probability = p;
|
e->probability = p;
|
}
|
}
|
|
|
/* Recount dominators for outer blocks. */
|
/* Recount dominators for outer blocks. */
|
iterate_fix_dominators (CDI_DOMINATORS, dom_bbs, n_dom_bbs);
|
iterate_fix_dominators (CDI_DOMINATORS, dom_bbs, n_dom_bbs);
|
|
|
/* And unroll loop. */
|
/* And unroll loop. */
|
|
|
sbitmap_ones (wont_exit);
|
sbitmap_ones (wont_exit);
|
RESET_BIT (wont_exit, may_exit_copy);
|
RESET_BIT (wont_exit, may_exit_copy);
|
opt_info_start_duplication (opt_info);
|
opt_info_start_duplication (opt_info);
|
|
|
ok = duplicate_loop_to_header_edge (loop, loop_latch_edge (loop),
|
ok = duplicate_loop_to_header_edge (loop, loop_latch_edge (loop),
|
loops, max_unroll,
|
loops, max_unroll,
|
wont_exit, desc->out_edge,
|
wont_exit, desc->out_edge,
|
remove_edges, &n_remove_edges,
|
remove_edges, &n_remove_edges,
|
DLTHE_FLAG_UPDATE_FREQ
|
DLTHE_FLAG_UPDATE_FREQ
|
| (opt_info
|
| (opt_info
|
? DLTHE_RECORD_COPY_NUMBER
|
? DLTHE_RECORD_COPY_NUMBER
|
: 0));
|
: 0));
|
gcc_assert (ok);
|
gcc_assert (ok);
|
|
|
if (opt_info)
|
if (opt_info)
|
{
|
{
|
apply_opt_in_copies (opt_info, max_unroll, true, true);
|
apply_opt_in_copies (opt_info, max_unroll, true, true);
|
free_opt_info (opt_info);
|
free_opt_info (opt_info);
|
}
|
}
|
|
|
free (wont_exit);
|
free (wont_exit);
|
|
|
if (exit_at_end)
|
if (exit_at_end)
|
{
|
{
|
basic_block exit_block = get_bb_copy (desc->in_edge->src);
|
basic_block exit_block = get_bb_copy (desc->in_edge->src);
|
/* Find a new in and out edge; they are in the last copy we have
|
/* Find a new in and out edge; they are in the last copy we have
|
made. */
|
made. */
|
|
|
if (EDGE_SUCC (exit_block, 0)->dest == desc->out_edge->dest)
|
if (EDGE_SUCC (exit_block, 0)->dest == desc->out_edge->dest)
|
{
|
{
|
desc->out_edge = EDGE_SUCC (exit_block, 0);
|
desc->out_edge = EDGE_SUCC (exit_block, 0);
|
desc->in_edge = EDGE_SUCC (exit_block, 1);
|
desc->in_edge = EDGE_SUCC (exit_block, 1);
|
}
|
}
|
else
|
else
|
{
|
{
|
desc->out_edge = EDGE_SUCC (exit_block, 1);
|
desc->out_edge = EDGE_SUCC (exit_block, 1);
|
desc->in_edge = EDGE_SUCC (exit_block, 0);
|
desc->in_edge = EDGE_SUCC (exit_block, 0);
|
}
|
}
|
}
|
}
|
|
|
/* Remove the edges. */
|
/* Remove the edges. */
|
for (i = 0; i < n_remove_edges; i++)
|
for (i = 0; i < n_remove_edges; i++)
|
remove_path (loops, remove_edges[i]);
|
remove_path (loops, remove_edges[i]);
|
free (remove_edges);
|
free (remove_edges);
|
|
|
/* We must be careful when updating the number of iterations due to
|
/* We must be careful when updating the number of iterations due to
|
preconditioning and the fact that the value must be valid at entry
|
preconditioning and the fact that the value must be valid at entry
|
of the loop. After passing through the above code, we see that
|
of the loop. After passing through the above code, we see that
|
the correct new number of iterations is this: */
|
the correct new number of iterations is this: */
|
gcc_assert (!desc->const_iter);
|
gcc_assert (!desc->const_iter);
|
desc->niter_expr =
|
desc->niter_expr =
|
simplify_gen_binary (UDIV, desc->mode, old_niter,
|
simplify_gen_binary (UDIV, desc->mode, old_niter,
|
GEN_INT (max_unroll + 1));
|
GEN_INT (max_unroll + 1));
|
desc->niter_max /= max_unroll + 1;
|
desc->niter_max /= max_unroll + 1;
|
if (exit_at_end)
|
if (exit_at_end)
|
{
|
{
|
desc->niter_expr =
|
desc->niter_expr =
|
simplify_gen_binary (MINUS, desc->mode, desc->niter_expr, const1_rtx);
|
simplify_gen_binary (MINUS, desc->mode, desc->niter_expr, const1_rtx);
|
desc->noloop_assumptions = NULL_RTX;
|
desc->noloop_assumptions = NULL_RTX;
|
desc->niter_max--;
|
desc->niter_max--;
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
";; Unrolled loop %d times, counting # of iterations "
|
";; Unrolled loop %d times, counting # of iterations "
|
"in runtime, %i insns\n",
|
"in runtime, %i insns\n",
|
max_unroll, num_loop_insns (loop));
|
max_unroll, num_loop_insns (loop));
|
|
|
if (dom_bbs)
|
if (dom_bbs)
|
free (dom_bbs);
|
free (dom_bbs);
|
}
|
}
|
|
|
/* Decide whether to simply peel LOOP and how much. */
|
/* Decide whether to simply peel LOOP and how much. */
|
static void
|
static void
|
decide_peel_simple (struct loop *loop, int flags)
|
decide_peel_simple (struct loop *loop, int flags)
|
{
|
{
|
unsigned npeel;
|
unsigned npeel;
|
struct niter_desc *desc;
|
struct niter_desc *desc;
|
|
|
if (!(flags & UAP_PEEL))
|
if (!(flags & UAP_PEEL))
|
{
|
{
|
/* We were not asked to, just return back silently. */
|
/* We were not asked to, just return back silently. */
|
return;
|
return;
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "\n;; Considering simply peeling loop\n");
|
fprintf (dump_file, "\n;; Considering simply peeling loop\n");
|
|
|
/* npeel = number of iterations to peel. */
|
/* npeel = number of iterations to peel. */
|
npeel = PARAM_VALUE (PARAM_MAX_PEELED_INSNS) / loop->ninsns;
|
npeel = PARAM_VALUE (PARAM_MAX_PEELED_INSNS) / loop->ninsns;
|
if (npeel > (unsigned) PARAM_VALUE (PARAM_MAX_PEEL_TIMES))
|
if (npeel > (unsigned) PARAM_VALUE (PARAM_MAX_PEEL_TIMES))
|
npeel = PARAM_VALUE (PARAM_MAX_PEEL_TIMES);
|
npeel = PARAM_VALUE (PARAM_MAX_PEEL_TIMES);
|
|
|
/* Skip big loops. */
|
/* Skip big loops. */
|
if (!npeel)
|
if (!npeel)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Not considering loop, is too big\n");
|
fprintf (dump_file, ";; Not considering loop, is too big\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* Check for simple loops. */
|
/* Check for simple loops. */
|
desc = get_simple_loop_desc (loop);
|
desc = get_simple_loop_desc (loop);
|
|
|
/* Check number of iterations. */
|
/* Check number of iterations. */
|
if (desc->simple_p && !desc->assumptions && desc->const_iter)
|
if (desc->simple_p && !desc->assumptions && desc->const_iter)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Loop iterates constant times\n");
|
fprintf (dump_file, ";; Loop iterates constant times\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* Do not simply peel loops with branches inside -- it increases number
|
/* Do not simply peel loops with branches inside -- it increases number
|
of mispredicts. */
|
of mispredicts. */
|
if (num_loop_branches (loop) > 1)
|
if (num_loop_branches (loop) > 1)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Not peeling, contains branches\n");
|
fprintf (dump_file, ";; Not peeling, contains branches\n");
|
return;
|
return;
|
}
|
}
|
|
|
if (loop->header->count)
|
if (loop->header->count)
|
{
|
{
|
unsigned niter = expected_loop_iterations (loop);
|
unsigned niter = expected_loop_iterations (loop);
|
if (niter + 1 > npeel)
|
if (niter + 1 > npeel)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, ";; Not peeling loop, rolls too much (");
|
fprintf (dump_file, ";; Not peeling loop, rolls too much (");
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC,
|
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC,
|
(HOST_WIDEST_INT) (niter + 1));
|
(HOST_WIDEST_INT) (niter + 1));
|
fprintf (dump_file, " iterations > %d [maximum peelings])\n",
|
fprintf (dump_file, " iterations > %d [maximum peelings])\n",
|
npeel);
|
npeel);
|
}
|
}
|
return;
|
return;
|
}
|
}
|
npeel = niter + 1;
|
npeel = niter + 1;
|
}
|
}
|
else
|
else
|
{
|
{
|
/* For now we have no good heuristics to decide whether loop peeling
|
/* For now we have no good heuristics to decide whether loop peeling
|
will be effective, so disable it. */
|
will be effective, so disable it. */
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
";; Not peeling loop, no evidence it will be profitable\n");
|
";; Not peeling loop, no evidence it will be profitable\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* Success. */
|
/* Success. */
|
loop->lpt_decision.decision = LPT_PEEL_SIMPLE;
|
loop->lpt_decision.decision = LPT_PEEL_SIMPLE;
|
loop->lpt_decision.times = npeel;
|
loop->lpt_decision.times = npeel;
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Decided to simply peel the loop, %d times.\n",
|
fprintf (dump_file, ";; Decided to simply peel the loop, %d times.\n",
|
loop->lpt_decision.times);
|
loop->lpt_decision.times);
|
}
|
}
|
|
|
/* Peel a LOOP LOOP->LPT_DECISION.TIMES times. The transformation:
|
/* Peel a LOOP LOOP->LPT_DECISION.TIMES times. The transformation:
|
while (cond)
|
while (cond)
|
body;
|
body;
|
|
|
==>
|
==>
|
|
|
if (!cond) goto end;
|
if (!cond) goto end;
|
body;
|
body;
|
if (!cond) goto end;
|
if (!cond) goto end;
|
body;
|
body;
|
while (cond)
|
while (cond)
|
body;
|
body;
|
end: ;
|
end: ;
|
*/
|
*/
|
static void
|
static void
|
peel_loop_simple (struct loops *loops, struct loop *loop)
|
peel_loop_simple (struct loops *loops, struct loop *loop)
|
{
|
{
|
sbitmap wont_exit;
|
sbitmap wont_exit;
|
unsigned npeel = loop->lpt_decision.times;
|
unsigned npeel = loop->lpt_decision.times;
|
struct niter_desc *desc = get_simple_loop_desc (loop);
|
struct niter_desc *desc = get_simple_loop_desc (loop);
|
struct opt_info *opt_info = NULL;
|
struct opt_info *opt_info = NULL;
|
bool ok;
|
bool ok;
|
|
|
if (flag_split_ivs_in_unroller && npeel > 1)
|
if (flag_split_ivs_in_unroller && npeel > 1)
|
opt_info = analyze_insns_in_loop (loop);
|
opt_info = analyze_insns_in_loop (loop);
|
|
|
wont_exit = sbitmap_alloc (npeel + 1);
|
wont_exit = sbitmap_alloc (npeel + 1);
|
sbitmap_zero (wont_exit);
|
sbitmap_zero (wont_exit);
|
|
|
opt_info_start_duplication (opt_info);
|
opt_info_start_duplication (opt_info);
|
|
|
ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
|
ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
|
loops, npeel, wont_exit,
|
loops, npeel, wont_exit,
|
NULL, NULL,
|
NULL, NULL,
|
NULL, DLTHE_FLAG_UPDATE_FREQ
|
NULL, DLTHE_FLAG_UPDATE_FREQ
|
| (opt_info
|
| (opt_info
|
? DLTHE_RECORD_COPY_NUMBER
|
? DLTHE_RECORD_COPY_NUMBER
|
: 0));
|
: 0));
|
gcc_assert (ok);
|
gcc_assert (ok);
|
|
|
free (wont_exit);
|
free (wont_exit);
|
|
|
if (opt_info)
|
if (opt_info)
|
{
|
{
|
apply_opt_in_copies (opt_info, npeel, false, false);
|
apply_opt_in_copies (opt_info, npeel, false, false);
|
free_opt_info (opt_info);
|
free_opt_info (opt_info);
|
}
|
}
|
|
|
if (desc->simple_p)
|
if (desc->simple_p)
|
{
|
{
|
if (desc->const_iter)
|
if (desc->const_iter)
|
{
|
{
|
desc->niter -= npeel;
|
desc->niter -= npeel;
|
desc->niter_expr = GEN_INT (desc->niter);
|
desc->niter_expr = GEN_INT (desc->niter);
|
desc->noloop_assumptions = NULL_RTX;
|
desc->noloop_assumptions = NULL_RTX;
|
}
|
}
|
else
|
else
|
{
|
{
|
/* We cannot just update niter_expr, as its value might be clobbered
|
/* We cannot just update niter_expr, as its value might be clobbered
|
inside loop. We could handle this by counting the number into
|
inside loop. We could handle this by counting the number into
|
temporary just like we do in runtime unrolling, but it does not
|
temporary just like we do in runtime unrolling, but it does not
|
seem worthwhile. */
|
seem worthwhile. */
|
free_simple_loop_desc (loop);
|
free_simple_loop_desc (loop);
|
}
|
}
|
}
|
}
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Peeling loop %d times\n", npeel);
|
fprintf (dump_file, ";; Peeling loop %d times\n", npeel);
|
}
|
}
|
|
|
/* Decide whether to unroll LOOP stupidly and how much. */
|
/* Decide whether to unroll LOOP stupidly and how much. */
|
static void
|
static void
|
decide_unroll_stupid (struct loop *loop, int flags)
|
decide_unroll_stupid (struct loop *loop, int flags)
|
{
|
{
|
unsigned nunroll, nunroll_by_av, i;
|
unsigned nunroll, nunroll_by_av, i;
|
struct niter_desc *desc;
|
struct niter_desc *desc;
|
|
|
if (!(flags & UAP_UNROLL_ALL))
|
if (!(flags & UAP_UNROLL_ALL))
|
{
|
{
|
/* We were not asked to, just return back silently. */
|
/* We were not asked to, just return back silently. */
|
return;
|
return;
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "\n;; Considering unrolling loop stupidly\n");
|
fprintf (dump_file, "\n;; Considering unrolling loop stupidly\n");
|
|
|
/* nunroll = total number of copies of the original loop body in
|
/* nunroll = total number of copies of the original loop body in
|
unrolled loop (i.e. if it is 2, we have to duplicate loop body once. */
|
unrolled loop (i.e. if it is 2, we have to duplicate loop body once. */
|
nunroll = PARAM_VALUE (PARAM_MAX_UNROLLED_INSNS) / loop->ninsns;
|
nunroll = PARAM_VALUE (PARAM_MAX_UNROLLED_INSNS) / loop->ninsns;
|
nunroll_by_av
|
nunroll_by_av
|
= PARAM_VALUE (PARAM_MAX_AVERAGE_UNROLLED_INSNS) / loop->av_ninsns;
|
= PARAM_VALUE (PARAM_MAX_AVERAGE_UNROLLED_INSNS) / loop->av_ninsns;
|
if (nunroll > nunroll_by_av)
|
if (nunroll > nunroll_by_av)
|
nunroll = nunroll_by_av;
|
nunroll = nunroll_by_av;
|
if (nunroll > (unsigned) PARAM_VALUE (PARAM_MAX_UNROLL_TIMES))
|
if (nunroll > (unsigned) PARAM_VALUE (PARAM_MAX_UNROLL_TIMES))
|
nunroll = PARAM_VALUE (PARAM_MAX_UNROLL_TIMES);
|
nunroll = PARAM_VALUE (PARAM_MAX_UNROLL_TIMES);
|
|
|
/* Skip big loops. */
|
/* Skip big loops. */
|
if (nunroll <= 1)
|
if (nunroll <= 1)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Not considering loop, is too big\n");
|
fprintf (dump_file, ";; Not considering loop, is too big\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* Check for simple loops. */
|
/* Check for simple loops. */
|
desc = get_simple_loop_desc (loop);
|
desc = get_simple_loop_desc (loop);
|
|
|
/* Check simpleness. */
|
/* Check simpleness. */
|
if (desc->simple_p && !desc->assumptions)
|
if (desc->simple_p && !desc->assumptions)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; The loop is simple\n");
|
fprintf (dump_file, ";; The loop is simple\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* Do not unroll loops with branches inside -- it increases number
|
/* Do not unroll loops with branches inside -- it increases number
|
of mispredicts. */
|
of mispredicts. */
|
if (num_loop_branches (loop) > 1)
|
if (num_loop_branches (loop) > 1)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Not unrolling, contains branches\n");
|
fprintf (dump_file, ";; Not unrolling, contains branches\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* If we have profile feedback, check whether the loop rolls. */
|
/* If we have profile feedback, check whether the loop rolls. */
|
if (loop->header->count
|
if (loop->header->count
|
&& expected_loop_iterations (loop) < 2 * nunroll)
|
&& expected_loop_iterations (loop) < 2 * nunroll)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Not unrolling loop, doesn't roll\n");
|
fprintf (dump_file, ";; Not unrolling loop, doesn't roll\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* Success. Now force nunroll to be power of 2, as it seems that this
|
/* Success. Now force nunroll to be power of 2, as it seems that this
|
improves results (partially because of better alignments, partially
|
improves results (partially because of better alignments, partially
|
because of some dark magic). */
|
because of some dark magic). */
|
for (i = 1; 2 * i <= nunroll; i *= 2)
|
for (i = 1; 2 * i <= nunroll; i *= 2)
|
continue;
|
continue;
|
|
|
loop->lpt_decision.decision = LPT_UNROLL_STUPID;
|
loop->lpt_decision.decision = LPT_UNROLL_STUPID;
|
loop->lpt_decision.times = i - 1;
|
loop->lpt_decision.times = i - 1;
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
";; Decided to unroll the loop stupidly, %d times.\n",
|
";; Decided to unroll the loop stupidly, %d times.\n",
|
loop->lpt_decision.times);
|
loop->lpt_decision.times);
|
}
|
}
|
|
|
/* Unroll a LOOP LOOP->LPT_DECISION.TIMES times. The transformation:
|
/* Unroll a LOOP LOOP->LPT_DECISION.TIMES times. The transformation:
|
while (cond)
|
while (cond)
|
body;
|
body;
|
|
|
==>
|
==>
|
|
|
while (cond)
|
while (cond)
|
{
|
{
|
body;
|
body;
|
if (!cond) break;
|
if (!cond) break;
|
body;
|
body;
|
if (!cond) break;
|
if (!cond) break;
|
body;
|
body;
|
if (!cond) break;
|
if (!cond) break;
|
body;
|
body;
|
}
|
}
|
*/
|
*/
|
static void
|
static void
|
unroll_loop_stupid (struct loops *loops, struct loop *loop)
|
unroll_loop_stupid (struct loops *loops, struct loop *loop)
|
{
|
{
|
sbitmap wont_exit;
|
sbitmap wont_exit;
|
unsigned nunroll = loop->lpt_decision.times;
|
unsigned nunroll = loop->lpt_decision.times;
|
struct niter_desc *desc = get_simple_loop_desc (loop);
|
struct niter_desc *desc = get_simple_loop_desc (loop);
|
struct opt_info *opt_info = NULL;
|
struct opt_info *opt_info = NULL;
|
bool ok;
|
bool ok;
|
|
|
if (flag_split_ivs_in_unroller
|
if (flag_split_ivs_in_unroller
|
|| flag_variable_expansion_in_unroller)
|
|| flag_variable_expansion_in_unroller)
|
opt_info = analyze_insns_in_loop (loop);
|
opt_info = analyze_insns_in_loop (loop);
|
|
|
|
|
wont_exit = sbitmap_alloc (nunroll + 1);
|
wont_exit = sbitmap_alloc (nunroll + 1);
|
sbitmap_zero (wont_exit);
|
sbitmap_zero (wont_exit);
|
opt_info_start_duplication (opt_info);
|
opt_info_start_duplication (opt_info);
|
|
|
ok = duplicate_loop_to_header_edge (loop, loop_latch_edge (loop),
|
ok = duplicate_loop_to_header_edge (loop, loop_latch_edge (loop),
|
loops, nunroll, wont_exit,
|
loops, nunroll, wont_exit,
|
NULL, NULL, NULL,
|
NULL, NULL, NULL,
|
DLTHE_FLAG_UPDATE_FREQ
|
DLTHE_FLAG_UPDATE_FREQ
|
| (opt_info
|
| (opt_info
|
? DLTHE_RECORD_COPY_NUMBER
|
? DLTHE_RECORD_COPY_NUMBER
|
: 0));
|
: 0));
|
gcc_assert (ok);
|
gcc_assert (ok);
|
|
|
if (opt_info)
|
if (opt_info)
|
{
|
{
|
apply_opt_in_copies (opt_info, nunroll, true, true);
|
apply_opt_in_copies (opt_info, nunroll, true, true);
|
free_opt_info (opt_info);
|
free_opt_info (opt_info);
|
}
|
}
|
|
|
free (wont_exit);
|
free (wont_exit);
|
|
|
if (desc->simple_p)
|
if (desc->simple_p)
|
{
|
{
|
/* We indeed may get here provided that there are nontrivial assumptions
|
/* We indeed may get here provided that there are nontrivial assumptions
|
for a loop to be really simple. We could update the counts, but the
|
for a loop to be really simple. We could update the counts, but the
|
problem is that we are unable to decide which exit will be taken
|
problem is that we are unable to decide which exit will be taken
|
(not really true in case the number of iterations is constant,
|
(not really true in case the number of iterations is constant,
|
but noone will do anything with this information, so we do not
|
but noone will do anything with this information, so we do not
|
worry about it). */
|
worry about it). */
|
desc->simple_p = false;
|
desc->simple_p = false;
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, ";; Unrolled loop %d times, %i insns\n",
|
fprintf (dump_file, ";; Unrolled loop %d times, %i insns\n",
|
nunroll, num_loop_insns (loop));
|
nunroll, num_loop_insns (loop));
|
}
|
}
|
|
|
/* A hash function for information about insns to split. */
|
/* A hash function for information about insns to split. */
|
|
|
static hashval_t
|
static hashval_t
|
si_info_hash (const void *ivts)
|
si_info_hash (const void *ivts)
|
{
|
{
|
return (hashval_t) INSN_UID (((struct iv_to_split *) ivts)->insn);
|
return (hashval_t) INSN_UID (((struct iv_to_split *) ivts)->insn);
|
}
|
}
|
|
|
/* An equality functions for information about insns to split. */
|
/* An equality functions for information about insns to split. */
|
|
|
static int
|
static int
|
si_info_eq (const void *ivts1, const void *ivts2)
|
si_info_eq (const void *ivts1, const void *ivts2)
|
{
|
{
|
const struct iv_to_split *i1 = ivts1;
|
const struct iv_to_split *i1 = ivts1;
|
const struct iv_to_split *i2 = ivts2;
|
const struct iv_to_split *i2 = ivts2;
|
|
|
return i1->insn == i2->insn;
|
return i1->insn == i2->insn;
|
}
|
}
|
|
|
/* Return a hash for VES, which is really a "var_to_expand *". */
|
/* Return a hash for VES, which is really a "var_to_expand *". */
|
|
|
static hashval_t
|
static hashval_t
|
ve_info_hash (const void *ves)
|
ve_info_hash (const void *ves)
|
{
|
{
|
return (hashval_t) INSN_UID (((struct var_to_expand *) ves)->insn);
|
return (hashval_t) INSN_UID (((struct var_to_expand *) ves)->insn);
|
}
|
}
|
|
|
/* Return true if IVTS1 and IVTS2 (which are really both of type
|
/* Return true if IVTS1 and IVTS2 (which are really both of type
|
"var_to_expand *") refer to the same instruction. */
|
"var_to_expand *") refer to the same instruction. */
|
|
|
static int
|
static int
|
ve_info_eq (const void *ivts1, const void *ivts2)
|
ve_info_eq (const void *ivts1, const void *ivts2)
|
{
|
{
|
const struct var_to_expand *i1 = ivts1;
|
const struct var_to_expand *i1 = ivts1;
|
const struct var_to_expand *i2 = ivts2;
|
const struct var_to_expand *i2 = ivts2;
|
|
|
return i1->insn == i2->insn;
|
return i1->insn == i2->insn;
|
}
|
}
|
|
|
/* Returns true if REG is referenced in one insn in LOOP. */
|
/* Returns true if REG is referenced in one insn in LOOP. */
|
|
|
bool
|
bool
|
referenced_in_one_insn_in_loop_p (struct loop *loop, rtx reg)
|
referenced_in_one_insn_in_loop_p (struct loop *loop, rtx reg)
|
{
|
{
|
basic_block *body, bb;
|
basic_block *body, bb;
|
unsigned i;
|
unsigned i;
|
int count_ref = 0;
|
int count_ref = 0;
|
rtx insn;
|
rtx insn;
|
|
|
body = get_loop_body (loop);
|
body = get_loop_body (loop);
|
for (i = 0; i < loop->num_nodes; i++)
|
for (i = 0; i < loop->num_nodes; i++)
|
{
|
{
|
bb = body[i];
|
bb = body[i];
|
|
|
FOR_BB_INSNS (bb, insn)
|
FOR_BB_INSNS (bb, insn)
|
{
|
{
|
if (rtx_referenced_p (reg, insn))
|
if (rtx_referenced_p (reg, insn))
|
count_ref++;
|
count_ref++;
|
}
|
}
|
}
|
}
|
return (count_ref == 1);
|
return (count_ref == 1);
|
}
|
}
|
|
|
/* Determine whether INSN contains an accumulator
|
/* Determine whether INSN contains an accumulator
|
which can be expanded into separate copies,
|
which can be expanded into separate copies,
|
one for each copy of the LOOP body.
|
one for each copy of the LOOP body.
|
|
|
for (i = 0 ; i < n; i++)
|
for (i = 0 ; i < n; i++)
|
sum += a[i];
|
sum += a[i];
|
|
|
==>
|
==>
|
|
|
sum += a[i]
|
sum += a[i]
|
....
|
....
|
i = i+1;
|
i = i+1;
|
sum1 += a[i]
|
sum1 += a[i]
|
....
|
....
|
i = i+1
|
i = i+1
|
sum2 += a[i];
|
sum2 += a[i];
|
....
|
....
|
|
|
Return NULL if INSN contains no opportunity for expansion of accumulator.
|
Return NULL if INSN contains no opportunity for expansion of accumulator.
|
Otherwise, allocate a VAR_TO_EXPAND structure, fill it with the relevant
|
Otherwise, allocate a VAR_TO_EXPAND structure, fill it with the relevant
|
information and return a pointer to it.
|
information and return a pointer to it.
|
*/
|
*/
|
|
|
static struct var_to_expand *
|
static struct var_to_expand *
|
analyze_insn_to_expand_var (struct loop *loop, rtx insn)
|
analyze_insn_to_expand_var (struct loop *loop, rtx insn)
|
{
|
{
|
rtx set, dest, src, op1;
|
rtx set, dest, src, op1;
|
struct var_to_expand *ves;
|
struct var_to_expand *ves;
|
enum machine_mode mode1, mode2;
|
enum machine_mode mode1, mode2;
|
|
|
set = single_set (insn);
|
set = single_set (insn);
|
if (!set)
|
if (!set)
|
return NULL;
|
return NULL;
|
|
|
dest = SET_DEST (set);
|
dest = SET_DEST (set);
|
src = SET_SRC (set);
|
src = SET_SRC (set);
|
|
|
if (GET_CODE (src) != PLUS
|
if (GET_CODE (src) != PLUS
|
&& GET_CODE (src) != MINUS
|
&& GET_CODE (src) != MINUS
|
&& GET_CODE (src) != MULT)
|
&& GET_CODE (src) != MULT)
|
return NULL;
|
return NULL;
|
|
|
/* Hmm, this is a bit paradoxical. We know that INSN is a valid insn
|
/* Hmm, this is a bit paradoxical. We know that INSN is a valid insn
|
in MD. But if there is no optab to generate the insn, we can not
|
in MD. But if there is no optab to generate the insn, we can not
|
perform the variable expansion. This can happen if an MD provides
|
perform the variable expansion. This can happen if an MD provides
|
an insn but not a named pattern to generate it, for example to avoid
|
an insn but not a named pattern to generate it, for example to avoid
|
producing code that needs additional mode switches like for x87/mmx.
|
producing code that needs additional mode switches like for x87/mmx.
|
|
|
So we check have_insn_for which looks for an optab for the operation
|
So we check have_insn_for which looks for an optab for the operation
|
in SRC. If it doesn't exist, we can't perform the expansion even
|
in SRC. If it doesn't exist, we can't perform the expansion even
|
though INSN is valid. */
|
though INSN is valid. */
|
if (!have_insn_for (GET_CODE (src), GET_MODE (src)))
|
if (!have_insn_for (GET_CODE (src), GET_MODE (src)))
|
return NULL;
|
return NULL;
|
|
|
if (!XEXP (src, 0))
|
if (!XEXP (src, 0))
|
return NULL;
|
return NULL;
|
|
|
op1 = XEXP (src, 0);
|
op1 = XEXP (src, 0);
|
|
|
if (!REG_P (dest)
|
if (!REG_P (dest)
|
&& !(GET_CODE (dest) == SUBREG
|
&& !(GET_CODE (dest) == SUBREG
|
&& REG_P (SUBREG_REG (dest))))
|
&& REG_P (SUBREG_REG (dest))))
|
return NULL;
|
return NULL;
|
|
|
if (!rtx_equal_p (dest, op1))
|
if (!rtx_equal_p (dest, op1))
|
return NULL;
|
return NULL;
|
|
|
if (!referenced_in_one_insn_in_loop_p (loop, dest))
|
if (!referenced_in_one_insn_in_loop_p (loop, dest))
|
return NULL;
|
return NULL;
|
|
|
if (rtx_referenced_p (dest, XEXP (src, 1)))
|
if (rtx_referenced_p (dest, XEXP (src, 1)))
|
return NULL;
|
return NULL;
|
|
|
mode1 = GET_MODE (dest);
|
mode1 = GET_MODE (dest);
|
mode2 = GET_MODE (XEXP (src, 1));
|
mode2 = GET_MODE (XEXP (src, 1));
|
if ((FLOAT_MODE_P (mode1)
|
if ((FLOAT_MODE_P (mode1)
|
|| FLOAT_MODE_P (mode2))
|
|| FLOAT_MODE_P (mode2))
|
&& !flag_unsafe_math_optimizations)
|
&& !flag_unsafe_math_optimizations)
|
return NULL;
|
return NULL;
|
|
|
/* Record the accumulator to expand. */
|
/* Record the accumulator to expand. */
|
ves = XNEW (struct var_to_expand);
|
ves = XNEW (struct var_to_expand);
|
ves->insn = insn;
|
ves->insn = insn;
|
ves->var_expansions = VEC_alloc (rtx, heap, 1);
|
ves->var_expansions = VEC_alloc (rtx, heap, 1);
|
ves->reg = copy_rtx (dest);
|
ves->reg = copy_rtx (dest);
|
ves->op = GET_CODE (src);
|
ves->op = GET_CODE (src);
|
ves->expansion_count = 0;
|
ves->expansion_count = 0;
|
ves->reuse_expansion = 0;
|
ves->reuse_expansion = 0;
|
return ves;
|
return ves;
|
}
|
}
|
|
|
/* Determine whether there is an induction variable in INSN that
|
/* Determine whether there is an induction variable in INSN that
|
we would like to split during unrolling.
|
we would like to split during unrolling.
|
|
|
I.e. replace
|
I.e. replace
|
|
|
i = i + 1;
|
i = i + 1;
|
...
|
...
|
i = i + 1;
|
i = i + 1;
|
...
|
...
|
i = i + 1;
|
i = i + 1;
|
...
|
...
|
|
|
type chains by
|
type chains by
|
|
|
i0 = i + 1
|
i0 = i + 1
|
...
|
...
|
i = i0 + 1
|
i = i0 + 1
|
...
|
...
|
i = i0 + 2
|
i = i0 + 2
|
...
|
...
|
|
|
Return NULL if INSN contains no interesting IVs. Otherwise, allocate
|
Return NULL if INSN contains no interesting IVs. Otherwise, allocate
|
an IV_TO_SPLIT structure, fill it with the relevant information and return a
|
an IV_TO_SPLIT structure, fill it with the relevant information and return a
|
pointer to it. */
|
pointer to it. */
|
|
|
static struct iv_to_split *
|
static struct iv_to_split *
|
analyze_iv_to_split_insn (rtx insn)
|
analyze_iv_to_split_insn (rtx insn)
|
{
|
{
|
rtx set, dest;
|
rtx set, dest;
|
struct rtx_iv iv;
|
struct rtx_iv iv;
|
struct iv_to_split *ivts;
|
struct iv_to_split *ivts;
|
bool ok;
|
bool ok;
|
|
|
/* For now we just split the basic induction variables. Later this may be
|
/* For now we just split the basic induction variables. Later this may be
|
extended for example by selecting also addresses of memory references. */
|
extended for example by selecting also addresses of memory references. */
|
set = single_set (insn);
|
set = single_set (insn);
|
if (!set)
|
if (!set)
|
return NULL;
|
return NULL;
|
|
|
dest = SET_DEST (set);
|
dest = SET_DEST (set);
|
if (!REG_P (dest))
|
if (!REG_P (dest))
|
return NULL;
|
return NULL;
|
|
|
if (!biv_p (insn, dest))
|
if (!biv_p (insn, dest))
|
return NULL;
|
return NULL;
|
|
|
ok = iv_analyze_result (insn, dest, &iv);
|
ok = iv_analyze_result (insn, dest, &iv);
|
|
|
/* This used to be an assert under the assumption that if biv_p returns
|
/* This used to be an assert under the assumption that if biv_p returns
|
true that iv_analyze_result must also return true. However, that
|
true that iv_analyze_result must also return true. However, that
|
assumption is not strictly correct as evidenced by pr25569.
|
assumption is not strictly correct as evidenced by pr25569.
|
|
|
Returning NULL when iv_analyze_result returns false is safe and
|
Returning NULL when iv_analyze_result returns false is safe and
|
avoids the problems in pr25569 until the iv_analyze_* routines
|
avoids the problems in pr25569 until the iv_analyze_* routines
|
can be fixed, which is apparently hard and time consuming
|
can be fixed, which is apparently hard and time consuming
|
according to their author. */
|
according to their author. */
|
if (! ok)
|
if (! ok)
|
return NULL;
|
return NULL;
|
|
|
if (iv.step == const0_rtx
|
if (iv.step == const0_rtx
|
|| iv.mode != iv.extend_mode)
|
|| iv.mode != iv.extend_mode)
|
return NULL;
|
return NULL;
|
|
|
/* Record the insn to split. */
|
/* Record the insn to split. */
|
ivts = XNEW (struct iv_to_split);
|
ivts = XNEW (struct iv_to_split);
|
ivts->insn = insn;
|
ivts->insn = insn;
|
ivts->base_var = NULL_RTX;
|
ivts->base_var = NULL_RTX;
|
ivts->step = iv.step;
|
ivts->step = iv.step;
|
ivts->n_loc = 1;
|
ivts->n_loc = 1;
|
ivts->loc[0] = 1;
|
ivts->loc[0] = 1;
|
|
|
return ivts;
|
return ivts;
|
}
|
}
|
|
|
/* Determines which of insns in LOOP can be optimized.
|
/* Determines which of insns in LOOP can be optimized.
|
Return a OPT_INFO struct with the relevant hash tables filled
|
Return a OPT_INFO struct with the relevant hash tables filled
|
with all insns to be optimized. The FIRST_NEW_BLOCK field
|
with all insns to be optimized. The FIRST_NEW_BLOCK field
|
is undefined for the return value. */
|
is undefined for the return value. */
|
|
|
static struct opt_info *
|
static struct opt_info *
|
analyze_insns_in_loop (struct loop *loop)
|
analyze_insns_in_loop (struct loop *loop)
|
{
|
{
|
basic_block *body, bb;
|
basic_block *body, bb;
|
unsigned i, num_edges = 0;
|
unsigned i, num_edges = 0;
|
struct opt_info *opt_info = XCNEW (struct opt_info);
|
struct opt_info *opt_info = XCNEW (struct opt_info);
|
rtx insn;
|
rtx insn;
|
struct iv_to_split *ivts = NULL;
|
struct iv_to_split *ivts = NULL;
|
struct var_to_expand *ves = NULL;
|
struct var_to_expand *ves = NULL;
|
PTR *slot1;
|
PTR *slot1;
|
PTR *slot2;
|
PTR *slot2;
|
edge *edges = get_loop_exit_edges (loop, &num_edges);
|
edge *edges = get_loop_exit_edges (loop, &num_edges);
|
bool can_apply = false;
|
bool can_apply = false;
|
|
|
iv_analysis_loop_init (loop);
|
iv_analysis_loop_init (loop);
|
|
|
body = get_loop_body (loop);
|
body = get_loop_body (loop);
|
|
|
if (flag_split_ivs_in_unroller)
|
if (flag_split_ivs_in_unroller)
|
opt_info->insns_to_split = htab_create (5 * loop->num_nodes,
|
opt_info->insns_to_split = htab_create (5 * loop->num_nodes,
|
si_info_hash, si_info_eq, free);
|
si_info_hash, si_info_eq, free);
|
|
|
/* Record the loop exit bb and loop preheader before the unrolling. */
|
/* Record the loop exit bb and loop preheader before the unrolling. */
|
if (!loop_preheader_edge (loop)->src)
|
if (!loop_preheader_edge (loop)->src)
|
{
|
{
|
loop_split_edge_with (loop_preheader_edge (loop), NULL_RTX);
|
loop_split_edge_with (loop_preheader_edge (loop), NULL_RTX);
|
opt_info->loop_preheader = loop_split_edge_with (loop_preheader_edge (loop), NULL_RTX);
|
opt_info->loop_preheader = loop_split_edge_with (loop_preheader_edge (loop), NULL_RTX);
|
}
|
}
|
else
|
else
|
opt_info->loop_preheader = loop_preheader_edge (loop)->src;
|
opt_info->loop_preheader = loop_preheader_edge (loop)->src;
|
|
|
if (num_edges == 1
|
if (num_edges == 1
|
&& !(edges[0]->flags & EDGE_COMPLEX))
|
&& !(edges[0]->flags & EDGE_COMPLEX))
|
{
|
{
|
opt_info->loop_exit = loop_split_edge_with (edges[0], NULL_RTX);
|
opt_info->loop_exit = loop_split_edge_with (edges[0], NULL_RTX);
|
can_apply = true;
|
can_apply = true;
|
}
|
}
|
|
|
if (flag_variable_expansion_in_unroller
|
if (flag_variable_expansion_in_unroller
|
&& can_apply)
|
&& can_apply)
|
opt_info->insns_with_var_to_expand = htab_create (5 * loop->num_nodes,
|
opt_info->insns_with_var_to_expand = htab_create (5 * loop->num_nodes,
|
ve_info_hash, ve_info_eq, free);
|
ve_info_hash, ve_info_eq, free);
|
|
|
for (i = 0; i < loop->num_nodes; i++)
|
for (i = 0; i < loop->num_nodes; i++)
|
{
|
{
|
bb = body[i];
|
bb = body[i];
|
if (!dominated_by_p (CDI_DOMINATORS, loop->latch, bb))
|
if (!dominated_by_p (CDI_DOMINATORS, loop->latch, bb))
|
continue;
|
continue;
|
|
|
FOR_BB_INSNS (bb, insn)
|
FOR_BB_INSNS (bb, insn)
|
{
|
{
|
if (!INSN_P (insn))
|
if (!INSN_P (insn))
|
continue;
|
continue;
|
|
|
if (opt_info->insns_to_split)
|
if (opt_info->insns_to_split)
|
ivts = analyze_iv_to_split_insn (insn);
|
ivts = analyze_iv_to_split_insn (insn);
|
|
|
if (ivts)
|
if (ivts)
|
{
|
{
|
slot1 = htab_find_slot (opt_info->insns_to_split, ivts, INSERT);
|
slot1 = htab_find_slot (opt_info->insns_to_split, ivts, INSERT);
|
*slot1 = ivts;
|
*slot1 = ivts;
|
continue;
|
continue;
|
}
|
}
|
|
|
if (opt_info->insns_with_var_to_expand)
|
if (opt_info->insns_with_var_to_expand)
|
ves = analyze_insn_to_expand_var (loop, insn);
|
ves = analyze_insn_to_expand_var (loop, insn);
|
|
|
if (ves)
|
if (ves)
|
{
|
{
|
slot2 = htab_find_slot (opt_info->insns_with_var_to_expand, ves, INSERT);
|
slot2 = htab_find_slot (opt_info->insns_with_var_to_expand, ves, INSERT);
|
*slot2 = ves;
|
*slot2 = ves;
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
free (edges);
|
free (edges);
|
free (body);
|
free (body);
|
return opt_info;
|
return opt_info;
|
}
|
}
|
|
|
/* Called just before loop duplication. Records start of duplicated area
|
/* Called just before loop duplication. Records start of duplicated area
|
to OPT_INFO. */
|
to OPT_INFO. */
|
|
|
static void
|
static void
|
opt_info_start_duplication (struct opt_info *opt_info)
|
opt_info_start_duplication (struct opt_info *opt_info)
|
{
|
{
|
if (opt_info)
|
if (opt_info)
|
opt_info->first_new_block = last_basic_block;
|
opt_info->first_new_block = last_basic_block;
|
}
|
}
|
|
|
/* Determine the number of iterations between initialization of the base
|
/* Determine the number of iterations between initialization of the base
|
variable and the current copy (N_COPY). N_COPIES is the total number
|
variable and the current copy (N_COPY). N_COPIES is the total number
|
of newly created copies. UNROLLING is true if we are unrolling
|
of newly created copies. UNROLLING is true if we are unrolling
|
(not peeling) the loop. */
|
(not peeling) the loop. */
|
|
|
static unsigned
|
static unsigned
|
determine_split_iv_delta (unsigned n_copy, unsigned n_copies, bool unrolling)
|
determine_split_iv_delta (unsigned n_copy, unsigned n_copies, bool unrolling)
|
{
|
{
|
if (unrolling)
|
if (unrolling)
|
{
|
{
|
/* If we are unrolling, initialization is done in the original loop
|
/* If we are unrolling, initialization is done in the original loop
|
body (number 0). */
|
body (number 0). */
|
return n_copy;
|
return n_copy;
|
}
|
}
|
else
|
else
|
{
|
{
|
/* If we are peeling, the copy in that the initialization occurs has
|
/* If we are peeling, the copy in that the initialization occurs has
|
number 1. The original loop (number 0) is the last. */
|
number 1. The original loop (number 0) is the last. */
|
if (n_copy)
|
if (n_copy)
|
return n_copy - 1;
|
return n_copy - 1;
|
else
|
else
|
return n_copies;
|
return n_copies;
|
}
|
}
|
}
|
}
|
|
|
/* Locate in EXPR the expression corresponding to the location recorded
|
/* Locate in EXPR the expression corresponding to the location recorded
|
in IVTS, and return a pointer to the RTX for this location. */
|
in IVTS, and return a pointer to the RTX for this location. */
|
|
|
static rtx *
|
static rtx *
|
get_ivts_expr (rtx expr, struct iv_to_split *ivts)
|
get_ivts_expr (rtx expr, struct iv_to_split *ivts)
|
{
|
{
|
unsigned i;
|
unsigned i;
|
rtx *ret = &expr;
|
rtx *ret = &expr;
|
|
|
for (i = 0; i < ivts->n_loc; i++)
|
for (i = 0; i < ivts->n_loc; i++)
|
ret = &XEXP (*ret, ivts->loc[i]);
|
ret = &XEXP (*ret, ivts->loc[i]);
|
|
|
return ret;
|
return ret;
|
}
|
}
|
|
|
/* Allocate basic variable for the induction variable chain. Callback for
|
/* Allocate basic variable for the induction variable chain. Callback for
|
htab_traverse. */
|
htab_traverse. */
|
|
|
static int
|
static int
|
allocate_basic_variable (void **slot, void *data ATTRIBUTE_UNUSED)
|
allocate_basic_variable (void **slot, void *data ATTRIBUTE_UNUSED)
|
{
|
{
|
struct iv_to_split *ivts = *slot;
|
struct iv_to_split *ivts = *slot;
|
rtx expr = *get_ivts_expr (single_set (ivts->insn), ivts);
|
rtx expr = *get_ivts_expr (single_set (ivts->insn), ivts);
|
|
|
ivts->base_var = gen_reg_rtx (GET_MODE (expr));
|
ivts->base_var = gen_reg_rtx (GET_MODE (expr));
|
|
|
return 1;
|
return 1;
|
}
|
}
|
|
|
/* Insert initialization of basic variable of IVTS before INSN, taking
|
/* Insert initialization of basic variable of IVTS before INSN, taking
|
the initial value from INSN. */
|
the initial value from INSN. */
|
|
|
static void
|
static void
|
insert_base_initialization (struct iv_to_split *ivts, rtx insn)
|
insert_base_initialization (struct iv_to_split *ivts, rtx insn)
|
{
|
{
|
rtx expr = copy_rtx (*get_ivts_expr (single_set (insn), ivts));
|
rtx expr = copy_rtx (*get_ivts_expr (single_set (insn), ivts));
|
rtx seq;
|
rtx seq;
|
|
|
start_sequence ();
|
start_sequence ();
|
expr = force_operand (expr, ivts->base_var);
|
expr = force_operand (expr, ivts->base_var);
|
if (expr != ivts->base_var)
|
if (expr != ivts->base_var)
|
emit_move_insn (ivts->base_var, expr);
|
emit_move_insn (ivts->base_var, expr);
|
seq = get_insns ();
|
seq = get_insns ();
|
end_sequence ();
|
end_sequence ();
|
|
|
emit_insn_before (seq, insn);
|
emit_insn_before (seq, insn);
|
}
|
}
|
|
|
/* Replace the use of induction variable described in IVTS in INSN
|
/* Replace the use of induction variable described in IVTS in INSN
|
by base variable + DELTA * step. */
|
by base variable + DELTA * step. */
|
|
|
static void
|
static void
|
split_iv (struct iv_to_split *ivts, rtx insn, unsigned delta)
|
split_iv (struct iv_to_split *ivts, rtx insn, unsigned delta)
|
{
|
{
|
rtx expr, *loc, seq, incr, var;
|
rtx expr, *loc, seq, incr, var;
|
enum machine_mode mode = GET_MODE (ivts->base_var);
|
enum machine_mode mode = GET_MODE (ivts->base_var);
|
rtx src, dest, set;
|
rtx src, dest, set;
|
|
|
/* Construct base + DELTA * step. */
|
/* Construct base + DELTA * step. */
|
if (!delta)
|
if (!delta)
|
expr = ivts->base_var;
|
expr = ivts->base_var;
|
else
|
else
|
{
|
{
|
incr = simplify_gen_binary (MULT, mode,
|
incr = simplify_gen_binary (MULT, mode,
|
ivts->step, gen_int_mode (delta, mode));
|
ivts->step, gen_int_mode (delta, mode));
|
expr = simplify_gen_binary (PLUS, GET_MODE (ivts->base_var),
|
expr = simplify_gen_binary (PLUS, GET_MODE (ivts->base_var),
|
ivts->base_var, incr);
|
ivts->base_var, incr);
|
}
|
}
|
|
|
/* Figure out where to do the replacement. */
|
/* Figure out where to do the replacement. */
|
loc = get_ivts_expr (single_set (insn), ivts);
|
loc = get_ivts_expr (single_set (insn), ivts);
|
|
|
/* If we can make the replacement right away, we're done. */
|
/* If we can make the replacement right away, we're done. */
|
if (validate_change (insn, loc, expr, 0))
|
if (validate_change (insn, loc, expr, 0))
|
return;
|
return;
|
|
|
/* Otherwise, force EXPR into a register and try again. */
|
/* Otherwise, force EXPR into a register and try again. */
|
start_sequence ();
|
start_sequence ();
|
var = gen_reg_rtx (mode);
|
var = gen_reg_rtx (mode);
|
expr = force_operand (expr, var);
|
expr = force_operand (expr, var);
|
if (expr != var)
|
if (expr != var)
|
emit_move_insn (var, expr);
|
emit_move_insn (var, expr);
|
seq = get_insns ();
|
seq = get_insns ();
|
end_sequence ();
|
end_sequence ();
|
emit_insn_before (seq, insn);
|
emit_insn_before (seq, insn);
|
|
|
if (validate_change (insn, loc, var, 0))
|
if (validate_change (insn, loc, var, 0))
|
return;
|
return;
|
|
|
/* The last chance. Try recreating the assignment in insn
|
/* The last chance. Try recreating the assignment in insn
|
completely from scratch. */
|
completely from scratch. */
|
set = single_set (insn);
|
set = single_set (insn);
|
gcc_assert (set);
|
gcc_assert (set);
|
|
|
start_sequence ();
|
start_sequence ();
|
*loc = var;
|
*loc = var;
|
src = copy_rtx (SET_SRC (set));
|
src = copy_rtx (SET_SRC (set));
|
dest = copy_rtx (SET_DEST (set));
|
dest = copy_rtx (SET_DEST (set));
|
src = force_operand (src, dest);
|
src = force_operand (src, dest);
|
if (src != dest)
|
if (src != dest)
|
emit_move_insn (dest, src);
|
emit_move_insn (dest, src);
|
seq = get_insns ();
|
seq = get_insns ();
|
end_sequence ();
|
end_sequence ();
|
|
|
emit_insn_before (seq, insn);
|
emit_insn_before (seq, insn);
|
delete_insn (insn);
|
delete_insn (insn);
|
}
|
}
|
|
|
|
|
/* Return one expansion of the accumulator recorded in struct VE. */
|
/* Return one expansion of the accumulator recorded in struct VE. */
|
|
|
static rtx
|
static rtx
|
get_expansion (struct var_to_expand *ve)
|
get_expansion (struct var_to_expand *ve)
|
{
|
{
|
rtx reg;
|
rtx reg;
|
|
|
if (ve->reuse_expansion == 0)
|
if (ve->reuse_expansion == 0)
|
reg = ve->reg;
|
reg = ve->reg;
|
else
|
else
|
reg = VEC_index (rtx, ve->var_expansions, ve->reuse_expansion - 1);
|
reg = VEC_index (rtx, ve->var_expansions, ve->reuse_expansion - 1);
|
|
|
if (VEC_length (rtx, ve->var_expansions) == (unsigned) ve->reuse_expansion)
|
if (VEC_length (rtx, ve->var_expansions) == (unsigned) ve->reuse_expansion)
|
ve->reuse_expansion = 0;
|
ve->reuse_expansion = 0;
|
else
|
else
|
ve->reuse_expansion++;
|
ve->reuse_expansion++;
|
|
|
return reg;
|
return reg;
|
}
|
}
|
|
|
|
|
/* Given INSN replace the uses of the accumulator recorded in VE
|
/* Given INSN replace the uses of the accumulator recorded in VE
|
with a new register. */
|
with a new register. */
|
|
|
static void
|
static void
|
expand_var_during_unrolling (struct var_to_expand *ve, rtx insn)
|
expand_var_during_unrolling (struct var_to_expand *ve, rtx insn)
|
{
|
{
|
rtx new_reg, set;
|
rtx new_reg, set;
|
bool really_new_expansion = false;
|
bool really_new_expansion = false;
|
|
|
set = single_set (insn);
|
set = single_set (insn);
|
gcc_assert (set);
|
gcc_assert (set);
|
|
|
/* Generate a new register only if the expansion limit has not been
|
/* Generate a new register only if the expansion limit has not been
|
reached. Else reuse an already existing expansion. */
|
reached. Else reuse an already existing expansion. */
|
if (PARAM_VALUE (PARAM_MAX_VARIABLE_EXPANSIONS) > ve->expansion_count)
|
if (PARAM_VALUE (PARAM_MAX_VARIABLE_EXPANSIONS) > ve->expansion_count)
|
{
|
{
|
really_new_expansion = true;
|
really_new_expansion = true;
|
new_reg = gen_reg_rtx (GET_MODE (ve->reg));
|
new_reg = gen_reg_rtx (GET_MODE (ve->reg));
|
}
|
}
|
else
|
else
|
new_reg = get_expansion (ve);
|
new_reg = get_expansion (ve);
|
|
|
validate_change (insn, &SET_DEST (set), new_reg, 1);
|
validate_change (insn, &SET_DEST (set), new_reg, 1);
|
validate_change (insn, &XEXP (SET_SRC (set), 0), new_reg, 1);
|
validate_change (insn, &XEXP (SET_SRC (set), 0), new_reg, 1);
|
|
|
if (apply_change_group ())
|
if (apply_change_group ())
|
if (really_new_expansion)
|
if (really_new_expansion)
|
{
|
{
|
VEC_safe_push (rtx, heap, ve->var_expansions, new_reg);
|
VEC_safe_push (rtx, heap, ve->var_expansions, new_reg);
|
ve->expansion_count++;
|
ve->expansion_count++;
|
}
|
}
|
}
|
}
|
|
|
/* Initialize the variable expansions in loop preheader.
|
/* Initialize the variable expansions in loop preheader.
|
Callbacks for htab_traverse. PLACE_P is the loop-preheader
|
Callbacks for htab_traverse. PLACE_P is the loop-preheader
|
basic block where the initialization of the expansions
|
basic block where the initialization of the expansions
|
should take place. */
|
should take place. */
|
|
|
static int
|
static int
|
insert_var_expansion_initialization (void **slot, void *place_p)
|
insert_var_expansion_initialization (void **slot, void *place_p)
|
{
|
{
|
struct var_to_expand *ve = *slot;
|
struct var_to_expand *ve = *slot;
|
basic_block place = (basic_block)place_p;
|
basic_block place = (basic_block)place_p;
|
rtx seq, var, zero_init, insn;
|
rtx seq, var, zero_init, insn;
|
unsigned i;
|
unsigned i;
|
|
|
if (VEC_length (rtx, ve->var_expansions) == 0)
|
if (VEC_length (rtx, ve->var_expansions) == 0)
|
return 1;
|
return 1;
|
|
|
start_sequence ();
|
start_sequence ();
|
if (ve->op == PLUS || ve->op == MINUS)
|
if (ve->op == PLUS || ve->op == MINUS)
|
for (i = 0; VEC_iterate (rtx, ve->var_expansions, i, var); i++)
|
for (i = 0; VEC_iterate (rtx, ve->var_expansions, i, var); i++)
|
{
|
{
|
zero_init = CONST0_RTX (GET_MODE (var));
|
zero_init = CONST0_RTX (GET_MODE (var));
|
emit_move_insn (var, zero_init);
|
emit_move_insn (var, zero_init);
|
}
|
}
|
else if (ve->op == MULT)
|
else if (ve->op == MULT)
|
for (i = 0; VEC_iterate (rtx, ve->var_expansions, i, var); i++)
|
for (i = 0; VEC_iterate (rtx, ve->var_expansions, i, var); i++)
|
{
|
{
|
zero_init = CONST1_RTX (GET_MODE (var));
|
zero_init = CONST1_RTX (GET_MODE (var));
|
emit_move_insn (var, zero_init);
|
emit_move_insn (var, zero_init);
|
}
|
}
|
|
|
seq = get_insns ();
|
seq = get_insns ();
|
end_sequence ();
|
end_sequence ();
|
|
|
insn = BB_HEAD (place);
|
insn = BB_HEAD (place);
|
while (!NOTE_INSN_BASIC_BLOCK_P (insn))
|
while (!NOTE_INSN_BASIC_BLOCK_P (insn))
|
insn = NEXT_INSN (insn);
|
insn = NEXT_INSN (insn);
|
|
|
emit_insn_after (seq, insn);
|
emit_insn_after (seq, insn);
|
/* Continue traversing the hash table. */
|
/* Continue traversing the hash table. */
|
return 1;
|
return 1;
|
}
|
}
|
|
|
/* Combine the variable expansions at the loop exit.
|
/* Combine the variable expansions at the loop exit.
|
Callbacks for htab_traverse. PLACE_P is the loop exit
|
Callbacks for htab_traverse. PLACE_P is the loop exit
|
basic block where the summation of the expansions should
|
basic block where the summation of the expansions should
|
take place. */
|
take place. */
|
|
|
static int
|
static int
|
combine_var_copies_in_loop_exit (void **slot, void *place_p)
|
combine_var_copies_in_loop_exit (void **slot, void *place_p)
|
{
|
{
|
struct var_to_expand *ve = *slot;
|
struct var_to_expand *ve = *slot;
|
basic_block place = (basic_block)place_p;
|
basic_block place = (basic_block)place_p;
|
rtx sum = ve->reg;
|
rtx sum = ve->reg;
|
rtx expr, seq, var, insn;
|
rtx expr, seq, var, insn;
|
unsigned i;
|
unsigned i;
|
|
|
if (VEC_length (rtx, ve->var_expansions) == 0)
|
if (VEC_length (rtx, ve->var_expansions) == 0)
|
return 1;
|
return 1;
|
|
|
start_sequence ();
|
start_sequence ();
|
if (ve->op == PLUS || ve->op == MINUS)
|
if (ve->op == PLUS || ve->op == MINUS)
|
for (i = 0; VEC_iterate (rtx, ve->var_expansions, i, var); i++)
|
for (i = 0; VEC_iterate (rtx, ve->var_expansions, i, var); i++)
|
{
|
{
|
sum = simplify_gen_binary (PLUS, GET_MODE (ve->reg),
|
sum = simplify_gen_binary (PLUS, GET_MODE (ve->reg),
|
var, sum);
|
var, sum);
|
}
|
}
|
else if (ve->op == MULT)
|
else if (ve->op == MULT)
|
for (i = 0; VEC_iterate (rtx, ve->var_expansions, i, var); i++)
|
for (i = 0; VEC_iterate (rtx, ve->var_expansions, i, var); i++)
|
{
|
{
|
sum = simplify_gen_binary (MULT, GET_MODE (ve->reg),
|
sum = simplify_gen_binary (MULT, GET_MODE (ve->reg),
|
var, sum);
|
var, sum);
|
}
|
}
|
|
|
expr = force_operand (sum, ve->reg);
|
expr = force_operand (sum, ve->reg);
|
if (expr != ve->reg)
|
if (expr != ve->reg)
|
emit_move_insn (ve->reg, expr);
|
emit_move_insn (ve->reg, expr);
|
seq = get_insns ();
|
seq = get_insns ();
|
end_sequence ();
|
end_sequence ();
|
|
|
insn = BB_HEAD (place);
|
insn = BB_HEAD (place);
|
while (!NOTE_INSN_BASIC_BLOCK_P (insn))
|
while (!NOTE_INSN_BASIC_BLOCK_P (insn))
|
insn = NEXT_INSN (insn);
|
insn = NEXT_INSN (insn);
|
|
|
emit_insn_after (seq, insn);
|
emit_insn_after (seq, insn);
|
|
|
/* Continue traversing the hash table. */
|
/* Continue traversing the hash table. */
|
return 1;
|
return 1;
|
}
|
}
|
|
|
/* Apply loop optimizations in loop copies using the
|
/* Apply loop optimizations in loop copies using the
|
data which gathered during the unrolling. Structure
|
data which gathered during the unrolling. Structure
|
OPT_INFO record that data.
|
OPT_INFO record that data.
|
|
|
UNROLLING is true if we unrolled (not peeled) the loop.
|
UNROLLING is true if we unrolled (not peeled) the loop.
|
REWRITE_ORIGINAL_BODY is true if we should also rewrite the original body of
|
REWRITE_ORIGINAL_BODY is true if we should also rewrite the original body of
|
the loop (as it should happen in complete unrolling, but not in ordinary
|
the loop (as it should happen in complete unrolling, but not in ordinary
|
peeling of the loop). */
|
peeling of the loop). */
|
|
|
static void
|
static void
|
apply_opt_in_copies (struct opt_info *opt_info,
|
apply_opt_in_copies (struct opt_info *opt_info,
|
unsigned n_copies, bool unrolling,
|
unsigned n_copies, bool unrolling,
|
bool rewrite_original_loop)
|
bool rewrite_original_loop)
|
{
|
{
|
unsigned i, delta;
|
unsigned i, delta;
|
basic_block bb, orig_bb;
|
basic_block bb, orig_bb;
|
rtx insn, orig_insn, next;
|
rtx insn, orig_insn, next;
|
struct iv_to_split ivts_templ, *ivts;
|
struct iv_to_split ivts_templ, *ivts;
|
struct var_to_expand ve_templ, *ves;
|
struct var_to_expand ve_templ, *ves;
|
|
|
/* Sanity check -- we need to put initialization in the original loop
|
/* Sanity check -- we need to put initialization in the original loop
|
body. */
|
body. */
|
gcc_assert (!unrolling || rewrite_original_loop);
|
gcc_assert (!unrolling || rewrite_original_loop);
|
|
|
/* Allocate the basic variables (i0). */
|
/* Allocate the basic variables (i0). */
|
if (opt_info->insns_to_split)
|
if (opt_info->insns_to_split)
|
htab_traverse (opt_info->insns_to_split, allocate_basic_variable, NULL);
|
htab_traverse (opt_info->insns_to_split, allocate_basic_variable, NULL);
|
|
|
for (i = opt_info->first_new_block; i < (unsigned) last_basic_block; i++)
|
for (i = opt_info->first_new_block; i < (unsigned) last_basic_block; i++)
|
{
|
{
|
bb = BASIC_BLOCK (i);
|
bb = BASIC_BLOCK (i);
|
orig_bb = get_bb_original (bb);
|
orig_bb = get_bb_original (bb);
|
|
|
/* bb->aux holds position in copy sequence initialized by
|
/* bb->aux holds position in copy sequence initialized by
|
duplicate_loop_to_header_edge. */
|
duplicate_loop_to_header_edge. */
|
delta = determine_split_iv_delta ((size_t)bb->aux, n_copies,
|
delta = determine_split_iv_delta ((size_t)bb->aux, n_copies,
|
unrolling);
|
unrolling);
|
bb->aux = 0;
|
bb->aux = 0;
|
orig_insn = BB_HEAD (orig_bb);
|
orig_insn = BB_HEAD (orig_bb);
|
for (insn = BB_HEAD (bb); insn != NEXT_INSN (BB_END (bb)); insn = next)
|
for (insn = BB_HEAD (bb); insn != NEXT_INSN (BB_END (bb)); insn = next)
|
{
|
{
|
next = NEXT_INSN (insn);
|
next = NEXT_INSN (insn);
|
if (!INSN_P (insn))
|
if (!INSN_P (insn))
|
continue;
|
continue;
|
|
|
while (!INSN_P (orig_insn))
|
while (!INSN_P (orig_insn))
|
orig_insn = NEXT_INSN (orig_insn);
|
orig_insn = NEXT_INSN (orig_insn);
|
|
|
ivts_templ.insn = orig_insn;
|
ivts_templ.insn = orig_insn;
|
ve_templ.insn = orig_insn;
|
ve_templ.insn = orig_insn;
|
|
|
/* Apply splitting iv optimization. */
|
/* Apply splitting iv optimization. */
|
if (opt_info->insns_to_split)
|
if (opt_info->insns_to_split)
|
{
|
{
|
ivts = htab_find (opt_info->insns_to_split, &ivts_templ);
|
ivts = htab_find (opt_info->insns_to_split, &ivts_templ);
|
|
|
if (ivts)
|
if (ivts)
|
{
|
{
|
gcc_assert (GET_CODE (PATTERN (insn))
|
gcc_assert (GET_CODE (PATTERN (insn))
|
== GET_CODE (PATTERN (orig_insn)));
|
== GET_CODE (PATTERN (orig_insn)));
|
|
|
if (!delta)
|
if (!delta)
|
insert_base_initialization (ivts, insn);
|
insert_base_initialization (ivts, insn);
|
split_iv (ivts, insn, delta);
|
split_iv (ivts, insn, delta);
|
}
|
}
|
}
|
}
|
/* Apply variable expansion optimization. */
|
/* Apply variable expansion optimization. */
|
if (unrolling && opt_info->insns_with_var_to_expand)
|
if (unrolling && opt_info->insns_with_var_to_expand)
|
{
|
{
|
ves = htab_find (opt_info->insns_with_var_to_expand, &ve_templ);
|
ves = htab_find (opt_info->insns_with_var_to_expand, &ve_templ);
|
if (ves)
|
if (ves)
|
{
|
{
|
gcc_assert (GET_CODE (PATTERN (insn))
|
gcc_assert (GET_CODE (PATTERN (insn))
|
== GET_CODE (PATTERN (orig_insn)));
|
== GET_CODE (PATTERN (orig_insn)));
|
expand_var_during_unrolling (ves, insn);
|
expand_var_during_unrolling (ves, insn);
|
}
|
}
|
}
|
}
|
orig_insn = NEXT_INSN (orig_insn);
|
orig_insn = NEXT_INSN (orig_insn);
|
}
|
}
|
}
|
}
|
|
|
if (!rewrite_original_loop)
|
if (!rewrite_original_loop)
|
return;
|
return;
|
|
|
/* Initialize the variable expansions in the loop preheader
|
/* Initialize the variable expansions in the loop preheader
|
and take care of combining them at the loop exit. */
|
and take care of combining them at the loop exit. */
|
if (opt_info->insns_with_var_to_expand)
|
if (opt_info->insns_with_var_to_expand)
|
{
|
{
|
htab_traverse (opt_info->insns_with_var_to_expand,
|
htab_traverse (opt_info->insns_with_var_to_expand,
|
insert_var_expansion_initialization,
|
insert_var_expansion_initialization,
|
opt_info->loop_preheader);
|
opt_info->loop_preheader);
|
htab_traverse (opt_info->insns_with_var_to_expand,
|
htab_traverse (opt_info->insns_with_var_to_expand,
|
combine_var_copies_in_loop_exit,
|
combine_var_copies_in_loop_exit,
|
opt_info->loop_exit);
|
opt_info->loop_exit);
|
}
|
}
|
|
|
/* Rewrite also the original loop body. Find them as originals of the blocks
|
/* Rewrite also the original loop body. Find them as originals of the blocks
|
in the last copied iteration, i.e. those that have
|
in the last copied iteration, i.e. those that have
|
get_bb_copy (get_bb_original (bb)) == bb. */
|
get_bb_copy (get_bb_original (bb)) == bb. */
|
for (i = opt_info->first_new_block; i < (unsigned) last_basic_block; i++)
|
for (i = opt_info->first_new_block; i < (unsigned) last_basic_block; i++)
|
{
|
{
|
bb = BASIC_BLOCK (i);
|
bb = BASIC_BLOCK (i);
|
orig_bb = get_bb_original (bb);
|
orig_bb = get_bb_original (bb);
|
if (get_bb_copy (orig_bb) != bb)
|
if (get_bb_copy (orig_bb) != bb)
|
continue;
|
continue;
|
|
|
delta = determine_split_iv_delta (0, n_copies, unrolling);
|
delta = determine_split_iv_delta (0, n_copies, unrolling);
|
for (orig_insn = BB_HEAD (orig_bb);
|
for (orig_insn = BB_HEAD (orig_bb);
|
orig_insn != NEXT_INSN (BB_END (bb));
|
orig_insn != NEXT_INSN (BB_END (bb));
|
orig_insn = next)
|
orig_insn = next)
|
{
|
{
|
next = NEXT_INSN (orig_insn);
|
next = NEXT_INSN (orig_insn);
|
|
|
if (!INSN_P (orig_insn))
|
if (!INSN_P (orig_insn))
|
continue;
|
continue;
|
|
|
ivts_templ.insn = orig_insn;
|
ivts_templ.insn = orig_insn;
|
if (opt_info->insns_to_split)
|
if (opt_info->insns_to_split)
|
{
|
{
|
ivts = htab_find (opt_info->insns_to_split, &ivts_templ);
|
ivts = htab_find (opt_info->insns_to_split, &ivts_templ);
|
if (ivts)
|
if (ivts)
|
{
|
{
|
if (!delta)
|
if (!delta)
|
insert_base_initialization (ivts, orig_insn);
|
insert_base_initialization (ivts, orig_insn);
|
split_iv (ivts, orig_insn, delta);
|
split_iv (ivts, orig_insn, delta);
|
continue;
|
continue;
|
}
|
}
|
}
|
}
|
|
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
/* Release the data structures used for the variable expansion
|
/* Release the data structures used for the variable expansion
|
optimization. Callbacks for htab_traverse. */
|
optimization. Callbacks for htab_traverse. */
|
|
|
static int
|
static int
|
release_var_copies (void **slot, void *data ATTRIBUTE_UNUSED)
|
release_var_copies (void **slot, void *data ATTRIBUTE_UNUSED)
|
{
|
{
|
struct var_to_expand *ve = *slot;
|
struct var_to_expand *ve = *slot;
|
|
|
VEC_free (rtx, heap, ve->var_expansions);
|
VEC_free (rtx, heap, ve->var_expansions);
|
|
|
/* Continue traversing the hash table. */
|
/* Continue traversing the hash table. */
|
return 1;
|
return 1;
|
}
|
}
|
|
|
/* Release OPT_INFO. */
|
/* Release OPT_INFO. */
|
|
|
static void
|
static void
|
free_opt_info (struct opt_info *opt_info)
|
free_opt_info (struct opt_info *opt_info)
|
{
|
{
|
if (opt_info->insns_to_split)
|
if (opt_info->insns_to_split)
|
htab_delete (opt_info->insns_to_split);
|
htab_delete (opt_info->insns_to_split);
|
if (opt_info->insns_with_var_to_expand)
|
if (opt_info->insns_with_var_to_expand)
|
{
|
{
|
htab_traverse (opt_info->insns_with_var_to_expand,
|
htab_traverse (opt_info->insns_with_var_to_expand,
|
release_var_copies, NULL);
|
release_var_copies, NULL);
|
htab_delete (opt_info->insns_with_var_to_expand);
|
htab_delete (opt_info->insns_with_var_to_expand);
|
}
|
}
|
free (opt_info);
|
free (opt_info);
|
}
|
}
|
|
|
