/* Branch prediction routines for the GNU compiler.
|
/* Branch prediction routines for the GNU compiler.
|
Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2007
|
Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2007
|
Free Software Foundation, Inc.
|
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/>. */
|
|
|
/* References:
|
/* References:
|
|
|
[1] "Branch Prediction for Free"
|
[1] "Branch Prediction for Free"
|
Ball and Larus; PLDI '93.
|
Ball and Larus; PLDI '93.
|
[2] "Static Branch Frequency and Program Profile Analysis"
|
[2] "Static Branch Frequency and Program Profile Analysis"
|
Wu and Larus; MICRO-27.
|
Wu and Larus; MICRO-27.
|
[3] "Corpus-based Static Branch Prediction"
|
[3] "Corpus-based Static Branch Prediction"
|
Calder, Grunwald, Lindsay, Martin, Mozer, and Zorn; PLDI '95. */
|
Calder, Grunwald, Lindsay, Martin, Mozer, and Zorn; PLDI '95. */
|
|
|
|
|
#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 "tree.h"
|
#include "tree.h"
|
#include "rtl.h"
|
#include "rtl.h"
|
#include "tm_p.h"
|
#include "tm_p.h"
|
#include "hard-reg-set.h"
|
#include "hard-reg-set.h"
|
#include "basic-block.h"
|
#include "basic-block.h"
|
#include "insn-config.h"
|
#include "insn-config.h"
|
#include "regs.h"
|
#include "regs.h"
|
#include "flags.h"
|
#include "flags.h"
|
#include "output.h"
|
#include "output.h"
|
#include "function.h"
|
#include "function.h"
|
#include "except.h"
|
#include "except.h"
|
#include "toplev.h"
|
#include "toplev.h"
|
#include "recog.h"
|
#include "recog.h"
|
#include "expr.h"
|
#include "expr.h"
|
#include "predict.h"
|
#include "predict.h"
|
#include "coverage.h"
|
#include "coverage.h"
|
#include "sreal.h"
|
#include "sreal.h"
|
#include "params.h"
|
#include "params.h"
|
#include "target.h"
|
#include "target.h"
|
#include "cfgloop.h"
|
#include "cfgloop.h"
|
#include "tree-flow.h"
|
#include "tree-flow.h"
|
#include "ggc.h"
|
#include "ggc.h"
|
#include "tree-dump.h"
|
#include "tree-dump.h"
|
#include "tree-pass.h"
|
#include "tree-pass.h"
|
#include "timevar.h"
|
#include "timevar.h"
|
#include "tree-scalar-evolution.h"
|
#include "tree-scalar-evolution.h"
|
#include "cfgloop.h"
|
#include "cfgloop.h"
|
|
|
/* real constants: 0, 1, 1-1/REG_BR_PROB_BASE, REG_BR_PROB_BASE,
|
/* real constants: 0, 1, 1-1/REG_BR_PROB_BASE, REG_BR_PROB_BASE,
|
1/REG_BR_PROB_BASE, 0.5, BB_FREQ_MAX. */
|
1/REG_BR_PROB_BASE, 0.5, BB_FREQ_MAX. */
|
static sreal real_zero, real_one, real_almost_one, real_br_prob_base,
|
static sreal real_zero, real_one, real_almost_one, real_br_prob_base,
|
real_inv_br_prob_base, real_one_half, real_bb_freq_max;
|
real_inv_br_prob_base, real_one_half, real_bb_freq_max;
|
|
|
/* Random guesstimation given names. */
|
/* Random guesstimation given names. */
|
#define PROB_VERY_UNLIKELY (REG_BR_PROB_BASE / 100 - 1)
|
#define PROB_VERY_UNLIKELY (REG_BR_PROB_BASE / 100 - 1)
|
#define PROB_EVEN (REG_BR_PROB_BASE / 2)
|
#define PROB_EVEN (REG_BR_PROB_BASE / 2)
|
#define PROB_VERY_LIKELY (REG_BR_PROB_BASE - PROB_VERY_UNLIKELY)
|
#define PROB_VERY_LIKELY (REG_BR_PROB_BASE - PROB_VERY_UNLIKELY)
|
#define PROB_ALWAYS (REG_BR_PROB_BASE)
|
#define PROB_ALWAYS (REG_BR_PROB_BASE)
|
|
|
static void combine_predictions_for_insn (rtx, basic_block);
|
static void combine_predictions_for_insn (rtx, basic_block);
|
static void dump_prediction (FILE *, enum br_predictor, int, basic_block, int);
|
static void dump_prediction (FILE *, enum br_predictor, int, basic_block, int);
|
static void estimate_loops_at_level (struct loop *, bitmap);
|
static void estimate_loops_at_level (struct loop *, bitmap);
|
static void propagate_freq (struct loop *, bitmap);
|
static void propagate_freq (struct loop *, bitmap);
|
static void estimate_bb_frequencies (struct loops *);
|
static void estimate_bb_frequencies (struct loops *);
|
static void predict_paths_leading_to (basic_block, int *, enum br_predictor, enum prediction);
|
static void predict_paths_leading_to (basic_block, int *, enum br_predictor, enum prediction);
|
static bool last_basic_block_p (basic_block);
|
static bool last_basic_block_p (basic_block);
|
static void compute_function_frequency (void);
|
static void compute_function_frequency (void);
|
static void choose_function_section (void);
|
static void choose_function_section (void);
|
static bool can_predict_insn_p (rtx);
|
static bool can_predict_insn_p (rtx);
|
|
|
/* Information we hold about each branch predictor.
|
/* Information we hold about each branch predictor.
|
Filled using information from predict.def. */
|
Filled using information from predict.def. */
|
|
|
struct predictor_info
|
struct predictor_info
|
{
|
{
|
const char *const name; /* Name used in the debugging dumps. */
|
const char *const name; /* Name used in the debugging dumps. */
|
const int hitrate; /* Expected hitrate used by
|
const int hitrate; /* Expected hitrate used by
|
predict_insn_def call. */
|
predict_insn_def call. */
|
const int flags;
|
const int flags;
|
};
|
};
|
|
|
/* Use given predictor without Dempster-Shaffer theory if it matches
|
/* Use given predictor without Dempster-Shaffer theory if it matches
|
using first_match heuristics. */
|
using first_match heuristics. */
|
#define PRED_FLAG_FIRST_MATCH 1
|
#define PRED_FLAG_FIRST_MATCH 1
|
|
|
/* Recompute hitrate in percent to our representation. */
|
/* Recompute hitrate in percent to our representation. */
|
|
|
#define HITRATE(VAL) ((int) ((VAL) * REG_BR_PROB_BASE + 50) / 100)
|
#define HITRATE(VAL) ((int) ((VAL) * REG_BR_PROB_BASE + 50) / 100)
|
|
|
#define DEF_PREDICTOR(ENUM, NAME, HITRATE, FLAGS) {NAME, HITRATE, FLAGS},
|
#define DEF_PREDICTOR(ENUM, NAME, HITRATE, FLAGS) {NAME, HITRATE, FLAGS},
|
static const struct predictor_info predictor_info[]= {
|
static const struct predictor_info predictor_info[]= {
|
#include "predict.def"
|
#include "predict.def"
|
|
|
/* Upper bound on predictors. */
|
/* Upper bound on predictors. */
|
{NULL, 0, 0}
|
{NULL, 0, 0}
|
};
|
};
|
#undef DEF_PREDICTOR
|
#undef DEF_PREDICTOR
|
|
|
/* Return true in case BB can be CPU intensive and should be optimized
|
/* Return true in case BB can be CPU intensive and should be optimized
|
for maximal performance. */
|
for maximal performance. */
|
|
|
bool
|
bool
|
maybe_hot_bb_p (basic_block bb)
|
maybe_hot_bb_p (basic_block bb)
|
{
|
{
|
if (profile_info && flag_branch_probabilities
|
if (profile_info && flag_branch_probabilities
|
&& (bb->count
|
&& (bb->count
|
< profile_info->sum_max / PARAM_VALUE (HOT_BB_COUNT_FRACTION)))
|
< profile_info->sum_max / PARAM_VALUE (HOT_BB_COUNT_FRACTION)))
|
return false;
|
return false;
|
if (bb->frequency < BB_FREQ_MAX / PARAM_VALUE (HOT_BB_FREQUENCY_FRACTION))
|
if (bb->frequency < BB_FREQ_MAX / PARAM_VALUE (HOT_BB_FREQUENCY_FRACTION))
|
return false;
|
return false;
|
return true;
|
return true;
|
}
|
}
|
|
|
/* Return true in case BB is cold and should be optimized for size. */
|
/* Return true in case BB is cold and should be optimized for size. */
|
|
|
bool
|
bool
|
probably_cold_bb_p (basic_block bb)
|
probably_cold_bb_p (basic_block bb)
|
{
|
{
|
if (profile_info && flag_branch_probabilities
|
if (profile_info && flag_branch_probabilities
|
&& (bb->count
|
&& (bb->count
|
< profile_info->sum_max / PARAM_VALUE (HOT_BB_COUNT_FRACTION)))
|
< profile_info->sum_max / PARAM_VALUE (HOT_BB_COUNT_FRACTION)))
|
return true;
|
return true;
|
if (bb->frequency < BB_FREQ_MAX / PARAM_VALUE (HOT_BB_FREQUENCY_FRACTION))
|
if (bb->frequency < BB_FREQ_MAX / PARAM_VALUE (HOT_BB_FREQUENCY_FRACTION))
|
return true;
|
return true;
|
return false;
|
return false;
|
}
|
}
|
|
|
/* Return true in case BB is probably never executed. */
|
/* Return true in case BB is probably never executed. */
|
bool
|
bool
|
probably_never_executed_bb_p (basic_block bb)
|
probably_never_executed_bb_p (basic_block bb)
|
{
|
{
|
if (profile_info && flag_branch_probabilities)
|
if (profile_info && flag_branch_probabilities)
|
return ((bb->count + profile_info->runs / 2) / profile_info->runs) == 0;
|
return ((bb->count + profile_info->runs / 2) / profile_info->runs) == 0;
|
return false;
|
return false;
|
}
|
}
|
|
|
/* Return true if the one of outgoing edges is already predicted by
|
/* Return true if the one of outgoing edges is already predicted by
|
PREDICTOR. */
|
PREDICTOR. */
|
|
|
bool
|
bool
|
rtl_predicted_by_p (basic_block bb, enum br_predictor predictor)
|
rtl_predicted_by_p (basic_block bb, enum br_predictor predictor)
|
{
|
{
|
rtx note;
|
rtx note;
|
if (!INSN_P (BB_END (bb)))
|
if (!INSN_P (BB_END (bb)))
|
return false;
|
return false;
|
for (note = REG_NOTES (BB_END (bb)); note; note = XEXP (note, 1))
|
for (note = REG_NOTES (BB_END (bb)); note; note = XEXP (note, 1))
|
if (REG_NOTE_KIND (note) == REG_BR_PRED
|
if (REG_NOTE_KIND (note) == REG_BR_PRED
|
&& INTVAL (XEXP (XEXP (note, 0), 0)) == (int)predictor)
|
&& INTVAL (XEXP (XEXP (note, 0), 0)) == (int)predictor)
|
return true;
|
return true;
|
return false;
|
return false;
|
}
|
}
|
|
|
/* Return true if the one of outgoing edges is already predicted by
|
/* Return true if the one of outgoing edges is already predicted by
|
PREDICTOR. */
|
PREDICTOR. */
|
|
|
bool
|
bool
|
tree_predicted_by_p (basic_block bb, enum br_predictor predictor)
|
tree_predicted_by_p (basic_block bb, enum br_predictor predictor)
|
{
|
{
|
struct edge_prediction *i;
|
struct edge_prediction *i;
|
for (i = bb->predictions; i; i = i->ep_next)
|
for (i = bb->predictions; i; i = i->ep_next)
|
if (i->ep_predictor == predictor)
|
if (i->ep_predictor == predictor)
|
return true;
|
return true;
|
return false;
|
return false;
|
}
|
}
|
|
|
/* Return true when the probability of edge is reliable.
|
/* Return true when the probability of edge is reliable.
|
|
|
The profile guessing code is good at predicting branch outcome (ie.
|
The profile guessing code is good at predicting branch outcome (ie.
|
taken/not taken), that is predicted right slightly over 75% of time.
|
taken/not taken), that is predicted right slightly over 75% of time.
|
It is however notoriously poor on predicting the probability itself.
|
It is however notoriously poor on predicting the probability itself.
|
In general the profile appear a lot flatter (with probabilities closer
|
In general the profile appear a lot flatter (with probabilities closer
|
to 50%) than the reality so it is bad idea to use it to drive optimization
|
to 50%) than the reality so it is bad idea to use it to drive optimization
|
such as those disabling dynamic branch prediction for well predictable
|
such as those disabling dynamic branch prediction for well predictable
|
branches.
|
branches.
|
|
|
There are two exceptions - edges leading to noreturn edges and edges
|
There are two exceptions - edges leading to noreturn edges and edges
|
predicted by number of iterations heuristics are predicted well. This macro
|
predicted by number of iterations heuristics are predicted well. This macro
|
should be able to distinguish those, but at the moment it simply check for
|
should be able to distinguish those, but at the moment it simply check for
|
noreturn heuristic that is only one giving probability over 99% or bellow
|
noreturn heuristic that is only one giving probability over 99% or bellow
|
1%. In future we might want to propagate reliability information across the
|
1%. In future we might want to propagate reliability information across the
|
CFG if we find this information useful on multiple places. */
|
CFG if we find this information useful on multiple places. */
|
static bool
|
static bool
|
probability_reliable_p (int prob)
|
probability_reliable_p (int prob)
|
{
|
{
|
return (profile_status == PROFILE_READ
|
return (profile_status == PROFILE_READ
|
|| (profile_status == PROFILE_GUESSED
|
|| (profile_status == PROFILE_GUESSED
|
&& (prob <= HITRATE (1) || prob >= HITRATE (99))));
|
&& (prob <= HITRATE (1) || prob >= HITRATE (99))));
|
}
|
}
|
|
|
/* Same predicate as above, working on edges. */
|
/* Same predicate as above, working on edges. */
|
bool
|
bool
|
edge_probability_reliable_p (edge e)
|
edge_probability_reliable_p (edge e)
|
{
|
{
|
return probability_reliable_p (e->probability);
|
return probability_reliable_p (e->probability);
|
}
|
}
|
|
|
/* Same predicate as edge_probability_reliable_p, working on notes. */
|
/* Same predicate as edge_probability_reliable_p, working on notes. */
|
bool
|
bool
|
br_prob_note_reliable_p (rtx note)
|
br_prob_note_reliable_p (rtx note)
|
{
|
{
|
gcc_assert (REG_NOTE_KIND (note) == REG_BR_PROB);
|
gcc_assert (REG_NOTE_KIND (note) == REG_BR_PROB);
|
return probability_reliable_p (INTVAL (XEXP (note, 0)));
|
return probability_reliable_p (INTVAL (XEXP (note, 0)));
|
}
|
}
|
|
|
static void
|
static void
|
predict_insn (rtx insn, enum br_predictor predictor, int probability)
|
predict_insn (rtx insn, enum br_predictor predictor, int probability)
|
{
|
{
|
gcc_assert (any_condjump_p (insn));
|
gcc_assert (any_condjump_p (insn));
|
if (!flag_guess_branch_prob)
|
if (!flag_guess_branch_prob)
|
return;
|
return;
|
|
|
REG_NOTES (insn)
|
REG_NOTES (insn)
|
= gen_rtx_EXPR_LIST (REG_BR_PRED,
|
= gen_rtx_EXPR_LIST (REG_BR_PRED,
|
gen_rtx_CONCAT (VOIDmode,
|
gen_rtx_CONCAT (VOIDmode,
|
GEN_INT ((int) predictor),
|
GEN_INT ((int) predictor),
|
GEN_INT ((int) probability)),
|
GEN_INT ((int) probability)),
|
REG_NOTES (insn));
|
REG_NOTES (insn));
|
}
|
}
|
|
|
/* Predict insn by given predictor. */
|
/* Predict insn by given predictor. */
|
|
|
void
|
void
|
predict_insn_def (rtx insn, enum br_predictor predictor,
|
predict_insn_def (rtx insn, enum br_predictor predictor,
|
enum prediction taken)
|
enum prediction taken)
|
{
|
{
|
int probability = predictor_info[(int) predictor].hitrate;
|
int probability = predictor_info[(int) predictor].hitrate;
|
|
|
if (taken != TAKEN)
|
if (taken != TAKEN)
|
probability = REG_BR_PROB_BASE - probability;
|
probability = REG_BR_PROB_BASE - probability;
|
|
|
predict_insn (insn, predictor, probability);
|
predict_insn (insn, predictor, probability);
|
}
|
}
|
|
|
/* Predict edge E with given probability if possible. */
|
/* Predict edge E with given probability if possible. */
|
|
|
void
|
void
|
rtl_predict_edge (edge e, enum br_predictor predictor, int probability)
|
rtl_predict_edge (edge e, enum br_predictor predictor, int probability)
|
{
|
{
|
rtx last_insn;
|
rtx last_insn;
|
last_insn = BB_END (e->src);
|
last_insn = BB_END (e->src);
|
|
|
/* We can store the branch prediction information only about
|
/* We can store the branch prediction information only about
|
conditional jumps. */
|
conditional jumps. */
|
if (!any_condjump_p (last_insn))
|
if (!any_condjump_p (last_insn))
|
return;
|
return;
|
|
|
/* We always store probability of branching. */
|
/* We always store probability of branching. */
|
if (e->flags & EDGE_FALLTHRU)
|
if (e->flags & EDGE_FALLTHRU)
|
probability = REG_BR_PROB_BASE - probability;
|
probability = REG_BR_PROB_BASE - probability;
|
|
|
predict_insn (last_insn, predictor, probability);
|
predict_insn (last_insn, predictor, probability);
|
}
|
}
|
|
|
/* Predict edge E with the given PROBABILITY. */
|
/* Predict edge E with the given PROBABILITY. */
|
void
|
void
|
tree_predict_edge (edge e, enum br_predictor predictor, int probability)
|
tree_predict_edge (edge e, enum br_predictor predictor, int probability)
|
{
|
{
|
gcc_assert (profile_status != PROFILE_GUESSED);
|
gcc_assert (profile_status != PROFILE_GUESSED);
|
if ((e->src != ENTRY_BLOCK_PTR && EDGE_COUNT (e->src->succs) > 1)
|
if ((e->src != ENTRY_BLOCK_PTR && EDGE_COUNT (e->src->succs) > 1)
|
&& flag_guess_branch_prob && optimize)
|
&& flag_guess_branch_prob && optimize)
|
{
|
{
|
struct edge_prediction *i = ggc_alloc (sizeof (struct edge_prediction));
|
struct edge_prediction *i = ggc_alloc (sizeof (struct edge_prediction));
|
|
|
i->ep_next = e->src->predictions;
|
i->ep_next = e->src->predictions;
|
e->src->predictions = i;
|
e->src->predictions = i;
|
i->ep_probability = probability;
|
i->ep_probability = probability;
|
i->ep_predictor = predictor;
|
i->ep_predictor = predictor;
|
i->ep_edge = e;
|
i->ep_edge = e;
|
}
|
}
|
}
|
}
|
|
|
/* Remove all predictions on given basic block that are attached
|
/* Remove all predictions on given basic block that are attached
|
to edge E. */
|
to edge E. */
|
void
|
void
|
remove_predictions_associated_with_edge (edge e)
|
remove_predictions_associated_with_edge (edge e)
|
{
|
{
|
if (e->src->predictions)
|
if (e->src->predictions)
|
{
|
{
|
struct edge_prediction **prediction = &e->src->predictions;
|
struct edge_prediction **prediction = &e->src->predictions;
|
while (*prediction)
|
while (*prediction)
|
{
|
{
|
if ((*prediction)->ep_edge == e)
|
if ((*prediction)->ep_edge == e)
|
*prediction = (*prediction)->ep_next;
|
*prediction = (*prediction)->ep_next;
|
else
|
else
|
prediction = &((*prediction)->ep_next);
|
prediction = &((*prediction)->ep_next);
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
/* Return true when we can store prediction on insn INSN.
|
/* Return true when we can store prediction on insn INSN.
|
At the moment we represent predictions only on conditional
|
At the moment we represent predictions only on conditional
|
jumps, not at computed jump or other complicated cases. */
|
jumps, not at computed jump or other complicated cases. */
|
static bool
|
static bool
|
can_predict_insn_p (rtx insn)
|
can_predict_insn_p (rtx insn)
|
{
|
{
|
return (JUMP_P (insn)
|
return (JUMP_P (insn)
|
&& any_condjump_p (insn)
|
&& any_condjump_p (insn)
|
&& EDGE_COUNT (BLOCK_FOR_INSN (insn)->succs) >= 2);
|
&& EDGE_COUNT (BLOCK_FOR_INSN (insn)->succs) >= 2);
|
}
|
}
|
|
|
/* Predict edge E by given predictor if possible. */
|
/* Predict edge E by given predictor if possible. */
|
|
|
void
|
void
|
predict_edge_def (edge e, enum br_predictor predictor,
|
predict_edge_def (edge e, enum br_predictor predictor,
|
enum prediction taken)
|
enum prediction taken)
|
{
|
{
|
int probability = predictor_info[(int) predictor].hitrate;
|
int probability = predictor_info[(int) predictor].hitrate;
|
|
|
if (taken != TAKEN)
|
if (taken != TAKEN)
|
probability = REG_BR_PROB_BASE - probability;
|
probability = REG_BR_PROB_BASE - probability;
|
|
|
predict_edge (e, predictor, probability);
|
predict_edge (e, predictor, probability);
|
}
|
}
|
|
|
/* Invert all branch predictions or probability notes in the INSN. This needs
|
/* Invert all branch predictions or probability notes in the INSN. This needs
|
to be done each time we invert the condition used by the jump. */
|
to be done each time we invert the condition used by the jump. */
|
|
|
void
|
void
|
invert_br_probabilities (rtx insn)
|
invert_br_probabilities (rtx insn)
|
{
|
{
|
rtx note;
|
rtx note;
|
|
|
for (note = REG_NOTES (insn); note; note = XEXP (note, 1))
|
for (note = REG_NOTES (insn); note; note = XEXP (note, 1))
|
if (REG_NOTE_KIND (note) == REG_BR_PROB)
|
if (REG_NOTE_KIND (note) == REG_BR_PROB)
|
XEXP (note, 0) = GEN_INT (REG_BR_PROB_BASE - INTVAL (XEXP (note, 0)));
|
XEXP (note, 0) = GEN_INT (REG_BR_PROB_BASE - INTVAL (XEXP (note, 0)));
|
else if (REG_NOTE_KIND (note) == REG_BR_PRED)
|
else if (REG_NOTE_KIND (note) == REG_BR_PRED)
|
XEXP (XEXP (note, 0), 1)
|
XEXP (XEXP (note, 0), 1)
|
= GEN_INT (REG_BR_PROB_BASE - INTVAL (XEXP (XEXP (note, 0), 1)));
|
= GEN_INT (REG_BR_PROB_BASE - INTVAL (XEXP (XEXP (note, 0), 1)));
|
}
|
}
|
|
|
/* Dump information about the branch prediction to the output file. */
|
/* Dump information about the branch prediction to the output file. */
|
|
|
static void
|
static void
|
dump_prediction (FILE *file, enum br_predictor predictor, int probability,
|
dump_prediction (FILE *file, enum br_predictor predictor, int probability,
|
basic_block bb, int used)
|
basic_block bb, int used)
|
{
|
{
|
edge e;
|
edge e;
|
edge_iterator ei;
|
edge_iterator ei;
|
|
|
if (!file)
|
if (!file)
|
return;
|
return;
|
|
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
if (! (e->flags & EDGE_FALLTHRU))
|
if (! (e->flags & EDGE_FALLTHRU))
|
break;
|
break;
|
|
|
fprintf (file, " %s heuristics%s: %.1f%%",
|
fprintf (file, " %s heuristics%s: %.1f%%",
|
predictor_info[predictor].name,
|
predictor_info[predictor].name,
|
used ? "" : " (ignored)", probability * 100.0 / REG_BR_PROB_BASE);
|
used ? "" : " (ignored)", probability * 100.0 / REG_BR_PROB_BASE);
|
|
|
if (bb->count)
|
if (bb->count)
|
{
|
{
|
fprintf (file, " exec ");
|
fprintf (file, " exec ");
|
fprintf (file, HOST_WIDEST_INT_PRINT_DEC, bb->count);
|
fprintf (file, HOST_WIDEST_INT_PRINT_DEC, bb->count);
|
if (e)
|
if (e)
|
{
|
{
|
fprintf (file, " hit ");
|
fprintf (file, " hit ");
|
fprintf (file, HOST_WIDEST_INT_PRINT_DEC, e->count);
|
fprintf (file, HOST_WIDEST_INT_PRINT_DEC, e->count);
|
fprintf (file, " (%.1f%%)", e->count * 100.0 / bb->count);
|
fprintf (file, " (%.1f%%)", e->count * 100.0 / bb->count);
|
}
|
}
|
}
|
}
|
|
|
fprintf (file, "\n");
|
fprintf (file, "\n");
|
}
|
}
|
|
|
/* We can not predict the probabilities of outgoing edges of bb. Set them
|
/* We can not predict the probabilities of outgoing edges of bb. Set them
|
evenly and hope for the best. */
|
evenly and hope for the best. */
|
static void
|
static void
|
set_even_probabilities (basic_block bb)
|
set_even_probabilities (basic_block bb)
|
{
|
{
|
int nedges = 0;
|
int nedges = 0;
|
edge e;
|
edge e;
|
edge_iterator ei;
|
edge_iterator ei;
|
|
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
if (!(e->flags & (EDGE_EH | EDGE_FAKE)))
|
if (!(e->flags & (EDGE_EH | EDGE_FAKE)))
|
nedges ++;
|
nedges ++;
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
if (!(e->flags & (EDGE_EH | EDGE_FAKE)))
|
if (!(e->flags & (EDGE_EH | EDGE_FAKE)))
|
e->probability = (REG_BR_PROB_BASE + nedges / 2) / nedges;
|
e->probability = (REG_BR_PROB_BASE + nedges / 2) / nedges;
|
else
|
else
|
e->probability = 0;
|
e->probability = 0;
|
}
|
}
|
|
|
/* Combine all REG_BR_PRED notes into single probability and attach REG_BR_PROB
|
/* Combine all REG_BR_PRED notes into single probability and attach REG_BR_PROB
|
note if not already present. Remove now useless REG_BR_PRED notes. */
|
note if not already present. Remove now useless REG_BR_PRED notes. */
|
|
|
static void
|
static void
|
combine_predictions_for_insn (rtx insn, basic_block bb)
|
combine_predictions_for_insn (rtx insn, basic_block bb)
|
{
|
{
|
rtx prob_note;
|
rtx prob_note;
|
rtx *pnote;
|
rtx *pnote;
|
rtx note;
|
rtx note;
|
int best_probability = PROB_EVEN;
|
int best_probability = PROB_EVEN;
|
int best_predictor = END_PREDICTORS;
|
int best_predictor = END_PREDICTORS;
|
int combined_probability = REG_BR_PROB_BASE / 2;
|
int combined_probability = REG_BR_PROB_BASE / 2;
|
int d;
|
int d;
|
bool first_match = false;
|
bool first_match = false;
|
bool found = false;
|
bool found = false;
|
|
|
if (!can_predict_insn_p (insn))
|
if (!can_predict_insn_p (insn))
|
{
|
{
|
set_even_probabilities (bb);
|
set_even_probabilities (bb);
|
return;
|
return;
|
}
|
}
|
|
|
prob_note = find_reg_note (insn, REG_BR_PROB, 0);
|
prob_note = find_reg_note (insn, REG_BR_PROB, 0);
|
pnote = ®_NOTES (insn);
|
pnote = ®_NOTES (insn);
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "Predictions for insn %i bb %i\n", INSN_UID (insn),
|
fprintf (dump_file, "Predictions for insn %i bb %i\n", INSN_UID (insn),
|
bb->index);
|
bb->index);
|
|
|
/* We implement "first match" heuristics and use probability guessed
|
/* We implement "first match" heuristics and use probability guessed
|
by predictor with smallest index. */
|
by predictor with smallest index. */
|
for (note = REG_NOTES (insn); note; note = XEXP (note, 1))
|
for (note = REG_NOTES (insn); note; note = XEXP (note, 1))
|
if (REG_NOTE_KIND (note) == REG_BR_PRED)
|
if (REG_NOTE_KIND (note) == REG_BR_PRED)
|
{
|
{
|
int predictor = INTVAL (XEXP (XEXP (note, 0), 0));
|
int predictor = INTVAL (XEXP (XEXP (note, 0), 0));
|
int probability = INTVAL (XEXP (XEXP (note, 0), 1));
|
int probability = INTVAL (XEXP (XEXP (note, 0), 1));
|
|
|
found = true;
|
found = true;
|
if (best_predictor > predictor)
|
if (best_predictor > predictor)
|
best_probability = probability, best_predictor = predictor;
|
best_probability = probability, best_predictor = predictor;
|
|
|
d = (combined_probability * probability
|
d = (combined_probability * probability
|
+ (REG_BR_PROB_BASE - combined_probability)
|
+ (REG_BR_PROB_BASE - combined_probability)
|
* (REG_BR_PROB_BASE - probability));
|
* (REG_BR_PROB_BASE - probability));
|
|
|
/* Use FP math to avoid overflows of 32bit integers. */
|
/* Use FP math to avoid overflows of 32bit integers. */
|
if (d == 0)
|
if (d == 0)
|
/* If one probability is 0% and one 100%, avoid division by zero. */
|
/* If one probability is 0% and one 100%, avoid division by zero. */
|
combined_probability = REG_BR_PROB_BASE / 2;
|
combined_probability = REG_BR_PROB_BASE / 2;
|
else
|
else
|
combined_probability = (((double) combined_probability) * probability
|
combined_probability = (((double) combined_probability) * probability
|
* REG_BR_PROB_BASE / d + 0.5);
|
* REG_BR_PROB_BASE / d + 0.5);
|
}
|
}
|
|
|
/* Decide which heuristic to use. In case we didn't match anything,
|
/* Decide which heuristic to use. In case we didn't match anything,
|
use no_prediction heuristic, in case we did match, use either
|
use no_prediction heuristic, in case we did match, use either
|
first match or Dempster-Shaffer theory depending on the flags. */
|
first match or Dempster-Shaffer theory depending on the flags. */
|
|
|
if (predictor_info [best_predictor].flags & PRED_FLAG_FIRST_MATCH)
|
if (predictor_info [best_predictor].flags & PRED_FLAG_FIRST_MATCH)
|
first_match = true;
|
first_match = true;
|
|
|
if (!found)
|
if (!found)
|
dump_prediction (dump_file, PRED_NO_PREDICTION,
|
dump_prediction (dump_file, PRED_NO_PREDICTION,
|
combined_probability, bb, true);
|
combined_probability, bb, true);
|
else
|
else
|
{
|
{
|
dump_prediction (dump_file, PRED_DS_THEORY, combined_probability,
|
dump_prediction (dump_file, PRED_DS_THEORY, combined_probability,
|
bb, !first_match);
|
bb, !first_match);
|
dump_prediction (dump_file, PRED_FIRST_MATCH, best_probability,
|
dump_prediction (dump_file, PRED_FIRST_MATCH, best_probability,
|
bb, first_match);
|
bb, first_match);
|
}
|
}
|
|
|
if (first_match)
|
if (first_match)
|
combined_probability = best_probability;
|
combined_probability = best_probability;
|
dump_prediction (dump_file, PRED_COMBINED, combined_probability, bb, true);
|
dump_prediction (dump_file, PRED_COMBINED, combined_probability, bb, true);
|
|
|
while (*pnote)
|
while (*pnote)
|
{
|
{
|
if (REG_NOTE_KIND (*pnote) == REG_BR_PRED)
|
if (REG_NOTE_KIND (*pnote) == REG_BR_PRED)
|
{
|
{
|
int predictor = INTVAL (XEXP (XEXP (*pnote, 0), 0));
|
int predictor = INTVAL (XEXP (XEXP (*pnote, 0), 0));
|
int probability = INTVAL (XEXP (XEXP (*pnote, 0), 1));
|
int probability = INTVAL (XEXP (XEXP (*pnote, 0), 1));
|
|
|
dump_prediction (dump_file, predictor, probability, bb,
|
dump_prediction (dump_file, predictor, probability, bb,
|
!first_match || best_predictor == predictor);
|
!first_match || best_predictor == predictor);
|
*pnote = XEXP (*pnote, 1);
|
*pnote = XEXP (*pnote, 1);
|
}
|
}
|
else
|
else
|
pnote = &XEXP (*pnote, 1);
|
pnote = &XEXP (*pnote, 1);
|
}
|
}
|
|
|
if (!prob_note)
|
if (!prob_note)
|
{
|
{
|
REG_NOTES (insn)
|
REG_NOTES (insn)
|
= gen_rtx_EXPR_LIST (REG_BR_PROB,
|
= gen_rtx_EXPR_LIST (REG_BR_PROB,
|
GEN_INT (combined_probability), REG_NOTES (insn));
|
GEN_INT (combined_probability), REG_NOTES (insn));
|
|
|
/* Save the prediction into CFG in case we are seeing non-degenerated
|
/* Save the prediction into CFG in case we are seeing non-degenerated
|
conditional jump. */
|
conditional jump. */
|
if (!single_succ_p (bb))
|
if (!single_succ_p (bb))
|
{
|
{
|
BRANCH_EDGE (bb)->probability = combined_probability;
|
BRANCH_EDGE (bb)->probability = combined_probability;
|
FALLTHRU_EDGE (bb)->probability
|
FALLTHRU_EDGE (bb)->probability
|
= REG_BR_PROB_BASE - combined_probability;
|
= REG_BR_PROB_BASE - combined_probability;
|
}
|
}
|
}
|
}
|
else if (!single_succ_p (bb))
|
else if (!single_succ_p (bb))
|
{
|
{
|
int prob = INTVAL (XEXP (prob_note, 0));
|
int prob = INTVAL (XEXP (prob_note, 0));
|
|
|
BRANCH_EDGE (bb)->probability = prob;
|
BRANCH_EDGE (bb)->probability = prob;
|
FALLTHRU_EDGE (bb)->probability = REG_BR_PROB_BASE - prob;
|
FALLTHRU_EDGE (bb)->probability = REG_BR_PROB_BASE - prob;
|
}
|
}
|
else
|
else
|
single_succ_edge (bb)->probability = REG_BR_PROB_BASE;
|
single_succ_edge (bb)->probability = REG_BR_PROB_BASE;
|
}
|
}
|
|
|
/* Combine predictions into single probability and store them into CFG.
|
/* Combine predictions into single probability and store them into CFG.
|
Remove now useless prediction entries. */
|
Remove now useless prediction entries. */
|
|
|
static void
|
static void
|
combine_predictions_for_bb (basic_block bb)
|
combine_predictions_for_bb (basic_block bb)
|
{
|
{
|
int best_probability = PROB_EVEN;
|
int best_probability = PROB_EVEN;
|
int best_predictor = END_PREDICTORS;
|
int best_predictor = END_PREDICTORS;
|
int combined_probability = REG_BR_PROB_BASE / 2;
|
int combined_probability = REG_BR_PROB_BASE / 2;
|
int d;
|
int d;
|
bool first_match = false;
|
bool first_match = false;
|
bool found = false;
|
bool found = false;
|
struct edge_prediction *pred;
|
struct edge_prediction *pred;
|
int nedges = 0;
|
int nedges = 0;
|
edge e, first = NULL, second = NULL;
|
edge e, first = NULL, second = NULL;
|
edge_iterator ei;
|
edge_iterator ei;
|
|
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
if (!(e->flags & (EDGE_EH | EDGE_FAKE)))
|
if (!(e->flags & (EDGE_EH | EDGE_FAKE)))
|
{
|
{
|
nedges ++;
|
nedges ++;
|
if (first && !second)
|
if (first && !second)
|
second = e;
|
second = e;
|
if (!first)
|
if (!first)
|
first = e;
|
first = e;
|
}
|
}
|
|
|
/* When there is no successor or only one choice, prediction is easy.
|
/* When there is no successor or only one choice, prediction is easy.
|
|
|
We are lazy for now and predict only basic blocks with two outgoing
|
We are lazy for now and predict only basic blocks with two outgoing
|
edges. It is possible to predict generic case too, but we have to
|
edges. It is possible to predict generic case too, but we have to
|
ignore first match heuristics and do more involved combining. Implement
|
ignore first match heuristics and do more involved combining. Implement
|
this later. */
|
this later. */
|
if (nedges != 2)
|
if (nedges != 2)
|
{
|
{
|
if (!bb->count)
|
if (!bb->count)
|
set_even_probabilities (bb);
|
set_even_probabilities (bb);
|
bb->predictions = NULL;
|
bb->predictions = NULL;
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "%i edges in bb %i predicted to even probabilities\n",
|
fprintf (dump_file, "%i edges in bb %i predicted to even probabilities\n",
|
nedges, bb->index);
|
nedges, bb->index);
|
return;
|
return;
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "Predictions for bb %i\n", bb->index);
|
fprintf (dump_file, "Predictions for bb %i\n", bb->index);
|
|
|
/* We implement "first match" heuristics and use probability guessed
|
/* We implement "first match" heuristics and use probability guessed
|
by predictor with smallest index. */
|
by predictor with smallest index. */
|
for (pred = bb->predictions; pred; pred = pred->ep_next)
|
for (pred = bb->predictions; pred; pred = pred->ep_next)
|
{
|
{
|
int predictor = pred->ep_predictor;
|
int predictor = pred->ep_predictor;
|
int probability = pred->ep_probability;
|
int probability = pred->ep_probability;
|
|
|
if (pred->ep_edge != first)
|
if (pred->ep_edge != first)
|
probability = REG_BR_PROB_BASE - probability;
|
probability = REG_BR_PROB_BASE - probability;
|
|
|
found = true;
|
found = true;
|
if (best_predictor > predictor)
|
if (best_predictor > predictor)
|
best_probability = probability, best_predictor = predictor;
|
best_probability = probability, best_predictor = predictor;
|
|
|
d = (combined_probability * probability
|
d = (combined_probability * probability
|
+ (REG_BR_PROB_BASE - combined_probability)
|
+ (REG_BR_PROB_BASE - combined_probability)
|
* (REG_BR_PROB_BASE - probability));
|
* (REG_BR_PROB_BASE - probability));
|
|
|
/* Use FP math to avoid overflows of 32bit integers. */
|
/* Use FP math to avoid overflows of 32bit integers. */
|
if (d == 0)
|
if (d == 0)
|
/* If one probability is 0% and one 100%, avoid division by zero. */
|
/* If one probability is 0% and one 100%, avoid division by zero. */
|
combined_probability = REG_BR_PROB_BASE / 2;
|
combined_probability = REG_BR_PROB_BASE / 2;
|
else
|
else
|
combined_probability = (((double) combined_probability) * probability
|
combined_probability = (((double) combined_probability) * probability
|
* REG_BR_PROB_BASE / d + 0.5);
|
* REG_BR_PROB_BASE / d + 0.5);
|
}
|
}
|
|
|
/* Decide which heuristic to use. In case we didn't match anything,
|
/* Decide which heuristic to use. In case we didn't match anything,
|
use no_prediction heuristic, in case we did match, use either
|
use no_prediction heuristic, in case we did match, use either
|
first match or Dempster-Shaffer theory depending on the flags. */
|
first match or Dempster-Shaffer theory depending on the flags. */
|
|
|
if (predictor_info [best_predictor].flags & PRED_FLAG_FIRST_MATCH)
|
if (predictor_info [best_predictor].flags & PRED_FLAG_FIRST_MATCH)
|
first_match = true;
|
first_match = true;
|
|
|
if (!found)
|
if (!found)
|
dump_prediction (dump_file, PRED_NO_PREDICTION, combined_probability, bb, true);
|
dump_prediction (dump_file, PRED_NO_PREDICTION, combined_probability, bb, true);
|
else
|
else
|
{
|
{
|
dump_prediction (dump_file, PRED_DS_THEORY, combined_probability, bb,
|
dump_prediction (dump_file, PRED_DS_THEORY, combined_probability, bb,
|
!first_match);
|
!first_match);
|
dump_prediction (dump_file, PRED_FIRST_MATCH, best_probability, bb,
|
dump_prediction (dump_file, PRED_FIRST_MATCH, best_probability, bb,
|
first_match);
|
first_match);
|
}
|
}
|
|
|
if (first_match)
|
if (first_match)
|
combined_probability = best_probability;
|
combined_probability = best_probability;
|
dump_prediction (dump_file, PRED_COMBINED, combined_probability, bb, true);
|
dump_prediction (dump_file, PRED_COMBINED, combined_probability, bb, true);
|
|
|
for (pred = bb->predictions; pred; pred = pred->ep_next)
|
for (pred = bb->predictions; pred; pred = pred->ep_next)
|
{
|
{
|
int predictor = pred->ep_predictor;
|
int predictor = pred->ep_predictor;
|
int probability = pred->ep_probability;
|
int probability = pred->ep_probability;
|
|
|
if (pred->ep_edge != EDGE_SUCC (bb, 0))
|
if (pred->ep_edge != EDGE_SUCC (bb, 0))
|
probability = REG_BR_PROB_BASE - probability;
|
probability = REG_BR_PROB_BASE - probability;
|
dump_prediction (dump_file, predictor, probability, bb,
|
dump_prediction (dump_file, predictor, probability, bb,
|
!first_match || best_predictor == predictor);
|
!first_match || best_predictor == predictor);
|
}
|
}
|
bb->predictions = NULL;
|
bb->predictions = NULL;
|
|
|
if (!bb->count)
|
if (!bb->count)
|
{
|
{
|
first->probability = combined_probability;
|
first->probability = combined_probability;
|
second->probability = REG_BR_PROB_BASE - combined_probability;
|
second->probability = REG_BR_PROB_BASE - combined_probability;
|
}
|
}
|
}
|
}
|
|
|
/* Predict edge probabilities by exploiting loop structure.
|
/* Predict edge probabilities by exploiting loop structure.
|
When RTLSIMPLELOOPS is set, attempt to count number of iterations by analyzing
|
When RTLSIMPLELOOPS is set, attempt to count number of iterations by analyzing
|
RTL otherwise use tree based approach. */
|
RTL otherwise use tree based approach. */
|
static void
|
static void
|
predict_loops (struct loops *loops_info, bool rtlsimpleloops)
|
predict_loops (struct loops *loops_info, bool rtlsimpleloops)
|
{
|
{
|
unsigned i;
|
unsigned i;
|
|
|
if (!rtlsimpleloops)
|
if (!rtlsimpleloops)
|
scev_initialize (loops_info);
|
scev_initialize (loops_info);
|
|
|
/* Try to predict out blocks in a loop that are not part of a
|
/* Try to predict out blocks in a loop that are not part of a
|
natural loop. */
|
natural loop. */
|
for (i = 1; i < loops_info->num; i++)
|
for (i = 1; i < loops_info->num; i++)
|
{
|
{
|
basic_block bb, *bbs;
|
basic_block bb, *bbs;
|
unsigned j;
|
unsigned j;
|
unsigned n_exits;
|
unsigned n_exits;
|
struct loop *loop = loops_info->parray[i];
|
struct loop *loop = loops_info->parray[i];
|
struct niter_desc desc;
|
struct niter_desc desc;
|
unsigned HOST_WIDE_INT niter;
|
unsigned HOST_WIDE_INT niter;
|
edge *exits;
|
edge *exits;
|
|
|
exits = get_loop_exit_edges (loop, &n_exits);
|
exits = get_loop_exit_edges (loop, &n_exits);
|
|
|
if (rtlsimpleloops)
|
if (rtlsimpleloops)
|
{
|
{
|
iv_analysis_loop_init (loop);
|
iv_analysis_loop_init (loop);
|
find_simple_exit (loop, &desc);
|
find_simple_exit (loop, &desc);
|
|
|
if (desc.simple_p && desc.const_iter)
|
if (desc.simple_p && desc.const_iter)
|
{
|
{
|
int prob;
|
int prob;
|
niter = desc.niter + 1;
|
niter = desc.niter + 1;
|
if (niter == 0) /* We might overflow here. */
|
if (niter == 0) /* We might overflow here. */
|
niter = desc.niter;
|
niter = desc.niter;
|
if (niter
|
if (niter
|
> (unsigned int) PARAM_VALUE (PARAM_MAX_PREDICTED_ITERATIONS))
|
> (unsigned int) PARAM_VALUE (PARAM_MAX_PREDICTED_ITERATIONS))
|
niter = PARAM_VALUE (PARAM_MAX_PREDICTED_ITERATIONS);
|
niter = PARAM_VALUE (PARAM_MAX_PREDICTED_ITERATIONS);
|
|
|
prob = (REG_BR_PROB_BASE
|
prob = (REG_BR_PROB_BASE
|
- (REG_BR_PROB_BASE + niter /2) / niter);
|
- (REG_BR_PROB_BASE + niter /2) / niter);
|
/* Branch prediction algorithm gives 0 frequency for everything
|
/* Branch prediction algorithm gives 0 frequency for everything
|
after the end of loop for loop having 0 probability to finish. */
|
after the end of loop for loop having 0 probability to finish. */
|
if (prob == REG_BR_PROB_BASE)
|
if (prob == REG_BR_PROB_BASE)
|
prob = REG_BR_PROB_BASE - 1;
|
prob = REG_BR_PROB_BASE - 1;
|
predict_edge (desc.in_edge, PRED_LOOP_ITERATIONS,
|
predict_edge (desc.in_edge, PRED_LOOP_ITERATIONS,
|
prob);
|
prob);
|
}
|
}
|
}
|
}
|
else
|
else
|
{
|
{
|
struct tree_niter_desc niter_desc;
|
struct tree_niter_desc niter_desc;
|
|
|
for (j = 0; j < n_exits; j++)
|
for (j = 0; j < n_exits; j++)
|
{
|
{
|
tree niter = NULL;
|
tree niter = NULL;
|
|
|
if (number_of_iterations_exit (loop, exits[j], &niter_desc, false))
|
if (number_of_iterations_exit (loop, exits[j], &niter_desc, false))
|
niter = niter_desc.niter;
|
niter = niter_desc.niter;
|
if (!niter || TREE_CODE (niter_desc.niter) != INTEGER_CST)
|
if (!niter || TREE_CODE (niter_desc.niter) != INTEGER_CST)
|
niter = loop_niter_by_eval (loop, exits[j]);
|
niter = loop_niter_by_eval (loop, exits[j]);
|
|
|
if (TREE_CODE (niter) == INTEGER_CST)
|
if (TREE_CODE (niter) == INTEGER_CST)
|
{
|
{
|
int probability;
|
int probability;
|
int max = PARAM_VALUE (PARAM_MAX_PREDICTED_ITERATIONS);
|
int max = PARAM_VALUE (PARAM_MAX_PREDICTED_ITERATIONS);
|
if (host_integerp (niter, 1)
|
if (host_integerp (niter, 1)
|
&& tree_int_cst_lt (niter,
|
&& tree_int_cst_lt (niter,
|
build_int_cstu (NULL_TREE, max - 1)))
|
build_int_cstu (NULL_TREE, max - 1)))
|
{
|
{
|
HOST_WIDE_INT nitercst = tree_low_cst (niter, 1) + 1;
|
HOST_WIDE_INT nitercst = tree_low_cst (niter, 1) + 1;
|
probability = ((REG_BR_PROB_BASE + nitercst / 2)
|
probability = ((REG_BR_PROB_BASE + nitercst / 2)
|
/ nitercst);
|
/ nitercst);
|
}
|
}
|
else
|
else
|
probability = ((REG_BR_PROB_BASE + max / 2) / max);
|
probability = ((REG_BR_PROB_BASE + max / 2) / max);
|
|
|
predict_edge (exits[j], PRED_LOOP_ITERATIONS, probability);
|
predict_edge (exits[j], PRED_LOOP_ITERATIONS, probability);
|
}
|
}
|
}
|
}
|
|
|
}
|
}
|
free (exits);
|
free (exits);
|
|
|
bbs = get_loop_body (loop);
|
bbs = get_loop_body (loop);
|
|
|
for (j = 0; j < loop->num_nodes; j++)
|
for (j = 0; j < loop->num_nodes; j++)
|
{
|
{
|
int header_found = 0;
|
int header_found = 0;
|
edge e;
|
edge e;
|
edge_iterator ei;
|
edge_iterator ei;
|
|
|
bb = bbs[j];
|
bb = bbs[j];
|
|
|
/* Bypass loop heuristics on continue statement. These
|
/* Bypass loop heuristics on continue statement. These
|
statements construct loops via "non-loop" constructs
|
statements construct loops via "non-loop" constructs
|
in the source language and are better to be handled
|
in the source language and are better to be handled
|
separately. */
|
separately. */
|
if ((rtlsimpleloops && !can_predict_insn_p (BB_END (bb)))
|
if ((rtlsimpleloops && !can_predict_insn_p (BB_END (bb)))
|
|| predicted_by_p (bb, PRED_CONTINUE))
|
|| predicted_by_p (bb, PRED_CONTINUE))
|
continue;
|
continue;
|
|
|
/* Loop branch heuristics - predict an edge back to a
|
/* Loop branch heuristics - predict an edge back to a
|
loop's head as taken. */
|
loop's head as taken. */
|
if (bb == loop->latch)
|
if (bb == loop->latch)
|
{
|
{
|
e = find_edge (loop->latch, loop->header);
|
e = find_edge (loop->latch, loop->header);
|
if (e)
|
if (e)
|
{
|
{
|
header_found = 1;
|
header_found = 1;
|
predict_edge_def (e, PRED_LOOP_BRANCH, TAKEN);
|
predict_edge_def (e, PRED_LOOP_BRANCH, TAKEN);
|
}
|
}
|
}
|
}
|
|
|
/* Loop exit heuristics - predict an edge exiting the loop if the
|
/* Loop exit heuristics - predict an edge exiting the loop if the
|
conditional has no loop header successors as not taken. */
|
conditional has no loop header successors as not taken. */
|
if (!header_found)
|
if (!header_found)
|
{
|
{
|
/* For loop with many exits we don't want to predict all exits
|
/* For loop with many exits we don't want to predict all exits
|
with the pretty large probability, because if all exits are
|
with the pretty large probability, because if all exits are
|
considered in row, the loop would be predicted to iterate
|
considered in row, the loop would be predicted to iterate
|
almost never. The code to divide probability by number of
|
almost never. The code to divide probability by number of
|
exits is very rough. It should compute the number of exits
|
exits is very rough. It should compute the number of exits
|
taken in each patch through function (not the overall number
|
taken in each patch through function (not the overall number
|
of exits that might be a lot higher for loops with wide switch
|
of exits that might be a lot higher for loops with wide switch
|
statements in them) and compute n-th square root.
|
statements in them) and compute n-th square root.
|
|
|
We limit the minimal probability by 2% to avoid
|
We limit the minimal probability by 2% to avoid
|
EDGE_PROBABILITY_RELIABLE from trusting the branch prediction
|
EDGE_PROBABILITY_RELIABLE from trusting the branch prediction
|
as this was causing regression in perl benchmark containing such
|
as this was causing regression in perl benchmark containing such
|
a wide loop. */
|
a wide loop. */
|
|
|
int probability = ((REG_BR_PROB_BASE
|
int probability = ((REG_BR_PROB_BASE
|
- predictor_info [(int) PRED_LOOP_EXIT].hitrate)
|
- predictor_info [(int) PRED_LOOP_EXIT].hitrate)
|
/ n_exits);
|
/ n_exits);
|
if (probability < HITRATE (2))
|
if (probability < HITRATE (2))
|
probability = HITRATE (2);
|
probability = HITRATE (2);
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
if (e->dest->index < NUM_FIXED_BLOCKS
|
if (e->dest->index < NUM_FIXED_BLOCKS
|
|| !flow_bb_inside_loop_p (loop, e->dest))
|
|| !flow_bb_inside_loop_p (loop, e->dest))
|
predict_edge (e, PRED_LOOP_EXIT, probability);
|
predict_edge (e, PRED_LOOP_EXIT, probability);
|
}
|
}
|
}
|
}
|
|
|
/* Free basic blocks from get_loop_body. */
|
/* Free basic blocks from get_loop_body. */
|
free (bbs);
|
free (bbs);
|
}
|
}
|
|
|
if (!rtlsimpleloops)
|
if (!rtlsimpleloops)
|
{
|
{
|
scev_finalize ();
|
scev_finalize ();
|
current_loops = NULL;
|
current_loops = NULL;
|
}
|
}
|
}
|
}
|
|
|
/* Attempt to predict probabilities of BB outgoing edges using local
|
/* Attempt to predict probabilities of BB outgoing edges using local
|
properties. */
|
properties. */
|
static void
|
static void
|
bb_estimate_probability_locally (basic_block bb)
|
bb_estimate_probability_locally (basic_block bb)
|
{
|
{
|
rtx last_insn = BB_END (bb);
|
rtx last_insn = BB_END (bb);
|
rtx cond;
|
rtx cond;
|
|
|
if (! can_predict_insn_p (last_insn))
|
if (! can_predict_insn_p (last_insn))
|
return;
|
return;
|
cond = get_condition (last_insn, NULL, false, false);
|
cond = get_condition (last_insn, NULL, false, false);
|
if (! cond)
|
if (! cond)
|
return;
|
return;
|
|
|
/* Try "pointer heuristic."
|
/* Try "pointer heuristic."
|
A comparison ptr == 0 is predicted as false.
|
A comparison ptr == 0 is predicted as false.
|
Similarly, a comparison ptr1 == ptr2 is predicted as false. */
|
Similarly, a comparison ptr1 == ptr2 is predicted as false. */
|
if (COMPARISON_P (cond)
|
if (COMPARISON_P (cond)
|
&& ((REG_P (XEXP (cond, 0)) && REG_POINTER (XEXP (cond, 0)))
|
&& ((REG_P (XEXP (cond, 0)) && REG_POINTER (XEXP (cond, 0)))
|
|| (REG_P (XEXP (cond, 1)) && REG_POINTER (XEXP (cond, 1)))))
|
|| (REG_P (XEXP (cond, 1)) && REG_POINTER (XEXP (cond, 1)))))
|
{
|
{
|
if (GET_CODE (cond) == EQ)
|
if (GET_CODE (cond) == EQ)
|
predict_insn_def (last_insn, PRED_POINTER, NOT_TAKEN);
|
predict_insn_def (last_insn, PRED_POINTER, NOT_TAKEN);
|
else if (GET_CODE (cond) == NE)
|
else if (GET_CODE (cond) == NE)
|
predict_insn_def (last_insn, PRED_POINTER, TAKEN);
|
predict_insn_def (last_insn, PRED_POINTER, TAKEN);
|
}
|
}
|
else
|
else
|
|
|
/* Try "opcode heuristic."
|
/* Try "opcode heuristic."
|
EQ tests are usually false and NE tests are usually true. Also,
|
EQ tests are usually false and NE tests are usually true. Also,
|
most quantities are positive, so we can make the appropriate guesses
|
most quantities are positive, so we can make the appropriate guesses
|
about signed comparisons against zero. */
|
about signed comparisons against zero. */
|
switch (GET_CODE (cond))
|
switch (GET_CODE (cond))
|
{
|
{
|
case CONST_INT:
|
case CONST_INT:
|
/* Unconditional branch. */
|
/* Unconditional branch. */
|
predict_insn_def (last_insn, PRED_UNCONDITIONAL,
|
predict_insn_def (last_insn, PRED_UNCONDITIONAL,
|
cond == const0_rtx ? NOT_TAKEN : TAKEN);
|
cond == const0_rtx ? NOT_TAKEN : TAKEN);
|
break;
|
break;
|
|
|
case EQ:
|
case EQ:
|
case UNEQ:
|
case UNEQ:
|
/* Floating point comparisons appears to behave in a very
|
/* Floating point comparisons appears to behave in a very
|
unpredictable way because of special role of = tests in
|
unpredictable way because of special role of = tests in
|
FP code. */
|
FP code. */
|
if (FLOAT_MODE_P (GET_MODE (XEXP (cond, 0))))
|
if (FLOAT_MODE_P (GET_MODE (XEXP (cond, 0))))
|
;
|
;
|
/* Comparisons with 0 are often used for booleans and there is
|
/* Comparisons with 0 are often used for booleans and there is
|
nothing useful to predict about them. */
|
nothing useful to predict about them. */
|
else if (XEXP (cond, 1) == const0_rtx
|
else if (XEXP (cond, 1) == const0_rtx
|
|| XEXP (cond, 0) == const0_rtx)
|
|| XEXP (cond, 0) == const0_rtx)
|
;
|
;
|
else
|
else
|
predict_insn_def (last_insn, PRED_OPCODE_NONEQUAL, NOT_TAKEN);
|
predict_insn_def (last_insn, PRED_OPCODE_NONEQUAL, NOT_TAKEN);
|
break;
|
break;
|
|
|
case NE:
|
case NE:
|
case LTGT:
|
case LTGT:
|
/* Floating point comparisons appears to behave in a very
|
/* Floating point comparisons appears to behave in a very
|
unpredictable way because of special role of = tests in
|
unpredictable way because of special role of = tests in
|
FP code. */
|
FP code. */
|
if (FLOAT_MODE_P (GET_MODE (XEXP (cond, 0))))
|
if (FLOAT_MODE_P (GET_MODE (XEXP (cond, 0))))
|
;
|
;
|
/* Comparisons with 0 are often used for booleans and there is
|
/* Comparisons with 0 are often used for booleans and there is
|
nothing useful to predict about them. */
|
nothing useful to predict about them. */
|
else if (XEXP (cond, 1) == const0_rtx
|
else if (XEXP (cond, 1) == const0_rtx
|
|| XEXP (cond, 0) == const0_rtx)
|
|| XEXP (cond, 0) == const0_rtx)
|
;
|
;
|
else
|
else
|
predict_insn_def (last_insn, PRED_OPCODE_NONEQUAL, TAKEN);
|
predict_insn_def (last_insn, PRED_OPCODE_NONEQUAL, TAKEN);
|
break;
|
break;
|
|
|
case ORDERED:
|
case ORDERED:
|
predict_insn_def (last_insn, PRED_FPOPCODE, TAKEN);
|
predict_insn_def (last_insn, PRED_FPOPCODE, TAKEN);
|
break;
|
break;
|
|
|
case UNORDERED:
|
case UNORDERED:
|
predict_insn_def (last_insn, PRED_FPOPCODE, NOT_TAKEN);
|
predict_insn_def (last_insn, PRED_FPOPCODE, NOT_TAKEN);
|
break;
|
break;
|
|
|
case LE:
|
case LE:
|
case LT:
|
case LT:
|
if (XEXP (cond, 1) == const0_rtx || XEXP (cond, 1) == const1_rtx
|
if (XEXP (cond, 1) == const0_rtx || XEXP (cond, 1) == const1_rtx
|
|| XEXP (cond, 1) == constm1_rtx)
|
|| XEXP (cond, 1) == constm1_rtx)
|
predict_insn_def (last_insn, PRED_OPCODE_POSITIVE, NOT_TAKEN);
|
predict_insn_def (last_insn, PRED_OPCODE_POSITIVE, NOT_TAKEN);
|
break;
|
break;
|
|
|
case GE:
|
case GE:
|
case GT:
|
case GT:
|
if (XEXP (cond, 1) == const0_rtx || XEXP (cond, 1) == const1_rtx
|
if (XEXP (cond, 1) == const0_rtx || XEXP (cond, 1) == const1_rtx
|
|| XEXP (cond, 1) == constm1_rtx)
|
|| XEXP (cond, 1) == constm1_rtx)
|
predict_insn_def (last_insn, PRED_OPCODE_POSITIVE, TAKEN);
|
predict_insn_def (last_insn, PRED_OPCODE_POSITIVE, TAKEN);
|
break;
|
break;
|
|
|
default:
|
default:
|
break;
|
break;
|
}
|
}
|
}
|
}
|
|
|
/* Set edge->probability for each successor edge of BB. */
|
/* Set edge->probability for each successor edge of BB. */
|
void
|
void
|
guess_outgoing_edge_probabilities (basic_block bb)
|
guess_outgoing_edge_probabilities (basic_block bb)
|
{
|
{
|
bb_estimate_probability_locally (bb);
|
bb_estimate_probability_locally (bb);
|
combine_predictions_for_insn (BB_END (bb), bb);
|
combine_predictions_for_insn (BB_END (bb), bb);
|
}
|
}
|
|
|
/* Return constant EXPR will likely have at execution time, NULL if unknown.
|
/* Return constant EXPR will likely have at execution time, NULL if unknown.
|
The function is used by builtin_expect branch predictor so the evidence
|
The function is used by builtin_expect branch predictor so the evidence
|
must come from this construct and additional possible constant folding.
|
must come from this construct and additional possible constant folding.
|
|
|
We may want to implement more involved value guess (such as value range
|
We may want to implement more involved value guess (such as value range
|
propagation based prediction), but such tricks shall go to new
|
propagation based prediction), but such tricks shall go to new
|
implementation. */
|
implementation. */
|
|
|
static tree
|
static tree
|
expr_expected_value (tree expr, bitmap visited)
|
expr_expected_value (tree expr, bitmap visited)
|
{
|
{
|
if (TREE_CONSTANT (expr))
|
if (TREE_CONSTANT (expr))
|
return expr;
|
return expr;
|
else if (TREE_CODE (expr) == SSA_NAME)
|
else if (TREE_CODE (expr) == SSA_NAME)
|
{
|
{
|
tree def = SSA_NAME_DEF_STMT (expr);
|
tree def = SSA_NAME_DEF_STMT (expr);
|
|
|
/* If we were already here, break the infinite cycle. */
|
/* If we were already here, break the infinite cycle. */
|
if (bitmap_bit_p (visited, SSA_NAME_VERSION (expr)))
|
if (bitmap_bit_p (visited, SSA_NAME_VERSION (expr)))
|
return NULL;
|
return NULL;
|
bitmap_set_bit (visited, SSA_NAME_VERSION (expr));
|
bitmap_set_bit (visited, SSA_NAME_VERSION (expr));
|
|
|
if (TREE_CODE (def) == PHI_NODE)
|
if (TREE_CODE (def) == PHI_NODE)
|
{
|
{
|
/* All the arguments of the PHI node must have the same constant
|
/* All the arguments of the PHI node must have the same constant
|
length. */
|
length. */
|
int i;
|
int i;
|
tree val = NULL, new_val;
|
tree val = NULL, new_val;
|
|
|
for (i = 0; i < PHI_NUM_ARGS (def); i++)
|
for (i = 0; i < PHI_NUM_ARGS (def); i++)
|
{
|
{
|
tree arg = PHI_ARG_DEF (def, i);
|
tree arg = PHI_ARG_DEF (def, i);
|
|
|
/* If this PHI has itself as an argument, we cannot
|
/* If this PHI has itself as an argument, we cannot
|
determine the string length of this argument. However,
|
determine the string length of this argument. However,
|
if we can find an expected constant value for the other
|
if we can find an expected constant value for the other
|
PHI args then we can still be sure that this is
|
PHI args then we can still be sure that this is
|
likely a constant. So be optimistic and just
|
likely a constant. So be optimistic and just
|
continue with the next argument. */
|
continue with the next argument. */
|
if (arg == PHI_RESULT (def))
|
if (arg == PHI_RESULT (def))
|
continue;
|
continue;
|
|
|
new_val = expr_expected_value (arg, visited);
|
new_val = expr_expected_value (arg, visited);
|
if (!new_val)
|
if (!new_val)
|
return NULL;
|
return NULL;
|
if (!val)
|
if (!val)
|
val = new_val;
|
val = new_val;
|
else if (!operand_equal_p (val, new_val, false))
|
else if (!operand_equal_p (val, new_val, false))
|
return NULL;
|
return NULL;
|
}
|
}
|
return val;
|
return val;
|
}
|
}
|
if (TREE_CODE (def) != MODIFY_EXPR || TREE_OPERAND (def, 0) != expr)
|
if (TREE_CODE (def) != MODIFY_EXPR || TREE_OPERAND (def, 0) != expr)
|
return NULL;
|
return NULL;
|
return expr_expected_value (TREE_OPERAND (def, 1), visited);
|
return expr_expected_value (TREE_OPERAND (def, 1), visited);
|
}
|
}
|
else if (TREE_CODE (expr) == CALL_EXPR)
|
else if (TREE_CODE (expr) == CALL_EXPR)
|
{
|
{
|
tree decl = get_callee_fndecl (expr);
|
tree decl = get_callee_fndecl (expr);
|
if (!decl)
|
if (!decl)
|
return NULL;
|
return NULL;
|
if (DECL_BUILT_IN_CLASS (decl) == BUILT_IN_NORMAL
|
if (DECL_BUILT_IN_CLASS (decl) == BUILT_IN_NORMAL
|
&& DECL_FUNCTION_CODE (decl) == BUILT_IN_EXPECT)
|
&& DECL_FUNCTION_CODE (decl) == BUILT_IN_EXPECT)
|
{
|
{
|
tree arglist = TREE_OPERAND (expr, 1);
|
tree arglist = TREE_OPERAND (expr, 1);
|
tree val;
|
tree val;
|
|
|
if (arglist == NULL_TREE
|
if (arglist == NULL_TREE
|
|| TREE_CHAIN (arglist) == NULL_TREE)
|
|| TREE_CHAIN (arglist) == NULL_TREE)
|
return NULL;
|
return NULL;
|
val = TREE_VALUE (TREE_CHAIN (TREE_OPERAND (expr, 1)));
|
val = TREE_VALUE (TREE_CHAIN (TREE_OPERAND (expr, 1)));
|
if (TREE_CONSTANT (val))
|
if (TREE_CONSTANT (val))
|
return val;
|
return val;
|
return TREE_VALUE (TREE_CHAIN (TREE_OPERAND (expr, 1)));
|
return TREE_VALUE (TREE_CHAIN (TREE_OPERAND (expr, 1)));
|
}
|
}
|
}
|
}
|
if (BINARY_CLASS_P (expr) || COMPARISON_CLASS_P (expr))
|
if (BINARY_CLASS_P (expr) || COMPARISON_CLASS_P (expr))
|
{
|
{
|
tree op0, op1, res;
|
tree op0, op1, res;
|
op0 = expr_expected_value (TREE_OPERAND (expr, 0), visited);
|
op0 = expr_expected_value (TREE_OPERAND (expr, 0), visited);
|
if (!op0)
|
if (!op0)
|
return NULL;
|
return NULL;
|
op1 = expr_expected_value (TREE_OPERAND (expr, 1), visited);
|
op1 = expr_expected_value (TREE_OPERAND (expr, 1), visited);
|
if (!op1)
|
if (!op1)
|
return NULL;
|
return NULL;
|
res = fold_build2 (TREE_CODE (expr), TREE_TYPE (expr), op0, op1);
|
res = fold_build2 (TREE_CODE (expr), TREE_TYPE (expr), op0, op1);
|
if (TREE_CONSTANT (res))
|
if (TREE_CONSTANT (res))
|
return res;
|
return res;
|
return NULL;
|
return NULL;
|
}
|
}
|
if (UNARY_CLASS_P (expr))
|
if (UNARY_CLASS_P (expr))
|
{
|
{
|
tree op0, res;
|
tree op0, res;
|
op0 = expr_expected_value (TREE_OPERAND (expr, 0), visited);
|
op0 = expr_expected_value (TREE_OPERAND (expr, 0), visited);
|
if (!op0)
|
if (!op0)
|
return NULL;
|
return NULL;
|
res = fold_build1 (TREE_CODE (expr), TREE_TYPE (expr), op0);
|
res = fold_build1 (TREE_CODE (expr), TREE_TYPE (expr), op0);
|
if (TREE_CONSTANT (res))
|
if (TREE_CONSTANT (res))
|
return res;
|
return res;
|
return NULL;
|
return NULL;
|
}
|
}
|
return NULL;
|
return NULL;
|
}
|
}
|
|
|
/* Get rid of all builtin_expect calls we no longer need. */
|
/* Get rid of all builtin_expect calls we no longer need. */
|
static void
|
static void
|
strip_builtin_expect (void)
|
strip_builtin_expect (void)
|
{
|
{
|
basic_block bb;
|
basic_block bb;
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
{
|
{
|
block_stmt_iterator bi;
|
block_stmt_iterator bi;
|
for (bi = bsi_start (bb); !bsi_end_p (bi); bsi_next (&bi))
|
for (bi = bsi_start (bb); !bsi_end_p (bi); bsi_next (&bi))
|
{
|
{
|
tree stmt = bsi_stmt (bi);
|
tree stmt = bsi_stmt (bi);
|
tree fndecl;
|
tree fndecl;
|
tree arglist;
|
tree arglist;
|
|
|
if (TREE_CODE (stmt) == MODIFY_EXPR
|
if (TREE_CODE (stmt) == MODIFY_EXPR
|
&& TREE_CODE (TREE_OPERAND (stmt, 1)) == CALL_EXPR
|
&& TREE_CODE (TREE_OPERAND (stmt, 1)) == CALL_EXPR
|
&& (fndecl = get_callee_fndecl (TREE_OPERAND (stmt, 1)))
|
&& (fndecl = get_callee_fndecl (TREE_OPERAND (stmt, 1)))
|
&& DECL_BUILT_IN_CLASS (fndecl) == BUILT_IN_NORMAL
|
&& DECL_BUILT_IN_CLASS (fndecl) == BUILT_IN_NORMAL
|
&& DECL_FUNCTION_CODE (fndecl) == BUILT_IN_EXPECT
|
&& DECL_FUNCTION_CODE (fndecl) == BUILT_IN_EXPECT
|
&& (arglist = TREE_OPERAND (TREE_OPERAND (stmt, 1), 1))
|
&& (arglist = TREE_OPERAND (TREE_OPERAND (stmt, 1), 1))
|
&& TREE_CHAIN (arglist))
|
&& TREE_CHAIN (arglist))
|
{
|
{
|
TREE_OPERAND (stmt, 1) = TREE_VALUE (arglist);
|
TREE_OPERAND (stmt, 1) = TREE_VALUE (arglist);
|
update_stmt (stmt);
|
update_stmt (stmt);
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
/* Predict using opcode of the last statement in basic block. */
|
/* Predict using opcode of the last statement in basic block. */
|
static void
|
static void
|
tree_predict_by_opcode (basic_block bb)
|
tree_predict_by_opcode (basic_block bb)
|
{
|
{
|
tree stmt = last_stmt (bb);
|
tree stmt = last_stmt (bb);
|
edge then_edge;
|
edge then_edge;
|
tree cond;
|
tree cond;
|
tree op0;
|
tree op0;
|
tree type;
|
tree type;
|
tree val;
|
tree val;
|
bitmap visited;
|
bitmap visited;
|
edge_iterator ei;
|
edge_iterator ei;
|
|
|
if (!stmt || TREE_CODE (stmt) != COND_EXPR)
|
if (!stmt || TREE_CODE (stmt) != COND_EXPR)
|
return;
|
return;
|
FOR_EACH_EDGE (then_edge, ei, bb->succs)
|
FOR_EACH_EDGE (then_edge, ei, bb->succs)
|
if (then_edge->flags & EDGE_TRUE_VALUE)
|
if (then_edge->flags & EDGE_TRUE_VALUE)
|
break;
|
break;
|
cond = TREE_OPERAND (stmt, 0);
|
cond = TREE_OPERAND (stmt, 0);
|
if (!COMPARISON_CLASS_P (cond))
|
if (!COMPARISON_CLASS_P (cond))
|
return;
|
return;
|
op0 = TREE_OPERAND (cond, 0);
|
op0 = TREE_OPERAND (cond, 0);
|
type = TREE_TYPE (op0);
|
type = TREE_TYPE (op0);
|
visited = BITMAP_ALLOC (NULL);
|
visited = BITMAP_ALLOC (NULL);
|
val = expr_expected_value (cond, visited);
|
val = expr_expected_value (cond, visited);
|
BITMAP_FREE (visited);
|
BITMAP_FREE (visited);
|
if (val)
|
if (val)
|
{
|
{
|
if (integer_zerop (val))
|
if (integer_zerop (val))
|
predict_edge_def (then_edge, PRED_BUILTIN_EXPECT, NOT_TAKEN);
|
predict_edge_def (then_edge, PRED_BUILTIN_EXPECT, NOT_TAKEN);
|
else
|
else
|
predict_edge_def (then_edge, PRED_BUILTIN_EXPECT, TAKEN);
|
predict_edge_def (then_edge, PRED_BUILTIN_EXPECT, TAKEN);
|
return;
|
return;
|
}
|
}
|
/* Try "pointer heuristic."
|
/* Try "pointer heuristic."
|
A comparison ptr == 0 is predicted as false.
|
A comparison ptr == 0 is predicted as false.
|
Similarly, a comparison ptr1 == ptr2 is predicted as false. */
|
Similarly, a comparison ptr1 == ptr2 is predicted as false. */
|
if (POINTER_TYPE_P (type))
|
if (POINTER_TYPE_P (type))
|
{
|
{
|
if (TREE_CODE (cond) == EQ_EXPR)
|
if (TREE_CODE (cond) == EQ_EXPR)
|
predict_edge_def (then_edge, PRED_TREE_POINTER, NOT_TAKEN);
|
predict_edge_def (then_edge, PRED_TREE_POINTER, NOT_TAKEN);
|
else if (TREE_CODE (cond) == NE_EXPR)
|
else if (TREE_CODE (cond) == NE_EXPR)
|
predict_edge_def (then_edge, PRED_TREE_POINTER, TAKEN);
|
predict_edge_def (then_edge, PRED_TREE_POINTER, TAKEN);
|
}
|
}
|
else
|
else
|
|
|
/* Try "opcode heuristic."
|
/* Try "opcode heuristic."
|
EQ tests are usually false and NE tests are usually true. Also,
|
EQ tests are usually false and NE tests are usually true. Also,
|
most quantities are positive, so we can make the appropriate guesses
|
most quantities are positive, so we can make the appropriate guesses
|
about signed comparisons against zero. */
|
about signed comparisons against zero. */
|
switch (TREE_CODE (cond))
|
switch (TREE_CODE (cond))
|
{
|
{
|
case EQ_EXPR:
|
case EQ_EXPR:
|
case UNEQ_EXPR:
|
case UNEQ_EXPR:
|
/* Floating point comparisons appears to behave in a very
|
/* Floating point comparisons appears to behave in a very
|
unpredictable way because of special role of = tests in
|
unpredictable way because of special role of = tests in
|
FP code. */
|
FP code. */
|
if (FLOAT_TYPE_P (type))
|
if (FLOAT_TYPE_P (type))
|
;
|
;
|
/* Comparisons with 0 are often used for booleans and there is
|
/* Comparisons with 0 are often used for booleans and there is
|
nothing useful to predict about them. */
|
nothing useful to predict about them. */
|
else if (integer_zerop (op0)
|
else if (integer_zerop (op0)
|
|| integer_zerop (TREE_OPERAND (cond, 1)))
|
|| integer_zerop (TREE_OPERAND (cond, 1)))
|
;
|
;
|
else
|
else
|
predict_edge_def (then_edge, PRED_TREE_OPCODE_NONEQUAL, NOT_TAKEN);
|
predict_edge_def (then_edge, PRED_TREE_OPCODE_NONEQUAL, NOT_TAKEN);
|
break;
|
break;
|
|
|
case NE_EXPR:
|
case NE_EXPR:
|
case LTGT_EXPR:
|
case LTGT_EXPR:
|
/* Floating point comparisons appears to behave in a very
|
/* Floating point comparisons appears to behave in a very
|
unpredictable way because of special role of = tests in
|
unpredictable way because of special role of = tests in
|
FP code. */
|
FP code. */
|
if (FLOAT_TYPE_P (type))
|
if (FLOAT_TYPE_P (type))
|
;
|
;
|
/* Comparisons with 0 are often used for booleans and there is
|
/* Comparisons with 0 are often used for booleans and there is
|
nothing useful to predict about them. */
|
nothing useful to predict about them. */
|
else if (integer_zerop (op0)
|
else if (integer_zerop (op0)
|
|| integer_zerop (TREE_OPERAND (cond, 1)))
|
|| integer_zerop (TREE_OPERAND (cond, 1)))
|
;
|
;
|
else
|
else
|
predict_edge_def (then_edge, PRED_TREE_OPCODE_NONEQUAL, TAKEN);
|
predict_edge_def (then_edge, PRED_TREE_OPCODE_NONEQUAL, TAKEN);
|
break;
|
break;
|
|
|
case ORDERED_EXPR:
|
case ORDERED_EXPR:
|
predict_edge_def (then_edge, PRED_TREE_FPOPCODE, TAKEN);
|
predict_edge_def (then_edge, PRED_TREE_FPOPCODE, TAKEN);
|
break;
|
break;
|
|
|
case UNORDERED_EXPR:
|
case UNORDERED_EXPR:
|
predict_edge_def (then_edge, PRED_TREE_FPOPCODE, NOT_TAKEN);
|
predict_edge_def (then_edge, PRED_TREE_FPOPCODE, NOT_TAKEN);
|
break;
|
break;
|
|
|
case LE_EXPR:
|
case LE_EXPR:
|
case LT_EXPR:
|
case LT_EXPR:
|
if (integer_zerop (TREE_OPERAND (cond, 1))
|
if (integer_zerop (TREE_OPERAND (cond, 1))
|
|| integer_onep (TREE_OPERAND (cond, 1))
|
|| integer_onep (TREE_OPERAND (cond, 1))
|
|| integer_all_onesp (TREE_OPERAND (cond, 1))
|
|| integer_all_onesp (TREE_OPERAND (cond, 1))
|
|| real_zerop (TREE_OPERAND (cond, 1))
|
|| real_zerop (TREE_OPERAND (cond, 1))
|
|| real_onep (TREE_OPERAND (cond, 1))
|
|| real_onep (TREE_OPERAND (cond, 1))
|
|| real_minus_onep (TREE_OPERAND (cond, 1)))
|
|| real_minus_onep (TREE_OPERAND (cond, 1)))
|
predict_edge_def (then_edge, PRED_TREE_OPCODE_POSITIVE, NOT_TAKEN);
|
predict_edge_def (then_edge, PRED_TREE_OPCODE_POSITIVE, NOT_TAKEN);
|
break;
|
break;
|
|
|
case GE_EXPR:
|
case GE_EXPR:
|
case GT_EXPR:
|
case GT_EXPR:
|
if (integer_zerop (TREE_OPERAND (cond, 1))
|
if (integer_zerop (TREE_OPERAND (cond, 1))
|
|| integer_onep (TREE_OPERAND (cond, 1))
|
|| integer_onep (TREE_OPERAND (cond, 1))
|
|| integer_all_onesp (TREE_OPERAND (cond, 1))
|
|| integer_all_onesp (TREE_OPERAND (cond, 1))
|
|| real_zerop (TREE_OPERAND (cond, 1))
|
|| real_zerop (TREE_OPERAND (cond, 1))
|
|| real_onep (TREE_OPERAND (cond, 1))
|
|| real_onep (TREE_OPERAND (cond, 1))
|
|| real_minus_onep (TREE_OPERAND (cond, 1)))
|
|| real_minus_onep (TREE_OPERAND (cond, 1)))
|
predict_edge_def (then_edge, PRED_TREE_OPCODE_POSITIVE, TAKEN);
|
predict_edge_def (then_edge, PRED_TREE_OPCODE_POSITIVE, TAKEN);
|
break;
|
break;
|
|
|
default:
|
default:
|
break;
|
break;
|
}
|
}
|
}
|
}
|
|
|
/* Try to guess whether the value of return means error code. */
|
/* Try to guess whether the value of return means error code. */
|
static enum br_predictor
|
static enum br_predictor
|
return_prediction (tree val, enum prediction *prediction)
|
return_prediction (tree val, enum prediction *prediction)
|
{
|
{
|
/* VOID. */
|
/* VOID. */
|
if (!val)
|
if (!val)
|
return PRED_NO_PREDICTION;
|
return PRED_NO_PREDICTION;
|
/* Different heuristics for pointers and scalars. */
|
/* Different heuristics for pointers and scalars. */
|
if (POINTER_TYPE_P (TREE_TYPE (val)))
|
if (POINTER_TYPE_P (TREE_TYPE (val)))
|
{
|
{
|
/* NULL is usually not returned. */
|
/* NULL is usually not returned. */
|
if (integer_zerop (val))
|
if (integer_zerop (val))
|
{
|
{
|
*prediction = NOT_TAKEN;
|
*prediction = NOT_TAKEN;
|
return PRED_NULL_RETURN;
|
return PRED_NULL_RETURN;
|
}
|
}
|
}
|
}
|
else if (INTEGRAL_TYPE_P (TREE_TYPE (val)))
|
else if (INTEGRAL_TYPE_P (TREE_TYPE (val)))
|
{
|
{
|
/* Negative return values are often used to indicate
|
/* Negative return values are often used to indicate
|
errors. */
|
errors. */
|
if (TREE_CODE (val) == INTEGER_CST
|
if (TREE_CODE (val) == INTEGER_CST
|
&& tree_int_cst_sgn (val) < 0)
|
&& tree_int_cst_sgn (val) < 0)
|
{
|
{
|
*prediction = NOT_TAKEN;
|
*prediction = NOT_TAKEN;
|
return PRED_NEGATIVE_RETURN;
|
return PRED_NEGATIVE_RETURN;
|
}
|
}
|
/* Constant return values seems to be commonly taken.
|
/* Constant return values seems to be commonly taken.
|
Zero/one often represent booleans so exclude them from the
|
Zero/one often represent booleans so exclude them from the
|
heuristics. */
|
heuristics. */
|
if (TREE_CONSTANT (val)
|
if (TREE_CONSTANT (val)
|
&& (!integer_zerop (val) && !integer_onep (val)))
|
&& (!integer_zerop (val) && !integer_onep (val)))
|
{
|
{
|
*prediction = TAKEN;
|
*prediction = TAKEN;
|
return PRED_NEGATIVE_RETURN;
|
return PRED_NEGATIVE_RETURN;
|
}
|
}
|
}
|
}
|
return PRED_NO_PREDICTION;
|
return PRED_NO_PREDICTION;
|
}
|
}
|
|
|
/* Find the basic block with return expression and look up for possible
|
/* Find the basic block with return expression and look up for possible
|
return value trying to apply RETURN_PREDICTION heuristics. */
|
return value trying to apply RETURN_PREDICTION heuristics. */
|
static void
|
static void
|
apply_return_prediction (int *heads)
|
apply_return_prediction (int *heads)
|
{
|
{
|
tree return_stmt = NULL;
|
tree return_stmt = NULL;
|
tree return_val;
|
tree return_val;
|
edge e;
|
edge e;
|
tree phi;
|
tree phi;
|
int phi_num_args, i;
|
int phi_num_args, i;
|
enum br_predictor pred;
|
enum br_predictor pred;
|
enum prediction direction;
|
enum prediction direction;
|
edge_iterator ei;
|
edge_iterator ei;
|
|
|
FOR_EACH_EDGE (e, ei, EXIT_BLOCK_PTR->preds)
|
FOR_EACH_EDGE (e, ei, EXIT_BLOCK_PTR->preds)
|
{
|
{
|
return_stmt = last_stmt (e->src);
|
return_stmt = last_stmt (e->src);
|
if (TREE_CODE (return_stmt) == RETURN_EXPR)
|
if (TREE_CODE (return_stmt) == RETURN_EXPR)
|
break;
|
break;
|
}
|
}
|
if (!e)
|
if (!e)
|
return;
|
return;
|
return_val = TREE_OPERAND (return_stmt, 0);
|
return_val = TREE_OPERAND (return_stmt, 0);
|
if (!return_val)
|
if (!return_val)
|
return;
|
return;
|
if (TREE_CODE (return_val) == MODIFY_EXPR)
|
if (TREE_CODE (return_val) == MODIFY_EXPR)
|
return_val = TREE_OPERAND (return_val, 1);
|
return_val = TREE_OPERAND (return_val, 1);
|
if (TREE_CODE (return_val) != SSA_NAME
|
if (TREE_CODE (return_val) != SSA_NAME
|
|| !SSA_NAME_DEF_STMT (return_val)
|
|| !SSA_NAME_DEF_STMT (return_val)
|
|| TREE_CODE (SSA_NAME_DEF_STMT (return_val)) != PHI_NODE)
|
|| TREE_CODE (SSA_NAME_DEF_STMT (return_val)) != PHI_NODE)
|
return;
|
return;
|
for (phi = SSA_NAME_DEF_STMT (return_val); phi; phi = PHI_CHAIN (phi))
|
for (phi = SSA_NAME_DEF_STMT (return_val); phi; phi = PHI_CHAIN (phi))
|
if (PHI_RESULT (phi) == return_val)
|
if (PHI_RESULT (phi) == return_val)
|
break;
|
break;
|
if (!phi)
|
if (!phi)
|
return;
|
return;
|
phi_num_args = PHI_NUM_ARGS (phi);
|
phi_num_args = PHI_NUM_ARGS (phi);
|
pred = return_prediction (PHI_ARG_DEF (phi, 0), &direction);
|
pred = return_prediction (PHI_ARG_DEF (phi, 0), &direction);
|
|
|
/* Avoid the degenerate case where all return values form the function
|
/* Avoid the degenerate case where all return values form the function
|
belongs to same category (ie they are all positive constants)
|
belongs to same category (ie they are all positive constants)
|
so we can hardly say something about them. */
|
so we can hardly say something about them. */
|
for (i = 1; i < phi_num_args; i++)
|
for (i = 1; i < phi_num_args; i++)
|
if (pred != return_prediction (PHI_ARG_DEF (phi, i), &direction))
|
if (pred != return_prediction (PHI_ARG_DEF (phi, i), &direction))
|
break;
|
break;
|
if (i != phi_num_args)
|
if (i != phi_num_args)
|
for (i = 0; i < phi_num_args; i++)
|
for (i = 0; i < phi_num_args; i++)
|
{
|
{
|
pred = return_prediction (PHI_ARG_DEF (phi, i), &direction);
|
pred = return_prediction (PHI_ARG_DEF (phi, i), &direction);
|
if (pred != PRED_NO_PREDICTION)
|
if (pred != PRED_NO_PREDICTION)
|
predict_paths_leading_to (PHI_ARG_EDGE (phi, i)->src, heads, pred,
|
predict_paths_leading_to (PHI_ARG_EDGE (phi, i)->src, heads, pred,
|
direction);
|
direction);
|
}
|
}
|
}
|
}
|
|
|
/* Look for basic block that contains unlikely to happen events
|
/* Look for basic block that contains unlikely to happen events
|
(such as noreturn calls) and mark all paths leading to execution
|
(such as noreturn calls) and mark all paths leading to execution
|
of this basic blocks as unlikely. */
|
of this basic blocks as unlikely. */
|
|
|
static void
|
static void
|
tree_bb_level_predictions (void)
|
tree_bb_level_predictions (void)
|
{
|
{
|
basic_block bb;
|
basic_block bb;
|
int *heads;
|
int *heads;
|
|
|
heads = XNEWVEC (int, last_basic_block);
|
heads = XNEWVEC (int, last_basic_block);
|
memset (heads, ENTRY_BLOCK, sizeof (int) * last_basic_block);
|
memset (heads, ENTRY_BLOCK, sizeof (int) * last_basic_block);
|
heads[ENTRY_BLOCK_PTR->next_bb->index] = last_basic_block;
|
heads[ENTRY_BLOCK_PTR->next_bb->index] = last_basic_block;
|
|
|
apply_return_prediction (heads);
|
apply_return_prediction (heads);
|
|
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
{
|
{
|
block_stmt_iterator bsi = bsi_last (bb);
|
block_stmt_iterator bsi = bsi_last (bb);
|
|
|
for (bsi = bsi_start (bb); !bsi_end_p (bsi); bsi_next (&bsi))
|
for (bsi = bsi_start (bb); !bsi_end_p (bsi); bsi_next (&bsi))
|
{
|
{
|
tree stmt = bsi_stmt (bsi);
|
tree stmt = bsi_stmt (bsi);
|
switch (TREE_CODE (stmt))
|
switch (TREE_CODE (stmt))
|
{
|
{
|
case MODIFY_EXPR:
|
case MODIFY_EXPR:
|
if (TREE_CODE (TREE_OPERAND (stmt, 1)) == CALL_EXPR)
|
if (TREE_CODE (TREE_OPERAND (stmt, 1)) == CALL_EXPR)
|
{
|
{
|
stmt = TREE_OPERAND (stmt, 1);
|
stmt = TREE_OPERAND (stmt, 1);
|
goto call_expr;
|
goto call_expr;
|
}
|
}
|
break;
|
break;
|
case CALL_EXPR:
|
case CALL_EXPR:
|
call_expr:;
|
call_expr:;
|
if (call_expr_flags (stmt) & ECF_NORETURN)
|
if (call_expr_flags (stmt) & ECF_NORETURN)
|
predict_paths_leading_to (bb, heads, PRED_NORETURN,
|
predict_paths_leading_to (bb, heads, PRED_NORETURN,
|
NOT_TAKEN);
|
NOT_TAKEN);
|
break;
|
break;
|
default:
|
default:
|
break;
|
break;
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
free (heads);
|
free (heads);
|
}
|
}
|
|
|
/* Predict branch probabilities and estimate profile of the tree CFG. */
|
/* Predict branch probabilities and estimate profile of the tree CFG. */
|
static unsigned int
|
static unsigned int
|
tree_estimate_probability (void)
|
tree_estimate_probability (void)
|
{
|
{
|
basic_block bb;
|
basic_block bb;
|
struct loops loops_info;
|
struct loops loops_info;
|
|
|
flow_loops_find (&loops_info);
|
flow_loops_find (&loops_info);
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
flow_loops_dump (&loops_info, dump_file, NULL, 0);
|
flow_loops_dump (&loops_info, dump_file, NULL, 0);
|
|
|
add_noreturn_fake_exit_edges ();
|
add_noreturn_fake_exit_edges ();
|
connect_infinite_loops_to_exit ();
|
connect_infinite_loops_to_exit ();
|
calculate_dominance_info (CDI_DOMINATORS);
|
calculate_dominance_info (CDI_DOMINATORS);
|
calculate_dominance_info (CDI_POST_DOMINATORS);
|
calculate_dominance_info (CDI_POST_DOMINATORS);
|
|
|
tree_bb_level_predictions ();
|
tree_bb_level_predictions ();
|
|
|
mark_irreducible_loops (&loops_info);
|
mark_irreducible_loops (&loops_info);
|
predict_loops (&loops_info, false);
|
predict_loops (&loops_info, false);
|
|
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
{
|
{
|
edge e;
|
edge e;
|
edge_iterator ei;
|
edge_iterator ei;
|
|
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
{
|
{
|
/* Predict early returns to be probable, as we've already taken
|
/* Predict early returns to be probable, as we've already taken
|
care for error returns and other cases are often used for
|
care for error returns and other cases are often used for
|
fast paths through function. */
|
fast paths through function. */
|
if (e->dest == EXIT_BLOCK_PTR
|
if (e->dest == EXIT_BLOCK_PTR
|
&& TREE_CODE (last_stmt (bb)) == RETURN_EXPR
|
&& TREE_CODE (last_stmt (bb)) == RETURN_EXPR
|
&& !single_pred_p (bb))
|
&& !single_pred_p (bb))
|
{
|
{
|
edge e1;
|
edge e1;
|
edge_iterator ei1;
|
edge_iterator ei1;
|
|
|
FOR_EACH_EDGE (e1, ei1, bb->preds)
|
FOR_EACH_EDGE (e1, ei1, bb->preds)
|
if (!predicted_by_p (e1->src, PRED_NULL_RETURN)
|
if (!predicted_by_p (e1->src, PRED_NULL_RETURN)
|
&& !predicted_by_p (e1->src, PRED_CONST_RETURN)
|
&& !predicted_by_p (e1->src, PRED_CONST_RETURN)
|
&& !predicted_by_p (e1->src, PRED_NEGATIVE_RETURN)
|
&& !predicted_by_p (e1->src, PRED_NEGATIVE_RETURN)
|
&& !last_basic_block_p (e1->src))
|
&& !last_basic_block_p (e1->src))
|
predict_edge_def (e1, PRED_TREE_EARLY_RETURN, NOT_TAKEN);
|
predict_edge_def (e1, PRED_TREE_EARLY_RETURN, NOT_TAKEN);
|
}
|
}
|
|
|
/* Look for block we are guarding (ie we dominate it,
|
/* Look for block we are guarding (ie we dominate it,
|
but it doesn't postdominate us). */
|
but it doesn't postdominate us). */
|
if (e->dest != EXIT_BLOCK_PTR && e->dest != bb
|
if (e->dest != EXIT_BLOCK_PTR && e->dest != bb
|
&& dominated_by_p (CDI_DOMINATORS, e->dest, e->src)
|
&& dominated_by_p (CDI_DOMINATORS, e->dest, e->src)
|
&& !dominated_by_p (CDI_POST_DOMINATORS, e->src, e->dest))
|
&& !dominated_by_p (CDI_POST_DOMINATORS, e->src, e->dest))
|
{
|
{
|
block_stmt_iterator bi;
|
block_stmt_iterator bi;
|
|
|
/* The call heuristic claims that a guarded function call
|
/* The call heuristic claims that a guarded function call
|
is improbable. This is because such calls are often used
|
is improbable. This is because such calls are often used
|
to signal exceptional situations such as printing error
|
to signal exceptional situations such as printing error
|
messages. */
|
messages. */
|
for (bi = bsi_start (e->dest); !bsi_end_p (bi);
|
for (bi = bsi_start (e->dest); !bsi_end_p (bi);
|
bsi_next (&bi))
|
bsi_next (&bi))
|
{
|
{
|
tree stmt = bsi_stmt (bi);
|
tree stmt = bsi_stmt (bi);
|
if ((TREE_CODE (stmt) == CALL_EXPR
|
if ((TREE_CODE (stmt) == CALL_EXPR
|
|| (TREE_CODE (stmt) == MODIFY_EXPR
|
|| (TREE_CODE (stmt) == MODIFY_EXPR
|
&& TREE_CODE (TREE_OPERAND (stmt, 1)) == CALL_EXPR))
|
&& TREE_CODE (TREE_OPERAND (stmt, 1)) == CALL_EXPR))
|
/* Constant and pure calls are hardly used to signalize
|
/* Constant and pure calls are hardly used to signalize
|
something exceptional. */
|
something exceptional. */
|
&& TREE_SIDE_EFFECTS (stmt))
|
&& TREE_SIDE_EFFECTS (stmt))
|
{
|
{
|
predict_edge_def (e, PRED_CALL, NOT_TAKEN);
|
predict_edge_def (e, PRED_CALL, NOT_TAKEN);
|
break;
|
break;
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
tree_predict_by_opcode (bb);
|
tree_predict_by_opcode (bb);
|
}
|
}
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
combine_predictions_for_bb (bb);
|
combine_predictions_for_bb (bb);
|
|
|
strip_builtin_expect ();
|
strip_builtin_expect ();
|
estimate_bb_frequencies (&loops_info);
|
estimate_bb_frequencies (&loops_info);
|
free_dominance_info (CDI_POST_DOMINATORS);
|
free_dominance_info (CDI_POST_DOMINATORS);
|
remove_fake_exit_edges ();
|
remove_fake_exit_edges ();
|
flow_loops_free (&loops_info);
|
flow_loops_free (&loops_info);
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
dump_tree_cfg (dump_file, dump_flags);
|
dump_tree_cfg (dump_file, dump_flags);
|
if (profile_status == PROFILE_ABSENT)
|
if (profile_status == PROFILE_ABSENT)
|
profile_status = PROFILE_GUESSED;
|
profile_status = PROFILE_GUESSED;
|
return 0;
|
return 0;
|
}
|
}
|
|
|
/* __builtin_expect dropped tokens into the insn stream describing expected
|
/* __builtin_expect dropped tokens into the insn stream describing expected
|
values of registers. Generate branch probabilities based off these
|
values of registers. Generate branch probabilities based off these
|
values. */
|
values. */
|
|
|
void
|
void
|
expected_value_to_br_prob (void)
|
expected_value_to_br_prob (void)
|
{
|
{
|
rtx insn, cond, ev = NULL_RTX, ev_reg = NULL_RTX;
|
rtx insn, cond, ev = NULL_RTX, ev_reg = NULL_RTX;
|
|
|
for (insn = get_insns (); insn ; insn = NEXT_INSN (insn))
|
for (insn = get_insns (); insn ; insn = NEXT_INSN (insn))
|
{
|
{
|
switch (GET_CODE (insn))
|
switch (GET_CODE (insn))
|
{
|
{
|
case NOTE:
|
case NOTE:
|
/* Look for expected value notes. */
|
/* Look for expected value notes. */
|
if (NOTE_LINE_NUMBER (insn) == NOTE_INSN_EXPECTED_VALUE)
|
if (NOTE_LINE_NUMBER (insn) == NOTE_INSN_EXPECTED_VALUE)
|
{
|
{
|
ev = NOTE_EXPECTED_VALUE (insn);
|
ev = NOTE_EXPECTED_VALUE (insn);
|
ev_reg = XEXP (ev, 0);
|
ev_reg = XEXP (ev, 0);
|
delete_insn (insn);
|
delete_insn (insn);
|
}
|
}
|
continue;
|
continue;
|
|
|
case CODE_LABEL:
|
case CODE_LABEL:
|
/* Never propagate across labels. */
|
/* Never propagate across labels. */
|
ev = NULL_RTX;
|
ev = NULL_RTX;
|
continue;
|
continue;
|
|
|
case JUMP_INSN:
|
case JUMP_INSN:
|
/* Look for simple conditional branches. If we haven't got an
|
/* Look for simple conditional branches. If we haven't got an
|
expected value yet, no point going further. */
|
expected value yet, no point going further. */
|
if (!JUMP_P (insn) || ev == NULL_RTX
|
if (!JUMP_P (insn) || ev == NULL_RTX
|
|| ! any_condjump_p (insn))
|
|| ! any_condjump_p (insn))
|
continue;
|
continue;
|
break;
|
break;
|
|
|
default:
|
default:
|
/* Look for insns that clobber the EV register. */
|
/* Look for insns that clobber the EV register. */
|
if (ev && reg_set_p (ev_reg, insn))
|
if (ev && reg_set_p (ev_reg, insn))
|
ev = NULL_RTX;
|
ev = NULL_RTX;
|
continue;
|
continue;
|
}
|
}
|
|
|
/* Collect the branch condition, hopefully relative to EV_REG. */
|
/* Collect the branch condition, hopefully relative to EV_REG. */
|
/* ??? At present we'll miss things like
|
/* ??? At present we'll miss things like
|
(expected_value (eq r70 0))
|
(expected_value (eq r70 0))
|
(set r71 -1)
|
(set r71 -1)
|
(set r80 (lt r70 r71))
|
(set r80 (lt r70 r71))
|
(set pc (if_then_else (ne r80 0) ...))
|
(set pc (if_then_else (ne r80 0) ...))
|
as canonicalize_condition will render this to us as
|
as canonicalize_condition will render this to us as
|
(lt r70, r71)
|
(lt r70, r71)
|
Could use cselib to try and reduce this further. */
|
Could use cselib to try and reduce this further. */
|
cond = XEXP (SET_SRC (pc_set (insn)), 0);
|
cond = XEXP (SET_SRC (pc_set (insn)), 0);
|
cond = canonicalize_condition (insn, cond, 0, NULL, ev_reg,
|
cond = canonicalize_condition (insn, cond, 0, NULL, ev_reg,
|
false, false);
|
false, false);
|
if (! cond || XEXP (cond, 0) != ev_reg
|
if (! cond || XEXP (cond, 0) != ev_reg
|
|| GET_CODE (XEXP (cond, 1)) != CONST_INT)
|
|| GET_CODE (XEXP (cond, 1)) != CONST_INT)
|
continue;
|
continue;
|
|
|
/* Substitute and simplify. Given that the expression we're
|
/* Substitute and simplify. Given that the expression we're
|
building involves two constants, we should wind up with either
|
building involves two constants, we should wind up with either
|
true or false. */
|
true or false. */
|
cond = gen_rtx_fmt_ee (GET_CODE (cond), VOIDmode,
|
cond = gen_rtx_fmt_ee (GET_CODE (cond), VOIDmode,
|
XEXP (ev, 1), XEXP (cond, 1));
|
XEXP (ev, 1), XEXP (cond, 1));
|
cond = simplify_rtx (cond);
|
cond = simplify_rtx (cond);
|
|
|
/* Turn the condition into a scaled branch probability. */
|
/* Turn the condition into a scaled branch probability. */
|
gcc_assert (cond == const_true_rtx || cond == const0_rtx);
|
gcc_assert (cond == const_true_rtx || cond == const0_rtx);
|
predict_insn_def (insn, PRED_BUILTIN_EXPECT,
|
predict_insn_def (insn, PRED_BUILTIN_EXPECT,
|
cond == const_true_rtx ? TAKEN : NOT_TAKEN);
|
cond == const_true_rtx ? TAKEN : NOT_TAKEN);
|
}
|
}
|
}
|
}
|
|
|
/* Check whether this is the last basic block of function. Commonly
|
/* Check whether this is the last basic block of function. Commonly
|
there is one extra common cleanup block. */
|
there is one extra common cleanup block. */
|
static bool
|
static bool
|
last_basic_block_p (basic_block bb)
|
last_basic_block_p (basic_block bb)
|
{
|
{
|
if (bb == EXIT_BLOCK_PTR)
|
if (bb == EXIT_BLOCK_PTR)
|
return false;
|
return false;
|
|
|
return (bb->next_bb == EXIT_BLOCK_PTR
|
return (bb->next_bb == EXIT_BLOCK_PTR
|
|| (bb->next_bb->next_bb == EXIT_BLOCK_PTR
|
|| (bb->next_bb->next_bb == EXIT_BLOCK_PTR
|
&& single_succ_p (bb)
|
&& single_succ_p (bb)
|
&& single_succ (bb)->next_bb == EXIT_BLOCK_PTR));
|
&& single_succ (bb)->next_bb == EXIT_BLOCK_PTR));
|
}
|
}
|
|
|
/* Sets branch probabilities according to PREDiction and
|
/* Sets branch probabilities according to PREDiction and
|
FLAGS. HEADS[bb->index] should be index of basic block in that we
|
FLAGS. HEADS[bb->index] should be index of basic block in that we
|
need to alter branch predictions (i.e. the first of our dominators
|
need to alter branch predictions (i.e. the first of our dominators
|
such that we do not post-dominate it) (but we fill this information
|
such that we do not post-dominate it) (but we fill this information
|
on demand, so -1 may be there in case this was not needed yet). */
|
on demand, so -1 may be there in case this was not needed yet). */
|
|
|
static void
|
static void
|
predict_paths_leading_to (basic_block bb, int *heads, enum br_predictor pred,
|
predict_paths_leading_to (basic_block bb, int *heads, enum br_predictor pred,
|
enum prediction taken)
|
enum prediction taken)
|
{
|
{
|
edge e;
|
edge e;
|
edge_iterator ei;
|
edge_iterator ei;
|
int y;
|
int y;
|
|
|
if (heads[bb->index] == ENTRY_BLOCK)
|
if (heads[bb->index] == ENTRY_BLOCK)
|
{
|
{
|
/* This is first time we need this field in heads array; so
|
/* This is first time we need this field in heads array; so
|
find first dominator that we do not post-dominate (we are
|
find first dominator that we do not post-dominate (we are
|
using already known members of heads array). */
|
using already known members of heads array). */
|
basic_block ai = bb;
|
basic_block ai = bb;
|
basic_block next_ai = get_immediate_dominator (CDI_DOMINATORS, bb);
|
basic_block next_ai = get_immediate_dominator (CDI_DOMINATORS, bb);
|
int head;
|
int head;
|
|
|
while (heads[next_ai->index] == ENTRY_BLOCK)
|
while (heads[next_ai->index] == ENTRY_BLOCK)
|
{
|
{
|
if (!dominated_by_p (CDI_POST_DOMINATORS, next_ai, bb))
|
if (!dominated_by_p (CDI_POST_DOMINATORS, next_ai, bb))
|
break;
|
break;
|
heads[next_ai->index] = ai->index;
|
heads[next_ai->index] = ai->index;
|
ai = next_ai;
|
ai = next_ai;
|
next_ai = get_immediate_dominator (CDI_DOMINATORS, next_ai);
|
next_ai = get_immediate_dominator (CDI_DOMINATORS, next_ai);
|
}
|
}
|
if (!dominated_by_p (CDI_POST_DOMINATORS, next_ai, bb))
|
if (!dominated_by_p (CDI_POST_DOMINATORS, next_ai, bb))
|
head = next_ai->index;
|
head = next_ai->index;
|
else
|
else
|
head = heads[next_ai->index];
|
head = heads[next_ai->index];
|
while (next_ai != bb)
|
while (next_ai != bb)
|
{
|
{
|
next_ai = ai;
|
next_ai = ai;
|
ai = BASIC_BLOCK (heads[ai->index]);
|
ai = BASIC_BLOCK (heads[ai->index]);
|
heads[next_ai->index] = head;
|
heads[next_ai->index] = head;
|
}
|
}
|
}
|
}
|
y = heads[bb->index];
|
y = heads[bb->index];
|
|
|
/* Now find the edge that leads to our branch and aply the prediction. */
|
/* Now find the edge that leads to our branch and aply the prediction. */
|
|
|
if (y == last_basic_block)
|
if (y == last_basic_block)
|
return;
|
return;
|
FOR_EACH_EDGE (e, ei, BASIC_BLOCK (y)->succs)
|
FOR_EACH_EDGE (e, ei, BASIC_BLOCK (y)->succs)
|
if (e->dest->index >= NUM_FIXED_BLOCKS
|
if (e->dest->index >= NUM_FIXED_BLOCKS
|
&& dominated_by_p (CDI_POST_DOMINATORS, e->dest, bb))
|
&& dominated_by_p (CDI_POST_DOMINATORS, e->dest, bb))
|
predict_edge_def (e, pred, taken);
|
predict_edge_def (e, pred, taken);
|
}
|
}
|
|
|
/* This is used to carry information about basic blocks. It is
|
/* This is used to carry information about basic blocks. It is
|
attached to the AUX field of the standard CFG block. */
|
attached to the AUX field of the standard CFG block. */
|
|
|
typedef struct block_info_def
|
typedef struct block_info_def
|
{
|
{
|
/* Estimated frequency of execution of basic_block. */
|
/* Estimated frequency of execution of basic_block. */
|
sreal frequency;
|
sreal frequency;
|
|
|
/* To keep queue of basic blocks to process. */
|
/* To keep queue of basic blocks to process. */
|
basic_block next;
|
basic_block next;
|
|
|
/* Number of predecessors we need to visit first. */
|
/* Number of predecessors we need to visit first. */
|
int npredecessors;
|
int npredecessors;
|
} *block_info;
|
} *block_info;
|
|
|
/* Similar information for edges. */
|
/* Similar information for edges. */
|
typedef struct edge_info_def
|
typedef struct edge_info_def
|
{
|
{
|
/* In case edge is a loopback edge, the probability edge will be reached
|
/* In case edge is a loopback edge, the probability edge will be reached
|
in case header is. Estimated number of iterations of the loop can be
|
in case header is. Estimated number of iterations of the loop can be
|
then computed as 1 / (1 - back_edge_prob). */
|
then computed as 1 / (1 - back_edge_prob). */
|
sreal back_edge_prob;
|
sreal back_edge_prob;
|
/* True if the edge is a loopback edge in the natural loop. */
|
/* True if the edge is a loopback edge in the natural loop. */
|
unsigned int back_edge:1;
|
unsigned int back_edge:1;
|
} *edge_info;
|
} *edge_info;
|
|
|
#define BLOCK_INFO(B) ((block_info) (B)->aux)
|
#define BLOCK_INFO(B) ((block_info) (B)->aux)
|
#define EDGE_INFO(E) ((edge_info) (E)->aux)
|
#define EDGE_INFO(E) ((edge_info) (E)->aux)
|
|
|
/* Helper function for estimate_bb_frequencies.
|
/* Helper function for estimate_bb_frequencies.
|
Propagate the frequencies for LOOP. */
|
Propagate the frequencies for LOOP. */
|
|
|
static void
|
static void
|
propagate_freq (struct loop *loop, bitmap tovisit)
|
propagate_freq (struct loop *loop, bitmap tovisit)
|
{
|
{
|
basic_block head = loop->header;
|
basic_block head = loop->header;
|
basic_block bb;
|
basic_block bb;
|
basic_block last;
|
basic_block last;
|
unsigned i;
|
unsigned i;
|
edge e;
|
edge e;
|
basic_block nextbb;
|
basic_block nextbb;
|
bitmap_iterator bi;
|
bitmap_iterator bi;
|
|
|
/* For each basic block we need to visit count number of his predecessors
|
/* For each basic block we need to visit count number of his predecessors
|
we need to visit first. */
|
we need to visit first. */
|
EXECUTE_IF_SET_IN_BITMAP (tovisit, 0, i, bi)
|
EXECUTE_IF_SET_IN_BITMAP (tovisit, 0, i, bi)
|
{
|
{
|
edge_iterator ei;
|
edge_iterator ei;
|
int count = 0;
|
int count = 0;
|
|
|
/* The outermost "loop" includes the exit block, which we can not
|
/* The outermost "loop" includes the exit block, which we can not
|
look up via BASIC_BLOCK. Detect this and use EXIT_BLOCK_PTR
|
look up via BASIC_BLOCK. Detect this and use EXIT_BLOCK_PTR
|
directly. Do the same for the entry block. */
|
directly. Do the same for the entry block. */
|
bb = BASIC_BLOCK (i);
|
bb = BASIC_BLOCK (i);
|
|
|
FOR_EACH_EDGE (e, ei, bb->preds)
|
FOR_EACH_EDGE (e, ei, bb->preds)
|
{
|
{
|
bool visit = bitmap_bit_p (tovisit, e->src->index);
|
bool visit = bitmap_bit_p (tovisit, e->src->index);
|
|
|
if (visit && !(e->flags & EDGE_DFS_BACK))
|
if (visit && !(e->flags & EDGE_DFS_BACK))
|
count++;
|
count++;
|
else if (visit && dump_file && !EDGE_INFO (e)->back_edge)
|
else if (visit && dump_file && !EDGE_INFO (e)->back_edge)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
"Irreducible region hit, ignoring edge to %i->%i\n",
|
"Irreducible region hit, ignoring edge to %i->%i\n",
|
e->src->index, bb->index);
|
e->src->index, bb->index);
|
}
|
}
|
BLOCK_INFO (bb)->npredecessors = count;
|
BLOCK_INFO (bb)->npredecessors = count;
|
}
|
}
|
|
|
memcpy (&BLOCK_INFO (head)->frequency, &real_one, sizeof (real_one));
|
memcpy (&BLOCK_INFO (head)->frequency, &real_one, sizeof (real_one));
|
last = head;
|
last = head;
|
for (bb = head; bb; bb = nextbb)
|
for (bb = head; bb; bb = nextbb)
|
{
|
{
|
edge_iterator ei;
|
edge_iterator ei;
|
sreal cyclic_probability, frequency;
|
sreal cyclic_probability, frequency;
|
|
|
memcpy (&cyclic_probability, &real_zero, sizeof (real_zero));
|
memcpy (&cyclic_probability, &real_zero, sizeof (real_zero));
|
memcpy (&frequency, &real_zero, sizeof (real_zero));
|
memcpy (&frequency, &real_zero, sizeof (real_zero));
|
|
|
nextbb = BLOCK_INFO (bb)->next;
|
nextbb = BLOCK_INFO (bb)->next;
|
BLOCK_INFO (bb)->next = NULL;
|
BLOCK_INFO (bb)->next = NULL;
|
|
|
/* Compute frequency of basic block. */
|
/* Compute frequency of basic block. */
|
if (bb != head)
|
if (bb != head)
|
{
|
{
|
#ifdef ENABLE_CHECKING
|
#ifdef ENABLE_CHECKING
|
FOR_EACH_EDGE (e, ei, bb->preds)
|
FOR_EACH_EDGE (e, ei, bb->preds)
|
gcc_assert (!bitmap_bit_p (tovisit, e->src->index)
|
gcc_assert (!bitmap_bit_p (tovisit, e->src->index)
|
|| (e->flags & EDGE_DFS_BACK));
|
|| (e->flags & EDGE_DFS_BACK));
|
#endif
|
#endif
|
|
|
FOR_EACH_EDGE (e, ei, bb->preds)
|
FOR_EACH_EDGE (e, ei, bb->preds)
|
if (EDGE_INFO (e)->back_edge)
|
if (EDGE_INFO (e)->back_edge)
|
{
|
{
|
sreal_add (&cyclic_probability, &cyclic_probability,
|
sreal_add (&cyclic_probability, &cyclic_probability,
|
&EDGE_INFO (e)->back_edge_prob);
|
&EDGE_INFO (e)->back_edge_prob);
|
}
|
}
|
else if (!(e->flags & EDGE_DFS_BACK))
|
else if (!(e->flags & EDGE_DFS_BACK))
|
{
|
{
|
sreal tmp;
|
sreal tmp;
|
|
|
/* frequency += (e->probability
|
/* frequency += (e->probability
|
* BLOCK_INFO (e->src)->frequency /
|
* BLOCK_INFO (e->src)->frequency /
|
REG_BR_PROB_BASE); */
|
REG_BR_PROB_BASE); */
|
|
|
sreal_init (&tmp, e->probability, 0);
|
sreal_init (&tmp, e->probability, 0);
|
sreal_mul (&tmp, &tmp, &BLOCK_INFO (e->src)->frequency);
|
sreal_mul (&tmp, &tmp, &BLOCK_INFO (e->src)->frequency);
|
sreal_mul (&tmp, &tmp, &real_inv_br_prob_base);
|
sreal_mul (&tmp, &tmp, &real_inv_br_prob_base);
|
sreal_add (&frequency, &frequency, &tmp);
|
sreal_add (&frequency, &frequency, &tmp);
|
}
|
}
|
|
|
if (sreal_compare (&cyclic_probability, &real_zero) == 0)
|
if (sreal_compare (&cyclic_probability, &real_zero) == 0)
|
{
|
{
|
memcpy (&BLOCK_INFO (bb)->frequency, &frequency,
|
memcpy (&BLOCK_INFO (bb)->frequency, &frequency,
|
sizeof (frequency));
|
sizeof (frequency));
|
}
|
}
|
else
|
else
|
{
|
{
|
if (sreal_compare (&cyclic_probability, &real_almost_one) > 0)
|
if (sreal_compare (&cyclic_probability, &real_almost_one) > 0)
|
{
|
{
|
memcpy (&cyclic_probability, &real_almost_one,
|
memcpy (&cyclic_probability, &real_almost_one,
|
sizeof (real_almost_one));
|
sizeof (real_almost_one));
|
}
|
}
|
|
|
/* BLOCK_INFO (bb)->frequency = frequency
|
/* BLOCK_INFO (bb)->frequency = frequency
|
/ (1 - cyclic_probability) */
|
/ (1 - cyclic_probability) */
|
|
|
sreal_sub (&cyclic_probability, &real_one, &cyclic_probability);
|
sreal_sub (&cyclic_probability, &real_one, &cyclic_probability);
|
sreal_div (&BLOCK_INFO (bb)->frequency,
|
sreal_div (&BLOCK_INFO (bb)->frequency,
|
&frequency, &cyclic_probability);
|
&frequency, &cyclic_probability);
|
}
|
}
|
}
|
}
|
|
|
bitmap_clear_bit (tovisit, bb->index);
|
bitmap_clear_bit (tovisit, bb->index);
|
|
|
e = find_edge (bb, head);
|
e = find_edge (bb, head);
|
if (e)
|
if (e)
|
{
|
{
|
sreal tmp;
|
sreal tmp;
|
|
|
/* EDGE_INFO (e)->back_edge_prob
|
/* EDGE_INFO (e)->back_edge_prob
|
= ((e->probability * BLOCK_INFO (bb)->frequency)
|
= ((e->probability * BLOCK_INFO (bb)->frequency)
|
/ REG_BR_PROB_BASE); */
|
/ REG_BR_PROB_BASE); */
|
|
|
sreal_init (&tmp, e->probability, 0);
|
sreal_init (&tmp, e->probability, 0);
|
sreal_mul (&tmp, &tmp, &BLOCK_INFO (bb)->frequency);
|
sreal_mul (&tmp, &tmp, &BLOCK_INFO (bb)->frequency);
|
sreal_mul (&EDGE_INFO (e)->back_edge_prob,
|
sreal_mul (&EDGE_INFO (e)->back_edge_prob,
|
&tmp, &real_inv_br_prob_base);
|
&tmp, &real_inv_br_prob_base);
|
}
|
}
|
|
|
/* Propagate to successor blocks. */
|
/* Propagate to successor blocks. */
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
if (!(e->flags & EDGE_DFS_BACK)
|
if (!(e->flags & EDGE_DFS_BACK)
|
&& BLOCK_INFO (e->dest)->npredecessors)
|
&& BLOCK_INFO (e->dest)->npredecessors)
|
{
|
{
|
BLOCK_INFO (e->dest)->npredecessors--;
|
BLOCK_INFO (e->dest)->npredecessors--;
|
if (!BLOCK_INFO (e->dest)->npredecessors)
|
if (!BLOCK_INFO (e->dest)->npredecessors)
|
{
|
{
|
if (!nextbb)
|
if (!nextbb)
|
nextbb = e->dest;
|
nextbb = e->dest;
|
else
|
else
|
BLOCK_INFO (last)->next = e->dest;
|
BLOCK_INFO (last)->next = e->dest;
|
|
|
last = e->dest;
|
last = e->dest;
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
/* Estimate probabilities of loopback edges in loops at same nest level. */
|
/* Estimate probabilities of loopback edges in loops at same nest level. */
|
|
|
static void
|
static void
|
estimate_loops_at_level (struct loop *first_loop, bitmap tovisit)
|
estimate_loops_at_level (struct loop *first_loop, bitmap tovisit)
|
{
|
{
|
struct loop *loop;
|
struct loop *loop;
|
|
|
for (loop = first_loop; loop; loop = loop->next)
|
for (loop = first_loop; loop; loop = loop->next)
|
{
|
{
|
edge e;
|
edge e;
|
basic_block *bbs;
|
basic_block *bbs;
|
unsigned i;
|
unsigned i;
|
|
|
estimate_loops_at_level (loop->inner, tovisit);
|
estimate_loops_at_level (loop->inner, tovisit);
|
|
|
/* Do not do this for dummy function loop. */
|
/* Do not do this for dummy function loop. */
|
if (EDGE_COUNT (loop->latch->succs) > 0)
|
if (EDGE_COUNT (loop->latch->succs) > 0)
|
{
|
{
|
/* Find current loop back edge and mark it. */
|
/* Find current loop back edge and mark it. */
|
e = loop_latch_edge (loop);
|
e = loop_latch_edge (loop);
|
EDGE_INFO (e)->back_edge = 1;
|
EDGE_INFO (e)->back_edge = 1;
|
}
|
}
|
|
|
bbs = get_loop_body (loop);
|
bbs = get_loop_body (loop);
|
for (i = 0; i < loop->num_nodes; i++)
|
for (i = 0; i < loop->num_nodes; i++)
|
bitmap_set_bit (tovisit, bbs[i]->index);
|
bitmap_set_bit (tovisit, bbs[i]->index);
|
free (bbs);
|
free (bbs);
|
propagate_freq (loop, tovisit);
|
propagate_freq (loop, tovisit);
|
}
|
}
|
}
|
}
|
|
|
/* Convert counts measured by profile driven feedback to frequencies.
|
/* Convert counts measured by profile driven feedback to frequencies.
|
Return nonzero iff there was any nonzero execution count. */
|
Return nonzero iff there was any nonzero execution count. */
|
|
|
int
|
int
|
counts_to_freqs (void)
|
counts_to_freqs (void)
|
{
|
{
|
gcov_type count_max, true_count_max = 0;
|
gcov_type count_max, true_count_max = 0;
|
basic_block bb;
|
basic_block bb;
|
|
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
true_count_max = MAX (bb->count, true_count_max);
|
true_count_max = MAX (bb->count, true_count_max);
|
|
|
count_max = MAX (true_count_max, 1);
|
count_max = MAX (true_count_max, 1);
|
FOR_BB_BETWEEN (bb, ENTRY_BLOCK_PTR, NULL, next_bb)
|
FOR_BB_BETWEEN (bb, ENTRY_BLOCK_PTR, NULL, next_bb)
|
bb->frequency = (bb->count * BB_FREQ_MAX + count_max / 2) / count_max;
|
bb->frequency = (bb->count * BB_FREQ_MAX + count_max / 2) / count_max;
|
return true_count_max;
|
return true_count_max;
|
}
|
}
|
|
|
/* Return true if function is likely to be expensive, so there is no point to
|
/* Return true if function is likely to be expensive, so there is no point to
|
optimize performance of prologue, epilogue or do inlining at the expense
|
optimize performance of prologue, epilogue or do inlining at the expense
|
of code size growth. THRESHOLD is the limit of number of instructions
|
of code size growth. THRESHOLD is the limit of number of instructions
|
function can execute at average to be still considered not expensive. */
|
function can execute at average to be still considered not expensive. */
|
|
|
bool
|
bool
|
expensive_function_p (int threshold)
|
expensive_function_p (int threshold)
|
{
|
{
|
unsigned int sum = 0;
|
unsigned int sum = 0;
|
basic_block bb;
|
basic_block bb;
|
unsigned int limit;
|
unsigned int limit;
|
|
|
/* We can not compute accurately for large thresholds due to scaled
|
/* We can not compute accurately for large thresholds due to scaled
|
frequencies. */
|
frequencies. */
|
gcc_assert (threshold <= BB_FREQ_MAX);
|
gcc_assert (threshold <= BB_FREQ_MAX);
|
|
|
/* Frequencies are out of range. This either means that function contains
|
/* Frequencies are out of range. This either means that function contains
|
internal loop executing more than BB_FREQ_MAX times or profile feedback
|
internal loop executing more than BB_FREQ_MAX times or profile feedback
|
is available and function has not been executed at all. */
|
is available and function has not been executed at all. */
|
if (ENTRY_BLOCK_PTR->frequency == 0)
|
if (ENTRY_BLOCK_PTR->frequency == 0)
|
return true;
|
return true;
|
|
|
/* Maximally BB_FREQ_MAX^2 so overflow won't happen. */
|
/* Maximally BB_FREQ_MAX^2 so overflow won't happen. */
|
limit = ENTRY_BLOCK_PTR->frequency * threshold;
|
limit = ENTRY_BLOCK_PTR->frequency * threshold;
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
{
|
{
|
rtx insn;
|
rtx insn;
|
|
|
for (insn = BB_HEAD (bb); insn != NEXT_INSN (BB_END (bb));
|
for (insn = BB_HEAD (bb); insn != NEXT_INSN (BB_END (bb));
|
insn = NEXT_INSN (insn))
|
insn = NEXT_INSN (insn))
|
if (active_insn_p (insn))
|
if (active_insn_p (insn))
|
{
|
{
|
sum += bb->frequency;
|
sum += bb->frequency;
|
if (sum > limit)
|
if (sum > limit)
|
return true;
|
return true;
|
}
|
}
|
}
|
}
|
|
|
return false;
|
return false;
|
}
|
}
|
|
|
/* Estimate basic blocks frequency by given branch probabilities. */
|
/* Estimate basic blocks frequency by given branch probabilities. */
|
|
|
static void
|
static void
|
estimate_bb_frequencies (struct loops *loops)
|
estimate_bb_frequencies (struct loops *loops)
|
{
|
{
|
basic_block bb;
|
basic_block bb;
|
sreal freq_max;
|
sreal freq_max;
|
|
|
if (!flag_branch_probabilities || !counts_to_freqs ())
|
if (!flag_branch_probabilities || !counts_to_freqs ())
|
{
|
{
|
static int real_values_initialized = 0;
|
static int real_values_initialized = 0;
|
bitmap tovisit;
|
bitmap tovisit;
|
|
|
if (!real_values_initialized)
|
if (!real_values_initialized)
|
{
|
{
|
real_values_initialized = 1;
|
real_values_initialized = 1;
|
sreal_init (&real_zero, 0, 0);
|
sreal_init (&real_zero, 0, 0);
|
sreal_init (&real_one, 1, 0);
|
sreal_init (&real_one, 1, 0);
|
sreal_init (&real_br_prob_base, REG_BR_PROB_BASE, 0);
|
sreal_init (&real_br_prob_base, REG_BR_PROB_BASE, 0);
|
sreal_init (&real_bb_freq_max, BB_FREQ_MAX, 0);
|
sreal_init (&real_bb_freq_max, BB_FREQ_MAX, 0);
|
sreal_init (&real_one_half, 1, -1);
|
sreal_init (&real_one_half, 1, -1);
|
sreal_div (&real_inv_br_prob_base, &real_one, &real_br_prob_base);
|
sreal_div (&real_inv_br_prob_base, &real_one, &real_br_prob_base);
|
sreal_sub (&real_almost_one, &real_one, &real_inv_br_prob_base);
|
sreal_sub (&real_almost_one, &real_one, &real_inv_br_prob_base);
|
}
|
}
|
|
|
mark_dfs_back_edges ();
|
mark_dfs_back_edges ();
|
|
|
single_succ_edge (ENTRY_BLOCK_PTR)->probability = REG_BR_PROB_BASE;
|
single_succ_edge (ENTRY_BLOCK_PTR)->probability = REG_BR_PROB_BASE;
|
|
|
/* Set up block info for each basic block. */
|
/* Set up block info for each basic block. */
|
tovisit = BITMAP_ALLOC (NULL);
|
tovisit = BITMAP_ALLOC (NULL);
|
alloc_aux_for_blocks (sizeof (struct block_info_def));
|
alloc_aux_for_blocks (sizeof (struct block_info_def));
|
alloc_aux_for_edges (sizeof (struct edge_info_def));
|
alloc_aux_for_edges (sizeof (struct edge_info_def));
|
FOR_BB_BETWEEN (bb, ENTRY_BLOCK_PTR, NULL, next_bb)
|
FOR_BB_BETWEEN (bb, ENTRY_BLOCK_PTR, NULL, next_bb)
|
{
|
{
|
edge e;
|
edge e;
|
edge_iterator ei;
|
edge_iterator ei;
|
|
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
{
|
{
|
sreal_init (&EDGE_INFO (e)->back_edge_prob, e->probability, 0);
|
sreal_init (&EDGE_INFO (e)->back_edge_prob, e->probability, 0);
|
sreal_mul (&EDGE_INFO (e)->back_edge_prob,
|
sreal_mul (&EDGE_INFO (e)->back_edge_prob,
|
&EDGE_INFO (e)->back_edge_prob,
|
&EDGE_INFO (e)->back_edge_prob,
|
&real_inv_br_prob_base);
|
&real_inv_br_prob_base);
|
}
|
}
|
}
|
}
|
|
|
/* First compute probabilities locally for each loop from innermost
|
/* First compute probabilities locally for each loop from innermost
|
to outermost to examine probabilities for back edges. */
|
to outermost to examine probabilities for back edges. */
|
estimate_loops_at_level (loops->tree_root, tovisit);
|
estimate_loops_at_level (loops->tree_root, tovisit);
|
|
|
memcpy (&freq_max, &real_zero, sizeof (real_zero));
|
memcpy (&freq_max, &real_zero, sizeof (real_zero));
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
if (sreal_compare (&freq_max, &BLOCK_INFO (bb)->frequency) < 0)
|
if (sreal_compare (&freq_max, &BLOCK_INFO (bb)->frequency) < 0)
|
memcpy (&freq_max, &BLOCK_INFO (bb)->frequency, sizeof (freq_max));
|
memcpy (&freq_max, &BLOCK_INFO (bb)->frequency, sizeof (freq_max));
|
|
|
sreal_div (&freq_max, &real_bb_freq_max, &freq_max);
|
sreal_div (&freq_max, &real_bb_freq_max, &freq_max);
|
FOR_BB_BETWEEN (bb, ENTRY_BLOCK_PTR, NULL, next_bb)
|
FOR_BB_BETWEEN (bb, ENTRY_BLOCK_PTR, NULL, next_bb)
|
{
|
{
|
sreal tmp;
|
sreal tmp;
|
|
|
sreal_mul (&tmp, &BLOCK_INFO (bb)->frequency, &freq_max);
|
sreal_mul (&tmp, &BLOCK_INFO (bb)->frequency, &freq_max);
|
sreal_add (&tmp, &tmp, &real_one_half);
|
sreal_add (&tmp, &tmp, &real_one_half);
|
bb->frequency = sreal_to_int (&tmp);
|
bb->frequency = sreal_to_int (&tmp);
|
}
|
}
|
|
|
free_aux_for_blocks ();
|
free_aux_for_blocks ();
|
free_aux_for_edges ();
|
free_aux_for_edges ();
|
BITMAP_FREE (tovisit);
|
BITMAP_FREE (tovisit);
|
}
|
}
|
compute_function_frequency ();
|
compute_function_frequency ();
|
if (flag_reorder_functions)
|
if (flag_reorder_functions)
|
choose_function_section ();
|
choose_function_section ();
|
}
|
}
|
|
|
/* Decide whether function is hot, cold or unlikely executed. */
|
/* Decide whether function is hot, cold or unlikely executed. */
|
static void
|
static void
|
compute_function_frequency (void)
|
compute_function_frequency (void)
|
{
|
{
|
basic_block bb;
|
basic_block bb;
|
|
|
if (!profile_info || !flag_branch_probabilities)
|
if (!profile_info || !flag_branch_probabilities)
|
return;
|
return;
|
cfun->function_frequency = FUNCTION_FREQUENCY_UNLIKELY_EXECUTED;
|
cfun->function_frequency = FUNCTION_FREQUENCY_UNLIKELY_EXECUTED;
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
{
|
{
|
if (maybe_hot_bb_p (bb))
|
if (maybe_hot_bb_p (bb))
|
{
|
{
|
cfun->function_frequency = FUNCTION_FREQUENCY_HOT;
|
cfun->function_frequency = FUNCTION_FREQUENCY_HOT;
|
return;
|
return;
|
}
|
}
|
if (!probably_never_executed_bb_p (bb))
|
if (!probably_never_executed_bb_p (bb))
|
cfun->function_frequency = FUNCTION_FREQUENCY_NORMAL;
|
cfun->function_frequency = FUNCTION_FREQUENCY_NORMAL;
|
}
|
}
|
}
|
}
|
|
|
/* Choose appropriate section for the function. */
|
/* Choose appropriate section for the function. */
|
static void
|
static void
|
choose_function_section (void)
|
choose_function_section (void)
|
{
|
{
|
if (DECL_SECTION_NAME (current_function_decl)
|
if (DECL_SECTION_NAME (current_function_decl)
|
|| !targetm.have_named_sections
|
|| !targetm.have_named_sections
|
/* Theoretically we can split the gnu.linkonce text section too,
|
/* Theoretically we can split the gnu.linkonce text section too,
|
but this requires more work as the frequency needs to match
|
but this requires more work as the frequency needs to match
|
for all generated objects so we need to merge the frequency
|
for all generated objects so we need to merge the frequency
|
of all instances. For now just never set frequency for these. */
|
of all instances. For now just never set frequency for these. */
|
|| DECL_ONE_ONLY (current_function_decl))
|
|| DECL_ONE_ONLY (current_function_decl))
|
return;
|
return;
|
|
|
/* If we are doing the partitioning optimization, let the optimization
|
/* If we are doing the partitioning optimization, let the optimization
|
choose the correct section into which to put things. */
|
choose the correct section into which to put things. */
|
|
|
if (flag_reorder_blocks_and_partition)
|
if (flag_reorder_blocks_and_partition)
|
return;
|
return;
|
|
|
if (cfun->function_frequency == FUNCTION_FREQUENCY_HOT)
|
if (cfun->function_frequency == FUNCTION_FREQUENCY_HOT)
|
DECL_SECTION_NAME (current_function_decl) =
|
DECL_SECTION_NAME (current_function_decl) =
|
build_string (strlen (HOT_TEXT_SECTION_NAME), HOT_TEXT_SECTION_NAME);
|
build_string (strlen (HOT_TEXT_SECTION_NAME), HOT_TEXT_SECTION_NAME);
|
if (cfun->function_frequency == FUNCTION_FREQUENCY_UNLIKELY_EXECUTED)
|
if (cfun->function_frequency == FUNCTION_FREQUENCY_UNLIKELY_EXECUTED)
|
DECL_SECTION_NAME (current_function_decl) =
|
DECL_SECTION_NAME (current_function_decl) =
|
build_string (strlen (UNLIKELY_EXECUTED_TEXT_SECTION_NAME),
|
build_string (strlen (UNLIKELY_EXECUTED_TEXT_SECTION_NAME),
|
UNLIKELY_EXECUTED_TEXT_SECTION_NAME);
|
UNLIKELY_EXECUTED_TEXT_SECTION_NAME);
|
}
|
}
|
|
|
static bool
|
static bool
|
gate_estimate_probability (void)
|
gate_estimate_probability (void)
|
{
|
{
|
return flag_guess_branch_prob;
|
return flag_guess_branch_prob;
|
}
|
}
|
|
|
struct tree_opt_pass pass_profile =
|
struct tree_opt_pass pass_profile =
|
{
|
{
|
"profile", /* name */
|
"profile", /* name */
|
gate_estimate_probability, /* gate */
|
gate_estimate_probability, /* gate */
|
tree_estimate_probability, /* execute */
|
tree_estimate_probability, /* execute */
|
NULL, /* sub */
|
NULL, /* sub */
|
NULL, /* next */
|
NULL, /* next */
|
0, /* static_pass_number */
|
0, /* static_pass_number */
|
TV_BRANCH_PROB, /* tv_id */
|
TV_BRANCH_PROB, /* tv_id */
|
PROP_cfg, /* properties_required */
|
PROP_cfg, /* properties_required */
|
0, /* properties_provided */
|
0, /* properties_provided */
|
0, /* properties_destroyed */
|
0, /* properties_destroyed */
|
0, /* todo_flags_start */
|
0, /* todo_flags_start */
|
TODO_ggc_collect | TODO_verify_ssa, /* todo_flags_finish */
|
TODO_ggc_collect | TODO_verify_ssa, /* todo_flags_finish */
|
0 /* letter */
|
0 /* letter */
|
};
|
};
|
|
|
