Skip to content

R
riscv-gcc-1
Commit 87022a6b by Jan Hubicka Committed by Jan Hubicka
Options

basic-block.h (guess_outgoing_edge_probabilities): Declare. 


	* basic-block.h (guess_outgoing_edge_probabilities): Declare.
	* cfgbuild.c (compute_outgoing_frequencies): When probability is missing,
	guess it.
	(find_many_sub_basic_blocks): Do update profile only when it is present.
	* predict.c (set_even_probabilities): Break out from ...
	(combine_predictions_for_insn): ... here; deal with !can_predict_insn_p insns.
	(combine_predictions_for_bb): Use set_even_probabilities.
	(bb_estimate_probability_locally): Break out from ....
	(estimate_probability): ... here.
	(guess_outgoing_edge_probabilities): New entry point.

From-SVN: r87234
parent 3b5edfee
Showing with 194 additions and 127 deletions
gcc/ChangeLog
2004-09-09 Jan Hubicka <jh@suse.cz>
* basic-block.h (guess_outgoing_edge_probabilities): Declare.
* cfgbuild.c (compute_outgoing_frequencies): When probability is missing,
guess it.
(find_many_sub_basic_blocks): Do update profile only when it is present.
* predict.c (set_even_probabilities): Break out from ...
(combine_predictions_for_insn): ... here; deal with !can_predict_insn_p insns.
(combine_predictions_for_bb): Use set_even_probabilities.
(bb_estimate_probability_locally): Break out from ....
(estimate_probability): ... here.
(guess_outgoing_edge_probabilities): New entry point.
2004-09-09 Nathan Sidwell <nathan@codesourcery.com>
* gcc.c (add_sysrooted_prefix, execute, do_self_spec, do_spec_1,
......
gcc/basic-block.h
......@@ -622,6 +622,7 @@ extern bool rtl_predicted_by_p (basic_block, enum br_predictor);
extern void tree_predict_edge (edge, enum br_predictor, int);
extern void rtl_predict_edge (edge, enum br_predictor, int);
extern void predict_edge_def (edge, enum br_predictor, enum prediction);
extern void guess_outgoing_edge_probabilities (basic_block);
/* In flow.c */
extern void init_flow (void);
......
gcc/cfgbuild.c
......@@ -646,17 +646,18 @@ compute_outgoing_frequencies (basic_block b)
rtx note = find_reg_note (BB_END (b), REG_BR_PROB, NULL);
int probability;
if (!note)
return;
probability = INTVAL (XEXP (note, 0));
e = BRANCH_EDGE (b);
e->probability = probability;
e->count = ((b->count * probability + REG_BR_PROB_BASE / 2)
/ REG_BR_PROB_BASE);
f = FALLTHRU_EDGE (b);
f->probability = REG_BR_PROB_BASE - probability;
f->count = b->count - e->count;
if (note)
{
probability = INTVAL (XEXP (note, 0));
e = BRANCH_EDGE (b);
e->probability = probability;
e->count = ((b->count * probability + REG_BR_PROB_BASE / 2)
/ REG_BR_PROB_BASE);
f = FALLTHRU_EDGE (b);
f->probability = REG_BR_PROB_BASE - probability;
f->count = b->count - e->count;
return;
}
}
if (b->succ && !b->succ->succ_next)
......@@ -664,7 +665,13 @@ compute_outgoing_frequencies (basic_block b)
e = b->succ;
e->probability = REG_BR_PROB_BASE;
e->count = b->count;
return;
}
guess_outgoing_edge_probabilities (b);
if (b->count)
for (e = b->succ; e; e = e->succ_next)
e->count = ((b->count * e->probability + REG_BR_PROB_BASE / 2)
/ REG_BR_PROB_BASE);
}
/* Assume that someone emitted code with control flow instructions to the
......@@ -698,25 +705,26 @@ find_many_sub_basic_blocks (sbitmap blocks)
/* Update branch probabilities. Expect only (un)conditional jumps
to be created with only the forward edges. */
FOR_BB_BETWEEN (bb, min, max->next_bb, next_bb)
{
edge e;
if (STATE (bb) == BLOCK_ORIGINAL)
continue;
if (STATE (bb) == BLOCK_NEW)
{
bb->count = 0;
bb->frequency = 0;
for (e = bb->pred; e; e = e->pred_next)
{
bb->count += e->count;
bb->frequency += EDGE_FREQUENCY (e);
}
}
if (profile_status != PROFILE_ABSENT)
FOR_BB_BETWEEN (bb, min, max->next_bb, next_bb)
{
edge e;
if (STATE (bb) == BLOCK_ORIGINAL)
continue;
if (STATE (bb) == BLOCK_NEW)
{
bb->count = 0;
bb->frequency = 0;
for (e = bb->pred; e; e = e->pred_next)
{
bb->count += e->count;
bb->frequency += EDGE_FREQUENCY (e);
}
}
compute_outgoing_frequencies (bb);
}
compute_outgoing_frequencies (bb);
}
FOR_EACH_BB (bb)
SET_STATE (bb, 0);
......
gcc/predict.c
......@@ -313,14 +313,32 @@ dump_prediction (FILE *file, enum br_predictor predictor, int probability,
fprintf (file, "\n");
}
/* We can not predict the probabilities of ougtoing edges of bb. Set them
evenly and hope for the best. */
static void
set_even_probabilities (basic_block bb)
{
int nedges = 0;
edge e;
for (e = bb->succ; e; e = e->succ_next)
if (!(e->flags & (EDGE_EH | EDGE_FAKE)))
nedges ++;
for (e = bb->succ; e; e = e->succ_next)
if (!(e->flags & (EDGE_EH | EDGE_FAKE)))
e->probability = (REG_BR_PROB_BASE + nedges / 2) / nedges;
else
e->probability = 0;
}
/* 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. */
static void
combine_predictions_for_insn (rtx insn, basic_block bb)
{
rtx prob_note = find_reg_note (insn, REG_BR_PROB, 0);
rtx *pnote = &REG_NOTES (insn);
rtx prob_note;
rtx *pnote;
rtx note;
int best_probability = PROB_EVEN;
int best_predictor = END_PREDICTORS;
......@@ -329,6 +347,14 @@ combine_predictions_for_insn (rtx insn, basic_block bb)
bool first_match = false;
bool found = false;
if (!can_predict_insn_p (insn))
{
set_even_probabilities (bb);
return;
}
prob_note = find_reg_note (insn, REG_BR_PROB, 0);
pnote = &REG_NOTES (insn);
if (dump_file)
fprintf (dump_file, "Predictions for insn %i bb %i\n", INSN_UID (insn),
bb->index);
......@@ -446,11 +472,8 @@ combine_predictions_for_bb (FILE *file, basic_block bb)
this later. */
if (nedges != 2)
{
for (e = bb->succ; e; e = e->succ_next)
if (!(e->flags & (EDGE_EH | EDGE_FAKE)))
e->probability = (REG_BR_PROB_BASE + nedges / 2) / nedges;
else
e->probability = 0;
if (!bb->count)
set_even_probabilities (bb);
bb_ann (bb)->predictions = NULL;
if (file)
fprintf (file, "%i edges in bb %i predicted to even probabilities\n",
......@@ -521,8 +544,11 @@ combine_predictions_for_bb (FILE *file, basic_block bb)
}
bb_ann (bb)->predictions = NULL;
first->probability = combined_probability;
second->probability = REG_BR_PROB_BASE - combined_probability;
if (!bb->count)
{
first->probability = combined_probability;
second->probability = REG_BR_PROB_BASE - combined_probability;
}
}
/* Predict edge probabilities by exploiting loop structure.
......@@ -615,6 +641,105 @@ predict_loops (struct loops *loops_info, bool simpleloops)
}
}
/* Attempt to predict probabilities of BB outgoing edges using local
properties. */
static void
bb_estimate_probability_locally (basic_block bb)
{
rtx last_insn = BB_END (bb);
rtx cond;
if (! can_predict_insn_p (last_insn))
return;
cond = get_condition (last_insn, NULL, false, false);
if (! cond)
return;
/* Try "pointer heuristic."
A comparison ptr == 0 is predicted as false.
Similarly, a comparison ptr1 == ptr2 is predicted as false. */
if (COMPARISON_P (cond)
&& ((REG_P (XEXP (cond, 0)) && REG_POINTER (XEXP (cond, 0)))
|| (REG_P (XEXP (cond, 1)) && REG_POINTER (XEXP (cond, 1)))))
{
if (GET_CODE (cond) == EQ)
predict_insn_def (last_insn, PRED_POINTER, NOT_TAKEN);
else if (GET_CODE (cond) == NE)
predict_insn_def (last_insn, PRED_POINTER, TAKEN);
}
else
/* Try "opcode heuristic."
EQ tests are usually false and NE tests are usually true. Also,
most quantities are positive, so we can make the appropriate guesses
about signed comparisons against zero. */
switch (GET_CODE (cond))
{
case CONST_INT:
/* Unconditional branch. */
predict_insn_def (last_insn, PRED_UNCONDITIONAL,
cond == const0_rtx ? NOT_TAKEN : TAKEN);
break;
case EQ:
case UNEQ:
/* Floating point comparisons appears to behave in a very
unpredictable way because of special role of = tests in
FP code. */
if (FLOAT_MODE_P (GET_MODE (XEXP (cond, 0))))
;
/* Comparisons with 0 are often used for booleans and there is
nothing useful to predict about them. */
else if (XEXP (cond, 1) == const0_rtx
|| XEXP (cond, 0) == const0_rtx)
;
else
predict_insn_def (last_insn, PRED_OPCODE_NONEQUAL, NOT_TAKEN);
break;
case NE:
case LTGT:
/* Floating point comparisons appears to behave in a very
unpredictable way because of special role of = tests in
FP code. */
if (FLOAT_MODE_P (GET_MODE (XEXP (cond, 0))))
;
/* Comparisons with 0 are often used for booleans and there is
nothing useful to predict about them. */
else if (XEXP (cond, 1) == const0_rtx
|| XEXP (cond, 0) == const0_rtx)
;
else
predict_insn_def (last_insn, PRED_OPCODE_NONEQUAL, TAKEN);
break;
case ORDERED:
predict_insn_def (last_insn, PRED_FPOPCODE, TAKEN);
break;
case UNORDERED:
predict_insn_def (last_insn, PRED_FPOPCODE, NOT_TAKEN);
break;
case LE:
case LT:
if (XEXP (cond, 1) == const0_rtx || XEXP (cond, 1) == const1_rtx
|| XEXP (cond, 1) == constm1_rtx)
predict_insn_def (last_insn, PRED_OPCODE_POSITIVE, NOT_TAKEN);
break;
case GE:
case GT:
if (XEXP (cond, 1) == const0_rtx || XEXP (cond, 1) == const1_rtx
|| XEXP (cond, 1) == constm1_rtx)
predict_insn_def (last_insn, PRED_OPCODE_POSITIVE, TAKEN);
break;
default:
break;
}
}
/* Statically estimate the probability that a branch will be taken and produce
estimated profile. When profile feedback is present never executed portions
of function gets estimated. */
......@@ -636,7 +761,6 @@ estimate_probability (struct loops *loops_info)
FOR_EACH_BB (bb)
{
rtx last_insn = BB_END (bb);
rtx cond;
edge e;
if (! can_predict_insn_p (last_insn))
......@@ -680,94 +804,7 @@ estimate_probability (struct loops *loops_info)
}
}
}
cond = get_condition (last_insn, NULL, false, false);
if (! cond)
continue;
/* Try "pointer heuristic."
A comparison ptr == 0 is predicted as false.
Similarly, a comparison ptr1 == ptr2 is predicted as false. */
if (COMPARISON_P (cond)
&& ((REG_P (XEXP (cond, 0)) && REG_POINTER (XEXP (cond, 0)))
|| (REG_P (XEXP (cond, 1)) && REG_POINTER (XEXP (cond, 1)))))
{
if (GET_CODE (cond) == EQ)
predict_insn_def (last_insn, PRED_POINTER, NOT_TAKEN);
else if (GET_CODE (cond) == NE)
predict_insn_def (last_insn, PRED_POINTER, TAKEN);
}
else
/* Try "opcode heuristic."
EQ tests are usually false and NE tests are usually true. Also,
most quantities are positive, so we can make the appropriate guesses
about signed comparisons against zero. */
switch (GET_CODE (cond))
{
case CONST_INT:
/* Unconditional branch. */
predict_insn_def (last_insn, PRED_UNCONDITIONAL,
cond == const0_rtx ? NOT_TAKEN : TAKEN);
break;
case EQ:
case UNEQ:
/* Floating point comparisons appears to behave in a very
unpredictable way because of special role of = tests in
FP code. */
if (FLOAT_MODE_P (GET_MODE (XEXP (cond, 0))))
;
/* Comparisons with 0 are often used for booleans and there is
nothing useful to predict about them. */
else if (XEXP (cond, 1) == const0_rtx
|| XEXP (cond, 0) == const0_rtx)
;
else
predict_insn_def (last_insn, PRED_OPCODE_NONEQUAL, NOT_TAKEN);
break;
case NE:
case LTGT:
/* Floating point comparisons appears to behave in a very
unpredictable way because of special role of = tests in
FP code. */
if (FLOAT_MODE_P (GET_MODE (XEXP (cond, 0))))
;
/* Comparisons with 0 are often used for booleans and there is
nothing useful to predict about them. */
else if (XEXP (cond, 1) == const0_rtx
|| XEXP (cond, 0) == const0_rtx)
;
else
predict_insn_def (last_insn, PRED_OPCODE_NONEQUAL, TAKEN);
break;
case ORDERED:
predict_insn_def (last_insn, PRED_FPOPCODE, TAKEN);
break;
case UNORDERED:
predict_insn_def (last_insn, PRED_FPOPCODE, NOT_TAKEN);
break;
case LE:
case LT:
if (XEXP (cond, 1) == const0_rtx || XEXP (cond, 1) == const1_rtx
|| XEXP (cond, 1) == constm1_rtx)
predict_insn_def (last_insn, PRED_OPCODE_POSITIVE, NOT_TAKEN);
break;
case GE:
case GT:
if (XEXP (cond, 1) == const0_rtx || XEXP (cond, 1) == const1_rtx
|| XEXP (cond, 1) == constm1_rtx)
predict_insn_def (last_insn, PRED_OPCODE_POSITIVE, TAKEN);
break;
default:
break;
}
bb_estimate_probability_locally (bb);
}
/* Attach the combined probability to each conditional jump. */
......@@ -808,6 +845,14 @@ estimate_probability (struct loops *loops_info)
if (profile_status == PROFILE_ABSENT)
profile_status = PROFILE_GUESSED;
}
/* Set edge->probability for each succestor edge of BB. */
void
guess_outgoing_edge_probabilities (basic_block bb)
{
bb_estimate_probability_locally (bb);
combine_predictions_for_insn (BB_END (bb), bb);
}
/* Predict using opcode of the last statement in basic block. */
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment