Makefile.in (value-prof.o): New dependencies on $(DIAGNOSTIC_H) $(TREE_H) and $(COVERAGE_H).
* Makefile.in (value-prof.o): New dependencies on $(DIAGNOSTIC_H) $(TREE_H) and $(COVERAGE_H). * coverage.c (compute_checksum): Use DECL_NAME not DECL_ASSEMBLER_NAME. * opts.c (common_handle_option): Enable tree-based value transforms. * toplev.c (process_options): Ditto. * value-prof.h (struct histogram_value_t): Redefine. "Adjust" below refers to references to this type. * tree-flow.h: (struct stmt_ann_d): Add histograms field. * rtl-profile.c (rtl_gen_interval_profiler): Adjust. Remove checks for may_be_more, may_be_less. (rtl_gen_pow2_profiler): Adjust. (rtl_gen_one_value_profiler_no_edge_manip): Adjust. (rtl_gen_one_value_profiler): Adjust. (rtl_gen_const_delta_profiler): Adjust. * tree-profile.c (tree_gen_interval_profiler): Implement. (tree_gen_pow2_profiler): Ditto. (tree_gen_one_value_profiler): Ditto. (tree_profiling): New. (pass_tree_profile): Reference it. * value-prof.c: Include tree-flow.h, tree-flow-inline.h, diagnostic.h, tree.h, gcov-io.h. (insn_divmod_values_to_profile): Rename to rtl_divmod_values_to_profile. Adjust. (insn_values_to_profile): Rename to rtl_values_to_profile. Adjust. (insn_prefetch_values_to_profile): Adjust. (rtl_value_profile_transformations): Adjust. (gen_divmod_fixed_value): Rename to rtl_divmod_fixed_value. (gen_mod_pow2): Rename to rtl_mod_pow2. (gen_mod_subtract): Rename to rtl_mod_subtract. (divmod_fixed_value_transform): Rename to rtl_divmod_fixed_value_transform. (mod_pow2_value_transform): Rename to rtl_mod_pow2_value_transform. (mod_subtract_transform): Rename to rtl_mod_subtract_transform. (rtl_find_values_to_profile): Adjust. (tree_value_profile_transformations): Implement. (tree_divmod_values_to_profile): New. (tree_values_to_profile): New. (tree_divmod_fixed_value): New. (tree_mod_pow2): New. (tree_mod_subtract): New. (tree_divmod_fixed_value_transform): New. (tree_mod_pow2_value_transform): New. (tree_mod_subtract_transform): New. (tree_find_values_to_profile): Implement. * profile.c (instrument_values): Free histograms. (compute_value_histograms): Adjust. Implement tree version. From-SVN: r97156
This commit is contained in:
Dale Johannesen
Jan Hubicka
parent
82498ed4b9
commit
1f1e85278a
|
|
@ -1,3 +1,52 @@
|
|||
2005-03-29 Dale Johannesen <dalej@apple.com>
|
||||
|
||||
* Makefile.in (value-prof.o): New dependencies on $(DIAGNOSTIC_H)
|
||||
$(TREE_H) and $(COVERAGE_H).
|
||||
* coverage.c (compute_checksum): Use DECL_NAME not DECL_ASSEMBLER_NAME.
|
||||
* opts.c (common_handle_option): Enable tree-based value transforms.
|
||||
* toplev.c (process_options): Ditto.
|
||||
* value-prof.h (struct histogram_value_t): Redefine. "Adjust" below
|
||||
refers to references to this type.
|
||||
* tree-flow.h: (struct stmt_ann_d): Add histograms field.
|
||||
* rtl-profile.c (rtl_gen_interval_profiler): Adjust. Remove checks
|
||||
for may_be_more, may_be_less.
|
||||
(rtl_gen_pow2_profiler): Adjust.
|
||||
(rtl_gen_one_value_profiler_no_edge_manip): Adjust.
|
||||
(rtl_gen_one_value_profiler): Adjust.
|
||||
(rtl_gen_const_delta_profiler): Adjust.
|
||||
* tree-profile.c (tree_gen_interval_profiler): Implement.
|
||||
(tree_gen_pow2_profiler): Ditto.
|
||||
(tree_gen_one_value_profiler): Ditto.
|
||||
(tree_profiling): New.
|
||||
(pass_tree_profile): Reference it.
|
||||
* value-prof.c: Include tree-flow.h, tree-flow-inline.h, diagnostic.h,
|
||||
tree.h, gcov-io.h.
|
||||
(insn_divmod_values_to_profile): Rename to
|
||||
rtl_divmod_values_to_profile. Adjust.
|
||||
(insn_values_to_profile): Rename to rtl_values_to_profile. Adjust.
|
||||
(insn_prefetch_values_to_profile): Adjust.
|
||||
(rtl_value_profile_transformations): Adjust.
|
||||
(gen_divmod_fixed_value): Rename to rtl_divmod_fixed_value.
|
||||
(gen_mod_pow2): Rename to rtl_mod_pow2.
|
||||
(gen_mod_subtract): Rename to rtl_mod_subtract.
|
||||
(divmod_fixed_value_transform): Rename to
|
||||
rtl_divmod_fixed_value_transform.
|
||||
(mod_pow2_value_transform): Rename to rtl_mod_pow2_value_transform.
|
||||
(mod_subtract_transform): Rename to rtl_mod_subtract_transform.
|
||||
(rtl_find_values_to_profile): Adjust.
|
||||
(tree_value_profile_transformations): Implement.
|
||||
(tree_divmod_values_to_profile): New.
|
||||
(tree_values_to_profile): New.
|
||||
(tree_divmod_fixed_value): New.
|
||||
(tree_mod_pow2): New.
|
||||
(tree_mod_subtract): New.
|
||||
(tree_divmod_fixed_value_transform): New.
|
||||
(tree_mod_pow2_value_transform): New.
|
||||
(tree_mod_subtract_transform): New.
|
||||
(tree_find_values_to_profile): Implement.
|
||||
* profile.c (instrument_values): Free histograms.
|
||||
(compute_value_histograms): Adjust. Implement tree version.
|
||||
|
||||
2005-03-29 Uros Bizjak <uros@kss-loka.si>
|
||||
|
||||
* reg-stack.c (subst_stack_regs_pat): Handle <UNSPEC_FIST> case.
|
||||
|
|
|
|||
|
|
@ -2023,7 +2023,8 @@ rtl-profile.o : tree-profile.c $(CONFIG_H) $(SYSTEM_H) coretypes.h \
|
|||
toplev.h $(BASIC_BLOCK_H) $(COVERAGE_H) $(TREE_FLOW_H) value-prof.h $(GGC_H)
|
||||
value-prof.o : value-prof.c $(CONFIG_H) $(SYSTEM_H) coretypes.h $(TM_H) $(RTL_H) \
|
||||
$(BASIC_BLOCK_H) hard-reg-set.h value-prof.h $(EXPR_H) output.h $(FLAGS_H) \
|
||||
$(RECOG_H) insn-config.h $(OPTABS_H) $(REGS_H) $(GGC_H)
|
||||
$(RECOG_H) insn-config.h $(OPTABS_H) $(REGS_H) $(GGC_H) $(DIAGNOSTIC_H) \
|
||||
$(TREE_H) $(COVERAGE_H)
|
||||
loop.o : loop.c $(CONFIG_H) $(SYSTEM_H) coretypes.h $(TM_H) $(RTL_H) $(FLAGS_H) \
|
||||
insn-config.h $(REGS_H) hard-reg-set.h $(RECOG_H) $(EXPR_H) \
|
||||
real.h $(PREDICT_H) $(BASIC_BLOCK_H) function.h $(CFGLOOP_H) \
|
||||
|
|
|
|||
|
|
@ -224,6 +224,7 @@ instrument_values (histogram_values values)
|
|||
abort ();
|
||||
}
|
||||
}
|
||||
VEC_free (histogram_value, values);
|
||||
}
|
||||
|
||||
|
||||
|
|
@ -641,7 +642,7 @@ compute_branch_probabilities (void)
|
|||
}
|
||||
|
||||
/* Load value histograms values whose description is stored in VALUES array
|
||||
from .da file. */
|
||||
from .gcda file. */
|
||||
|
||||
static void
|
||||
compute_value_histograms (histogram_values values)
|
||||
|
|
@ -688,21 +689,32 @@ compute_value_histograms (histogram_values values)
|
|||
hist = VEC_index (histogram_value, values, i);
|
||||
t = (int) hist->type;
|
||||
|
||||
/* FIXME: make this work for trees. */
|
||||
aact_count = act_count[t];
|
||||
act_count[t] += hist->n_counters;
|
||||
|
||||
if (!ir_type ())
|
||||
{
|
||||
aact_count = act_count[t];
|
||||
act_count[t] += hist->n_counters;
|
||||
for (j = hist->n_counters; j > 0; j--)
|
||||
hist_list = alloc_EXPR_LIST (0, GEN_INT (aact_count[j - 1]),
|
||||
hist_list);
|
||||
hist_list = alloc_EXPR_LIST (0,
|
||||
copy_rtx ((rtx) hist->value), hist_list);
|
||||
hist_list = alloc_EXPR_LIST (0,
|
||||
copy_rtx (hist->hvalue.rtl.value), hist_list);
|
||||
hist_list = alloc_EXPR_LIST (0, GEN_INT (hist->type), hist_list);
|
||||
REG_NOTES ((rtx) hist->insn) =
|
||||
alloc_EXPR_LIST (REG_VALUE_PROFILE, hist_list,
|
||||
REG_NOTES ((rtx) hist->insn));
|
||||
REG_NOTES (hist->hvalue.rtl.insn) =
|
||||
alloc_EXPR_LIST (REG_VALUE_PROFILE, hist_list,
|
||||
REG_NOTES (hist->hvalue.rtl.insn));
|
||||
}
|
||||
else
|
||||
{
|
||||
tree stmt = hist->hvalue.tree.stmt;
|
||||
stmt_ann_t ann = get_stmt_ann (stmt);
|
||||
hist->hvalue.tree.next = ann->histograms;
|
||||
ann->histograms = hist;
|
||||
hist->hvalue.tree.counters =
|
||||
xmalloc (sizeof (gcov_type) * hist->n_counters);
|
||||
for (j = 0; j < hist->n_counters; j++)
|
||||
hist->hvalue.tree.counters[j] = aact_count[j];
|
||||
}
|
||||
}
|
||||
|
||||
for (t = 0; t < GCOV_N_VALUE_COUNTERS; t++)
|
||||
|
|
|
|||
|
|
@ -23,30 +23,7 @@ Software Foundation, 59 Temple Place - Suite 330, Boston, MA
|
|||
02111-1307, USA. */
|
||||
|
||||
/* Generate basic block profile instrumentation and auxiliary files.
|
||||
Profile generation is optimized, so that not all arcs in the basic
|
||||
block graph need instrumenting. First, the BB graph is closed with
|
||||
one entry (function start), and one exit (function exit). Any
|
||||
ABNORMAL_EDGE cannot be instrumented (because there is no control
|
||||
path to place the code). We close the graph by inserting fake
|
||||
EDGE_FAKE edges to the EXIT_BLOCK, from the sources of abnormal
|
||||
edges that do not go to the exit_block. We ignore such abnormal
|
||||
edges. Naturally these fake edges are never directly traversed,
|
||||
and so *cannot* be directly instrumented. Some other graph
|
||||
massaging is done. To optimize the instrumentation we generate the
|
||||
BB minimal span tree, only edges that are not on the span tree
|
||||
(plus the entry point) need instrumenting. From that information
|
||||
all other edge counts can be deduced. By construction all fake
|
||||
edges must be on the spanning tree. We also attempt to place
|
||||
EDGE_CRITICAL edges on the spanning tree.
|
||||
|
||||
The auxiliary file generated is <dumpbase>.bbg. The format is
|
||||
described in full in gcov-io.h. */
|
||||
|
||||
/* ??? Register allocation should use basic block execution counts to
|
||||
give preference to the most commonly executed blocks. */
|
||||
|
||||
/* ??? Should calculate branch probabilities before instrumenting code, since
|
||||
then we can use arc counts to help decide which arcs to instrument. */
|
||||
RTL-based version. See profile.c for overview. */
|
||||
|
||||
#include "config.h"
|
||||
#include "system.h"
|
||||
|
|
@ -114,33 +91,33 @@ rtl_gen_interval_profiler (histogram_value value, unsigned tag, unsigned base)
|
|||
rtx less_label = gen_label_rtx ();
|
||||
rtx end_of_code_label = gen_label_rtx ();
|
||||
int per_counter = gcov_size / BITS_PER_UNIT;
|
||||
edge e = split_block (BLOCK_FOR_INSN ((rtx)value->insn),
|
||||
PREV_INSN ((rtx)value->insn));
|
||||
edge e = split_block (BLOCK_FOR_INSN (value->hvalue.rtl.insn),
|
||||
PREV_INSN (value->hvalue.rtl.insn));
|
||||
|
||||
start_sequence ();
|
||||
|
||||
if (value->seq)
|
||||
emit_insn (value->seq);
|
||||
if (value->hvalue.rtl.seq)
|
||||
emit_insn (value->hvalue.rtl.seq);
|
||||
|
||||
mr = gen_reg_rtx (Pmode);
|
||||
|
||||
tmp = rtl_coverage_counter_ref (tag, base);
|
||||
tmp = force_reg (Pmode, XEXP (tmp, 0));
|
||||
|
||||
val = expand_simple_binop (value->mode, MINUS,
|
||||
copy_rtx (value->value),
|
||||
val = expand_simple_binop (value->hvalue.rtl.mode, MINUS,
|
||||
copy_rtx (value->hvalue.rtl.value),
|
||||
GEN_INT (value->hdata.intvl.int_start),
|
||||
NULL_RTX, 0, OPTAB_WIDEN);
|
||||
|
||||
if (value->hdata.intvl.may_be_more)
|
||||
do_compare_rtx_and_jump (copy_rtx (val), GEN_INT (value->hdata.intvl.steps),
|
||||
GE, 0, value->mode, NULL_RTX, NULL_RTX, more_label);
|
||||
if (value->hdata.intvl.may_be_less)
|
||||
do_compare_rtx_and_jump (copy_rtx (val), const0_rtx, LT, 0, value->mode,
|
||||
GE, 0, value->hvalue.rtl.mode, NULL_RTX, NULL_RTX,
|
||||
more_label);
|
||||
do_compare_rtx_and_jump (copy_rtx (val), const0_rtx, LT, 0,
|
||||
value->hvalue.rtl.mode,
|
||||
NULL_RTX, NULL_RTX, less_label);
|
||||
|
||||
/* We are in range. */
|
||||
tmp1 = expand_simple_binop (value->mode, MULT,
|
||||
tmp1 = expand_simple_binop (value->hvalue.rtl.mode, MULT,
|
||||
copy_rtx (val), GEN_INT (per_counter),
|
||||
NULL_RTX, 0, OPTAB_WIDEN);
|
||||
tmp1 = expand_simple_binop (Pmode, PLUS, copy_rtx (tmp), tmp1, mr,
|
||||
|
|
@ -148,43 +125,27 @@ rtl_gen_interval_profiler (histogram_value value, unsigned tag, unsigned base)
|
|||
if (tmp1 != mr)
|
||||
emit_move_insn (copy_rtx (mr), tmp1);
|
||||
|
||||
if (value->hdata.intvl.may_be_more
|
||||
|| value->hdata.intvl.may_be_less)
|
||||
{
|
||||
emit_jump_insn (gen_jump (end_of_code_label));
|
||||
emit_barrier ();
|
||||
}
|
||||
|
||||
/* Above the interval. */
|
||||
if (value->hdata.intvl.may_be_more)
|
||||
{
|
||||
emit_label (more_label);
|
||||
tmp1 = expand_simple_binop (Pmode, PLUS, copy_rtx (tmp),
|
||||
GEN_INT (per_counter * value->hdata.intvl.steps),
|
||||
mr, 0, OPTAB_WIDEN);
|
||||
if (tmp1 != mr)
|
||||
emit_move_insn (copy_rtx (mr), tmp1);
|
||||
if (value->hdata.intvl.may_be_less)
|
||||
{
|
||||
emit_jump_insn (gen_jump (end_of_code_label));
|
||||
emit_barrier ();
|
||||
}
|
||||
}
|
||||
|
||||
/* Below the interval. */
|
||||
if (value->hdata.intvl.may_be_less)
|
||||
{
|
||||
emit_label (less_label);
|
||||
tmp1 = expand_simple_binop (Pmode, PLUS, copy_rtx (tmp),
|
||||
GEN_INT (per_counter * (value->hdata.intvl.steps
|
||||
+ (value->hdata.intvl.may_be_more ? 1 : 0))),
|
||||
GEN_INT (per_counter * (value->hdata.intvl.steps +1)),
|
||||
mr, 0, OPTAB_WIDEN);
|
||||
if (tmp1 != mr)
|
||||
emit_move_insn (copy_rtx (mr), tmp1);
|
||||
}
|
||||
|
||||
if (value->hdata.intvl.may_be_more
|
||||
|| value->hdata.intvl.may_be_less)
|
||||
emit_label (end_of_code_label);
|
||||
|
||||
mem_ref = validize_mem (gen_rtx_MEM (mode, mr));
|
||||
|
|
@ -215,32 +176,32 @@ rtl_gen_pow2_profiler (histogram_value value, unsigned tag, unsigned base)
|
|||
rtx end_of_code_label = gen_label_rtx ();
|
||||
rtx loop_label = gen_label_rtx ();
|
||||
int per_counter = gcov_size / BITS_PER_UNIT;
|
||||
edge e = split_block (BLOCK_FOR_INSN ((rtx)value->insn),
|
||||
PREV_INSN ((rtx)value->insn));
|
||||
edge e = split_block (BLOCK_FOR_INSN (value->hvalue.rtl.insn),
|
||||
PREV_INSN (value->hvalue.rtl.insn));
|
||||
|
||||
start_sequence ();
|
||||
|
||||
if (value->seq)
|
||||
emit_insn (value->seq);
|
||||
if (value->hvalue.rtl.seq)
|
||||
emit_insn (value->hvalue.rtl.seq);
|
||||
|
||||
mr = gen_reg_rtx (Pmode);
|
||||
tmp = rtl_coverage_counter_ref (tag, base);
|
||||
tmp = force_reg (Pmode, XEXP (tmp, 0));
|
||||
emit_move_insn (mr, tmp);
|
||||
|
||||
uval = gen_reg_rtx (value->mode);
|
||||
emit_move_insn (uval, copy_rtx (value->value));
|
||||
uval = gen_reg_rtx (value->hvalue.rtl.mode);
|
||||
emit_move_insn (uval, copy_rtx (value->hvalue.rtl.value));
|
||||
|
||||
/* Check for non-power of 2. */
|
||||
if (value->hdata.pow2.may_be_other)
|
||||
{
|
||||
do_compare_rtx_and_jump (copy_rtx (uval), const0_rtx, LE, 0, value->mode,
|
||||
do_compare_rtx_and_jump (copy_rtx (uval), const0_rtx, LE, 0, value->hvalue.rtl.mode,
|
||||
NULL_RTX, NULL_RTX, end_of_code_label);
|
||||
tmp = expand_simple_binop (value->mode, PLUS, copy_rtx (uval),
|
||||
tmp = expand_simple_binop (value->hvalue.rtl.mode, PLUS, copy_rtx (uval),
|
||||
constm1_rtx, NULL_RTX, 0, OPTAB_WIDEN);
|
||||
tmp = expand_simple_binop (value->mode, AND, copy_rtx (uval), tmp,
|
||||
tmp = expand_simple_binop (value->hvalue.rtl.mode, AND, copy_rtx (uval), tmp,
|
||||
NULL_RTX, 0, OPTAB_WIDEN);
|
||||
do_compare_rtx_and_jump (tmp, const0_rtx, NE, 0, value->mode, NULL_RTX,
|
||||
do_compare_rtx_and_jump (tmp, const0_rtx, NE, 0, value->hvalue.rtl.mode, NULL_RTX,
|
||||
NULL_RTX, end_of_code_label);
|
||||
}
|
||||
|
||||
|
|
@ -251,12 +212,12 @@ rtl_gen_pow2_profiler (histogram_value value, unsigned tag, unsigned base)
|
|||
if (tmp != mr)
|
||||
emit_move_insn (copy_rtx (mr), tmp);
|
||||
|
||||
tmp = expand_simple_binop (value->mode, ASHIFTRT, copy_rtx (uval), const1_rtx,
|
||||
tmp = expand_simple_binop (value->hvalue.rtl.mode, ASHIFTRT, copy_rtx (uval), const1_rtx,
|
||||
uval, 0, OPTAB_WIDEN);
|
||||
if (tmp != uval)
|
||||
emit_move_insn (copy_rtx (uval), tmp);
|
||||
|
||||
do_compare_rtx_and_jump (copy_rtx (uval), const0_rtx, NE, 0, value->mode,
|
||||
do_compare_rtx_and_jump (copy_rtx (uval), const0_rtx, NE, 0, value->hvalue.rtl.mode,
|
||||
NULL_RTX, NULL_RTX, loop_label);
|
||||
|
||||
/* Increase the counter. */
|
||||
|
|
@ -295,8 +256,8 @@ rtl_gen_one_value_profiler_no_edge_manipulation (histogram_value value,
|
|||
|
||||
start_sequence ();
|
||||
|
||||
if (value->seq)
|
||||
emit_insn (value->seq);
|
||||
if (value->hvalue.rtl.seq)
|
||||
emit_insn (value->hvalue.rtl.seq);
|
||||
|
||||
stored_value_ref = rtl_coverage_counter_ref (tag, base);
|
||||
counter_ref = rtl_coverage_counter_ref (tag, base + 1);
|
||||
|
|
@ -306,7 +267,7 @@ rtl_gen_one_value_profiler_no_edge_manipulation (histogram_value value,
|
|||
all = validize_mem (all_ref);
|
||||
|
||||
uval = gen_reg_rtx (mode);
|
||||
convert_move (uval, copy_rtx (value->value), 0);
|
||||
convert_move (uval, copy_rtx (value->hvalue.rtl.value), 0);
|
||||
|
||||
/* Check if the stored value matches. */
|
||||
do_compare_rtx_and_jump (copy_rtx (uval), copy_rtx (stored_value), EQ,
|
||||
|
|
@ -362,8 +323,8 @@ rtl_gen_one_value_profiler_no_edge_manipulation (histogram_value value,
|
|||
static void
|
||||
rtl_gen_one_value_profiler (histogram_value value, unsigned tag, unsigned base)
|
||||
{
|
||||
edge e = split_block (BLOCK_FOR_INSN ((rtx)value->insn),
|
||||
PREV_INSN ((rtx)value->insn));
|
||||
edge e = split_block (BLOCK_FOR_INSN (value->hvalue.rtl.insn),
|
||||
PREV_INSN (value->hvalue.rtl.insn));
|
||||
rtx sequence = rtl_gen_one_value_profiler_no_edge_manipulation (value,
|
||||
tag, base);
|
||||
rebuild_jump_labels (sequence);
|
||||
|
|
@ -383,28 +344,28 @@ rtl_gen_const_delta_profiler (histogram_value value, unsigned tag, unsigned base
|
|||
enum machine_mode mode = mode_for_size (gcov_size, MODE_INT, 0);
|
||||
rtx stored_value_ref, stored_value, tmp, uval;
|
||||
rtx sequence;
|
||||
edge e = split_block (BLOCK_FOR_INSN ((rtx)value->insn),
|
||||
PREV_INSN ((rtx)value->insn));
|
||||
edge e = split_block (BLOCK_FOR_INSN (value->hvalue.rtl.insn),
|
||||
PREV_INSN (value->hvalue.rtl.insn));
|
||||
|
||||
start_sequence ();
|
||||
|
||||
if (value->seq)
|
||||
emit_insn (value->seq);
|
||||
if (value->hvalue.rtl.seq)
|
||||
emit_insn (value->hvalue.rtl.seq);
|
||||
|
||||
stored_value_ref = rtl_coverage_counter_ref (tag, base);
|
||||
stored_value = validize_mem (stored_value_ref);
|
||||
|
||||
uval = gen_reg_rtx (mode);
|
||||
convert_move (uval, copy_rtx (value->value), 0);
|
||||
convert_move (uval, copy_rtx (value->hvalue.rtl.value), 0);
|
||||
tmp = expand_simple_binop (mode, MINUS,
|
||||
copy_rtx (uval), copy_rtx (stored_value),
|
||||
NULL_RTX, 0, OPTAB_WIDEN);
|
||||
|
||||
one_value_delta = ggc_alloc (sizeof (*one_value_delta));
|
||||
one_value_delta->value = tmp;
|
||||
one_value_delta->mode = mode;
|
||||
one_value_delta->seq = NULL_RTX;
|
||||
one_value_delta->insn = value->insn;
|
||||
one_value_delta->hvalue.rtl.value = tmp;
|
||||
one_value_delta->hvalue.rtl.mode = mode;
|
||||
one_value_delta->hvalue.rtl.seq = NULL_RTX;
|
||||
one_value_delta->hvalue.rtl.insn = value->hvalue.rtl.insn;
|
||||
one_value_delta->type = HIST_TYPE_SINGLE_VALUE;
|
||||
emit_insn (rtl_gen_one_value_profiler_no_edge_manipulation (one_value_delta,
|
||||
tag, base + 1));
|
||||
|
|
|
|||
|
|
@ -1773,9 +1773,6 @@ process_options (void)
|
|||
warning ("this target machine does not have delayed branches");
|
||||
#endif
|
||||
|
||||
if (flag_tree_based_profiling && flag_profile_values)
|
||||
sorry ("value-based profiling not yet implemented in trees.");
|
||||
|
||||
user_label_prefix = USER_LABEL_PREFIX;
|
||||
if (flag_leading_underscore != -1)
|
||||
{
|
||||
|
|
|
|||
|
|
@ -309,6 +309,12 @@ struct stmt_ann_d GTY(())
|
|||
by each pass on an as-needed basis in any order convenient for the
|
||||
pass which needs statement UIDs. */
|
||||
unsigned int uid;
|
||||
|
||||
/* Linked list of histograms for value-based profiling. This is really a
|
||||
struct histogram_value*. We use void* to avoid having to export that
|
||||
everywhere, and to avoid having to put it in GC memory. */
|
||||
|
||||
void * GTY ((skip (""))) histograms;
|
||||
};
|
||||
|
||||
union tree_ann_d GTY((desc ("ann_type ((tree_ann_t)&%h)")))
|
||||
|
|
|
|||
|
|
@ -24,30 +24,7 @@ Software Foundation, 59 Temple Place - Suite 330, Boston, MA
|
|||
02111-1307, USA. */
|
||||
|
||||
/* Generate basic block profile instrumentation and auxiliary files.
|
||||
Profile generation is optimized, so that not all arcs in the basic
|
||||
block graph need instrumenting. First, the BB graph is closed with
|
||||
one entry (function start), and one exit (function exit). Any
|
||||
ABNORMAL_EDGE cannot be instrumented (because there is no control
|
||||
path to place the code). We close the graph by inserting fake
|
||||
EDGE_FAKE edges to the EXIT_BLOCK, from the sources of abnormal
|
||||
edges that do not go to the exit_block. We ignore such abnormal
|
||||
edges. Naturally these fake edges are never directly traversed,
|
||||
and so *cannot* be directly instrumented. Some other graph
|
||||
massaging is done. To optimize the instrumentation we generate the
|
||||
BB minimal span tree, only edges that are not on the span tree
|
||||
(plus the entry point) need instrumenting. From that information
|
||||
all other edge counts can be deduced. By construction all fake
|
||||
edges must be on the spanning tree. We also attempt to place
|
||||
EDGE_CRITICAL edges on the spanning tree.
|
||||
|
||||
The auxiliary file generated is <dumpbase>.bbg. The format is
|
||||
described in full in gcov-io.h. */
|
||||
|
||||
/* ??? Register allocation should use basic block execution counts to
|
||||
give preference to the most commonly executed blocks. */
|
||||
|
||||
/* ??? Should calculate branch probabilities before instrumenting code, since
|
||||
then we can use arc counts to help decide which arcs to instrument. */
|
||||
Tree-based version. See profile.c for overview. */
|
||||
|
||||
#include "config.h"
|
||||
#include "system.h"
|
||||
|
|
@ -102,15 +79,163 @@ tree_gen_edge_profiler (int edgeno, edge e)
|
|||
tag of the section for counters, BASE is offset of the counter position. */
|
||||
|
||||
static void
|
||||
tree_gen_interval_profiler (histogram_value value ATTRIBUTE_UNUSED,
|
||||
unsigned tag ATTRIBUTE_UNUSED,
|
||||
unsigned base ATTRIBUTE_UNUSED)
|
||||
tree_gen_interval_profiler (histogram_value value, unsigned tag, unsigned base)
|
||||
{
|
||||
/* FIXME implement this. */
|
||||
#ifdef ENABLE_CHECKING
|
||||
internal_error ("unimplemented functionality");
|
||||
#endif
|
||||
gcc_unreachable ();
|
||||
tree op, op1, op2, op1copy, op2copy;
|
||||
tree tmp1, tmp2, tmp3, val, index;
|
||||
tree label_decl2, label_decl3, label_decl4, label_decl5, label_decl6;
|
||||
edge e12, e23, e34, e45, e56;
|
||||
tree label2, label3, label4, label5, label6;
|
||||
tree stmt1, stmt2, stmt3, stmt4;
|
||||
/* Initializations are to prevent bogus uninitialized warnings. */
|
||||
tree bb1end = NULL_TREE, bb2end = NULL_TREE, bb3end = NULL_TREE;
|
||||
tree bb4end = NULL_TREE, bb5end = NULL_TREE;
|
||||
tree ref = tree_coverage_counter_ref (tag, base), ref2;
|
||||
basic_block bb2, bb3, bb4, bb5, bb6;
|
||||
tree stmt = value->hvalue.tree.stmt;
|
||||
block_stmt_iterator bsi = bsi_for_stmt (stmt);
|
||||
basic_block bb = bb_for_stmt (stmt);
|
||||
tree optype;
|
||||
|
||||
op = stmt;
|
||||
if (TREE_CODE (stmt) == RETURN_EXPR
|
||||
&& TREE_OPERAND (stmt, 0)
|
||||
&& TREE_CODE (TREE_OPERAND (stmt, 0)) == MODIFY_EXPR)
|
||||
op = TREE_OPERAND (stmt, 0);
|
||||
/* op == MODIFY_EXPR */
|
||||
op = TREE_OPERAND (op, 1);
|
||||
/* op == TRUNC_DIV or TRUNC_MOD */
|
||||
op1 = TREE_OPERAND (op, 0);
|
||||
op2 = TREE_OPERAND (op, 1);
|
||||
optype = TREE_TYPE (op);
|
||||
|
||||
/* Blocks:
|
||||
Original = 1
|
||||
For 2nd compare = 2
|
||||
Normal case, neither more nor less = 3
|
||||
More = 4
|
||||
Less = 5
|
||||
End = 6 */
|
||||
label_decl2 = create_artificial_label ();
|
||||
label_decl3 = create_artificial_label ();
|
||||
label_decl4 = create_artificial_label ();
|
||||
label_decl5 = create_artificial_label ();
|
||||
label_decl6 = create_artificial_label ();
|
||||
|
||||
/* Do not evaluate op1 or op2 more than once. Probably
|
||||
volatile loads are the only things that could cause
|
||||
a problem, but this is harmless in any case. */
|
||||
op1copy = create_tmp_var (optype, "PROF");
|
||||
op2copy = create_tmp_var (optype, "PROF");
|
||||
stmt1 = build2 (MODIFY_EXPR, optype, op1copy, op1);
|
||||
stmt2 = build2 (MODIFY_EXPR, optype, op2copy, op2);
|
||||
TREE_OPERAND (op, 0) = op1copy;
|
||||
TREE_OPERAND (op, 1) = op2copy;
|
||||
|
||||
val = create_tmp_var (optype, "PROF");
|
||||
stmt3 = build2 (MODIFY_EXPR, optype, val,
|
||||
build2 (TRUNC_DIV_EXPR, optype, op1copy, op2copy));
|
||||
stmt4 = build2 (MODIFY_EXPR, optype, val,
|
||||
build2 (MINUS_EXPR, optype, val,
|
||||
build_int_cst (optype, value->hdata.intvl.int_start)));
|
||||
bsi_insert_before (&bsi, stmt1, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt2, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt3, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt4, BSI_SAME_STMT);
|
||||
|
||||
index = create_tmp_var (GCOV_TYPE_NODE, "PROF");
|
||||
|
||||
/* Check for too big. */
|
||||
stmt1 = build3 (COND_EXPR, void_type_node,
|
||||
build2 (GE_EXPR, boolean_type_node, val,
|
||||
build_int_cst (optype, value->hdata.intvl.steps)),
|
||||
build1 (GOTO_EXPR, void_type_node, label_decl4),
|
||||
build1 (GOTO_EXPR, void_type_node, label_decl2));
|
||||
bsi_insert_before (&bsi, stmt1, BSI_SAME_STMT);
|
||||
bb1end = stmt1;
|
||||
|
||||
/* Check for too small. */
|
||||
label2 = build1 (LABEL_EXPR, void_type_node, label_decl2);
|
||||
bsi_insert_before (&bsi, label2, BSI_SAME_STMT);
|
||||
stmt1 = build3 (COND_EXPR, void_type_node,
|
||||
build2 (LT_EXPR, boolean_type_node, val, integer_zero_node),
|
||||
build1 (GOTO_EXPR, void_type_node, label_decl5),
|
||||
build1 (GOTO_EXPR, void_type_node, label_decl3));
|
||||
bsi_insert_before (&bsi, stmt1, BSI_SAME_STMT);
|
||||
bb2end = stmt1;
|
||||
|
||||
/* Normal case, within range. */
|
||||
label3 = build1 (LABEL_EXPR, void_type_node, label_decl3);
|
||||
bsi_insert_before (&bsi, label3, BSI_SAME_STMT);
|
||||
stmt1 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, index,
|
||||
build1 (NOP_EXPR, GCOV_TYPE_NODE, val));
|
||||
bsi_insert_before (&bsi, stmt1, BSI_SAME_STMT);
|
||||
bb3end = stmt1;
|
||||
|
||||
/* Too big */
|
||||
label4 = build1 (LABEL_EXPR, void_type_node, label_decl4);
|
||||
stmt1 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, index,
|
||||
build_int_cst (GCOV_TYPE_NODE, value->hdata.intvl.steps));
|
||||
bsi_insert_before (&bsi, label4, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt1, BSI_SAME_STMT);
|
||||
bb4end = stmt1;
|
||||
|
||||
/* Too small */
|
||||
label5 = build1 (LABEL_EXPR, void_type_node, label_decl5);
|
||||
stmt1 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, index,
|
||||
build_int_cst (GCOV_TYPE_NODE, value->hdata.intvl.steps + 1));
|
||||
bsi_insert_before (&bsi, label5, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt1, BSI_SAME_STMT);
|
||||
bb5end = stmt1;
|
||||
|
||||
/* Increment appropriate counter. */
|
||||
label6 = build1 (LABEL_EXPR, void_type_node, label_decl6);
|
||||
bsi_insert_before (&bsi, label6, BSI_SAME_STMT);
|
||||
|
||||
tmp1 = create_tmp_var (GCOV_TYPE_NODE, "PROF");
|
||||
tmp2 = create_tmp_var (GCOV_TYPE_NODE, "PROF");
|
||||
tmp3 = create_tmp_var (GCOV_TYPE_NODE, "PROF");
|
||||
stmt1 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, tmp1,
|
||||
build2 (PLUS_EXPR, GCOV_TYPE_NODE, index,
|
||||
TREE_OPERAND (ref, 1)));
|
||||
TREE_OPERAND (ref, 1) = tmp1;
|
||||
/* Make a copy to avoid sharing complaints. */
|
||||
ref2 = build4 (ARRAY_REF, TREE_TYPE (ref), TREE_OPERAND (ref, 0),
|
||||
TREE_OPERAND (ref, 1), TREE_OPERAND (ref, 2),
|
||||
TREE_OPERAND (ref, 3));
|
||||
|
||||
stmt2 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, tmp2, ref);
|
||||
stmt3 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, tmp3,
|
||||
build2 (PLUS_EXPR, GCOV_TYPE_NODE, tmp2, integer_one_node));
|
||||
stmt4 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, ref2, tmp3);
|
||||
bsi_insert_before (&bsi, stmt1, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt2, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt3, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt4, BSI_SAME_STMT);
|
||||
|
||||
/* Now fix up the CFG. */
|
||||
/* 1->2,4; 2->3,5; 3->6; 4->6; 5->6 */
|
||||
e12 = split_block (bb, bb1end);
|
||||
bb2 = e12->dest;
|
||||
e23 = split_block (bb2, bb2end);
|
||||
bb3 = e23->dest;
|
||||
e34 = split_block (bb3, bb3end);
|
||||
bb4 = e34->dest;
|
||||
e45 = split_block (bb4, bb4end);
|
||||
bb5 = e45->dest;
|
||||
e56 = split_block (bb5, bb5end);
|
||||
bb6 = e56->dest;
|
||||
|
||||
e12->flags &= ~EDGE_FALLTHRU;
|
||||
e12->flags |= EDGE_FALSE_VALUE;
|
||||
make_edge (bb, bb4, EDGE_TRUE_VALUE);
|
||||
e23->flags &= ~EDGE_FALLTHRU;
|
||||
e23->flags |= EDGE_FALSE_VALUE;
|
||||
make_edge (bb2, bb5, EDGE_TRUE_VALUE);
|
||||
remove_edge (e34);
|
||||
make_edge (bb3, bb6, EDGE_FALLTHRU);
|
||||
remove_edge (e45);
|
||||
make_edge (bb4, bb6, EDGE_FALLTHRU);
|
||||
}
|
||||
|
||||
/* Output instructions as GIMPLE trees to increment the power of two histogram
|
||||
|
|
@ -118,15 +243,162 @@ tree_gen_interval_profiler (histogram_value value ATTRIBUTE_UNUSED,
|
|||
of the section for counters, BASE is offset of the counter position. */
|
||||
|
||||
static void
|
||||
tree_gen_pow2_profiler (histogram_value value ATTRIBUTE_UNUSED,
|
||||
unsigned tag ATTRIBUTE_UNUSED,
|
||||
unsigned base ATTRIBUTE_UNUSED)
|
||||
tree_gen_pow2_profiler (histogram_value value, unsigned tag, unsigned base)
|
||||
{
|
||||
/* FIXME implement this. */
|
||||
#ifdef ENABLE_CHECKING
|
||||
internal_error ("unimplemented functionality");
|
||||
#endif
|
||||
gcc_unreachable ();
|
||||
tree op;
|
||||
tree tmp1, tmp2, tmp3;
|
||||
tree index, denom;
|
||||
tree label_decl1 = create_artificial_label ();
|
||||
tree label_decl2 = create_artificial_label ();
|
||||
tree label_decl3 = create_artificial_label ();
|
||||
tree label1, label2, label3;
|
||||
tree stmt1, stmt2, stmt3, stmt4;
|
||||
tree bb1end, bb2end, bb3end;
|
||||
tree ref = tree_coverage_counter_ref (tag, base), ref2;
|
||||
basic_block bb2, bb3, bb4;
|
||||
tree stmt = value->hvalue.tree.stmt;
|
||||
block_stmt_iterator bsi = bsi_for_stmt (stmt);
|
||||
basic_block bb = bb_for_stmt (stmt);
|
||||
tree optype, optypesigned, optypeunsigned;
|
||||
|
||||
op = stmt;
|
||||
if (TREE_CODE (stmt) == RETURN_EXPR
|
||||
&& TREE_OPERAND (stmt, 0)
|
||||
&& TREE_CODE (TREE_OPERAND (stmt, 0)) == MODIFY_EXPR)
|
||||
op = TREE_OPERAND (stmt, 0);
|
||||
/* op == MODIFY_EXPR */
|
||||
op = TREE_OPERAND (op, 1);
|
||||
/* op == TRUNC_DIV or TRUNC_MOD */
|
||||
op = TREE_OPERAND (op, 1);
|
||||
/* op == denominator */
|
||||
optype = TREE_TYPE (op);
|
||||
if (TYPE_UNSIGNED (optype))
|
||||
{
|
||||
/* Right shift must be unsigned. */
|
||||
optypeunsigned = optype;
|
||||
optypesigned = build_distinct_type_copy (optype);
|
||||
TYPE_UNSIGNED (optypesigned) = false;
|
||||
}
|
||||
else
|
||||
{
|
||||
/* Compare to zero must be signed. */
|
||||
optypesigned = optype;
|
||||
optypeunsigned = build_distinct_type_copy (optype);
|
||||
TYPE_UNSIGNED (optypeunsigned) = true;
|
||||
}
|
||||
|
||||
/* Set up variables and check if denominator is negative when considered
|
||||
as signed. */
|
||||
index = create_tmp_var (GCOV_TYPE_NODE, "PROF");
|
||||
denom = create_tmp_var (optype, "PROF");
|
||||
stmt1 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, index, integer_zero_node);
|
||||
stmt2 = build2 (MODIFY_EXPR, optype, denom, op);
|
||||
if (optypesigned == optype)
|
||||
{
|
||||
tmp1 = denom;
|
||||
stmt3 = NULL_TREE;
|
||||
}
|
||||
else
|
||||
{
|
||||
tmp1 = create_tmp_var (optypesigned, "PROF");
|
||||
stmt3 = build2 (MODIFY_EXPR, optypesigned, tmp1,
|
||||
build1 (NOP_EXPR, optypesigned, denom));
|
||||
}
|
||||
stmt4 = build3 (COND_EXPR, void_type_node,
|
||||
build2 (LE_EXPR, boolean_type_node, tmp1, integer_zero_node),
|
||||
build1 (GOTO_EXPR, void_type_node, label_decl3),
|
||||
build1 (GOTO_EXPR, void_type_node, label_decl1));
|
||||
bsi_insert_before (&bsi, stmt1, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt2, BSI_SAME_STMT);
|
||||
if (stmt3)
|
||||
bsi_insert_before (&bsi, stmt3, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt4, BSI_SAME_STMT);
|
||||
bb1end = stmt4;
|
||||
|
||||
/* Nonnegative. Check if denominator is power of 2. */
|
||||
label1 = build1 (LABEL_EXPR, void_type_node, label_decl1);
|
||||
tmp1 = create_tmp_var (optype, "PROF");
|
||||
tmp2 = create_tmp_var (optype, "PROF");
|
||||
stmt1 = build2 (MODIFY_EXPR, optype, tmp1,
|
||||
build2 (PLUS_EXPR, optype, denom, integer_minus_one_node));
|
||||
stmt2 = build2 (MODIFY_EXPR, optype, tmp2,
|
||||
build2 (BIT_AND_EXPR, optype, tmp1, denom));
|
||||
stmt3 = build3 (COND_EXPR, void_type_node,
|
||||
build2 (NE_EXPR, boolean_type_node, tmp2, integer_zero_node),
|
||||
build1 (GOTO_EXPR, void_type_node, label_decl3),
|
||||
build1 (GOTO_EXPR, void_type_node, label_decl2));
|
||||
bsi_insert_before (&bsi, label1, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt1, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt2, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt3, BSI_SAME_STMT);
|
||||
bb2end = stmt3;
|
||||
|
||||
/* Loop. Increment index, shift denominator, repeat if denominator nonzero. */
|
||||
label2 = build1 (LABEL_EXPR, void_type_node, label_decl2);
|
||||
stmt1 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, index,
|
||||
build2 (PLUS_EXPR, GCOV_TYPE_NODE, index, integer_one_node));
|
||||
if (optypeunsigned == optype)
|
||||
{
|
||||
tmp1 = denom;
|
||||
stmt2 = NULL_TREE;
|
||||
}
|
||||
else
|
||||
{
|
||||
tmp1 = create_tmp_var (optypeunsigned, "PROF");
|
||||
stmt2 = build2 (MODIFY_EXPR, optypeunsigned, tmp1,
|
||||
build1 (NOP_EXPR, optypeunsigned, denom));
|
||||
}
|
||||
stmt3 = build2 (MODIFY_EXPR, optype, denom,
|
||||
build2 (RSHIFT_EXPR, optype, tmp1, integer_one_node));
|
||||
stmt4 = build3 (COND_EXPR, void_type_node,
|
||||
build2 (NE_EXPR, boolean_type_node, denom, integer_zero_node),
|
||||
build1 (GOTO_EXPR, void_type_node, label_decl2),
|
||||
build1 (GOTO_EXPR, void_type_node, label_decl3));
|
||||
bsi_insert_before (&bsi, label2, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt1, BSI_SAME_STMT);
|
||||
if (stmt2)
|
||||
bsi_insert_before (&bsi, stmt2, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt3, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt4, BSI_SAME_STMT);
|
||||
bb3end = stmt4;
|
||||
|
||||
/* Increment the appropriate counter. */
|
||||
label3 = build1 (LABEL_EXPR, void_type_node, label_decl3);
|
||||
tmp1 = create_tmp_var (GCOV_TYPE_NODE, "PROF");
|
||||
tmp2 = create_tmp_var (GCOV_TYPE_NODE, "PROF");
|
||||
tmp3 = create_tmp_var (GCOV_TYPE_NODE, "PROF");
|
||||
stmt1 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, tmp1,
|
||||
build2 (PLUS_EXPR, GCOV_TYPE_NODE, index, TREE_OPERAND (ref, 1)));
|
||||
TREE_OPERAND (ref, 1) = tmp1;
|
||||
/* Make a copy to avoid sharing complaints. */
|
||||
ref2 = build4 (ARRAY_REF, TREE_TYPE (ref), TREE_OPERAND (ref, 0),
|
||||
TREE_OPERAND (ref, 1), TREE_OPERAND (ref, 2), TREE_OPERAND (ref, 3));
|
||||
stmt2 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, tmp2, ref);
|
||||
stmt3 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, tmp3,
|
||||
build2 (PLUS_EXPR, GCOV_TYPE_NODE, tmp2, integer_one_node));
|
||||
stmt4 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, ref2, tmp3);
|
||||
bsi_insert_before (&bsi, label3, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt1, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt2, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt3, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt4, BSI_SAME_STMT);
|
||||
|
||||
/* Now fix up the CFG. */
|
||||
bb2 = (split_block (bb, bb1end))->dest;
|
||||
bb3 = (split_block (bb2, bb2end))->dest;
|
||||
bb4 = (split_block (bb3, bb3end))->dest;
|
||||
|
||||
EDGE_SUCC (bb, 0)->flags &= ~EDGE_FALLTHRU;
|
||||
EDGE_SUCC (bb, 0)->flags |= EDGE_FALSE_VALUE;
|
||||
make_edge (bb, bb4, EDGE_TRUE_VALUE);
|
||||
|
||||
EDGE_SUCC (bb2, 0)->flags &= ~EDGE_FALLTHRU;
|
||||
EDGE_SUCC (bb2, 0)->flags |= EDGE_FALSE_VALUE;
|
||||
make_edge (bb2, bb4, EDGE_TRUE_VALUE);
|
||||
|
||||
EDGE_SUCC (bb3, 0)->flags &= ~EDGE_FALLTHRU;
|
||||
EDGE_SUCC (bb3, 0)->flags |= EDGE_FALSE_VALUE;
|
||||
make_edge (bb3, bb3, EDGE_TRUE_VALUE);
|
||||
}
|
||||
|
||||
/* Output instructions as GIMPLE trees for code to find the most common value.
|
||||
|
|
@ -134,15 +406,150 @@ tree_gen_pow2_profiler (histogram_value value ATTRIBUTE_UNUSED,
|
|||
section for counters, BASE is offset of the counter position. */
|
||||
|
||||
static void
|
||||
tree_gen_one_value_profiler (histogram_value value ATTRIBUTE_UNUSED,
|
||||
unsigned tag ATTRIBUTE_UNUSED,
|
||||
unsigned base ATTRIBUTE_UNUSED)
|
||||
tree_gen_one_value_profiler (histogram_value value, unsigned tag, unsigned base)
|
||||
{
|
||||
/* FIXME implement this. */
|
||||
#ifdef ENABLE_CHECKING
|
||||
internal_error ("unimplemented functionality");
|
||||
#endif
|
||||
gcc_unreachable ();
|
||||
tree op;
|
||||
tree tmp1, tmp2, tmp3;
|
||||
tree label_decl1 = create_artificial_label ();
|
||||
tree label_decl2 = create_artificial_label ();
|
||||
tree label_decl3 = create_artificial_label ();
|
||||
tree label_decl4 = create_artificial_label ();
|
||||
tree label_decl5 = create_artificial_label ();
|
||||
tree label1, label2, label3, label4, label5;
|
||||
tree stmt1, stmt2, stmt3, stmt4;
|
||||
tree bb1end, bb2end, bb3end, bb4end, bb5end;
|
||||
tree ref1 = tree_coverage_counter_ref (tag, base);
|
||||
tree ref2 = tree_coverage_counter_ref (tag, base + 1);
|
||||
tree ref3 = tree_coverage_counter_ref (tag, base + 2);
|
||||
basic_block bb2, bb3, bb4, bb5, bb6;
|
||||
tree stmt = value->hvalue.tree.stmt;
|
||||
block_stmt_iterator bsi = bsi_for_stmt (stmt);
|
||||
basic_block bb = bb_for_stmt (stmt);
|
||||
tree optype;
|
||||
|
||||
op = stmt;
|
||||
if (TREE_CODE (stmt) == RETURN_EXPR
|
||||
&& TREE_OPERAND (stmt, 0)
|
||||
&& TREE_CODE (TREE_OPERAND (stmt, 0)) == MODIFY_EXPR)
|
||||
op = TREE_OPERAND (stmt, 0);
|
||||
/* op == MODIFY_EXPR */
|
||||
op = TREE_OPERAND (op, 1);
|
||||
/* op == TRUNC_DIV or TRUNC_MOD */
|
||||
op = TREE_OPERAND (op, 1);
|
||||
/* op == denominator */
|
||||
optype = TREE_TYPE (op);
|
||||
|
||||
/* Check if the stored value matches. */
|
||||
tmp1 = create_tmp_var (GCOV_TYPE_NODE, "PROF");
|
||||
tmp2 = create_tmp_var (optype, "PROF");
|
||||
tmp3 = create_tmp_var (optype, "PROF");
|
||||
stmt1 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, tmp1, ref1);
|
||||
stmt2 = build2 (MODIFY_EXPR, optype, tmp2,
|
||||
build1 (NOP_EXPR, optype, tmp1));
|
||||
stmt3 = build2 (MODIFY_EXPR, optype, tmp3, op);
|
||||
stmt4 = build3 (COND_EXPR, void_type_node,
|
||||
build2 (EQ_EXPR, boolean_type_node, tmp2, tmp3),
|
||||
build1 (GOTO_EXPR, void_type_node, label_decl4),
|
||||
build1 (GOTO_EXPR, void_type_node, label_decl1));
|
||||
bsi_insert_before (&bsi, stmt1, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt2, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt3, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt4, BSI_SAME_STMT);
|
||||
bb1end = stmt4;
|
||||
|
||||
/* Does not match; check whether the counter is zero. */
|
||||
label1 = build1 (LABEL_EXPR, void_type_node, label_decl1);
|
||||
tmp1 = create_tmp_var (GCOV_TYPE_NODE, "PROF");
|
||||
stmt1 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, tmp1, ref2);
|
||||
stmt2 = build3 (COND_EXPR, void_type_node,
|
||||
build2 (EQ_EXPR, boolean_type_node, tmp1, integer_zero_node),
|
||||
build1 (GOTO_EXPR, void_type_node, label_decl3),
|
||||
build1 (GOTO_EXPR, void_type_node, label_decl2));
|
||||
bsi_insert_before (&bsi, label1, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt1, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt2, BSI_SAME_STMT);
|
||||
bb2end = stmt2;
|
||||
|
||||
/* Counter is not zero yet, decrement. */
|
||||
label2 = build1 (LABEL_EXPR, void_type_node, label_decl2);
|
||||
tmp1 = create_tmp_var (GCOV_TYPE_NODE, "PROF");
|
||||
tmp2 = create_tmp_var (GCOV_TYPE_NODE, "PROF");
|
||||
stmt1 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, tmp1, ref2);
|
||||
stmt2 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, tmp2,
|
||||
build (MINUS_EXPR, GCOV_TYPE_NODE,
|
||||
tmp1, integer_one_node));
|
||||
stmt3 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, ref2, tmp2);
|
||||
bsi_insert_before (&bsi, label2, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt1, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt2, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt3, BSI_SAME_STMT);
|
||||
bb3end = stmt3;
|
||||
|
||||
/* Counter was zero, store new value. */
|
||||
label3 = build1 (LABEL_EXPR, void_type_node, label_decl3);
|
||||
tmp1 = create_tmp_var (optype, "PROF");
|
||||
tmp2 = create_tmp_var (GCOV_TYPE_NODE, "PROF");
|
||||
stmt1 = build2 (MODIFY_EXPR, optype, tmp1, op);
|
||||
stmt2 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, tmp2,
|
||||
build1 (NOP_EXPR, GCOV_TYPE_NODE, tmp1));
|
||||
stmt3 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, ref1, tmp2);
|
||||
bsi_insert_before (&bsi, label3, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt1, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt2, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt3, BSI_SAME_STMT);
|
||||
bb4end = stmt3;
|
||||
/* (fall through) */
|
||||
|
||||
/* Increment counter. */
|
||||
label4 = build1 (LABEL_EXPR, void_type_node, label_decl4);
|
||||
tmp1 = create_tmp_var (GCOV_TYPE_NODE, "PROF");
|
||||
tmp2 = create_tmp_var (GCOV_TYPE_NODE, "PROF");
|
||||
stmt1 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, tmp1, ref2);
|
||||
stmt2 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, tmp2,
|
||||
build (PLUS_EXPR, GCOV_TYPE_NODE,
|
||||
tmp1, integer_one_node));
|
||||
stmt3 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, ref2, tmp2);
|
||||
bsi_insert_before (&bsi, label4, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt1, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt2, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt3, BSI_SAME_STMT);
|
||||
bb5end = stmt3;
|
||||
|
||||
/* Increment the counter of all executions; this seems redundant given
|
||||
that we have counts for edges in cfg, but it may happen that some
|
||||
optimization will change the counts for the block (either because
|
||||
it is unable to update them correctly, or because it will duplicate
|
||||
the block or its part). */
|
||||
label5 = build1 (LABEL_EXPR, void_type_node, label_decl5);
|
||||
tmp1 = create_tmp_var (GCOV_TYPE_NODE, "PROF");
|
||||
tmp2 = create_tmp_var (GCOV_TYPE_NODE, "PROF");
|
||||
stmt1 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, tmp1, ref3);
|
||||
stmt2 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, tmp2,
|
||||
build (PLUS_EXPR, GCOV_TYPE_NODE,
|
||||
tmp1, integer_one_node));
|
||||
stmt3 = build2 (MODIFY_EXPR, GCOV_TYPE_NODE, ref3, tmp2);
|
||||
bsi_insert_before (&bsi, label5, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt1, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt2, BSI_SAME_STMT);
|
||||
bsi_insert_before (&bsi, stmt3, BSI_SAME_STMT);
|
||||
|
||||
/* Now fix up the CFG. */
|
||||
bb2 = (split_block (bb, bb1end))->dest;
|
||||
bb3 = (split_block (bb2, bb2end))->dest;
|
||||
bb4 = (split_block (bb3, bb3end))->dest;
|
||||
bb5 = (split_block (bb4, bb4end))->dest;
|
||||
bb6 = (split_block (bb5, bb5end))->dest;
|
||||
|
||||
EDGE_SUCC (bb, 0)->flags &= ~EDGE_FALLTHRU;
|
||||
EDGE_SUCC (bb, 0)->flags |= EDGE_FALSE_VALUE;
|
||||
make_edge (bb, bb5, EDGE_TRUE_VALUE);
|
||||
|
||||
EDGE_SUCC (bb2, 0)->flags &= ~EDGE_FALLTHRU;
|
||||
EDGE_SUCC (bb2, 0)->flags |= EDGE_FALSE_VALUE;
|
||||
make_edge (bb2, bb4, EDGE_TRUE_VALUE);
|
||||
|
||||
remove_edge (EDGE_SUCC (bb3, 0));
|
||||
make_edge (bb3, bb6, EDGE_FALLTHRU);
|
||||
}
|
||||
|
||||
/* Output instructions as GIMPLE trees for code to find the most common value
|
||||
|
|
@ -166,7 +573,8 @@ tree_gen_const_delta_profiler (histogram_value value ATTRIBUTE_UNUSED,
|
|||
If it is, set up hooks for tree-based profiling.
|
||||
Gate for pass_tree_profile. */
|
||||
|
||||
static bool do_tree_profiling (void)
|
||||
static bool
|
||||
do_tree_profiling (void)
|
||||
{
|
||||
if (flag_tree_based_profiling
|
||||
&& (profile_arc_flag || flag_test_coverage || flag_branch_probabilities))
|
||||
|
|
@ -184,11 +592,26 @@ static FILE *tree_profile_dump_file (void) {
|
|||
return dump_file;
|
||||
}
|
||||
|
||||
static void
|
||||
tree_profiling (void)
|
||||
{
|
||||
branch_prob ();
|
||||
if (flag_branch_probabilities
|
||||
&& flag_profile_values
|
||||
&& flag_value_profile_transformations)
|
||||
value_profile_transformations ();
|
||||
/* The above could hose dominator info. Currently there is
|
||||
none coming in, this is a safety valve. It should be
|
||||
easy to adjust it, if and when there is some. */
|
||||
free_dominance_info (CDI_DOMINATORS);
|
||||
free_dominance_info (CDI_POST_DOMINATORS);
|
||||
}
|
||||
|
||||
struct tree_opt_pass pass_tree_profile =
|
||||
{
|
||||
"tree_profile", /* name */
|
||||
do_tree_profiling, /* gate */
|
||||
branch_prob, /* execute */
|
||||
tree_profiling, /* execute */
|
||||
NULL, /* sub */
|
||||
NULL, /* next */
|
||||
0, /* static_pass_number */
|
||||
|
|
|
|||
905
gcc/value-prof.c
905
gcc/value-prof.c
File diff suppressed because it is too large
Load Diff
|
|
@ -38,14 +38,25 @@ enum hist_type
|
|||
((enum hist_type) ((COUNTER) - GCOV_FIRST_VALUE_COUNTER))
|
||||
|
||||
/* The value to measure. */
|
||||
/* The void *'s are either rtx or tree, depending on which IR is in use. */
|
||||
struct histogram_value_t GTY(())
|
||||
struct histogram_value_t
|
||||
{
|
||||
PTR GTY ((skip (""))) value; /* The value to profile. */
|
||||
enum machine_mode mode; /* And its mode. */
|
||||
PTR GTY ((skip (""))) seq; /* Insns required to count the
|
||||
profiled value. */
|
||||
PTR GTY ((skip (""))) insn; /* Insn before that to measure. */
|
||||
union
|
||||
{
|
||||
struct
|
||||
{
|
||||
rtx value; /* The value to profile. */
|
||||
rtx seq; /* Insns required to count the profiled value. */
|
||||
rtx insn; /* Insn before that to measure. */
|
||||
enum machine_mode mode; /* Mode of value to profile. */
|
||||
} rtl;
|
||||
struct
|
||||
{
|
||||
tree value; /* The value to profile. */
|
||||
tree stmt; /* Insn containing the value. */
|
||||
gcov_type *counters; /* Pointer to first counter. */
|
||||
struct histogram_value_t *next; /* Linked list pointer. */
|
||||
} tree;
|
||||
} hvalue;
|
||||
enum hist_type type; /* Type of information to measure. */
|
||||
unsigned n_counters; /* Number of required counters. */
|
||||
union
|
||||
|
|
@ -53,9 +64,7 @@ struct histogram_value_t GTY(())
|
|||
struct
|
||||
{
|
||||
int int_start; /* First value in interval. */
|
||||
int steps; /* Number of values in it. */
|
||||
int may_be_less; /* May the value be below? */
|
||||
int may_be_more; /* Or above. */
|
||||
unsigned int steps; /* Number of values in it. */
|
||||
} intvl; /* Interval histogram data. */
|
||||
struct
|
||||
{
|
||||
|
|
@ -66,7 +75,7 @@ struct histogram_value_t GTY(())
|
|||
|
||||
typedef struct histogram_value_t *histogram_value;
|
||||
|
||||
DEF_VEC_GC_P(histogram_value);
|
||||
DEF_VEC_MALLOC_P(histogram_value);
|
||||
|
||||
typedef VEC(histogram_value) *histogram_values;
|
||||
|
||||
|
|
|
|||
Loading…
Reference in New Issue