/* Transformations based on profile information for values.
- Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008 Free Software
- Foundation, Inc.
+ Copyright (C) 2003-2013 Free Software Foundation, Inc.
This file is part of GCC.
#include "system.h"
#include "coretypes.h"
#include "tm.h"
+#include "tree.h"
#include "rtl.h"
#include "expr.h"
#include "hard-reg-set.h"
#include "basic-block.h"
#include "value-prof.h"
-#include "output.h"
#include "flags.h"
#include "insn-config.h"
#include "recog.h"
#include "optabs.h"
#include "regs.h"
#include "ggc.h"
-#include "tree-flow.h"
-#include "tree-flow-inline.h"
+#include "gimple.h"
+#include "gimplify.h"
+#include "gimple-iterator.h"
+#include "gimple-ssa.h"
+#include "tree-cfg.h"
+#include "tree-phinodes.h"
+#include "ssa-iterators.h"
+#include "tree-ssanames.h"
#include "diagnostic.h"
+#include "gimple-pretty-print.h"
#include "coverage.h"
#include "tree.h"
#include "gcov-io.h"
-#include "cgraph.h"
#include "timevar.h"
-#include "tree-pass.h"
-#include "toplev.h"
+#include "dumpfile.h"
#include "pointer-set.h"
-
-static struct value_prof_hooks *value_prof_hooks;
+#include "profile.h"
+#include "data-streamer.h"
+#include "builtins.h"
+#include "tree-nested.h"
/* In this file value profile based optimizations are placed. Currently the
following optimizations are implemented (for more detailed descriptions
1) Division/modulo specialization. Provided that we can determine that the
operands of the division have some special properties, we may use it to
produce more effective code.
- 2) Speculative prefetching. If we are able to determine that the difference
- between addresses accessed by a memory reference is usually constant, we
- may add the prefetch instructions.
- FIXME: This transformation was removed together with RTL based value
- profiling.
- 3) Indirect/virtual call specialization. If we can determine most
+ 2) Indirect/virtual call specialization. If we can determine most
common function callee in indirect/virtual call. We can use this
information to improve code effectiveness (especially info for
- inliner).
+ the inliner).
- Every such optimization should add its requirements for profiled values to
- insn_values_to_profile function. This function is called from branch_prob
- in profile.c and the requested values are instrumented by it in the first
- compilation with -fprofile-arcs. The optimization may then read the
- gathered data in the second compilation with -fbranch-probabilities.
+ 3) Speculative prefetching. If we are able to determine that the difference
+ between addresses accessed by a memory reference is usually constant, we
+ may add the prefetch instructions.
+ FIXME: This transformation was removed together with RTL based value
+ profiling.
- The measured data is pointed to from the histograms
- field of the statement annotation of the instrumented insns. It is
- kept as a linked list of struct histogram_value_t's, which contain the
- same information as above. */
+ Value profiling internals
+ ==========================
+
+ Every value profiling transformation starts with defining what values
+ to profile. There are different histogram types (see HIST_TYPE_* in
+ value-prof.h) and each transformation can request one or more histogram
+ types per GIMPLE statement. The function gimple_find_values_to_profile()
+ collects the values to profile in a vec, and adds the number of counters
+ required for the different histogram types.
+
+ For a -fprofile-generate run, the statements for which values should be
+ recorded, are instrumented in instrument_values(). The instrumentation
+ is done by helper functions that can be found in tree-profile.c, where
+ new types of histograms can be added if necessary.
+
+ After a -fprofile-use, the value profiling data is read back in by
+ compute_value_histograms() that translates the collected data to
+ histograms and attaches them to the profiled statements via
+ gimple_add_histogram_value(). Histograms are stored in a hash table
+ that is attached to every intrumented function, see VALUE_HISTOGRAMS
+ in function.h.
+
+ The value-profile transformations driver is the function
+ gimple_value_profile_transformations(). It traverses all statements in
+ the to-be-transformed function, and looks for statements with one or
+ more histograms attached to it. If a statement has histograms, the
+ transformation functions are called on the statement.
+
+ Limitations / FIXME / TODO:
+ * Only one histogram of each type can be associated with a statement.
+ * Currently, HIST_TYPE_CONST_DELTA is not implemented.
+ (This type of histogram was originally used to implement a form of
+ stride profiling based speculative prefetching to improve SPEC2000
+ scores for memory-bound benchmarks, mcf and equake. However, this
+ was an RTL value-profiling transformation, and those have all been
+ removed.)
+ * Some value profile transformations are done in builtins.c (?!)
+ * Updating of histograms needs some TLC.
+ * The value profiling code could be used to record analysis results
+ from non-profiling (e.g. VRP).
+ * Adding new profilers should be simplified, starting with a cleanup
+ of what-happens-where andwith making gimple_find_values_to_profile
+ and gimple_value_profile_transformations table-driven, perhaps...
+*/
static tree gimple_divmod_fixed_value (gimple, tree, int, gcov_type, gcov_type);
static tree gimple_mod_pow2 (gimple, int, gcov_type, gcov_type);
static bool gimple_mod_pow2_value_transform (gimple_stmt_iterator *);
static bool gimple_mod_subtract_transform (gimple_stmt_iterator *);
static bool gimple_stringops_transform (gimple_stmt_iterator *);
-static bool gimple_ic_transform (gimple);
+static bool gimple_ic_transform (gimple_stmt_iterator *);
/* Allocate histogram value. */
return htab_hash_pointer (((const_histogram_value)x)->hvalue.stmt);
}
-/* Return nonzero if decl_id of die_struct X is the same as UID of decl *Y. */
+/* Return nonzero if statement for histogram_value X is Y. */
static int
histogram_eq (const void *x, const void *y)
{
hist->hvalue.next = gimple_histogram_value (fun, stmt);
set_histogram_value (fun, stmt, hist);
+ hist->fun = fun;
}
if (hist->hvalue.counters)
{
unsigned int i;
- fprintf(dump_file, " [");
+ fprintf (dump_file, " [");
for (i = 0; i < hist->hdata.intvl.steps; i++)
fprintf (dump_file, " %d:"HOST_WIDEST_INT_PRINT_DEC,
hist->hdata.intvl.int_start + i,
}
fprintf (dump_file, ".\n");
break;
+ case HIST_TYPE_TIME_PROFILE:
+ fprintf (dump_file, "Time profile ");
+ if (hist->hvalue.counters)
+ {
+ fprintf (dump_file, "time:"HOST_WIDEST_INT_PRINT_DEC,
+ (HOST_WIDEST_INT) hist->hvalue.counters[0]);
+ }
+ fprintf (dump_file, ".\n");
+ break;
+ case HIST_TYPE_MAX:
+ gcc_unreachable ();
}
}
+/* Dump information about HIST to DUMP_FILE. */
+
+void
+stream_out_histogram_value (struct output_block *ob, histogram_value hist)
+{
+ struct bitpack_d bp;
+ unsigned int i;
+
+ bp = bitpack_create (ob->main_stream);
+ bp_pack_enum (&bp, hist_type, HIST_TYPE_MAX, hist->type);
+ bp_pack_value (&bp, hist->hvalue.next != NULL, 1);
+ streamer_write_bitpack (&bp);
+ switch (hist->type)
+ {
+ case HIST_TYPE_INTERVAL:
+ streamer_write_hwi (ob, hist->hdata.intvl.int_start);
+ streamer_write_uhwi (ob, hist->hdata.intvl.steps);
+ break;
+ default:
+ break;
+ }
+ for (i = 0; i < hist->n_counters; i++)
+ streamer_write_gcov_count (ob, hist->hvalue.counters[i]);
+ if (hist->hvalue.next)
+ stream_out_histogram_value (ob, hist->hvalue.next);
+}
+/* Dump information about HIST to DUMP_FILE. */
+
+void
+stream_in_histogram_value (struct lto_input_block *ib, gimple stmt)
+{
+ enum hist_type type;
+ unsigned int ncounters = 0;
+ struct bitpack_d bp;
+ unsigned int i;
+ histogram_value new_val;
+ bool next;
+ histogram_value *next_p = NULL;
+
+ do
+ {
+ bp = streamer_read_bitpack (ib);
+ type = bp_unpack_enum (&bp, hist_type, HIST_TYPE_MAX);
+ next = bp_unpack_value (&bp, 1);
+ new_val = gimple_alloc_histogram_value (cfun, type, stmt, NULL);
+ switch (type)
+ {
+ case HIST_TYPE_INTERVAL:
+ new_val->hdata.intvl.int_start = streamer_read_hwi (ib);
+ new_val->hdata.intvl.steps = streamer_read_uhwi (ib);
+ ncounters = new_val->hdata.intvl.steps + 2;
+ break;
+
+ case HIST_TYPE_POW2:
+ case HIST_TYPE_AVERAGE:
+ ncounters = 2;
+ break;
+
+ case HIST_TYPE_SINGLE_VALUE:
+ case HIST_TYPE_INDIR_CALL:
+ ncounters = 3;
+ break;
+
+ case HIST_TYPE_CONST_DELTA:
+ ncounters = 4;
+ break;
+
+ case HIST_TYPE_IOR:
+ case HIST_TYPE_TIME_PROFILE:
+ ncounters = 1;
+ break;
+ case HIST_TYPE_MAX:
+ gcc_unreachable ();
+ }
+ new_val->hvalue.counters = XNEWVAR (gcov_type, sizeof (*new_val->hvalue.counters) * ncounters);
+ new_val->n_counters = ncounters;
+ for (i = 0; i < ncounters; i++)
+ new_val->hvalue.counters[i] = streamer_read_gcov_count (ib);
+ if (!next_p)
+ gimple_add_histogram_value (cfun, stmt, new_val);
+ else
+ *next_p = new_val;
+ next_p = &new_val->hvalue.next;
+ }
+ while (next);
+}
+
/* Dump all histograms attached to STMT to DUMP_FILE. */
void
{
histogram_value hist;
for (hist = gimple_histogram_value (fun, stmt); hist; hist = hist->hvalue.next)
- dump_histogram_value (dump_file, hist);
+ dump_histogram_value (dump_file, hist);
}
/* Remove all histograms associated with STMT. */
{
struct pointer_set_t *visited = (struct pointer_set_t *) data;
histogram_value hist = *(histogram_value *) slot;
- if (!pointer_set_contains (visited, hist))
+
+ if (!pointer_set_contains (visited, hist)
+ && hist->type != HIST_TYPE_TIME_PROFILE)
{
- error ("Dead histogram");
+ error ("dead histogram");
dump_histogram_value (stderr, hist);
debug_gimple_stmt (hist->hvalue.stmt);
error_found = true;
/* Verify sanity of the histograms. */
-void
+DEBUG_FUNCTION void
verify_histograms (void)
{
basic_block bb;
: DECL_SOURCE_LOCATION (current_function_decl);
if (flag_profile_correction)
{
- inform (locus, "Correcting inconsistent value profile: "
- "%s profiler overall count (%d) does not match BB count "
- "(%d)", name, (int)*all, (int)bb_count);
+ if (dump_enabled_p ())
+ dump_printf_loc (MSG_MISSED_OPTIMIZATION, locus,
+ "correcting inconsistent value profile: %s "
+ "profiler overall count (%d) does not match BB "
+ "count (%d)\n", name, (int)*all, (int)bb_count);
*all = bb_count;
if (*count > *all)
*count = *all;
}
else
{
- error ("%HCorrupted value profile: %s profiler overall count (%d) "
- "does not match BB count (%d)", &locus, name, (int)*all,
- (int)bb_count);
+ error_at (locus, "corrupted value profile: %s "
+ "profile counter (%d out of %d) inconsistent with "
+ "basic-block count (%d)",
+ name,
+ (int) *count,
+ (int) *all,
+ (int) bb_count);
return true;
}
}
/* GIMPLE based transformations. */
-static bool
+bool
gimple_value_profile_transformations (void)
{
basic_block bb;
will be added before the current statement, and that the
current statement remain valid (although possibly
modified) upon return. */
- if (flag_value_profile_transformations
- && (gimple_mod_subtract_transform (&gsi)
- || gimple_divmod_fixed_value_transform (&gsi)
- || gimple_mod_pow2_value_transform (&gsi)
- || gimple_stringops_transform (&gsi)
- || gimple_ic_transform (stmt)))
+ if (gimple_mod_subtract_transform (&gsi)
+ || gimple_divmod_fixed_value_transform (&gsi)
+ || gimple_mod_pow2_value_transform (&gsi)
+ || gimple_stringops_transform (&gsi)
+ || gimple_ic_transform (&gsi))
{
stmt = gsi_stmt (gsi);
changed = true;
gcov_type all)
{
gimple stmt1, stmt2, stmt3;
- tree tmp1, tmp2, tmpv;
+ tree tmp0, tmp1, tmp2;
gimple bb1end, bb2end, bb3end;
basic_block bb, bb2, bb3, bb4;
tree optype, op1, op2;
bb = gimple_bb (stmt);
gsi = gsi_for_stmt (stmt);
- tmpv = create_tmp_var (optype, "PROF");
- tmp1 = create_tmp_var (optype, "PROF");
- stmt1 = gimple_build_assign (tmpv, fold_convert (optype, value));
+ tmp0 = make_temp_ssa_name (optype, NULL, "PROF");
+ tmp1 = make_temp_ssa_name (optype, NULL, "PROF");
+ stmt1 = gimple_build_assign (tmp0, fold_convert (optype, value));
stmt2 = gimple_build_assign (tmp1, op2);
- stmt3 = gimple_build_cond (NE_EXPR, tmp1, tmpv, NULL_TREE, NULL_TREE);
+ stmt3 = gimple_build_cond (NE_EXPR, tmp1, tmp0, NULL_TREE, NULL_TREE);
gsi_insert_before (&gsi, stmt1, GSI_SAME_STMT);
gsi_insert_before (&gsi, stmt2, GSI_SAME_STMT);
gsi_insert_before (&gsi, stmt3, GSI_SAME_STMT);
bb1end = stmt3;
- tmp2 = create_tmp_var (optype, "PROF");
+ tmp2 = create_tmp_reg (optype, "PROF");
stmt1 = gimple_build_assign_with_ops (gimple_assign_rhs_code (stmt), tmp2,
- op1, tmpv);
+ op1, tmp0);
gsi_insert_before (&gsi, stmt1, GSI_SAME_STMT);
bb2end = stmt1;
e13->count = all - count;
remove_edge (e23);
-
+
e24 = make_edge (bb2, bb4, EDGE_FALLTHRU);
e24->probability = REG_BR_PROB_BASE;
e24->count = count;
return false;
code = gimple_assign_rhs_code (stmt);
-
+
if (code != TRUNC_DIV_EXPR && code != TRUNC_MOD_EXPR)
return false;
at least 50% of time (and 75% gives the guarantee of usage). */
if (simple_cst_equal (gimple_assign_rhs2 (stmt), value) != 1
|| 2 * count < all
- || !maybe_hot_bb_p (gimple_bb (stmt)))
+ || optimize_bb_for_size_p (gimple_bb (stmt)))
return false;
if (check_counter (stmt, "value", &count, &all, gimple_bb (stmt)->count))
/* Compute probability of taking the optimal path. */
if (all > 0)
- prob = (count * REG_BR_PROB_BASE + all / 2) / all;
+ prob = GCOV_COMPUTE_SCALE (count, all);
else
prob = 0;
}
gimple_assign_set_rhs_from_tree (si, result);
+ update_stmt (gsi_stmt (*si));
return true;
}
/* Generate code for transformation 2 (with parent gimple assign STMT and
probability of taking the optimal path PROB, which is equivalent to COUNT/ALL
- within roundoff error). This generates the result into a temp and returns
+ within roundoff error). This generates the result into a temp and returns
the temp; it does not replace or alter the original STMT. */
static tree
gimple_mod_pow2 (gimple stmt, int prob, gcov_type count, gcov_type all)
bb = gimple_bb (stmt);
gsi = gsi_for_stmt (stmt);
- result = create_tmp_var (optype, "PROF");
- tmp2 = create_tmp_var (optype, "PROF");
- tmp3 = create_tmp_var (optype, "PROF");
+ result = create_tmp_reg (optype, "PROF");
+ tmp2 = make_temp_ssa_name (optype, NULL, "PROF");
+ tmp3 = make_temp_ssa_name (optype, NULL, "PROF");
stmt2 = gimple_build_assign_with_ops (PLUS_EXPR, tmp2, op2,
build_int_cst (optype, -1));
stmt3 = gimple_build_assign_with_ops (BIT_AND_EXPR, tmp3, tmp2, op2);
e13->count = all - count;
remove_edge (e23);
-
+
e24 = make_edge (bb2, bb4, EDGE_FALLTHRU);
e24->probability = REG_BR_PROB_BASE;
e24->count = count;
return false;
code = gimple_assign_rhs_code (stmt);
-
+
if (code != TRUNC_MOD_EXPR || !TYPE_UNSIGNED (lhs_type))
return false;
/* We require that we hit a power of 2 at least half of all evaluations. */
if (simple_cst_equal (gimple_assign_rhs2 (stmt), value) != 1
|| count < wrong_values
- || !maybe_hot_bb_p (gimple_bb (stmt)))
+ || optimize_bb_for_size_p (gimple_bb (stmt)))
return false;
if (dump_file)
return false;
if (all > 0)
- prob = (count * REG_BR_PROB_BASE + all / 2) / all;
+ prob = GCOV_COMPUTE_SCALE (count, all);
else
prob = 0;
result = gimple_mod_pow2 (stmt, prob, count, all);
gimple_assign_set_rhs_from_tree (si, result);
+ update_stmt (gsi_stmt (*si));
return true;
}
NCOUNTS the number of cases to support. Currently only NCOUNTS==0 or 1 is
supported and this is built into this interface. The probabilities of taking
the optimal paths are PROB1 and PROB2, which are equivalent to COUNT1/ALL and
- COUNT2/ALL respectively within roundoff error). This generates the
- result into a temp and returns the temp; it does not replace or alter
+ COUNT2/ALL respectively within roundoff error). This generates the
+ result into a temp and returns the temp; it does not replace or alter
the original STMT. */
/* FIXME: Generalize the interface to handle NCOUNTS > 1. */
bb = gimple_bb (stmt);
gsi = gsi_for_stmt (stmt);
- result = create_tmp_var (optype, "PROF");
- tmp1 = create_tmp_var (optype, "PROF");
+ result = create_tmp_reg (optype, "PROF");
+ tmp1 = make_temp_ssa_name (optype, NULL, "PROF");
stmt1 = gimple_build_assign (result, op1);
stmt2 = gimple_build_assign (tmp1, op2);
stmt3 = gimple_build_cond (LT_EXPR, result, tmp1, NULL_TREE, NULL_TREE);
e12 = split_block (bb, bb1end);
bb2 = e12->dest;
bb2->count = all - count1;
-
+
if (ncounts) /* Assumed to be 0 or 1. */
{
e23 = split_block (bb2, bb2end);
histogram_value histogram;
enum tree_code code;
gcov_type count, wrong_values, all;
- tree lhs_type, result, value;
+ tree lhs_type, result;
gcov_type prob1, prob2;
unsigned int i, steps;
gcov_type count1, count2;
return false;
code = gimple_assign_rhs_code (stmt);
-
+
if (code != TRUNC_MOD_EXPR || !TYPE_UNSIGNED (lhs_type))
return false;
if (!histogram)
return false;
- value = histogram->hvalue.value;
all = 0;
wrong_values = 0;
for (i = 0; i < histogram->hdata.intvl.steps; i++)
break;
}
if (i == steps
- || !maybe_hot_bb_p (gimple_bb (stmt)))
+ || optimize_bb_for_size_p (gimple_bb (stmt)))
return false;
gimple_remove_histogram_value (cfun, stmt, histogram);
/* Compute probability of taking the optimal path(s). */
if (all > 0)
{
- prob1 = (count1 * REG_BR_PROB_BASE + all / 2) / all;
- prob2 = (count2 * REG_BR_PROB_BASE + all / 2) / all;
+ prob1 = GCOV_COMPUTE_SCALE (count1, all);
+ prob2 = GCOV_COMPUTE_SCALE (count2, all);
}
else
{
result = gimple_mod_subtract (stmt, prob1, prob2, i, count1, count2, all);
gimple_assign_set_rhs_from_tree (si, result);
+ update_stmt (gsi_stmt (*si));
return true;
}
-static struct cgraph_node** pid_map = NULL;
+static pointer_map_t *cgraph_node_map;
-/* Initialize map of pids (pid -> cgraph node) */
+/* Initialize map from PROFILE_ID to CGRAPH_NODE.
+ When LOCAL is true, the PROFILE_IDs are computed. when it is false we assume
+ that the PROFILE_IDs was already assigned. */
-static void
-init_pid_map (void)
+void
+init_node_map (bool local)
{
struct cgraph_node *n;
+ cgraph_node_map = pointer_map_create ();
- if (pid_map != NULL)
- return;
+ FOR_EACH_DEFINED_FUNCTION (n)
+ if (cgraph_function_with_gimple_body_p (n)
+ && !cgraph_only_called_directly_p (n))
+ {
+ void **val;
+ if (local)
+ {
+ n->profile_id = coverage_compute_profile_id (n);
+ while ((val = pointer_map_contains (cgraph_node_map,
+ (void *)(size_t)n->profile_id))
+ || !n->profile_id)
+ {
+ if (dump_file)
+ fprintf (dump_file, "Local profile-id %i conflict"
+ " with nodes %s/%i %s/%i\n",
+ n->profile_id,
+ cgraph_node_name (n),
+ n->order,
+ symtab_node_name (*(symtab_node **)val),
+ (*(symtab_node **)val)->order);
+ n->profile_id = (n->profile_id + 1) & 0x7fffffff;
+ }
+ }
+ else if (!n->profile_id)
+ {
+ if (dump_file)
+ fprintf (dump_file,
+ "Node %s/%i has no profile-id"
+ " (profile feedback missing?)\n",
+ cgraph_node_name (n),
+ n->order);
+ continue;
+ }
+ else if ((val = pointer_map_contains (cgraph_node_map,
+ (void *)(size_t)n->profile_id)))
+ {
+ if (dump_file)
+ fprintf (dump_file,
+ "Node %s/%i has IP profile-id %i conflict. "
+ "Giving up.\n",
+ cgraph_node_name (n),
+ n->order,
+ n->profile_id);
+ *val = NULL;
+ continue;
+ }
+ *pointer_map_insert (cgraph_node_map,
+ (void *)(size_t)n->profile_id) = (void *)n;
+ }
+}
- pid_map
- = (struct cgraph_node**) xmalloc (sizeof (struct cgraph_node*) * cgraph_max_pid);
+/* Delete the CGRAPH_NODE_MAP. */
- for (n = cgraph_nodes; n; n = n->next)
- {
- if (n->pid != -1)
- pid_map [n->pid] = n;
- }
+void
+del_node_map (void)
+{
+ pointer_map_destroy (cgraph_node_map);
}
/* Return cgraph node for function with pid */
-static inline struct cgraph_node*
-find_func_by_pid (int pid)
+struct cgraph_node*
+find_func_by_profile_id (int profile_id)
{
- init_pid_map ();
+ void **val = pointer_map_contains (cgraph_node_map,
+ (void *)(size_t)profile_id);
+ if (val)
+ return (struct cgraph_node *)*val;
+ else
+ return NULL;
+}
- return pid_map [pid];
+/* Perform sanity check on the indirect call target. Due to race conditions,
+ false function target may be attributed to an indirect call site. If the
+ call expression type mismatches with the target function's type, expand_call
+ may ICE. Here we only do very minimal sanity check just to make compiler happy.
+ Returns true if TARGET is considered ok for call CALL_STMT. */
+
+static bool
+check_ic_target (gimple call_stmt, struct cgraph_node *target)
+{
+ location_t locus;
+ if (gimple_check_call_matching_types (call_stmt, target->decl, true))
+ return true;
+
+ locus = gimple_location (call_stmt);
+ if (dump_enabled_p ())
+ dump_printf_loc (MSG_MISSED_OPTIMIZATION, locus,
+ "Skipping target %s with mismatching types for icall\n",
+ cgraph_node_name (target));
+ return false;
}
/* Do transformation
old call
*/
-static gimple
-gimple_ic (gimple stmt, gimple call, struct cgraph_node *direct_call,
+gimple
+gimple_ic (gimple icall_stmt, struct cgraph_node *direct_call,
int prob, gcov_type count, gcov_type all)
{
- gimple stmt1, stmt2, stmt3;
- tree tmp1, tmpv, tmp;
- gimple bb1end, bb2end, bb3end;
- basic_block bb, bb2, bb3, bb4;
+ gimple dcall_stmt, load_stmt, cond_stmt;
+ tree tmp0, tmp1, tmp;
+ basic_block cond_bb, dcall_bb, icall_bb, join_bb = NULL;
tree optype = build_pointer_type (void_type_node);
- edge e12, e13, e23, e24, e34;
+ edge e_cd, e_ci, e_di, e_dj = NULL, e_ij;
gimple_stmt_iterator gsi;
- int region;
+ int lp_nr, dflags;
+ edge e_eh, e;
+ edge_iterator ei;
+ gimple_stmt_iterator psi;
- bb = gimple_bb (stmt);
- gsi = gsi_for_stmt (stmt);
+ cond_bb = gimple_bb (icall_stmt);
+ gsi = gsi_for_stmt (icall_stmt);
- tmpv = create_tmp_var (optype, "PROF");
- tmp1 = create_tmp_var (optype, "PROF");
- stmt1 = gimple_build_assign (tmpv, unshare_expr (gimple_call_fn (call)));
+ tmp0 = make_temp_ssa_name (optype, NULL, "PROF");
+ tmp1 = make_temp_ssa_name (optype, NULL, "PROF");
+ tmp = unshare_expr (gimple_call_fn (icall_stmt));
+ load_stmt = gimple_build_assign (tmp0, tmp);
+ gsi_insert_before (&gsi, load_stmt, GSI_SAME_STMT);
- tmp = fold_convert (optype, build_addr (direct_call->decl,
+ tmp = fold_convert (optype, build_addr (direct_call->decl,
current_function_decl));
- stmt2 = gimple_build_assign (tmp1, tmp);
- stmt3 = gimple_build_cond (NE_EXPR, tmp1, tmpv, NULL_TREE, NULL_TREE);
- gsi_insert_before (&gsi, stmt1, GSI_SAME_STMT);
- gsi_insert_before (&gsi, stmt2, GSI_SAME_STMT);
- gsi_insert_before (&gsi, stmt3, GSI_SAME_STMT);
- bb1end = stmt3;
-
- stmt1 = gimple_copy (stmt);
- gimple_call_set_fn (stmt,
- build_addr (direct_call->decl, current_function_decl));
- gsi_insert_before (&gsi, stmt1, GSI_SAME_STMT);
- bb2end = stmt1;
- bb3end = stmt;
+ load_stmt = gimple_build_assign (tmp1, tmp);
+ gsi_insert_before (&gsi, load_stmt, GSI_SAME_STMT);
+
+ cond_stmt = gimple_build_cond (EQ_EXPR, tmp1, tmp0, NULL_TREE, NULL_TREE);
+ gsi_insert_before (&gsi, cond_stmt, GSI_SAME_STMT);
+
+ gimple_set_vdef (icall_stmt, NULL_TREE);
+ gimple_set_vuse (icall_stmt, NULL_TREE);
+ update_stmt (icall_stmt);
+ dcall_stmt = gimple_copy (icall_stmt);
+ gimple_call_set_fndecl (dcall_stmt, direct_call->decl);
+ dflags = flags_from_decl_or_type (direct_call->decl);
+ if ((dflags & ECF_NORETURN) != 0)
+ gimple_call_set_lhs (dcall_stmt, NULL_TREE);
+ gsi_insert_before (&gsi, dcall_stmt, GSI_SAME_STMT);
/* Fix CFG. */
- /* Edge e23 connects bb2 to bb3, etc. */
- e12 = split_block (bb, bb1end);
- bb2 = e12->dest;
- bb2->count = count;
- e23 = split_block (bb2, bb2end);
- bb3 = e23->dest;
- bb3->count = all - count;
- e34 = split_block (bb3, bb3end);
- bb4 = e34->dest;
- bb4->count = all;
+ /* Edge e_cd connects cond_bb to dcall_bb, etc; note the first letters. */
+ e_cd = split_block (cond_bb, cond_stmt);
+ dcall_bb = e_cd->dest;
+ dcall_bb->count = count;
+
+ e_di = split_block (dcall_bb, dcall_stmt);
+ icall_bb = e_di->dest;
+ icall_bb->count = all - count;
+
+ /* Do not disturb existing EH edges from the indirect call. */
+ if (!stmt_ends_bb_p (icall_stmt))
+ e_ij = split_block (icall_bb, icall_stmt);
+ else
+ {
+ e_ij = find_fallthru_edge (icall_bb->succs);
+ /* The indirect call might be noreturn. */
+ if (e_ij != NULL)
+ {
+ e_ij->probability = REG_BR_PROB_BASE;
+ e_ij->count = all - count;
+ e_ij = single_pred_edge (split_edge (e_ij));
+ }
+ }
+ if (e_ij != NULL)
+ {
+ join_bb = e_ij->dest;
+ join_bb->count = all;
+ }
- e12->flags &= ~EDGE_FALLTHRU;
- e12->flags |= EDGE_FALSE_VALUE;
- e12->probability = prob;
- e12->count = count;
+ e_cd->flags = (e_cd->flags & ~EDGE_FALLTHRU) | EDGE_TRUE_VALUE;
+ e_cd->probability = prob;
+ e_cd->count = count;
- e13 = make_edge (bb, bb3, EDGE_TRUE_VALUE);
- e13->probability = REG_BR_PROB_BASE - prob;
- e13->count = all - count;
+ e_ci = make_edge (cond_bb, icall_bb, EDGE_FALSE_VALUE);
+ e_ci->probability = REG_BR_PROB_BASE - prob;
+ e_ci->count = all - count;
- remove_edge (e23);
-
- e24 = make_edge (bb2, bb4, EDGE_FALLTHRU);
- e24->probability = REG_BR_PROB_BASE;
- e24->count = count;
- e34->probability = REG_BR_PROB_BASE;
- e34->count = all - count;
+ remove_edge (e_di);
- /* Fix eh edges */
- region = lookup_stmt_eh_region (stmt);
- if (region >= 0 && stmt_could_throw_p (stmt1))
+ if (e_ij != NULL)
{
- add_stmt_to_eh_region (stmt1, region);
- make_eh_edges (stmt1);
+ if ((dflags & ECF_NORETURN) != 0)
+ e_ij->count = all;
+ else
+ {
+ e_dj = make_edge (dcall_bb, join_bb, EDGE_FALLTHRU);
+ e_dj->probability = REG_BR_PROB_BASE;
+ e_dj->count = count;
+
+ e_ij->count = all - count;
+ }
+ e_ij->probability = REG_BR_PROB_BASE;
}
- if (region >= 0 && stmt_could_throw_p (stmt))
+ /* Insert PHI node for the call result if necessary. */
+ if (gimple_call_lhs (icall_stmt)
+ && TREE_CODE (gimple_call_lhs (icall_stmt)) == SSA_NAME
+ && (dflags & ECF_NORETURN) == 0)
{
- gimple_purge_dead_eh_edges (bb4);
- make_eh_edges (stmt);
+ tree result = gimple_call_lhs (icall_stmt);
+ gimple phi = create_phi_node (result, join_bb);
+ gimple_call_set_lhs (icall_stmt,
+ duplicate_ssa_name (result, icall_stmt));
+ add_phi_arg (phi, gimple_call_lhs (icall_stmt), e_ij, UNKNOWN_LOCATION);
+ gimple_call_set_lhs (dcall_stmt,
+ duplicate_ssa_name (result, dcall_stmt));
+ add_phi_arg (phi, gimple_call_lhs (dcall_stmt), e_dj, UNKNOWN_LOCATION);
}
- return stmt1;
+ /* Build an EH edge for the direct call if necessary. */
+ lp_nr = lookup_stmt_eh_lp (icall_stmt);
+ if (lp_nr > 0 && stmt_could_throw_p (dcall_stmt))
+ {
+ add_stmt_to_eh_lp (dcall_stmt, lp_nr);
+ }
+
+ FOR_EACH_EDGE (e_eh, ei, icall_bb->succs)
+ if (e_eh->flags & (EDGE_EH | EDGE_ABNORMAL))
+ {
+ e = make_edge (dcall_bb, e_eh->dest, e_eh->flags);
+ for (psi = gsi_start_phis (e_eh->dest);
+ !gsi_end_p (psi); gsi_next (&psi))
+ {
+ gimple phi = gsi_stmt (psi);
+ SET_USE (PHI_ARG_DEF_PTR_FROM_EDGE (phi, e),
+ PHI_ARG_DEF_FROM_EDGE (phi, e_eh));
+ }
+ }
+ return dcall_stmt;
}
/*
*/
static bool
-gimple_ic_transform (gimple stmt)
+gimple_ic_transform (gimple_stmt_iterator *gsi)
{
+ gimple stmt = gsi_stmt (*gsi);
histogram_value histogram;
gcov_type val, count, all, bb_all;
- gcov_type prob;
- tree callee;
- gimple modify;
struct cgraph_node *direct_call;
-
+
if (gimple_code (stmt) != GIMPLE_CALL)
return false;
- callee = gimple_call_fn (stmt);
+ if (gimple_call_fndecl (stmt) != NULL_TREE)
+ return false;
- if (TREE_CODE (callee) == FUNCTION_DECL)
+ if (gimple_call_internal_p (stmt))
return false;
histogram = gimple_histogram_value_of_type (cfun, stmt, HIST_TYPE_INDIR_CALL);
val = histogram->hvalue.counters [0];
count = histogram->hvalue.counters [1];
all = histogram->hvalue.counters [2];
- gimple_remove_histogram_value (cfun, stmt, histogram);
-
- if (4 * count <= 3 * all)
- return false;
bb_all = gimple_bb (stmt)->count;
- /* The order of CHECK_COUNTER calls is important -
+ /* The order of CHECK_COUNTER calls is important -
since check_counter can correct the third parameter
and we want to make count <= all <= bb_all. */
if ( check_counter (stmt, "ic", &all, &bb_all, bb_all)
|| check_counter (stmt, "ic", &count, &all, all))
+ {
+ gimple_remove_histogram_value (cfun, stmt, histogram);
+ return false;
+ }
+
+ if (4 * count <= 3 * all)
return false;
- if (all > 0)
- prob = (count * REG_BR_PROB_BASE + all / 2) / all;
- else
- prob = 0;
- direct_call = find_func_by_pid ((int)val);
+ direct_call = find_func_by_profile_id ((int)val);
if (direct_call == NULL)
- return false;
+ {
+ if (val)
+ {
+ if (dump_file)
+ {
+ fprintf (dump_file, "Indirect call -> direct call from other module");
+ print_generic_expr (dump_file, gimple_call_fn (stmt), TDF_SLIM);
+ fprintf (dump_file, "=> %i (will resolve only with LTO)\n", (int)val);
+ }
+ }
+ return false;
+ }
- modify = gimple_ic (stmt, stmt, direct_call, prob, count, all);
+ if (!check_ic_target (stmt, direct_call))
+ {
+ if (dump_file)
+ {
+ fprintf (dump_file, "Indirect call -> direct call ");
+ print_generic_expr (dump_file, gimple_call_fn (stmt), TDF_SLIM);
+ fprintf (dump_file, "=> ");
+ print_generic_expr (dump_file, direct_call->decl, TDF_SLIM);
+ fprintf (dump_file, " transformation skipped because of type mismatch");
+ print_gimple_stmt (dump_file, stmt, 0, TDF_SLIM);
+ }
+ gimple_remove_histogram_value (cfun, stmt, histogram);
+ return false;
+ }
if (dump_file)
{
print_generic_expr (dump_file, gimple_call_fn (stmt), TDF_SLIM);
fprintf (dump_file, "=> ");
print_generic_expr (dump_file, direct_call->decl, TDF_SLIM);
- fprintf (dump_file, " transformation on insn ");
+ fprintf (dump_file, " transformation on insn postponned to ipa-profile");
print_gimple_stmt (dump_file, stmt, 0, TDF_SLIM);
- fprintf (dump_file, " to ");
- print_gimple_stmt (dump_file, modify, 0, TDF_SLIM);
fprintf (dump_file, "hist->count "HOST_WIDEST_INT_PRINT_DEC
" hist->all "HOST_WIDEST_INT_PRINT_DEC"\n", count, all);
}
return true;
}
-/* Return true if the stringop CALL with FNDECL shall be profiled. */
+/* Return true if the stringop CALL with FNDECL shall be profiled.
+ SIZE_ARG be set to the argument index for the size of the string
+ operation.
+*/
static bool
-interesting_stringop_to_profile_p (tree fndecl, gimple call)
+interesting_stringop_to_profile_p (tree fndecl, gimple call, int *size_arg)
{
enum built_in_function fcode = DECL_FUNCTION_CODE (fndecl);
{
case BUILT_IN_MEMCPY:
case BUILT_IN_MEMPCPY:
+ *size_arg = 2;
return validate_gimple_arglist (call, POINTER_TYPE, POINTER_TYPE,
INTEGER_TYPE, VOID_TYPE);
case BUILT_IN_MEMSET:
+ *size_arg = 2;
return validate_gimple_arglist (call, POINTER_TYPE, INTEGER_TYPE,
INTEGER_TYPE, VOID_TYPE);
case BUILT_IN_BZERO:
+ *size_arg = 1;
return validate_gimple_arglist (call, POINTER_TYPE, INTEGER_TYPE,
VOID_TYPE);
default:
}
}
-/* Convert stringop (..., size)
- into
- if (size == VALUE)
- stringop (...., VALUE);
+/* Convert stringop (..., vcall_size)
+ into
+ if (vcall_size == icall_size)
+ stringop (..., icall_size);
else
- stringop (...., size);
- assuming constant propagation of VALUE will happen later.
-*/
+ stringop (..., vcall_size);
+ assuming we'll propagate a true constant into ICALL_SIZE later. */
+
static void
-gimple_stringop_fixed_value (gimple stmt, tree value, int prob, gcov_type count,
- gcov_type all)
+gimple_stringop_fixed_value (gimple vcall_stmt, tree icall_size, int prob,
+ gcov_type count, gcov_type all)
{
- gimple stmt1, stmt2, stmt3;
- tree tmp1, tmpv;
- gimple bb1end, bb2end;
- basic_block bb, bb2, bb3, bb4;
- edge e12, e13, e23, e24, e34;
+ gimple tmp_stmt, cond_stmt, icall_stmt;
+ tree tmp0, tmp1, vcall_size, optype;
+ basic_block cond_bb, icall_bb, vcall_bb, join_bb;
+ edge e_ci, e_cv, e_iv, e_ij, e_vj;
gimple_stmt_iterator gsi;
- tree blck_size = gimple_call_arg (stmt, 2);
- tree optype = TREE_TYPE (blck_size);
- int region;
+ tree fndecl;
+ int size_arg;
- bb = gimple_bb (stmt);
- gsi = gsi_for_stmt (stmt);
+ fndecl = gimple_call_fndecl (vcall_stmt);
+ if (!interesting_stringop_to_profile_p (fndecl, vcall_stmt, &size_arg))
+ gcc_unreachable ();
- if (gsi_end_p (gsi))
- {
- edge_iterator ei;
- for (ei = ei_start (bb->succs); (e34 = ei_safe_edge (ei)); )
- if (!e34->flags & EDGE_ABNORMAL)
- break;
- }
- else
- {
- e34 = split_block (bb, stmt);
- gsi = gsi_for_stmt (stmt);
- }
- bb4 = e34->dest;
+ cond_bb = gimple_bb (vcall_stmt);
+ gsi = gsi_for_stmt (vcall_stmt);
- tmpv = create_tmp_var (optype, "PROF");
- tmp1 = create_tmp_var (optype, "PROF");
- stmt1 = gimple_build_assign (tmpv, fold_convert (optype, value));
- stmt2 = gimple_build_assign (tmp1, blck_size);
- stmt3 = gimple_build_cond (NE_EXPR, tmp1, tmpv, NULL_TREE, NULL_TREE);
- gsi_insert_before (&gsi, stmt1, GSI_SAME_STMT);
- gsi_insert_before (&gsi, stmt2, GSI_SAME_STMT);
- gsi_insert_before (&gsi, stmt3, GSI_SAME_STMT);
- bb1end = stmt3;
+ vcall_size = gimple_call_arg (vcall_stmt, size_arg);
+ optype = TREE_TYPE (vcall_size);
- stmt1 = gimple_copy (stmt);
- gimple_call_set_arg (stmt1, 2, value);
- gsi_insert_before (&gsi, stmt1, GSI_SAME_STMT);
- region = lookup_stmt_eh_region (stmt);
- if (region >= 0)
- add_stmt_to_eh_region (stmt1, region);
- bb2end = stmt1;
+ tmp0 = make_temp_ssa_name (optype, NULL, "PROF");
+ tmp1 = make_temp_ssa_name (optype, NULL, "PROF");
+ tmp_stmt = gimple_build_assign (tmp0, fold_convert (optype, icall_size));
+ gsi_insert_before (&gsi, tmp_stmt, GSI_SAME_STMT);
+
+ tmp_stmt = gimple_build_assign (tmp1, vcall_size);
+ gsi_insert_before (&gsi, tmp_stmt, GSI_SAME_STMT);
+
+ cond_stmt = gimple_build_cond (EQ_EXPR, tmp1, tmp0, NULL_TREE, NULL_TREE);
+ gsi_insert_before (&gsi, cond_stmt, GSI_SAME_STMT);
+
+ gimple_set_vdef (vcall_stmt, NULL);
+ gimple_set_vuse (vcall_stmt, NULL);
+ update_stmt (vcall_stmt);
+ icall_stmt = gimple_copy (vcall_stmt);
+ gimple_call_set_arg (icall_stmt, size_arg, icall_size);
+ gsi_insert_before (&gsi, icall_stmt, GSI_SAME_STMT);
/* Fix CFG. */
- /* Edge e23 connects bb2 to bb3, etc. */
- e12 = split_block (bb, bb1end);
- bb2 = e12->dest;
- bb2->count = count;
- e23 = split_block (bb2, bb2end);
- bb3 = e23->dest;
- bb3->count = all - count;
+ /* Edge e_ci connects cond_bb to icall_bb, etc. */
+ e_ci = split_block (cond_bb, cond_stmt);
+ icall_bb = e_ci->dest;
+ icall_bb->count = count;
- e12->flags &= ~EDGE_FALLTHRU;
- e12->flags |= EDGE_FALSE_VALUE;
- e12->probability = prob;
- e12->count = count;
+ e_iv = split_block (icall_bb, icall_stmt);
+ vcall_bb = e_iv->dest;
+ vcall_bb->count = all - count;
- e13 = make_edge (bb, bb3, EDGE_TRUE_VALUE);
- e13->probability = REG_BR_PROB_BASE - prob;
- e13->count = all - count;
+ e_vj = split_block (vcall_bb, vcall_stmt);
+ join_bb = e_vj->dest;
+ join_bb->count = all;
- remove_edge (e23);
-
- e24 = make_edge (bb2, bb4, EDGE_FALLTHRU);
- e24->probability = REG_BR_PROB_BASE;
- e24->count = count;
+ e_ci->flags = (e_ci->flags & ~EDGE_FALLTHRU) | EDGE_TRUE_VALUE;
+ e_ci->probability = prob;
+ e_ci->count = count;
- e34->probability = REG_BR_PROB_BASE;
- e34->count = all - count;
+ e_cv = make_edge (cond_bb, vcall_bb, EDGE_FALSE_VALUE);
+ e_cv->probability = REG_BR_PROB_BASE - prob;
+ e_cv->count = all - count;
+
+ remove_edge (e_iv);
+
+ e_ij = make_edge (icall_bb, join_bb, EDGE_FALLTHRU);
+ e_ij->probability = REG_BR_PROB_BASE;
+ e_ij->count = count;
+
+ e_vj->probability = REG_BR_PROB_BASE;
+ e_vj->count = all - count;
+
+ /* Insert PHI node for the call result if necessary. */
+ if (gimple_call_lhs (vcall_stmt)
+ && TREE_CODE (gimple_call_lhs (vcall_stmt)) == SSA_NAME)
+ {
+ tree result = gimple_call_lhs (vcall_stmt);
+ gimple phi = create_phi_node (result, join_bb);
+ gimple_call_set_lhs (vcall_stmt,
+ duplicate_ssa_name (result, vcall_stmt));
+ add_phi_arg (phi, gimple_call_lhs (vcall_stmt), e_vj, UNKNOWN_LOCATION);
+ gimple_call_set_lhs (icall_stmt,
+ duplicate_ssa_name (result, icall_stmt));
+ add_phi_arg (phi, gimple_call_lhs (icall_stmt), e_ij, UNKNOWN_LOCATION);
+ }
+
+ /* Because these are all string op builtins, they're all nothrow. */
+ gcc_assert (!stmt_could_throw_p (vcall_stmt));
+ gcc_assert (!stmt_could_throw_p (icall_stmt));
}
/* Find values inside STMT for that we want to measure histograms for
enum built_in_function fcode;
histogram_value histogram;
gcov_type count, all, val;
- tree value;
tree dest, src;
unsigned int dest_align, src_align;
gcov_type prob;
tree tree_val;
+ int size_arg;
if (gimple_code (stmt) != GIMPLE_CALL)
return false;
if (!fndecl)
return false;
fcode = DECL_FUNCTION_CODE (fndecl);
- if (!interesting_stringop_to_profile_p (fndecl, stmt))
+ if (!interesting_stringop_to_profile_p (fndecl, stmt, &size_arg))
return false;
- if (fcode == BUILT_IN_BZERO)
- blck_size = gimple_call_arg (stmt, 1);
- else
- blck_size = gimple_call_arg (stmt, 2);
+ blck_size = gimple_call_arg (stmt, size_arg);
if (TREE_CODE (blck_size) == INTEGER_CST)
return false;
histogram = gimple_histogram_value_of_type (cfun, stmt, HIST_TYPE_SINGLE_VALUE);
if (!histogram)
return false;
- value = histogram->hvalue.value;
val = histogram->hvalue.counters[0];
count = histogram->hvalue.counters[1];
all = histogram->hvalue.counters[2];
/* We require that count is at least half of all; this means
that for the transformation to fire the value must be constant
at least 80% of time. */
- if ((6 * count / 5) < all || !maybe_hot_bb_p (gimple_bb (stmt)))
+ if ((6 * count / 5) < all || optimize_bb_for_size_p (gimple_bb (stmt)))
return false;
if (check_counter (stmt, "value", &count, &all, gimple_bb (stmt)->count))
return false;
if (all > 0)
- prob = (count * REG_BR_PROB_BASE + all / 2) / all;
+ prob = GCOV_COMPUTE_SCALE (count, all);
else
prob = 0;
dest = gimple_call_arg (stmt, 0);
- dest_align = get_pointer_alignment (dest, BIGGEST_ALIGNMENT);
+ dest_align = get_pointer_alignment (dest);
switch (fcode)
{
case BUILT_IN_MEMCPY:
case BUILT_IN_MEMPCPY:
src = gimple_call_arg (stmt, 1);
- src_align = get_pointer_alignment (src, BIGGEST_ALIGNMENT);
+ src_align = get_pointer_alignment (src);
if (!can_move_by_pieces (val, MIN (dest_align, src_align)))
return false;
break;
print_gimple_stmt (dump_file, stmt, 0, TDF_SLIM);
}
gimple_stringop_fixed_value (stmt, tree_val, prob, count, all);
-
+
return true;
}
}
}
-struct value_prof_hooks {
- /* Find list of values for which we want to measure histograms. */
- void (*find_values_to_profile) (histogram_values *);
-
- /* Identify and exploit properties of values that are hard to analyze
- statically. See value-prof.c for more detail. */
- bool (*value_profile_transformations) (void);
-};
\f
/* Find values inside STMT for that we want to measure histograms for
division/modulo optimization. */
divisor = gimple_assign_rhs2 (stmt);
op0 = gimple_assign_rhs1 (stmt);
- VEC_reserve (histogram_value, heap, *values, 3);
+ values->reserve (3);
- if (is_gimple_reg (divisor))
+ if (TREE_CODE (divisor) == SSA_NAME)
/* Check for the case where the divisor is the same value most
of the time. */
- VEC_quick_push (histogram_value, *values,
- gimple_alloc_histogram_value (cfun,
+ values->quick_push (gimple_alloc_histogram_value (cfun,
HIST_TYPE_SINGLE_VALUE,
stmt, divisor));
{
tree val;
/* Check for a special case where the divisor is power of 2. */
- VEC_quick_push (histogram_value, *values,
- gimple_alloc_histogram_value (cfun, HIST_TYPE_POW2,
- stmt, divisor));
+ values->quick_push (gimple_alloc_histogram_value (cfun,
+ HIST_TYPE_POW2,
+ stmt, divisor));
val = build2 (TRUNC_DIV_EXPR, type, op0, divisor);
hist = gimple_alloc_histogram_value (cfun, HIST_TYPE_INTERVAL,
stmt, val);
hist->hdata.intvl.int_start = 0;
hist->hdata.intvl.steps = 2;
- VEC_quick_push (histogram_value, *values, hist);
+ values->quick_push (hist);
}
return;
}
}
-/* Find calls inside STMT for that we want to measure histograms for
- indirect/virtual call optimization. */
+/* Find calls inside STMT for that we want to measure histograms for
+ indirect/virtual call optimization. */
static void
gimple_indirect_call_to_profile (gimple stmt, histogram_values *values)
tree callee;
if (gimple_code (stmt) != GIMPLE_CALL
+ || gimple_call_internal_p (stmt)
|| gimple_call_fndecl (stmt) != NULL_TREE)
return;
callee = gimple_call_fn (stmt);
- VEC_reserve (histogram_value, heap, *values, 3);
+ values->reserve (3);
- VEC_quick_push (histogram_value, *values,
- gimple_alloc_histogram_value (cfun, HIST_TYPE_INDIR_CALL,
- stmt, callee));
+ values->quick_push (gimple_alloc_histogram_value (cfun, HIST_TYPE_INDIR_CALL,
+ stmt, callee));
return;
}
tree fndecl;
tree blck_size;
tree dest;
- enum built_in_function fcode;
+ int size_arg;
if (gimple_code (stmt) != GIMPLE_CALL)
return;
fndecl = gimple_call_fndecl (stmt);
if (!fndecl)
return;
- fcode = DECL_FUNCTION_CODE (fndecl);
- if (!interesting_stringop_to_profile_p (fndecl, stmt))
+ if (!interesting_stringop_to_profile_p (fndecl, stmt, &size_arg))
return;
dest = gimple_call_arg (stmt, 0);
- if (fcode == BUILT_IN_BZERO)
- blck_size = gimple_call_arg (stmt, 1);
- else
- blck_size = gimple_call_arg (stmt, 2);
+ blck_size = gimple_call_arg (stmt, size_arg);
if (TREE_CODE (blck_size) != INTEGER_CST)
{
- VEC_safe_push (histogram_value, heap, *values,
- gimple_alloc_histogram_value (cfun, HIST_TYPE_SINGLE_VALUE,
- stmt, blck_size));
- VEC_safe_push (histogram_value, heap, *values,
- gimple_alloc_histogram_value (cfun, HIST_TYPE_AVERAGE,
- stmt, blck_size));
+ values->safe_push (gimple_alloc_histogram_value (cfun,
+ HIST_TYPE_SINGLE_VALUE,
+ stmt, blck_size));
+ values->safe_push (gimple_alloc_histogram_value (cfun, HIST_TYPE_AVERAGE,
+ stmt, blck_size));
}
if (TREE_CODE (blck_size) != INTEGER_CST)
- VEC_safe_push (histogram_value, heap, *values,
- gimple_alloc_histogram_value (cfun, HIST_TYPE_IOR,
- stmt, dest));
+ values->safe_push (gimple_alloc_histogram_value (cfun, HIST_TYPE_IOR,
+ stmt, dest));
}
/* Find values inside STMT for that we want to measure histograms and adds
static void
gimple_values_to_profile (gimple stmt, histogram_values *values)
{
- if (flag_value_profile_transformations)
- {
- gimple_divmod_values_to_profile (stmt, values);
- gimple_stringops_values_to_profile (stmt, values);
- gimple_indirect_call_to_profile (stmt, values);
- }
+ gimple_divmod_values_to_profile (stmt, values);
+ gimple_stringops_values_to_profile (stmt, values);
+ gimple_indirect_call_to_profile (stmt, values);
}
-static void
+void
gimple_find_values_to_profile (histogram_values *values)
{
basic_block bb;
gimple_stmt_iterator gsi;
unsigned i;
histogram_value hist = NULL;
+ values->create (0);
- *values = NULL;
FOR_EACH_BB (bb)
for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
gimple_values_to_profile (gsi_stmt (gsi), values);
-
- for (i = 0; VEC_iterate (histogram_value, *values, i, hist); i++)
+
+ values->safe_push (gimple_alloc_histogram_value (cfun, HIST_TYPE_TIME_PROFILE, 0, 0));
+
+ FOR_EACH_VEC_ELT (*values, i, hist)
{
switch (hist->type)
{
hist->n_counters = 3;
break;
+ case HIST_TYPE_TIME_PROFILE:
+ hist->n_counters = 1;
+ break;
+
case HIST_TYPE_AVERAGE:
hist->n_counters = 2;
break;
}
}
}
-
-static struct value_prof_hooks gimple_value_prof_hooks = {
- gimple_find_values_to_profile,
- gimple_value_profile_transformations
-};
-
-void
-gimple_register_value_prof_hooks (void)
-{
- gcc_assert (current_ir_type () == IR_GIMPLE);
- value_prof_hooks = &gimple_value_prof_hooks;
-}
-\f
-/* IR-independent entry points. */
-void
-find_values_to_profile (histogram_values *values)
-{
- (value_prof_hooks->find_values_to_profile) (values);
-}
-
-bool
-value_profile_transformations (void)
-{
- return (value_prof_hooks->value_profile_transformations) ();
-}
-\f