gcc/0025-AutoPrefetch-Support-cache-misses-profile.patch

From 26e4ba63112f55c27b7dd3d5f8c4497ef9a2f459 Mon Sep 17 00:00:00 2001
From: benniaobufeijiushiji <linda7@huawei.com>
Date: Thu, 6 Jan 2022 15:33:29 +0800
Subject: [PATCH 25/28] [AutoPrefetch] Support cache misses profile

Add pass ex-afdo after pass afdo in auto-profile.c.
Add flag -fcache-misses-profile.
Read profile of different types of perf events and build maps for
function and gimple location to its count of each perf event.
Currently, instruction execution and cahce misses are supported.
---
 gcc/auto-profile.c | 415 +++++++++++++++++++++++++++++++++++++++++++++
 gcc/auto-profile.h |  28 +++
 gcc/common.opt     |  14 ++
 gcc/opts.c         |  26 +++
 gcc/passes.def     |   1 +
 gcc/timevar.def    |   1 +
 gcc/toplev.c       |   6 +
 gcc/tree-pass.h    |   2 +
 8 files changed, 493 insertions(+)

diff --git a/gcc/auto-profile.c b/gcc/auto-profile.c
index 7d09887c9..aced8fca5 100644
--- a/gcc/auto-profile.c
+++ b/gcc/auto-profile.c
@@ -49,6 +49,9 @@ along with GCC; see the file COPYING3.  If not see
 #include "auto-profile.h"
 #include "tree-pretty-print.h"
 #include "gimple-pretty-print.h"
+#include <map>
+#include <vector>
+#include <algorithm>
 
 /* The following routines implements AutoFDO optimization.
 
@@ -95,6 +98,7 @@ along with GCC; see the file COPYING3.  If not see
 */
 
 #define DEFAULT_AUTO_PROFILE_FILE "fbdata.afdo"
+#define DEFAULT_CACHE_MISSES_PROFILE_FILE "cmsdata.gcov"
 #define AUTO_PROFILE_VERSION 1
 
 namespace autofdo
@@ -117,6 +121,14 @@ private:
   bool annotated_;
 };
 
+/* pair <func_decl, count>  */
+static bool
+event_count_cmp (std::pair<unsigned, gcov_type> &a,
+		 std::pair<unsigned, gcov_type> &b)
+{
+  return a.second > b.second;
+}
+
 /* Represent a source location: (function_decl, lineno).  */
 typedef std::pair<tree, unsigned> decl_lineno;
 
@@ -338,6 +350,206 @@ static autofdo_source_profile *afdo_source_profile;
 /* gcov_summary structure to store the profile_info.  */
 static gcov_summary *afdo_profile_info;
 
+/* Check opts->x_flags and put file name into EVENT_FILES.  */
+
+static bool
+get_all_profile_names (const char **event_files)
+{
+  if (!(flag_auto_profile || flag_cache_misses_profile))
+    {
+      return false;
+    }
+
+  event_files[INST_EXEC] = auto_profile_file;
+
+  if (cache_misses_profile_file == NULL)
+    {
+      cache_misses_profile_file = DEFAULT_CACHE_MISSES_PROFILE_FILE;
+    }
+  event_files[CACHE_MISSES] = cache_misses_profile_file;
+
+  return true;
+}
+
+static void read_profile (void);
+
+/* Maintain multiple profile data of different events with event_loc_count_map
+   and event_func_count_map.  */
+
+class extend_auto_profile
+{
+public:
+  bool auto_profile_exist (enum event_type type);
+  gcov_type get_loc_count (location_t, event_type);
+  gcov_type get_func_count (unsigned, event_type);
+  struct rank_info get_func_rank (unsigned, enum event_type);
+  /* There should be only one instance of class EXTEND_AUTO_PROFILE.  */
+  static extend_auto_profile *create ()
+    {
+      extend_auto_profile *map = new extend_auto_profile ();
+      if (map->read ())
+	{
+	  return map;
+	}
+      delete map;
+      return NULL;
+    }
+private:
+  /* Basic maps of extend_auto_profile.  */
+  typedef std::map<location_t, gcov_type> loc_count_map;
+  typedef std::map<unsigned, gcov_type> func_count_map;
+
+  /* Map of function_uid to its descending order rank of counts.  */
+  typedef std::map<unsigned, unsigned> rank_map;
+
+  /* Mapping hardware events to corresponding basic maps.  */
+  typedef std::map<event_type, loc_count_map> event_loc_count_map;
+  typedef std::map<event_type, func_count_map> event_func_count_map;
+  typedef std::map<event_type, rank_map> event_rank_map;
+
+  extend_auto_profile () {}
+  bool read ();
+  void set_loc_count ();
+  void process_extend_source_profile ();
+  void read_extend_afdo_file (const char*, event_type);
+  void rank_all_func ();
+  void dump_event ();
+  event_loc_count_map event_loc_map;
+  event_func_count_map event_func_map;
+  event_rank_map func_rank;
+  event_type profile_type;
+};
+
+/* Member functions for extend_auto_profile.  */
+
+bool
+extend_auto_profile::auto_profile_exist (enum event_type type)
+{
+  switch (type)
+    {
+      case INST_EXEC:
+	return event_func_map.count (INST_EXEC) != 0
+	       || event_loc_map.count (INST_EXEC) != 0;
+      case CACHE_MISSES:
+	return event_func_map.count (CACHE_MISSES) != 0
+	       || event_loc_map.count (CACHE_MISSES) != 0;
+      default:
+	  return false;
+    }
+}
+
+void
+extend_auto_profile::dump_event ()
+{
+  if (dump_file)
+    {
+      switch (profile_type)
+	{
+	  case INST_EXEC:
+	    fprintf (dump_file, "Processing event instruction execution.\n");
+	    break;
+	  case CACHE_MISSES:
+	    fprintf (dump_file, "Processing event cache misses.\n");
+	    break;
+	  default:
+	    break;
+	}
+    }
+}
+
+/* Return true if any profile data was read.  */
+
+bool
+extend_auto_profile::read ()
+{
+  const char *event_files[EVENT_NUMBER] = {NULL};
+  if (!get_all_profile_names (event_files))
+    {
+      return false;
+    }
+
+  /* Backup AFDO_STRING_TABLE and AFDO_SOURCE_PROFILE since we will create
+     new ones for each event_type.  */
+  autofdo::string_table *string_table_afdo = afdo_string_table;
+  autofdo::autofdo_source_profile *source_profile_afdo = afdo_source_profile;
+
+  for (unsigned i = 0; i < EVENT_NUMBER; i++)
+    {
+      if (event_files[i] == NULL)
+	{
+	  continue;
+	}
+      profile_type = (enum event_type) i;
+      dump_event ();
+      gcov_close ();
+      auto_profile_file = event_files[i];
+      read_profile ();
+      gcov_close ();
+
+      process_extend_source_profile ();
+
+      delete afdo_source_profile;
+      delete afdo_string_table;
+    }
+
+  /* Restore AFDO_STRING_TABLE and AFDO_SOURCE_PROFILE.  Function
+     END_AUTO_PROFILE will free them at the end of compilation.  */
+  afdo_string_table = string_table_afdo;
+  afdo_source_profile = source_profile_afdo;
+  return true;
+}
+
+/* Helper functions.  */
+
+gcov_type
+extend_auto_profile::get_loc_count (location_t loc, event_type type)
+{
+  event_loc_count_map::iterator event_iter = event_loc_map.find (type);
+  if (event_iter != event_loc_map.end ())
+    {
+      loc_count_map::iterator loc_iter = event_iter->second.find (loc);
+      if (loc_iter != event_iter->second.end ())
+	{
+	  return loc_iter->second;
+	}
+    }
+  return 0;
+}
+
+struct rank_info
+extend_auto_profile::get_func_rank (unsigned decl_uid, enum event_type type)
+{
+  struct rank_info info = {0, 0};
+  event_rank_map::iterator event_iter = func_rank.find (type);
+  if (event_iter != func_rank.end ())
+    {
+      rank_map::iterator func_iter = event_iter->second.find (decl_uid);
+      if (func_iter != event_iter->second.end ())
+	{
+	  info.rank = func_iter->second;
+	  info.total = event_iter->second.size ();
+	}
+    }
+  return info;
+}
+
+gcov_type
+extend_auto_profile::get_func_count (unsigned decl_uid, event_type type)
+{
+  event_func_count_map::iterator event_iter = event_func_map.find (type);
+  if (event_iter != event_func_map.end ())
+    {
+      func_count_map::iterator func_iter = event_iter->second.find (decl_uid);
+      if (func_iter != event_iter->second.end ())
+	{
+	  return func_iter->second;
+	}
+    }
+  return 0;
+}
+
+static extend_auto_profile *extend_profile;
+
 /* Helper functions.  */
 
 /* Return the original name of NAME: strip the suffix that starts
@@ -1654,6 +1866,131 @@ auto_profile (void)
 
   return TODO_rebuild_cgraph_edges;
 }
+
+void
+extend_auto_profile::rank_all_func ()
+{
+  std::vector<std::pair<unsigned, gcov_type> > func_sorted;
+  event_func_count_map::iterator event_iter
+				 = event_func_map.find (profile_type);
+  if (event_iter != event_func_map.end ())
+    {
+      func_count_map::iterator func_iter;
+      for (func_iter = event_iter->second.begin ();
+	   func_iter != event_iter->second.end (); func_iter++)
+	{
+	  func_sorted.push_back (std::make_pair (func_iter->first,
+						 func_iter->second));
+	}
+
+      std::sort (func_sorted.begin (), func_sorted.end (), event_count_cmp);
+
+      for (unsigned i = 0; i < func_sorted.size (); ++i)
+	{
+	  func_rank[profile_type][func_sorted[i].first] = i + 1;
+	}
+    }
+}
+
+/* Iterate stmts in cfun and maintain its count to EVENT_LOC_MAP.  */
+
+void
+extend_auto_profile::set_loc_count ()
+{
+  basic_block bb;
+  FOR_EACH_BB_FN (bb, cfun)
+    {
+      gimple_stmt_iterator gsi;
+      for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
+	{
+	  count_info info;
+	  gimple *stmt = gsi_stmt (gsi);
+	  if (gimple_clobber_p (stmt) || is_gimple_debug (stmt))
+	    {
+	      continue;
+	    }
+	  if (afdo_source_profile->get_count_info (stmt, &info))
+	    {
+	      location_t loc = gimple_location (stmt);
+	      event_loc_map[profile_type][loc] += info.count;
+	      if (dump_file && (dump_flags & TDF_DETAILS))
+		{
+		  fprintf (dump_file, "stmt ");
+		  print_gimple_stmt (dump_file, stmt, 0, TDF_SLIM);
+		  fprintf (dump_file, "counts %ld\n",
+			   event_loc_map[profile_type][loc]);
+		}
+	    }
+	}
+    }
+}
+
+/* Process data in extend_auto_source_profile, save them into two maps.
+   1. gimple_location to count.
+   2. function_index to count.  */
+void
+extend_auto_profile::process_extend_source_profile ()
+{
+  struct cgraph_node *node;
+  if (symtab->state == FINISHED)
+    {
+      return;
+    }
+  FOR_EACH_FUNCTION (node)
+    {
+      if (!gimple_has_body_p (node->decl) || node->inlined_to)
+	{
+	  continue;
+	}
+
+      /* Don't profile functions produced for builtin stuff.  */
+      if (DECL_SOURCE_LOCATION (node->decl) == BUILTINS_LOCATION)
+	{
+	  continue;
+	}
+
+      function *fn = DECL_STRUCT_FUNCTION (node->decl);
+      push_cfun (fn);
+
+      const function_instance *s
+      = afdo_source_profile->get_function_instance_by_decl (
+	  current_function_decl);
+
+      if (s == NULL)
+	{
+	  pop_cfun ();
+	  continue;
+	}
+      unsigned int decl_uid = DECL_UID (current_function_decl);
+      gcov_type count = s->total_count ();
+      if (dump_file)
+	{
+	  fprintf (dump_file, "Extend auto-profile for function %s.\n",
+			       node->dump_name ());
+	}
+      event_func_map[profile_type][decl_uid] += count;
+      set_loc_count ();
+      pop_cfun ();
+    }
+  rank_all_func ();
+}
+
+/* Main entry of extend_auto_profile.  */
+
+static void
+extend_source_profile ()
+{
+  extend_profile = autofdo::extend_auto_profile::create ();
+  if (dump_file)
+    {
+      if (extend_profile == NULL)
+	{
+	  fprintf (dump_file, "No profile file is found.\n");
+	  return;
+	}
+      fprintf (dump_file, "Extend profile info generated.\n");
+    }
+}
 } /* namespace autofdo.  */
 
 /* Read the profile from the profile data file.  */
@@ -1682,6 +2019,42 @@ end_auto_profile (void)
   profile_info = NULL;
 }
 
+/* Extern function to get profile info in other passes.  */
+
+bool
+profile_exist (enum event_type type)
+{
+  return autofdo::extend_profile != NULL
+	 && autofdo::extend_profile->auto_profile_exist (type);
+}
+
+gcov_type
+event_get_loc_count (location_t loc, event_type type)
+{
+  return autofdo::extend_profile->get_loc_count (loc, type);
+}
+
+gcov_type
+event_get_func_count (unsigned decl_uid, event_type type)
+{
+  return autofdo::extend_profile->get_func_count (decl_uid, type);
+}
+
+struct rank_info
+event_get_func_rank (unsigned decl_uid, enum event_type type)
+{
+  return autofdo::extend_profile->get_func_rank (decl_uid, type);
+}
+
+void
+free_extend_profile_info ()
+{
+  if (autofdo::extend_profile != NULL)
+    {
+      delete autofdo::extend_profile;
+    }
+}
+
 /* Returns TRUE if EDGE is hot enough to be inlined early.  */
 
 bool
@@ -1743,8 +2116,50 @@ public:
 
 } // anon namespace
 
+namespace
+{
+const pass_data pass_data_ipa_extend_auto_profile =
+{
+  SIMPLE_IPA_PASS, /* type */
+  "ex-afdo", /* name */
+  OPTGROUP_NONE, /* optinfo_flags */
+  TV_IPA_EXTEND_AUTO_PROFILE, /* tv_id */
+  0, /* properties_required */
+  0, /* properties_provided */
+  0, /* properties_destroyed */
+  0, /* todo_flags_start */
+  0, /* todo_flags_finish */
+};
+
+class pass_ipa_extend_auto_profile : public simple_ipa_opt_pass
+{
+public:
+  pass_ipa_extend_auto_profile (gcc::context *ctxt)
+    : simple_ipa_opt_pass (pass_data_ipa_extend_auto_profile, ctxt)
+  {}
+
+  /* opt_pass methods: */
+  virtual bool gate (function *) {return (flag_ipa_extend_auto_profile > 0);}
+  virtual unsigned int execute (function *);
+
+};
+
+unsigned int
+pass_ipa_extend_auto_profile::execute (function *fun)
+{
+  autofdo::extend_source_profile ();
+  return 0;
+}
+} // anon namespace
+
 simple_ipa_opt_pass *
 make_pass_ipa_auto_profile (gcc::context *ctxt)
 {
   return new pass_ipa_auto_profile (ctxt);
 }
+
+simple_ipa_opt_pass *
+make_pass_ipa_extend_auto_profile (gcc::context *ctxt)
+{
+  return new pass_ipa_extend_auto_profile (ctxt);
+}
\ No newline at end of file
diff --git a/gcc/auto-profile.h b/gcc/auto-profile.h
index f5cff091d..230d7e68a 100644
--- a/gcc/auto-profile.h
+++ b/gcc/auto-profile.h
@@ -21,6 +21,13 @@ along with GCC; see the file COPYING3.  If not see
 #ifndef AUTO_PROFILE_H
 #define AUTO_PROFILE_H
 
+enum event_type
+{
+  INST_EXEC = 0,
+  CACHE_MISSES,
+  EVENT_NUMBER
+};
+
 /* Read, process, finalize AutoFDO data structures.  */
 extern void read_autofdo_file (void);
 extern void end_auto_profile (void);
@@ -28,4 +35,25 @@ extern void end_auto_profile (void);
 /* Returns TRUE if EDGE is hot enough to be inlined early.  */
 extern bool afdo_callsite_hot_enough_for_early_inline (struct cgraph_edge *);
 
+/* Chcek if profile exists before using this profile.  */
+extern bool profile_exist (enum event_type);
+
+/* Given func decl_uid or gimple location and event_type, return count.
+   Count is 0 if function or gimple is not sampled.  */
+extern gcov_type event_get_func_count (unsigned, enum event_type);
+extern gcov_type event_get_loc_count (location_t, enum event_type);
+
+struct rank_info
+{
+  unsigned total;
+  unsigned rank;
+};
+
+/* Given function decl_uid and event type, return rank_info.  Rank_info
+   is {0, 0} if function was not sampled.  */
+extern struct rank_info event_get_func_rank (unsigned, enum event_type);
+
+/* Free memory allocated by autofdo::extern_profile.  */
+extern void free_extend_profile_info ();
+
 #endif /* AUTO_PROFILE_H */
diff --git a/gcc/common.opt b/gcc/common.opt
index 73c24f28d..37cbbd8c0 100644
--- a/gcc/common.opt
+++ b/gcc/common.opt
@@ -1074,6 +1074,16 @@ Common Joined RejectNegative Var(auto_profile_file)
 Use sample profile information for call graph node weights. The profile
 file is specified in the argument.
 
+fcache-misses-profile
+Common Report Var(flag_cache_misses_profile)
+Use sample profile information for source code cache miss count.  The default
+profile file is cmsdata.gcov in `pwd`.
+
+fcache-misses-profile=
+Common Joined RejectNegative Var(cache_misses_profile_file)
+Use sample profile information for source code cache miss count.  The profile
+file is specified in the argument.
+
 ; -fcheck-bounds causes gcc to generate array bounds checks.
 ; For C, C++ and ObjC: defaults off.
 ; For Java: defaults to on.
@@ -1873,6 +1883,10 @@ fipa-struct-reorg
 Common Report Var(flag_ipa_struct_reorg) Init(0) Optimization
 Perform structure layout optimizations.
 
+fipa-extend-auto-profile
+Common Report Var(flag_ipa_extend_auto_profile)
+Use sample profile information for source code.
+
 fipa-vrp
 Common Report Var(flag_ipa_vrp) Optimization
 Perform IPA Value Range Propagation.
diff --git a/gcc/opts.c b/gcc/opts.c
index 6924a973a..642327296 100644
--- a/gcc/opts.c
+++ b/gcc/opts.c
@@ -1742,6 +1742,13 @@ enable_fdo_optimizations (struct gcc_options *opts,
   SET_OPTION_IF_UNSET (opts, opts_set, flag_tree_loop_distribution, value);
 }
 
+static void
+set_cache_misses_profile_params (struct gcc_options *opts,
+			       struct gcc_options *opts_set)
+{
+  SET_OPTION_IF_UNSET (opts, opts_set, flag_prefetch_loop_arrays, 1);
+}
+
 /* -f{,no-}sanitize{,-recover}= suboptions.  */
 const struct sanitizer_opts_s sanitizer_opts[] =
 {
@@ -2604,6 +2611,25 @@ common_handle_option (struct gcc_options *opts,
 			   param_early_inliner_max_iterations, 10);
       break;
 
+    case OPT_fipa_extend_auto_profile:
+      opts->x_flag_ipa_extend_auto_profile = opts->x_flag_cache_misses_profile
+					     ? true : value;
+      break;
+
+    case OPT_fcache_misses_profile_:
+      opts->x_cache_misses_profile_file = xstrdup (arg);
+      opts->x_flag_cache_misses_profile = true;
+      value = true;
+      /* No break here - do -fcache-misses-profile processing. */
+      /* FALLTHRU */
+    case OPT_fcache_misses_profile:
+      opts->x_flag_ipa_extend_auto_profile = value;
+      if (value)
+	{
+	  set_cache_misses_profile_params (opts, opts_set);
+	}
+      break;
+
     case OPT_fprofile_generate_:
       opts->x_profile_data_prefix = xstrdup (arg);
       value = true;
diff --git a/gcc/passes.def b/gcc/passes.def
index 63303ab65..e9c91d26e 100644
--- a/gcc/passes.def
+++ b/gcc/passes.def
@@ -133,6 +133,7 @@ along with GCC; see the file COPYING3.  If not see
 
   NEXT_PASS (pass_target_clone);
   NEXT_PASS (pass_ipa_auto_profile);
+  NEXT_PASS (pass_ipa_extend_auto_profile);
   NEXT_PASS (pass_ipa_tree_profile);
   PUSH_INSERT_PASSES_WITHIN (pass_ipa_tree_profile)
       NEXT_PASS (pass_feedback_split_functions);
diff --git a/gcc/timevar.def b/gcc/timevar.def
index ee25eccbb..e873747a8 100644
--- a/gcc/timevar.def
+++ b/gcc/timevar.def
@@ -82,6 +82,7 @@ DEFTIMEVAR (TV_IPA_FNSPLIT           , "ipa function splitting")
 DEFTIMEVAR (TV_IPA_COMDATS	     , "ipa comdats")
 DEFTIMEVAR (TV_IPA_REORDER_FIELDS    , "ipa struct reorder fields optimization")
 DEFTIMEVAR (TV_IPA_STRUCT_REORG	     , "ipa struct reorg optimization")
+DEFTIMEVAR (TV_IPA_EXTEND_AUTO_PROFILE, "ipa extend auto profile")
 DEFTIMEVAR (TV_IPA_OPT		     , "ipa various optimizations")
 DEFTIMEVAR (TV_IPA_LTO_DECOMPRESS    , "lto stream decompression")
 DEFTIMEVAR (TV_IPA_LTO_COMPRESS      , "lto stream compression")
diff --git a/gcc/toplev.c b/gcc/toplev.c
index eaed6f6c7..51e6bd400 100644
--- a/gcc/toplev.c
+++ b/gcc/toplev.c
@@ -577,6 +577,12 @@ compile_file (void)
       targetm.asm_out.output_ident (ident_str);
     }
 
+  /* Extend auto profile finalization.  */
+  if (flag_ipa_extend_auto_profile)
+    {
+      free_extend_profile_info ();
+    }
+
   /* Auto profile finalization. */
   if (flag_auto_profile)
     end_auto_profile ();
diff --git a/gcc/tree-pass.h b/gcc/tree-pass.h
index eb32c5d44..be6387768 100644
--- a/gcc/tree-pass.h
+++ b/gcc/tree-pass.h
@@ -511,6 +511,8 @@ extern ipa_opt_pass_d *make_pass_ipa_hsa (gcc::context *ctxt);
 extern ipa_opt_pass_d *make_pass_ipa_pure_const (gcc::context *ctxt);
 extern simple_ipa_opt_pass *make_pass_ipa_reorder_fields (gcc::context *ctxt);
 extern simple_ipa_opt_pass *make_pass_ipa_struct_reorg (gcc::context *ctxt);
+extern simple_ipa_opt_pass *make_pass_ipa_extend_auto_profile (gcc::context
+							       *ctxt);
 extern simple_ipa_opt_pass *make_pass_ipa_pta (gcc::context *ctxt);
 extern simple_ipa_opt_pass *make_pass_ipa_tm (gcc::context *ctxt);
 extern simple_ipa_opt_pass *make_pass_target_clone (gcc::context *ctxt);
-- 
2.27.0.windows.1