gcc/ipa-hsa.c

   1 /* Callgraph based analysis of static variables.
   2    Copyright (C) 2015-2016 Free Software Foundation, Inc.
   3    Contributed by Martin Liska <mliska@suse.cz>
   4
   5 This file is part of GCC.
   6
   7 GCC is free software; you can redistribute it and/or modify it under
   8 the terms of the GNU General Public License as published by the Free
   9 Software Foundation; either version 3, or (at your option) any later
  10 version.
  11
  12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
  13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
  14 FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  15 for more details.
  16
  17 You should have received a copy of the GNU General Public License
  18 along with GCC; see the file COPYING3.  If not see
  19 <http://www.gnu.org/licenses/>.  */
  20
  21 /* Interprocedural HSA pass is responsible for creation of HSA clones.
  22    For all these HSA clones, we emit HSAIL instructions and pass processing
  23    is terminated.  */
  24
  25 #include "config.h"
  26 #include "system.h"
  27 #include "coretypes.h"
  28 #include "tm.h"
  29 #include "is-a.h"
  30 #include "hash-set.h"
  31 #include "vec.h"
  32 #include "tree.h"
  33 #include "tree-pass.h"
  34 #include "function.h"
  35 #include "basic-block.h"
  36 #include "gimple.h"
  37 #include "dumpfile.h"
  38 #include "gimple-pretty-print.h"
  39 #include "tree-streamer.h"
  40 #include "stringpool.h"
  41 #include "cgraph.h"
  42 #include "print-tree.h"
  43 #include "symbol-summary.h"
  44 #include "hsa.h"
  45
  46 namespace {
  47
  48 /* If NODE is not versionable, warn about not emiting HSAIL and return false.
  49    Otherwise return true.  */
  50
  51 static bool
  52 check_warn_node_versionable (cgraph_node *node)
  53 {
  54   if (!node->local.versionable)
  55     {
  56       warning_at (EXPR_LOCATION (node->decl), OPT_Whsa,
  57                   "could not emit HSAIL for function %s: function cannot be "
  58                   "cloned", node->name ());
  59       return false;
  60     }
  61   return true;
  62 }
  63
  64 /* The function creates HSA clones for all functions that were either
  65    marked as HSA kernels or are callable HSA functions.  Apart from that,
  66    we redirect all edges that come from an HSA clone and end in another
  67    HSA clone to connect these two functions.  */
  68
  69 static unsigned int
  70 process_hsa_functions (void)
  71 {
  72   struct cgraph_node *node;
  73
  74   if (hsa_summaries == NULL)
  75     hsa_summaries = new hsa_summary_t (symtab);
  76
  77   FOR_EACH_DEFINED_FUNCTION (node)
  78     {
  79       hsa_function_summary *s = hsa_summaries->get (node);
  80
  81       /* A linked function is skipped.  */
  82       if (s->m_bound_function != NULL)
  83         continue;
  84
  85       if (s->m_kind != HSA_NONE)
  86         {
  87           if (!check_warn_node_versionable (node))
  88             continue;
  89           cgraph_node *clone
  90             = node->create_virtual_clone (vec <cgraph_edge *> (),
  91                                           NULL, NULL, "hsa");
  92           TREE_PUBLIC (clone->decl) = TREE_PUBLIC (node->decl);
  93           clone->externally_visible = node->externally_visible;
  94
  95           clone->force_output = true;
  96           hsa_summaries->link_functions (clone, node, s->m_kind, false);
  97
  98           if (dump_file)
  99             fprintf (dump_file, "Created a new HSA clone: %s, type: %s\n",
 100                      clone->name (),
 101                      s->m_kind == HSA_KERNEL ? "kernel" : "function");
 102         }
 103       else if (hsa_callable_function_p (node->decl))
 104         {
 105           if (!check_warn_node_versionable (node))
 106             continue;
 107           cgraph_node *clone
 108             = node->create_virtual_clone (vec <cgraph_edge *> (),
 109                                           NULL, NULL, "hsa");
 110           TREE_PUBLIC (clone->decl) = TREE_PUBLIC (node->decl);
 111           clone->externally_visible = node->externally_visible;
 112
 113           if (!cgraph_local_p (node))
 114             clone->force_output = true;
 115           hsa_summaries->link_functions (clone, node, HSA_FUNCTION, false);
 116
 117           if (dump_file)
 118             fprintf (dump_file, "Created a new HSA function clone: %s\n",
 119                      clone->name ());
 120         }
 121     }
 122
 123   /* Redirect all edges that are between HSA clones.  */
 124   FOR_EACH_DEFINED_FUNCTION (node)
 125     {
 126       cgraph_edge *e = node->callees;
 127
 128       while (e)
 129         {
 130           hsa_function_summary *src = hsa_summaries->get (node);
 131           if (src->m_kind != HSA_NONE && src->m_gpu_implementation_p)
 132             {
 133               hsa_function_summary *dst = hsa_summaries->get (e->callee);
 134               if (dst->m_kind != HSA_NONE && !dst->m_gpu_implementation_p)
 135                 {
 136                   e->redirect_callee (dst->m_bound_function);
 137                   if (dump_file)
 138                     fprintf (dump_file,
 139                              "Redirecting edge to HSA function: %s->%s\n",
 140                              xstrdup_for_dump (e->caller->name ()),
 141                              xstrdup_for_dump (e->callee->name ()));
 142                 }
 143             }
 144
 145           e = e->next_callee;
 146         }
 147     }
 148
 149   return 0;
 150 }
 151
 152 /* Iterate all HSA functions and stream out HSA function summary.  */
 153
 154 static void
 155 ipa_hsa_write_summary (void)
 156 {
 157   struct bitpack_d bp;
 158   struct cgraph_node *node;
 159   struct output_block *ob;
 160   unsigned int count = 0;
 161   lto_symtab_encoder_iterator lsei;
 162   lto_symtab_encoder_t encoder;
 163
 164   if (!hsa_summaries)
 165     return;
 166
 167   ob = create_output_block (LTO_section_ipa_hsa);
 168   encoder = ob->decl_state->symtab_node_encoder;
 169   ob->symbol = NULL;
 170   for (lsei = lsei_start_function_in_partition (encoder); !lsei_end_p (lsei);
 171        lsei_next_function_in_partition (&lsei))
 172     {
 173       node = lsei_cgraph_node (lsei);
 174       hsa_function_summary *s = hsa_summaries->get (node);
 175
 176       if (s->m_kind != HSA_NONE)
 177         count++;
 178     }
 179
 180   streamer_write_uhwi (ob, count);
 181
 182   /* Process all of the functions.  */
 183   for (lsei = lsei_start_function_in_partition (encoder); !lsei_end_p (lsei);
 184        lsei_next_function_in_partition (&lsei))
 185     {
 186       node = lsei_cgraph_node (lsei);
 187       hsa_function_summary *s = hsa_summaries->get (node);
 188
 189       if (s->m_kind != HSA_NONE)
 190         {
 191           encoder = ob->decl_state->symtab_node_encoder;
 192           int node_ref = lto_symtab_encoder_encode (encoder, node);
 193           streamer_write_uhwi (ob, node_ref);
 194
 195           bp = bitpack_create (ob->main_stream);
 196           bp_pack_value (&bp, s->m_kind, 2);
 197           bp_pack_value (&bp, s->m_gpu_implementation_p, 1);
 198           bp_pack_value (&bp, s->m_bound_function != NULL, 1);
 199           streamer_write_bitpack (&bp);
 200           if (s->m_bound_function)
 201             stream_write_tree (ob, s->m_bound_function->decl, true);
 202         }
 203     }
 204
 205   streamer_write_char_stream (ob->main_stream, 0);
 206   produce_asm (ob, NULL);
 207   destroy_output_block (ob);
 208 }
 209
 210 /* Read section in file FILE_DATA of length LEN with data DATA.  */
 211
 212 static void
 213 ipa_hsa_read_section (struct lto_file_decl_data *file_data, const char *data,
 214                        size_t len)
 215 {
 216   const struct lto_function_header *header
 217     = (const struct lto_function_header *) data;
 218   const int cfg_offset = sizeof (struct lto_function_header);
 219   const int main_offset = cfg_offset + header->cfg_size;
 220   const int string_offset = main_offset + header->main_size;
 221   struct data_in *data_in;
 222   unsigned int i;
 223   unsigned int count;
 224
 225   lto_input_block ib_main ((const char *) data + main_offset,
 226                            header->main_size, file_data->mode_table);
 227
 228   data_in
 229     = lto_data_in_create (file_data, (const char *) data + string_offset,
 230                           header->string_size, vNULL);
 231   count = streamer_read_uhwi (&ib_main);
 232
 233   for (i = 0; i < count; i++)
 234     {
 235       unsigned int index;
 236       struct cgraph_node *node;
 237       lto_symtab_encoder_t encoder;
 238
 239       index = streamer_read_uhwi (&ib_main);
 240       encoder = file_data->symtab_node_encoder;
 241       node = dyn_cast<cgraph_node *> (lto_symtab_encoder_deref (encoder,
 242                                                                 index));
 243       gcc_assert (node->definition);
 244       hsa_function_summary *s = hsa_summaries->get (node);
 245
 246       struct bitpack_d bp = streamer_read_bitpack (&ib_main);
 247       s->m_kind = (hsa_function_kind) bp_unpack_value (&bp, 2);
 248       s->m_gpu_implementation_p = bp_unpack_value (&bp, 1);
 249       bool has_tree = bp_unpack_value (&bp, 1);
 250
 251       if (has_tree)
 252         {
 253           tree decl = stream_read_tree (&ib_main, data_in);
 254           s->m_bound_function = cgraph_node::get_create (decl);
 255         }
 256     }
 257   lto_free_section_data (file_data, LTO_section_ipa_hsa, NULL, data,
 258                          len);
 259   lto_data_in_delete (data_in);
 260 }
 261
 262 /* Load streamed HSA functions summary and assign the summary to a function.  */
 263
 264 static void
 265 ipa_hsa_read_summary (void)
 266 {
 267   struct lto_file_decl_data **file_data_vec = lto_get_file_decl_data ();
 268   struct lto_file_decl_data *file_data;
 269   unsigned int j = 0;
 270
 271   if (hsa_summaries == NULL)
 272     hsa_summaries = new hsa_summary_t (symtab);
 273
 274   while ((file_data = file_data_vec[j++]))
 275     {
 276       size_t len;
 277       const char *data = lto_get_section_data (file_data, LTO_section_ipa_hsa,
 278                                                NULL, &len);
 279
 280       if (data)
 281         ipa_hsa_read_section (file_data, data, len);
 282     }
 283 }
 284
 285 const pass_data pass_data_ipa_hsa =
 286 {
 287   IPA_PASS, /* type */
 288   "hsa", /* name */
 289   OPTGROUP_NONE, /* optinfo_flags */
 290   TV_IPA_HSA, /* tv_id */
 291   0, /* properties_required */
 292   0, /* properties_provided */
 293   0, /* properties_destroyed */
 294   0, /* todo_flags_start */
 295   TODO_dump_symtab, /* todo_flags_finish */
 296 };
 297
 298 class pass_ipa_hsa : public ipa_opt_pass_d
 299 {
 300 public:
 301   pass_ipa_hsa (gcc::context *ctxt)
 302     : ipa_opt_pass_d (pass_data_ipa_hsa, ctxt,
 303                       NULL, /* generate_summary */
 304                       ipa_hsa_write_summary, /* write_summary */
 305                       ipa_hsa_read_summary, /* read_summary */
 306                       ipa_hsa_write_summary, /* write_optimization_summary */
 307                       ipa_hsa_read_summary, /* read_optimization_summary */
 308                       NULL, /* stmt_fixup */
 309                       0, /* function_transform_todo_flags_start */
 310                       NULL, /* function_transform */
 311                       NULL) /* variable_transform */
 312     {}
 313
 314   /* opt_pass methods: */
 315   virtual bool gate (function *);
 316
 317   virtual unsigned int execute (function *) { return process_hsa_functions (); }
 318
 319 }; // class pass_ipa_reference
 320
 321 bool
 322 pass_ipa_hsa::gate (function *)
 323 {
 324   return hsa_gen_requested_p ();
 325 }
 326
 327 } // anon namespace
 328
 329 ipa_opt_pass_d *
 330 make_pass_ipa_hsa (gcc::context *ctxt)
 331 {
 332   return new pass_ipa_hsa (ctxt);
 333 }