gcc/loop-doloop.c

   1 /* Perform doloop optimizations
   2    Copyright (C) 2004, 2005 Free Software Foundation, Inc.
   3    Based on code by Michael P. Hayes (m.hayes@elec.canterbury.ac.nz)
   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 2, 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 COPYING.  If not, write to the Free
  19 Software Foundation, 51 Franklin Street, Fifth Floor, Boston, MA
  20 02110-1301, USA.  */
  21
  22 #include "config.h"
  23 #include "system.h"
  24 #include "coretypes.h"
  25 #include "tm.h"
  26 #include "rtl.h"
  27 #include "flags.h"
  28 #include "expr.h"
  29 #include "hard-reg-set.h"
  30 #include "basic-block.h"
  31 #include "toplev.h"
  32 #include "tm_p.h"
  33 #include "cfgloop.h"
  34 #include "output.h"
  35 #include "params.h"
  36 #include "target.h"
  37
  38 /* This module is used to modify loops with a determinable number of
  39    iterations to use special low-overhead looping instructions.
  40
  41    It first validates whether the loop is well behaved and has a
  42    determinable number of iterations (either at compile or run-time).
  43    It then modifies the loop to use a low-overhead looping pattern as
  44    follows:
  45
  46    1. A pseudo register is allocated as the loop iteration counter.
  47
  48    2. The number of loop iterations is calculated and is stored
  49       in the loop counter.
  50
  51    3. At the end of the loop, the jump insn is replaced by the
  52       doloop_end pattern.  The compare must remain because it might be
  53       used elsewhere.  If the loop-variable or condition register are
  54       used elsewhere, they will be eliminated by flow.
  55
  56    4. An optional doloop_begin pattern is inserted at the top of the
  57       loop.
  58
  59    TODO The optimization should only performed when either the biv used for exit
  60    condition is unused at all except for the exit test, or if we do not have to
  61    change its value, since otherwise we have to add a new induction variable,
  62    which usually will not pay up (unless the cost of the doloop pattern is
  63    somehow extremely lower than the cost of compare & jump, or unless the bct
  64    register cannot be used for anything else but doloop -- ??? detect these
  65    cases).  */
  66
  67 #ifdef HAVE_doloop_end
  68
  69 /* Return the loop termination condition for PATTERN or zero
  70    if it is not a decrement and branch jump insn.  */
  71
  72 rtx
  73 doloop_condition_get (rtx pattern)
  74 {
  75   rtx cmp;
  76   rtx inc;
  77   rtx reg;
  78   rtx inc_src;
  79   rtx condition;
  80
  81   /* The canonical doloop pattern we expect is:
  82
  83      (parallel [(set (pc) (if_then_else (condition)
  84                                         (label_ref (label))
  85                                         (pc)))
  86                 (set (reg) (plus (reg) (const_int -1)))
  87                 (additional clobbers and uses)])
  88
  89      Some targets (IA-64) wrap the set of the loop counter in
  90      an if_then_else too.
  91
  92      In summary, the branch must be the first entry of the
  93      parallel (also required by jump.c), and the second
  94      entry of the parallel must be a set of the loop counter
  95      register.  */
  96
  97   if (GET_CODE (pattern) != PARALLEL)
  98     return 0;
  99
 100   cmp = XVECEXP (pattern, 0, 0);
 101   inc = XVECEXP (pattern, 0, 1);
 102
 103   /* Check for (set (reg) (something)).  */
 104   if (GET_CODE (inc) != SET)
 105     return 0;
 106   reg = SET_DEST (inc);
 107   if (! REG_P (reg))
 108     return 0;
 109
 110   /* Check if something = (plus (reg) (const_int -1)).
 111      On IA-64, this decrement is wrapped in an if_then_else.  */
 112   inc_src = SET_SRC (inc);
 113   if (GET_CODE (inc_src) == IF_THEN_ELSE)
 114     inc_src = XEXP (inc_src, 1);
 115   if (GET_CODE (inc_src) != PLUS
 116       || XEXP (inc_src, 0) != reg
 117       || XEXP (inc_src, 1) != constm1_rtx)
 118     return 0;
 119
 120   /* Check for (set (pc) (if_then_else (condition)
 121                                        (label_ref (label))
 122                                        (pc))).  */
 123   if (GET_CODE (cmp) != SET
 124       || SET_DEST (cmp) != pc_rtx
 125       || GET_CODE (SET_SRC (cmp)) != IF_THEN_ELSE
 126       || GET_CODE (XEXP (SET_SRC (cmp), 1)) != LABEL_REF
 127       || XEXP (SET_SRC (cmp), 2) != pc_rtx)
 128     return 0;
 129
 130   /* Extract loop termination condition.  */
 131   condition = XEXP (SET_SRC (cmp), 0);
 132
 133   /* We expect a GE or NE comparison with 0 or 1.  */
 134   if ((GET_CODE (condition) != GE
 135        && GET_CODE (condition) != NE)
 136       || (XEXP (condition, 1) != const0_rtx
 137           && XEXP (condition, 1) != const1_rtx))
 138     return 0;
 139
 140   if ((XEXP (condition, 0) == reg)
 141       || (GET_CODE (XEXP (condition, 0)) == PLUS
 142                    && XEXP (XEXP (condition, 0), 0) == reg))
 143     return condition;
 144
 145   /* ??? If a machine uses a funny comparison, we could return a
 146      canonicalized form here.  */
 147
 148   return 0;
 149 }
 150
 151 /* Return nonzero if the loop specified by LOOP is suitable for
 152    the use of special low-overhead looping instructions.  DESC
 153    describes the number of iterations of the loop.  */
 154
 155 static bool
 156 doloop_valid_p (struct loop *loop, struct niter_desc *desc)
 157 {
 158   basic_block *body = get_loop_body (loop), bb;
 159   rtx insn;
 160   unsigned i;
 161   bool result = true;
 162
 163   /* Check for loops that may not terminate under special conditions.  */
 164   if (!desc->simple_p
 165       || desc->assumptions
 166       || desc->infinite)
 167     {
 168       /* There are some cases that would require a special attention.
 169          For example if the comparison is LEU and the comparison value
 170          is UINT_MAX then the loop will not terminate.  Similarly, if the
 171          comparison code is GEU and the comparison value is 0, the
 172          loop will not terminate.
 173
 174          If the absolute increment is not 1, the loop can be infinite
 175          even with LTU/GTU, e.g. for (i = 3; i > 0; i -= 2)
 176
 177          ??? We could compute these conditions at run-time and have a
 178          additional jump around the loop to ensure an infinite loop.
 179          However, it is very unlikely that this is the intended
 180          behavior of the loop and checking for these rare boundary
 181          conditions would pessimize all other code.
 182
 183          If the loop is executed only a few times an extra check to
 184          restart the loop could use up most of the benefits of using a
 185          count register loop.  Note however, that normally, this
 186          restart branch would never execute, so it could be predicted
 187          well by the CPU.  We should generate the pessimistic code by
 188          default, and have an option, e.g. -funsafe-loops that would
 189          enable count-register loops in this case.  */
 190       if (dump_file)
 191         fprintf (dump_file, "Doloop: Possible infinite iteration case.\n");
 192       result = false;
 193       goto cleanup;
 194     }
 195
 196   for (i = 0; i < loop->num_nodes; i++)
 197     {
 198       bb = body[i];
 199
 200       for (insn = BB_HEAD (bb);
 201            insn != NEXT_INSN (BB_END (bb));
 202            insn = NEXT_INSN (insn))
 203         {
 204           /* Different targets have different necessities for low-overhead
 205              looping.  Call the back end for each instruction within the loop
 206              to let it decide whether the insn prohibits a low-overhead loop.
 207              It will then return the cause for it to emit to the dump file.  */
 208           const char * invalid = targetm.invalid_within_doloop (insn);
 209           if (invalid)
 210             {
 211               if (dump_file)
 212                 fprintf (dump_file, "Doloop: %s\n", invalid);
 213               result = false;
 214               goto cleanup;
 215             }
 216         }
 217     }
 218   result = true;
 219
 220 cleanup:
 221   free (body);
 222
 223   return result;
 224 }
 225
 226 /* Adds test of COND jumping to DEST to the end of BB.  */
 227
 228 static void
 229 add_test (rtx cond, basic_block bb, basic_block dest)
 230 {
 231   rtx seq, jump, label;
 232   enum machine_mode mode;
 233   rtx op0 = XEXP (cond, 0), op1 = XEXP (cond, 1);
 234   enum rtx_code code = GET_CODE (cond);
 235
 236   mode = GET_MODE (XEXP (cond, 0));
 237   if (mode == VOIDmode)
 238     mode = GET_MODE (XEXP (cond, 1));
 239
 240   start_sequence ();
 241   op0 = force_operand (op0, NULL_RTX);
 242   op1 = force_operand (op1, NULL_RTX);
 243   label = block_label (dest);
 244   do_compare_rtx_and_jump (op0, op1, code, 0, mode, NULL_RTX, NULL_RTX, label);
 245
 246   jump = get_last_insn ();
 247   JUMP_LABEL (jump) = label;
 248
 249   /* The jump is supposed to handle an unlikely special case.  */
 250   REG_NOTES (jump)
 251           = gen_rtx_EXPR_LIST (REG_BR_PROB,
 252                                const0_rtx, REG_NOTES (jump));
 253
 254   LABEL_NUSES (label)++;
 255
 256   seq = get_insns ();
 257   end_sequence ();
 258   emit_insn_after (seq, BB_END (bb));
 259 }
 260
 261 /* Modify the loop to use the low-overhead looping insn where LOOP
 262    describes the loop, DESC describes the number of iterations of the
 263    loop, and DOLOOP_INSN is the low-overhead looping insn to emit at the
 264    end of the loop.  CONDITION is the condition separated from the
 265    DOLOOP_SEQ.  COUNT is the number of iterations of the LOOP.  */
 266
 267 static void
 268 doloop_modify (struct loop *loop, struct niter_desc *desc,
 269                rtx doloop_seq, rtx condition, rtx count)
 270 {
 271   rtx counter_reg;
 272   rtx tmp, noloop = NULL_RTX;
 273   rtx sequence;
 274   rtx jump_insn;
 275   rtx jump_label;
 276   int nonneg = 0, irr;
 277   bool increment_count;
 278   basic_block loop_end = desc->out_edge->src;
 279   enum machine_mode mode;
 280
 281   jump_insn = BB_END (loop_end);
 282
 283   if (dump_file)
 284     {
 285       fprintf (dump_file, "Doloop: Inserting doloop pattern (");
 286       if (desc->const_iter)
 287         fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC, desc->niter);
 288       else
 289         fputs ("runtime", dump_file);
 290       fputs (" iterations).\n", dump_file);
 291     }
 292
 293   /* Discard original jump to continue loop.  The original compare
 294      result may still be live, so it cannot be discarded explicitly.  */
 295   delete_insn (jump_insn);
 296
 297   counter_reg = XEXP (condition, 0);
 298   if (GET_CODE (counter_reg) == PLUS)
 299     counter_reg = XEXP (counter_reg, 0);
 300   mode = GET_MODE (counter_reg);
 301
 302   increment_count = false;
 303   switch (GET_CODE (condition))
 304     {
 305     case NE:
 306       /* Currently only NE tests against zero and one are supported.  */
 307       noloop = XEXP (condition, 1);
 308       if (noloop != const0_rtx)
 309         {
 310           gcc_assert (noloop == const1_rtx);
 311           increment_count = true;
 312         }
 313       break;
 314
 315     case GE:
 316       /* Currently only GE tests against zero are supported.  */
 317       gcc_assert (XEXP (condition, 1) == const0_rtx);
 318
 319       noloop = constm1_rtx;
 320
 321       /* The iteration count does not need incrementing for a GE test.  */
 322       increment_count = false;
 323
 324       /* Determine if the iteration counter will be non-negative.
 325          Note that the maximum value loaded is iterations_max - 1.  */
 326       if (desc->niter_max
 327           <= ((unsigned HOST_WIDEST_INT) 1
 328               << (GET_MODE_BITSIZE (mode) - 1)))
 329         nonneg = 1;
 330       break;
 331
 332       /* Abort if an invalid doloop pattern has been generated.  */
 333     default:
 334       gcc_unreachable ();
 335     }
 336
 337   if (increment_count)
 338     count = simplify_gen_binary (PLUS, mode, count, const1_rtx);
 339
 340   /* Insert initialization of the count register into the loop header.  */
 341   start_sequence ();
 342   tmp = force_operand (count, counter_reg);
 343   convert_move (counter_reg, tmp, 1);
 344   sequence = get_insns ();
 345   end_sequence ();
 346   emit_insn_after (sequence, BB_END (loop_preheader_edge (loop)->src));
 347
 348   if (desc->noloop_assumptions)
 349     {
 350       rtx ass = copy_rtx (desc->noloop_assumptions);
 351       basic_block preheader = loop_preheader_edge (loop)->src;
 352       basic_block set_zero
 353               = loop_split_edge_with (loop_preheader_edge (loop), NULL_RTX);
 354       basic_block new_preheader
 355               = loop_split_edge_with (loop_preheader_edge (loop), NULL_RTX);
 356       basic_block bb;
 357       edge te;
 358       gcov_type cnt;
 359
 360       /* Expand the condition testing the assumptions and if it does not pass,
 361          reset the count register to 0.  */
 362       add_test (XEXP (ass, 0), preheader, set_zero);
 363       single_succ_edge (preheader)->flags &= ~EDGE_FALLTHRU;
 364       cnt = single_succ_edge (preheader)->count;
 365       single_succ_edge (preheader)->probability = 0;
 366       single_succ_edge (preheader)->count = 0;
 367       irr = single_succ_edge (preheader)->flags & EDGE_IRREDUCIBLE_LOOP;
 368       te = make_edge (preheader, new_preheader, EDGE_FALLTHRU | irr);
 369       te->probability = REG_BR_PROB_BASE;
 370       te->count = cnt;
 371       set_immediate_dominator (CDI_DOMINATORS, new_preheader, preheader);
 372
 373       set_zero->count = 0;
 374       set_zero->frequency = 0;
 375
 376       for (ass = XEXP (ass, 1); ass; ass = XEXP (ass, 1))
 377         {
 378           bb = loop_split_edge_with (te, NULL_RTX);
 379           te = single_succ_edge (bb);
 380           add_test (XEXP (ass, 0), bb, set_zero);
 381           make_edge (bb, set_zero, irr);
 382         }
 383
 384       start_sequence ();
 385       convert_move (counter_reg, noloop, 0);
 386       sequence = get_insns ();
 387       end_sequence ();
 388       emit_insn_after (sequence, BB_END (set_zero));
 389     }
 390
 391   /* Some targets (eg, C4x) need to initialize special looping
 392      registers.  */
 393 #ifdef HAVE_doloop_begin
 394   {
 395     rtx init;
 396     unsigned level = get_loop_level (loop) + 1;
 397     init = gen_doloop_begin (counter_reg,
 398                              desc->const_iter ? desc->niter_expr : const0_rtx,
 399                              desc->niter_max,
 400                              GEN_INT (level));
 401     if (init)
 402       {
 403         start_sequence ();
 404         emit_insn (init);
 405         sequence = get_insns ();
 406         end_sequence ();
 407         emit_insn_after (sequence, BB_END (loop_preheader_edge (loop)->src));
 408       }
 409   }
 410 #endif
 411
 412   /* Insert the new low-overhead looping insn.  */
 413   emit_jump_insn_after (doloop_seq, BB_END (loop_end));
 414   jump_insn = BB_END (loop_end);
 415   jump_label = block_label (desc->in_edge->dest);
 416   JUMP_LABEL (jump_insn) = jump_label;
 417   LABEL_NUSES (jump_label)++;
 418
 419   /* Ensure the right fallthru edge is marked, for case we have reversed
 420      the condition.  */
 421   desc->in_edge->flags &= ~EDGE_FALLTHRU;
 422   desc->out_edge->flags |= EDGE_FALLTHRU;
 423
 424   /* Add a REG_NONNEG note if the actual or estimated maximum number
 425      of iterations is non-negative.  */
 426   if (nonneg)
 427     {
 428       REG_NOTES (jump_insn)
 429         = gen_rtx_EXPR_LIST (REG_NONNEG, NULL_RTX, REG_NOTES (jump_insn));
 430     }
 431 }
 432
 433 /* Process loop described by LOOP validating that the loop is suitable for
 434    conversion to use a low overhead looping instruction, replacing the jump
 435    insn where suitable.  Returns true if the loop was successfully
 436    modified.  */
 437
 438 static bool
 439 doloop_optimize (struct loop *loop)
 440 {
 441   enum machine_mode mode;
 442   rtx doloop_seq, doloop_pat, doloop_reg;
 443   rtx iterations, count;
 444   rtx iterations_max;
 445   rtx start_label;
 446   rtx condition;
 447   unsigned level, est_niter;
 448   struct niter_desc *desc;
 449   unsigned word_mode_size;
 450   unsigned HOST_WIDE_INT word_mode_max;
 451
 452   if (dump_file)
 453     fprintf (dump_file, "Doloop: Processing loop %d.\n", loop->num);
 454
 455   iv_analysis_loop_init (loop);
 456
 457   /* Find the simple exit of a LOOP.  */
 458   desc = get_simple_loop_desc (loop);
 459
 460   /* Check that loop is a candidate for a low-overhead looping insn.  */
 461   if (!doloop_valid_p (loop, desc))
 462     {
 463       if (dump_file)
 464         fprintf (dump_file,
 465                  "Doloop: The loop is not suitable.\n");
 466       return false;
 467     }
 468   mode = desc->mode;
 469
 470   est_niter = 3;
 471   if (desc->const_iter)
 472     est_niter = desc->niter;
 473   /* If the estimate on number of iterations is reliable (comes from profile
 474      feedback), use it.  Do not use it normally, since the expected number
 475      of iterations of an unrolled loop is 2.  */
 476   if (loop->header->count)
 477     est_niter = expected_loop_iterations (loop);
 478
 479   if (est_niter < 3)
 480     {
 481       if (dump_file)
 482         fprintf (dump_file,
 483                  "Doloop: Too few iterations (%u) to be profitable.\n",
 484                  est_niter);
 485       return false;
 486     }
 487
 488   count = copy_rtx (desc->niter_expr);
 489   iterations = desc->const_iter ? desc->niter_expr : const0_rtx;
 490   iterations_max = GEN_INT (desc->niter_max);
 491   level = get_loop_level (loop) + 1;
 492
 493   /* Generate looping insn.  If the pattern FAILs then give up trying
 494      to modify the loop since there is some aspect the back-end does
 495      not like.  */
 496   start_label = block_label (desc->in_edge->dest);
 497   doloop_reg = gen_reg_rtx (mode);
 498   doloop_seq = gen_doloop_end (doloop_reg, iterations, iterations_max,
 499                                GEN_INT (level), start_label);
 500
 501   word_mode_size = GET_MODE_BITSIZE (word_mode);
 502   word_mode_max
 503           = ((unsigned HOST_WIDE_INT) 1 << (word_mode_size - 1) << 1) - 1;
 504   if (! doloop_seq
 505       && mode != word_mode
 506       /* Before trying mode different from the one in that # of iterations is
 507          computed, we must be sure that the number of iterations fits into
 508          the new mode.  */
 509       && (word_mode_size >= GET_MODE_BITSIZE (mode)
 510           || desc->niter_max <= word_mode_max))
 511     {
 512       if (word_mode_size > GET_MODE_BITSIZE (mode))
 513         {
 514           count = simplify_gen_unary (ZERO_EXTEND, word_mode,
 515                                       count, mode);
 516           iterations = simplify_gen_unary (ZERO_EXTEND, word_mode,
 517                                            iterations, mode);
 518           iterations_max = simplify_gen_unary (ZERO_EXTEND, word_mode,
 519                                                iterations_max, mode);
 520         }
 521       else
 522         {
 523           count = lowpart_subreg (word_mode, count, mode);
 524           iterations = lowpart_subreg (word_mode, iterations, mode);
 525           iterations_max = lowpart_subreg (word_mode, iterations_max, mode);
 526         }
 527       PUT_MODE (doloop_reg, word_mode);
 528       doloop_seq = gen_doloop_end (doloop_reg, iterations, iterations_max,
 529                                    GEN_INT (level), start_label);
 530     }
 531   if (! doloop_seq)
 532     {
 533       if (dump_file)
 534         fprintf (dump_file,
 535                  "Doloop: Target unwilling to use doloop pattern!\n");
 536       return false;
 537     }
 538
 539   /* If multiple instructions were created, the last must be the
 540      jump instruction.  Also, a raw define_insn may yield a plain
 541      pattern.  */
 542   doloop_pat = doloop_seq;
 543   if (INSN_P (doloop_pat))
 544     {
 545       while (NEXT_INSN (doloop_pat) != NULL_RTX)
 546         doloop_pat = NEXT_INSN (doloop_pat);
 547       if (JUMP_P (doloop_pat))
 548         doloop_pat = PATTERN (doloop_pat);
 549       else
 550         doloop_pat = NULL_RTX;
 551     }
 552
 553   if (! doloop_pat
 554       || ! (condition = doloop_condition_get (doloop_pat)))
 555     {
 556       if (dump_file)
 557         fprintf (dump_file, "Doloop: Unrecognizable doloop pattern!\n");
 558       return false;
 559     }
 560
 561   doloop_modify (loop, desc, doloop_seq, condition, count);
 562   return true;
 563 }
 564
 565 /* This is the main entry point.  Process all LOOPS using doloop_optimize.  */
 566
 567 void
 568 doloop_optimize_loops (struct loops *loops)
 569 {
 570   unsigned i;
 571   struct loop *loop;
 572
 573   for (i = 1; i < loops->num; i++)
 574     {
 575       loop = loops->parray[i];
 576       if (!loop)
 577         continue;
 578
 579       doloop_optimize (loop);
 580     }
 581
 582   iv_analysis_done ();
 583
 584 #ifdef ENABLE_CHECKING
 585   verify_dominators (CDI_DOMINATORS);
 586   verify_loop_structure (loops);
 587 #endif
 588 }
 589 #endif /* HAVE_doloop_end */
 590