src/compiler/glsl/loop_unroll.cpp

   1 /*
   2  * Copyright © 2010 Intel Corporation
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  21  * DEALINGS IN THE SOFTWARE.
  22  */
  23
  24 #include "compiler/glsl_types.h"
  25 #include "loop_analysis.h"
  26 #include "ir_hierarchical_visitor.h"
  27
  28 #include "main/mtypes.h"
  29
  30 namespace {
  31
  32 class loop_unroll_visitor : public ir_hierarchical_visitor {
  33 public:
  34    loop_unroll_visitor(loop_state *state,
  35                        const struct gl_shader_compiler_options *options)
  36    {
  37       this->state = state;
  38       this->progress = false;
  39       this->options = options;
  40    }
  41
  42    virtual ir_visitor_status visit_leave(ir_loop *ir);
  43    void simple_unroll(ir_loop *ir, int iterations);
  44    void complex_unroll(ir_loop *ir, int iterations,
  45                        bool continue_from_then_branch,
  46                        bool limiting_term_first,
  47                        bool lt_continue_from_then_branch);
  48    void splice_post_if_instructions(ir_if *ir_if, exec_list *splice_dest);
  49
  50    loop_state *state;
  51
  52    bool progress;
  53    const struct gl_shader_compiler_options *options;
  54 };
  55
  56 } /* anonymous namespace */
  57
  58 class loop_unroll_count : public ir_hierarchical_visitor {
  59 public:
  60    int nodes;
  61    bool unsupported_variable_indexing;
  62    bool array_indexed_by_induction_var_with_exact_iterations;
  63    /* If there are nested loops, the node count will be inaccurate. */
  64    bool nested_loop;
  65
  66    loop_unroll_count(exec_list *list, loop_variable_state *ls,
  67                      const struct gl_shader_compiler_options *options)
  68       : ls(ls), options(options)
  69    {
  70       nodes = 0;
  71       nested_loop = false;
  72       unsupported_variable_indexing = false;
  73       array_indexed_by_induction_var_with_exact_iterations = false;
  74
  75       run(list);
  76    }
  77
  78    virtual ir_visitor_status visit_enter(ir_assignment *)
  79    {
  80       nodes++;
  81       return visit_continue;
  82    }
  83
  84    virtual ir_visitor_status visit_enter(ir_expression *)
  85    {
  86       nodes++;
  87       return visit_continue;
  88    }
  89
  90    virtual ir_visitor_status visit_enter(ir_loop *)
  91    {
  92       nested_loop = true;
  93       return visit_continue;
  94    }
  95
  96    virtual ir_visitor_status visit_enter(ir_dereference_array *ir)
  97    {
  98       /* Force unroll in case of dynamic indexing with sampler arrays
  99        * when EmitNoIndirectSampler is set.
 100        */
 101       if (options->EmitNoIndirectSampler) {
 102          if ((ir->array->type->is_array() &&
 103               ir->array->type->contains_sampler()) &&
 104              !ir->array_index->constant_expression_value(ralloc_parent(ir))) {
 105             unsupported_variable_indexing = true;
 106             return visit_continue;
 107          }
 108       }
 109
 110       /* Check for arrays variably-indexed by a loop induction variable.
 111        * Unrolling the loop may convert that access into constant-indexing.
 112        *
 113        * Many drivers don't support particular kinds of variable indexing,
 114        * and have to resort to using lower_variable_index_to_cond_assign to
 115        * handle it.  This results in huge amounts of horrible code, so we'd
 116        * like to avoid that if possible.  Here, we just note that it will
 117        * happen.
 118        */
 119       if ((ir->array->type->is_array() || ir->array->type->is_matrix()) &&
 120           !ir->array_index->as_constant()) {
 121          ir_variable *array = ir->array->variable_referenced();
 122          loop_variable *lv = ls->get(ir->array_index->variable_referenced());
 123          if (array && lv && lv->is_induction_var()) {
 124             /* If an array is indexed by a loop induction variable, and the
 125              * array size is exactly the number of loop iterations, this is
 126              * probably a simple for-loop trying to access each element in
 127              * turn; the application may expect it to be unrolled.
 128              */
 129             if (int(array->type->length) == ls->limiting_terminator->iterations)
 130                array_indexed_by_induction_var_with_exact_iterations = true;
 131
 132             switch (array->data.mode) {
 133             case ir_var_auto:
 134             case ir_var_temporary:
 135             case ir_var_const_in:
 136             case ir_var_function_in:
 137             case ir_var_function_out:
 138             case ir_var_function_inout:
 139                if (options->EmitNoIndirectTemp)
 140                   unsupported_variable_indexing = true;
 141                break;
 142             case ir_var_uniform:
 143             case ir_var_shader_storage:
 144                if (options->EmitNoIndirectUniform)
 145                   unsupported_variable_indexing = true;
 146                break;
 147             case ir_var_shader_in:
 148                if (options->EmitNoIndirectInput)
 149                   unsupported_variable_indexing = true;
 150                break;
 151             case ir_var_shader_out:
 152                if (options->EmitNoIndirectOutput)
 153                   unsupported_variable_indexing = true;
 154                break;
 155             }
 156          }
 157       }
 158       return visit_continue;
 159    }
 160
 161 private:
 162    loop_variable_state *ls;
 163    const struct gl_shader_compiler_options *options;
 164 };
 165
 166
 167 /**
 168  * Unroll a loop which does not contain any jumps.  For example, if the input
 169  * is:
 170  *
 171  *     (loop (...) ...instrs...)
 172  *
 173  * And the iteration count is 3, the output will be:
 174  *
 175  *     ...instrs... ...instrs... ...instrs...
 176  */
 177 void
 178 loop_unroll_visitor::simple_unroll(ir_loop *ir, int iterations)
 179 {
 180    void *const mem_ctx = ralloc_parent(ir);
 181    loop_variable_state *const ls = this->state->get(ir);
 182
 183    ir_instruction *first_ir =
 184       (ir_instruction *) ir->body_instructions.get_head();
 185
 186    if (!first_ir) {
 187       /* The loop is empty remove it and return */
 188       ir->remove();
 189       return;
 190    }
 191
 192    ir_if *limit_if = NULL;
 193    bool exit_branch_has_instructions = false;
 194    if (ls->limiting_terminator) {
 195       limit_if = ls->limiting_terminator->ir;
 196       ir_instruction *ir_if_last = (ir_instruction *)
 197          limit_if->then_instructions.get_tail();
 198
 199       if (is_break(ir_if_last)) {
 200          if (ir_if_last != limit_if->then_instructions.get_head())
 201             exit_branch_has_instructions = true;
 202
 203          splice_post_if_instructions(limit_if, &limit_if->else_instructions);
 204          ir_if_last->remove();
 205       } else {
 206          ir_if_last = (ir_instruction *)
 207             limit_if->else_instructions.get_tail();
 208          assert(is_break(ir_if_last));
 209
 210          if (ir_if_last != limit_if->else_instructions.get_head())
 211             exit_branch_has_instructions = true;
 212
 213          splice_post_if_instructions(limit_if, &limit_if->then_instructions);
 214          ir_if_last->remove();
 215       }
 216    }
 217
 218    /* Because 'iterations' is the number of times we pass over the *entire*
 219     * loop body before hitting the first break, we need to bump the number of
 220     * iterations if the limiting terminator is not the first instruction in
 221     * the loop, or it the exit branch contains instructions. This ensures we
 222     * execute any instructions before the terminator or in its exit branch.
 223     */
 224    if (limit_if != first_ir->as_if() || exit_branch_has_instructions)
 225       iterations++;
 226
 227    for (int i = 0; i < iterations; i++) {
 228       exec_list copy_list;
 229
 230       copy_list.make_empty();
 231       clone_ir_list(mem_ctx, &copy_list, &ir->body_instructions);
 232
 233       ir->insert_before(&copy_list);
 234    }
 235
 236    /* The loop has been replaced by the unrolled copies.  Remove the original
 237     * loop from the IR sequence.
 238     */
 239    ir->remove();
 240
 241    this->progress = true;
 242 }
 243
 244
 245 /**
 246  * Unroll a loop whose last statement is an ir_if.  If \c
 247  * continue_from_then_branch is true, the loop is repeated only when the
 248  * "then" branch of the if is taken; otherwise it is repeated only when the
 249  * "else" branch of the if is taken.
 250  *
 251  * For example, if the input is:
 252  *
 253  *     (loop (...)
 254  *      ...body...
 255  *      (if (cond)
 256  *          (...then_instrs...)
 257  *        (...else_instrs...)))
 258  *
 259  * And the iteration count is 3, and \c continue_from_then_branch is true,
 260  * then the output will be:
 261  *
 262  *     ...body...
 263  *     (if (cond)
 264  *         (...then_instrs...
 265  *          ...body...
 266  *          (if (cond)
 267  *              (...then_instrs...
 268  *               ...body...
 269  *               (if (cond)
 270  *                   (...then_instrs...)
 271  *                 (...else_instrs...)))
 272  *            (...else_instrs...)))
 273  *       (...else_instrs))
 274  */
 275 void
 276 loop_unroll_visitor::complex_unroll(ir_loop *ir, int iterations,
 277                                     bool second_term_then_continue,
 278                                     bool extra_iteration_required,
 279                                     bool first_term_then_continue)
 280 {
 281    void *const mem_ctx = ralloc_parent(ir);
 282    ir_instruction *ir_to_replace = ir;
 283
 284    /* Because 'iterations' is the number of times we pass over the *entire*
 285     * loop body before hitting the first break, we need to bump the number of
 286     * iterations if the limiting terminator is not the first instruction in
 287     * the loop, or it the exit branch contains instructions. This ensures we
 288     * execute any instructions before the terminator or in its exit branch.
 289     */
 290    if (extra_iteration_required)
 291       iterations++;
 292
 293    for (int i = 0; i < iterations; i++) {
 294       exec_list copy_list;
 295
 296       copy_list.make_empty();
 297       clone_ir_list(mem_ctx, &copy_list, &ir->body_instructions);
 298
 299       ir_if *ir_if = ((ir_instruction *) copy_list.get_tail())->as_if();
 300       assert(ir_if != NULL);
 301
 302       exec_list *const first_list = first_term_then_continue
 303          ? &ir_if->then_instructions : &ir_if->else_instructions;
 304       ir_if = ((ir_instruction *) first_list->get_tail())->as_if();
 305
 306       ir_to_replace->insert_before(&copy_list);
 307       ir_to_replace->remove();
 308
 309       /* placeholder that will be removed in the next iteration */
 310       ir_to_replace =
 311          new(mem_ctx) ir_loop_jump(ir_loop_jump::jump_continue);
 312
 313       exec_list *const second_term_continue_list = second_term_then_continue
 314          ? &ir_if->then_instructions : &ir_if->else_instructions;
 315
 316       second_term_continue_list->push_tail(ir_to_replace);
 317    }
 318
 319    ir_to_replace->remove();
 320
 321    this->progress = true;
 322 }
 323
 324
 325 /**
 326  * Move all of the instructions which follow \c ir_if to the end of
 327  * \c splice_dest.
 328  *
 329  * For example, in the code snippet:
 330  *
 331  *     (if (cond)
 332  *         (...then_instructions...
 333  *          break)
 334  *       (...else_instructions...))
 335  *     ...post_if_instructions...
 336  *
 337  * If \c ir_if points to the "if" instruction, and \c splice_dest points to
 338  * (...else_instructions...), the code snippet is transformed into:
 339  *
 340  *     (if (cond)
 341  *         (...then_instructions...
 342  *          break)
 343  *       (...else_instructions...
 344  *        ...post_if_instructions...))
 345  */
 346 void
 347 loop_unroll_visitor::splice_post_if_instructions(ir_if *ir_if,
 348                                                  exec_list *splice_dest)
 349 {
 350    while (!ir_if->get_next()->is_tail_sentinel()) {
 351       ir_instruction *move_ir = (ir_instruction *) ir_if->get_next();
 352
 353       move_ir->remove();
 354       splice_dest->push_tail(move_ir);
 355    }
 356 }
 357
 358 static bool
 359 exit_branch_has_instructions(ir_if *term_if, bool lt_then_continue)
 360 {
 361    if (lt_then_continue) {
 362       if (term_if->else_instructions.get_head() ==
 363           term_if->else_instructions.get_tail())
 364          return false;
 365    } else {
 366       if (term_if->then_instructions.get_head() ==
 367           term_if->then_instructions.get_tail())
 368          return false;
 369    }
 370
 371    return true;
 372 }
 373
 374 ir_visitor_status
 375 loop_unroll_visitor::visit_leave(ir_loop *ir)
 376 {
 377    loop_variable_state *const ls = this->state->get(ir);
 378
 379    /* If we've entered a loop that hasn't been analyzed, something really,
 380     * really bad has happened.
 381     */
 382    if (ls == NULL) {
 383       assert(ls != NULL);
 384       return visit_continue;
 385    }
 386
 387    if (ls->limiting_terminator != NULL) {
 388       /* If the limiting terminator has an iteration count of zero, then we've
 389        * proven that the loop cannot run, so delete it.
 390        */
 391       int iterations = ls->limiting_terminator->iterations;
 392       if (iterations == 0) {
 393          ir->remove();
 394          this->progress = true;
 395          return visit_continue;
 396       }
 397    }
 398
 399    /* Remove the conditional break statements associated with all terminators
 400     * that are associated with a fixed iteration count, except for the one
 401     * associated with the limiting terminator--that one needs to stay, since
 402     * it terminates the loop.  Exception: if the loop still has a normative
 403     * bound, then that terminates the loop, so we don't even need the limiting
 404     * terminator.
 405     */
 406    foreach_in_list_safe(loop_terminator, t, &ls->terminators) {
 407       if (t->iterations < 0)
 408          continue;
 409
 410       exec_list *branch_instructions;
 411       if (t != ls->limiting_terminator) {
 412          ir_instruction *ir_if_last = (ir_instruction *)
 413             t->ir->then_instructions.get_tail();
 414          if (is_break(ir_if_last)) {
 415             branch_instructions = &t->ir->else_instructions;
 416          } else {
 417             branch_instructions = &t->ir->then_instructions;
 418             assert(is_break((ir_instruction *)
 419                             t->ir->else_instructions.get_tail()));
 420          }
 421
 422          exec_list copy_list;
 423          copy_list.make_empty();
 424          clone_ir_list(ir, &copy_list, branch_instructions);
 425
 426          t->ir->insert_before(&copy_list);
 427          t->ir->remove();
 428
 429          assert(ls->num_loop_jumps > 0);
 430          ls->num_loop_jumps--;
 431
 432          /* Also remove it from the terminator list */
 433          t->remove();
 434
 435          this->progress = true;
 436       }
 437    }
 438
 439    if (ls->limiting_terminator == NULL) {
 440       ir_instruction *last_ir =
 441          (ir_instruction *) ir->body_instructions.get_tail();
 442
 443       /* If a loop has no induction variable and the last instruction is
 444        * a break, unroll the loop with a count of 1.  This is the classic
 445        *
 446        *    do {
 447        *        // ...
 448        *    } while (false)
 449        *
 450        * that is used to wrap multi-line macros.
 451        *
 452        * If num_loop_jumps is not zero, last_ir cannot be NULL... there has to
 453        * be at least num_loop_jumps instructions in the loop.
 454        */
 455       if (ls->num_loop_jumps == 1 && is_break(last_ir)) {
 456          last_ir->remove();
 457
 458          simple_unroll(ir, 1);
 459       }
 460
 461       /* Don't try to unroll loops where the number of iterations is not known
 462        * at compile-time.
 463        */
 464       return visit_continue;
 465    }
 466
 467    int iterations = ls->limiting_terminator->iterations;
 468
 469    const int max_iterations = options->MaxUnrollIterations;
 470
 471    /* Don't try to unroll loops that have zillions of iterations either.
 472     */
 473    if (iterations > max_iterations)
 474       return visit_continue;
 475
 476    /* Don't try to unroll nested loops and loops with a huge body.
 477     */
 478    loop_unroll_count count(&ir->body_instructions, ls, options);
 479
 480    bool loop_too_large =
 481       count.nested_loop || count.nodes * iterations > max_iterations * 5;
 482
 483    if (loop_too_large && !count.unsupported_variable_indexing &&
 484        !count.array_indexed_by_induction_var_with_exact_iterations)
 485       return visit_continue;
 486
 487    /* Note: the limiting terminator contributes 1 to ls->num_loop_jumps.
 488     * We'll be removing the limiting terminator before we unroll.
 489     */
 490    assert(ls->num_loop_jumps > 0);
 491    unsigned predicted_num_loop_jumps = ls->num_loop_jumps - 1;
 492
 493    if (predicted_num_loop_jumps > 1)
 494       return visit_continue;
 495
 496    if (predicted_num_loop_jumps == 0) {
 497       simple_unroll(ir, iterations);
 498       return visit_continue;
 499    }
 500
 501    ir_instruction *last_ir = (ir_instruction *) ir->body_instructions.get_tail();
 502    assert(last_ir != NULL);
 503
 504    if (is_break(last_ir)) {
 505       /* If the only loop-jump is a break at the end of the loop, the loop
 506        * will execute exactly once.  Remove the break and use the simple
 507        * unroller with an iteration count of 1.
 508        */
 509       last_ir->remove();
 510
 511       simple_unroll(ir, 1);
 512       return visit_continue;
 513    }
 514
 515    /* Complex unrolling can only handle two terminators. One with an unknown
 516     * iteration count and one with a known iteration count. We have already
 517     * made sure we have a known iteration count above and removed any
 518     * unreachable terminators with a known count. Here we make sure there
 519     * isn't any additional unknown terminators, or any other jumps nested
 520     * inside futher ifs.
 521     */
 522    if (ls->num_loop_jumps != 2 || ls->terminators.length() != 2)
 523       return visit_continue;
 524
 525    ir_instruction *first_ir =
 526       (ir_instruction *) ir->body_instructions.get_head();
 527
 528    unsigned term_count = 0;
 529    bool first_term_then_continue = false;
 530    foreach_in_list(loop_terminator, t, &ls->terminators) {
 531       assert(term_count < 2);
 532
 533       ir_if *ir_if = t->ir->as_if();
 534       assert(ir_if != NULL);
 535
 536       ir_instruction *ir_if_last =
 537          (ir_instruction *) ir_if->then_instructions.get_tail();
 538
 539       if (is_break(ir_if_last)) {
 540          splice_post_if_instructions(ir_if, &ir_if->else_instructions);
 541          ir_if_last->remove();
 542          if (term_count == 1) {
 543             bool ebi =
 544                exit_branch_has_instructions(ls->limiting_terminator->ir,
 545                                             first_term_then_continue);
 546             complex_unroll(ir, iterations, false,
 547                            first_ir->as_if() != ls->limiting_terminator->ir ||
 548                            ebi,
 549                            first_term_then_continue);
 550             return visit_continue;
 551          }
 552       } else {
 553          ir_if_last =
 554             (ir_instruction *) ir_if->else_instructions.get_tail();
 555
 556          assert(is_break(ir_if_last));
 557          if (is_break(ir_if_last)) {
 558             splice_post_if_instructions(ir_if, &ir_if->then_instructions);
 559             ir_if_last->remove();
 560             if (term_count == 1) {
 561                bool ebi =
 562                   exit_branch_has_instructions(ls->limiting_terminator->ir,
 563                                                first_term_then_continue);
 564                complex_unroll(ir, iterations, true,
 565                               first_ir->as_if() != ls->limiting_terminator->ir ||
 566                               ebi,
 567                               first_term_then_continue);
 568                return visit_continue;
 569             } else {
 570                first_term_then_continue = true;
 571             }
 572          }
 573       }
 574
 575       term_count++;
 576    }
 577
 578    /* Did not find the break statement.  It must be in a complex if-nesting,
 579     * so don't try to unroll.
 580     */
 581    return visit_continue;
 582 }
 583
 584
 585 bool
 586 unroll_loops(exec_list *instructions, loop_state *ls,
 587              const struct gl_shader_compiler_options *options)
 588 {
 589    loop_unroll_visitor v(ls, options);
 590
 591    v.run(instructions);
 592
 593    return v.progress;
 594 }