src/compiler/nir/nir_loop_analyze.c

   1 /*
   2  * Copyright © 2015 Thomas Helland
   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 DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 #include "nir.h"
  25 #include "nir_constant_expressions.h"
  26 #include "nir_loop_analyze.h"
  27
  28 typedef enum {
  29    undefined,
  30    invariant,
  31    not_invariant,
  32    basic_induction
  33 } nir_loop_variable_type;
  34
  35 struct nir_basic_induction_var;
  36
  37 typedef struct {
  38    /* A link for the work list */
  39    struct list_head process_link;
  40
  41    bool in_loop;
  42
  43    /* The ssa_def associated with this info */
  44    nir_ssa_def *def;
  45
  46    /* The type of this ssa_def */
  47    nir_loop_variable_type type;
  48
  49    /* If this is of type basic_induction */
  50    struct nir_basic_induction_var *ind;
  51
  52    /* True if variable is in an if branch or a nested loop */
  53    bool in_control_flow;
  54
  55 } nir_loop_variable;
  56
  57 typedef struct nir_basic_induction_var {
  58    nir_op alu_op;                           /* The type of alu-operation    */
  59    nir_loop_variable *alu_def;              /* The def of the alu-operation */
  60    nir_loop_variable *invariant;            /* The invariant alu-operand    */
  61    nir_loop_variable *def_outside_loop;     /* The phi-src outside the loop */
  62 } nir_basic_induction_var;
  63
  64 typedef struct {
  65    /* The loop we store information for */
  66    nir_loop *loop;
  67
  68    /* Loop_variable for all ssa_defs in function */
  69    nir_loop_variable *loop_vars;
  70
  71    /* A list of the loop_vars to analyze */
  72    struct list_head process_list;
  73
  74    nir_variable_mode indirect_mask;
  75
  76 } loop_info_state;
  77
  78 static nir_loop_variable *
  79 get_loop_var(nir_ssa_def *value, loop_info_state *state)
  80 {
  81    return &(state->loop_vars[value->index]);
  82 }
  83
  84 typedef struct {
  85    loop_info_state *state;
  86    bool in_control_flow;
  87 } init_loop_state;
  88
  89 static bool
  90 init_loop_def(nir_ssa_def *def, void *void_init_loop_state)
  91 {
  92    init_loop_state *loop_init_state = void_init_loop_state;
  93    nir_loop_variable *var = get_loop_var(def, loop_init_state->state);
  94
  95    if (loop_init_state->in_control_flow) {
  96       var->in_control_flow = true;
  97    } else {
  98       /* Add to the tail of the list. That way we start at the beginning of
  99        * the defs in the loop instead of the end when walking the list. This
 100        * means less recursive calls. Only add defs that are not in nested
 101        * loops or conditional blocks.
 102        */
 103       list_addtail(&var->process_link, &loop_init_state->state->process_list);
 104    }
 105
 106    var->in_loop = true;
 107
 108    return true;
 109 }
 110
 111 static bool
 112 init_loop_block(nir_block *block, loop_info_state *state,
 113                 bool in_control_flow)
 114 {
 115    init_loop_state init_state = {.in_control_flow = in_control_flow,
 116                                  .state = state };
 117
 118    nir_foreach_instr(instr, block) {
 119       if (instr->type == nir_instr_type_intrinsic ||
 120           instr->type == nir_instr_type_alu ||
 121           instr->type == nir_instr_type_tex) {
 122          state->loop->info->num_instructions++;
 123       }
 124
 125       nir_foreach_ssa_def(instr, init_loop_def, &init_state);
 126    }
 127
 128    return true;
 129 }
 130
 131 static inline bool
 132 is_var_alu(nir_loop_variable *var)
 133 {
 134    return var->def->parent_instr->type == nir_instr_type_alu;
 135 }
 136
 137 static inline bool
 138 is_var_constant(nir_loop_variable *var)
 139 {
 140    return var->def->parent_instr->type == nir_instr_type_load_const;
 141 }
 142
 143 static inline bool
 144 is_var_phi(nir_loop_variable *var)
 145 {
 146    return var->def->parent_instr->type == nir_instr_type_phi;
 147 }
 148
 149 static inline bool
 150 mark_invariant(nir_ssa_def *def, loop_info_state *state)
 151 {
 152    nir_loop_variable *var = get_loop_var(def, state);
 153
 154    if (var->type == invariant)
 155       return true;
 156
 157    if (!var->in_loop) {
 158       var->type = invariant;
 159       return true;
 160    }
 161
 162    if (var->type == not_invariant)
 163       return false;
 164
 165    if (is_var_alu(var)) {
 166       nir_alu_instr *alu = nir_instr_as_alu(def->parent_instr);
 167
 168       for (unsigned i = 0; i < nir_op_infos[alu->op].num_inputs; i++) {
 169          if (!mark_invariant(alu->src[i].src.ssa, state)) {
 170             var->type = not_invariant;
 171             return false;
 172          }
 173       }
 174       var->type = invariant;
 175       return true;
 176    }
 177
 178    /* Phis shouldn't be invariant except if one operand is invariant, and the
 179     * other is the phi itself. These should be removed by opt_remove_phis.
 180     * load_consts are already set to invariant and constant during init,
 181     * and so should return earlier. Remaining op_codes are set undefined.
 182     */
 183    var->type = not_invariant;
 184    return false;
 185 }
 186
 187 static void
 188 compute_invariance_information(loop_info_state *state)
 189 {
 190    /* An expression is invariant in a loop L if:
 191     *  (base cases)
 192     *    – it’s a constant
 193     *    – it’s a variable use, all of whose single defs are outside of L
 194     *  (inductive cases)
 195     *    – it’s a pure computation all of whose args are loop invariant
 196     *    – it’s a variable use whose single reaching def, and the
 197     *      rhs of that def is loop-invariant
 198     */
 199    list_for_each_entry_safe(nir_loop_variable, var, &state->process_list,
 200                             process_link) {
 201       assert(!var->in_control_flow);
 202
 203       if (mark_invariant(var->def, state))
 204          list_del(&var->process_link);
 205    }
 206 }
 207
 208 static bool
 209 compute_induction_information(loop_info_state *state)
 210 {
 211    bool found_induction_var = false;
 212    list_for_each_entry_safe(nir_loop_variable, var, &state->process_list,
 213                             process_link) {
 214
 215       /* It can't be an induction variable if it is invariant. Invariants and
 216        * things in nested loops or conditionals should have been removed from
 217        * the list by compute_invariance_information().
 218        */
 219       assert(!var->in_control_flow && var->type != invariant);
 220
 221       /* We are only interested in checking phis for the basic induction
 222        * variable case as its simple to detect. All basic induction variables
 223        * have a phi node
 224        */
 225       if (!is_var_phi(var))
 226          continue;
 227
 228       nir_phi_instr *phi = nir_instr_as_phi(var->def->parent_instr);
 229       nir_basic_induction_var *biv = rzalloc(state, nir_basic_induction_var);
 230
 231       nir_foreach_phi_src(src, phi) {
 232          nir_loop_variable *src_var = get_loop_var(src->src.ssa, state);
 233
 234          /* If one of the sources is in a conditional or nested block then
 235           * panic.
 236           */
 237          if (src_var->in_control_flow)
 238             break;
 239
 240          if (!src_var->in_loop) {
 241             biv->def_outside_loop = src_var;
 242          } else if (is_var_alu(src_var)) {
 243             nir_alu_instr *alu = nir_instr_as_alu(src_var->def->parent_instr);
 244
 245             if (nir_op_infos[alu->op].num_inputs == 2) {
 246                biv->alu_def = src_var;
 247                biv->alu_op = alu->op;
 248
 249                for (unsigned i = 0; i < 2; i++) {
 250                   /* Is one of the operands const, and the other the phi */
 251                   if (alu->src[i].src.ssa->parent_instr->type == nir_instr_type_load_const &&
 252                       alu->src[1-i].src.ssa == &phi->dest.ssa)
 253                      biv->invariant = get_loop_var(alu->src[i].src.ssa, state);
 254                }
 255             }
 256          }
 257       }
 258
 259       if (biv->alu_def && biv->def_outside_loop && biv->invariant &&
 260           is_var_constant(biv->def_outside_loop)) {
 261          assert(is_var_constant(biv->invariant));
 262          biv->alu_def->type = basic_induction;
 263          biv->alu_def->ind = biv;
 264          var->type = basic_induction;
 265          var->ind = biv;
 266
 267          found_induction_var = true;
 268       } else {
 269          ralloc_free(biv);
 270       }
 271    }
 272    return found_induction_var;
 273 }
 274
 275 static bool
 276 initialize_ssa_def(nir_ssa_def *def, void *void_state)
 277 {
 278    loop_info_state *state = void_state;
 279    nir_loop_variable *var = get_loop_var(def, state);
 280
 281    var->in_loop = false;
 282    var->def = def;
 283
 284    if (def->parent_instr->type == nir_instr_type_load_const) {
 285       var->type = invariant;
 286    } else {
 287       var->type = undefined;
 288    }
 289
 290    return true;
 291 }
 292
 293 static bool
 294 find_loop_terminators(loop_info_state *state)
 295 {
 296    bool success = false;
 297    foreach_list_typed_safe(nir_cf_node, node, node, &state->loop->body) {
 298       if (node->type == nir_cf_node_if) {
 299          nir_if *nif = nir_cf_node_as_if(node);
 300
 301          nir_block *break_blk = NULL;
 302          nir_block *continue_from_blk = NULL;
 303          bool continue_from_then = true;
 304
 305          nir_block *last_then = nir_if_last_then_block(nif);
 306          nir_block *last_else = nir_if_last_else_block(nif);
 307          if (nir_block_ends_in_break(last_then)) {
 308             break_blk = last_then;
 309             continue_from_blk = last_else;
 310             continue_from_then = false;
 311          } else if (nir_block_ends_in_break(last_else)) {
 312             break_blk = last_else;
 313             continue_from_blk = last_then;
 314          }
 315
 316          /* If there is a break then we should find a terminator. If we can
 317           * not find a loop terminator, but there is a break-statement then
 318           * we should return false so that we do not try to find trip-count
 319           */
 320          if (!nir_is_trivial_loop_if(nif, break_blk))
 321             return false;
 322
 323          /* Continue if the if contained no jumps at all */
 324          if (!break_blk)
 325             continue;
 326
 327          if (nif->condition.ssa->parent_instr->type == nir_instr_type_phi)
 328             return false;
 329
 330          nir_loop_terminator *terminator =
 331             rzalloc(state->loop->info, nir_loop_terminator);
 332
 333          list_addtail(&terminator->loop_terminator_link,
 334                       &state->loop->info->loop_terminator_list);
 335
 336          terminator->nif = nif;
 337          terminator->break_block = break_blk;
 338          terminator->continue_from_block = continue_from_blk;
 339          terminator->continue_from_then = continue_from_then;
 340          terminator->conditional_instr = nif->condition.ssa->parent_instr;
 341
 342          success = true;
 343       }
 344    }
 345
 346    return success;
 347 }
 348
 349 static int32_t
 350 get_iteration(nir_op cond_op, nir_const_value *initial, nir_const_value *step,
 351               nir_const_value *limit)
 352 {
 353    int32_t iter;
 354
 355    switch (cond_op) {
 356    case nir_op_ige:
 357    case nir_op_ilt:
 358    case nir_op_ieq:
 359    case nir_op_ine: {
 360       int32_t initial_val = initial->i32[0];
 361       int32_t span = limit->i32[0] - initial_val;
 362       iter = span / step->i32[0];
 363       break;
 364    }
 365    case nir_op_uge:
 366    case nir_op_ult: {
 367       uint32_t initial_val = initial->u32[0];
 368       uint32_t span = limit->u32[0] - initial_val;
 369       iter = span / step->u32[0];
 370       break;
 371    }
 372    case nir_op_fge:
 373    case nir_op_flt:
 374    case nir_op_feq:
 375    case nir_op_fne: {
 376       float initial_val = initial->f32[0];
 377       float span = limit->f32[0] - initial_val;
 378       iter = span / step->f32[0];
 379       break;
 380    }
 381    default:
 382       return -1;
 383    }
 384
 385    return iter;
 386 }
 387
 388 static bool
 389 test_iterations(int32_t iter_int, nir_const_value *step,
 390                 nir_const_value *limit, nir_op cond_op, unsigned bit_size,
 391                 nir_alu_type induction_base_type,
 392                 nir_const_value *initial, bool limit_rhs, bool invert_cond)
 393 {
 394    assert(nir_op_infos[cond_op].num_inputs == 2);
 395
 396    nir_const_value iter_src = { {0, } };
 397    nir_op mul_op;
 398    nir_op add_op;
 399    switch (induction_base_type) {
 400    case nir_type_float:
 401       iter_src.f32[0] = (float) iter_int;
 402       mul_op = nir_op_fmul;
 403       add_op = nir_op_fadd;
 404       break;
 405    case nir_type_int:
 406    case nir_type_uint:
 407       iter_src.i32[0] = iter_int;
 408       mul_op = nir_op_imul;
 409       add_op = nir_op_iadd;
 410       break;
 411    default:
 412       unreachable("Unhandled induction variable base type!");
 413    }
 414
 415    /* Multiple the iteration count we are testing by the number of times we
 416     * step the induction variable each iteration.
 417     */
 418    nir_const_value mul_src[2] = { iter_src, *step };
 419    nir_const_value mul_result =
 420       nir_eval_const_opcode(mul_op, 1, bit_size, mul_src);
 421
 422    /* Add the initial value to the accumulated induction variable total */
 423    nir_const_value add_src[2] = { mul_result, *initial };
 424    nir_const_value add_result =
 425       nir_eval_const_opcode(add_op, 1, bit_size, add_src);
 426
 427    nir_const_value src[2] = { { {0, } }, { {0, } } };
 428    src[limit_rhs ? 0 : 1] = add_result;
 429    src[limit_rhs ? 1 : 0] = *limit;
 430
 431    /* Evaluate the loop exit condition */
 432    nir_const_value result = nir_eval_const_opcode(cond_op, 1, bit_size, src);
 433
 434    return invert_cond ? (result.u32[0] == 0) : (result.u32[0] != 0);
 435 }
 436
 437 static int
 438 calculate_iterations(nir_const_value *initial, nir_const_value *step,
 439                      nir_const_value *limit, nir_loop_variable *alu_def,
 440                      nir_alu_instr *cond_alu, bool limit_rhs, bool invert_cond)
 441 {
 442    assert(initial != NULL && step != NULL && limit != NULL);
 443
 444    nir_alu_instr *alu = nir_instr_as_alu(alu_def->def->parent_instr);
 445
 446    /* nir_op_isub should have been lowered away by this point */
 447    assert(alu->op != nir_op_isub);
 448
 449    /* Make sure the alu type for our induction variable is compatible with the
 450     * conditional alus input type. If its not something has gone really wrong.
 451     */
 452    nir_alu_type induction_base_type =
 453       nir_alu_type_get_base_type(nir_op_infos[alu->op].output_type);
 454    if (induction_base_type == nir_type_int || induction_base_type == nir_type_uint) {
 455       assert(nir_alu_type_get_base_type(nir_op_infos[cond_alu->op].input_types[1]) == nir_type_int ||
 456              nir_alu_type_get_base_type(nir_op_infos[cond_alu->op].input_types[1]) == nir_type_uint);
 457    } else {
 458       assert(nir_alu_type_get_base_type(nir_op_infos[cond_alu->op].input_types[0]) ==
 459              induction_base_type);
 460    }
 461
 462    /* Check for nsupported alu operations */
 463    if (alu->op != nir_op_iadd && alu->op != nir_op_fadd)
 464       return -1;
 465
 466    /* do-while loops can increment the starting value before the condition is
 467     * checked. e.g.
 468     *
 469     *    do {
 470     *        ndx++;
 471     *     } while (ndx < 3);
 472     *
 473     * Here we check if the induction variable is used directly by the loop
 474     * condition and if so we assume we need to step the initial value.
 475     */
 476    unsigned trip_offset = 0;
 477    if (cond_alu->src[0].src.ssa == alu_def->def ||
 478        cond_alu->src[1].src.ssa == alu_def->def) {
 479       trip_offset = 1;
 480    }
 481
 482    int iter_int = get_iteration(cond_alu->op, initial, step, limit);
 483
 484    /* If iter_int is negative the loop is ill-formed or is the conditional is
 485     * unsigned with a huge iteration count so don't bother going any further.
 486     */
 487    if (iter_int < 0)
 488       return -1;
 489
 490    /* An explanation from the GLSL unrolling pass:
 491     *
 492     * Make sure that the calculated number of iterations satisfies the exit
 493     * condition.  This is needed to catch off-by-one errors and some types of
 494     * ill-formed loops.  For example, we need to detect that the following
 495     * loop does not have a maximum iteration count.
 496     *
 497     *    for (float x = 0.0; x != 0.9; x += 0.2);
 498     */
 499    assert(nir_src_bit_size(alu->src[0].src) ==
 500           nir_src_bit_size(alu->src[1].src));
 501    unsigned bit_size = nir_src_bit_size(alu->src[0].src);
 502    for (int bias = -1; bias <= 1; bias++) {
 503       const int iter_bias = iter_int + bias;
 504
 505       if (test_iterations(iter_bias, step, limit, cond_alu->op, bit_size,
 506                           induction_base_type, initial,
 507                           limit_rhs, invert_cond)) {
 508          return iter_bias > 0 ? iter_bias - trip_offset : iter_bias;
 509       }
 510    }
 511
 512    return -1;
 513 }
 514
 515 /* Run through each of the terminators of the loop and try to infer a possible
 516  * trip-count. We need to check them all, and set the lowest trip-count as the
 517  * trip-count of our loop. If one of the terminators has an undecidable
 518  * trip-count we can not safely assume anything about the duration of the
 519  * loop.
 520  */
 521 static void
 522 find_trip_count(loop_info_state *state)
 523 {
 524    bool trip_count_known = true;
 525    nir_loop_terminator *limiting_terminator = NULL;
 526    int min_trip_count = -1;
 527
 528    list_for_each_entry(nir_loop_terminator, terminator,
 529                        &state->loop->info->loop_terminator_list,
 530                        loop_terminator_link) {
 531
 532       if (terminator->conditional_instr->type != nir_instr_type_alu) {
 533          /* If we get here the loop is dead and will get cleaned up by the
 534           * nir_opt_dead_cf pass.
 535           */
 536          trip_count_known = false;
 537          continue;
 538       }
 539
 540       nir_alu_instr *alu = nir_instr_as_alu(terminator->conditional_instr);
 541       nir_loop_variable *basic_ind = NULL;
 542       nir_loop_variable *limit = NULL;
 543       bool limit_rhs = true;
 544
 545       switch (alu->op) {
 546       case nir_op_fge:      case nir_op_ige:      case nir_op_uge:
 547       case nir_op_flt:      case nir_op_ilt:      case nir_op_ult:
 548       case nir_op_feq:      case nir_op_ieq:
 549       case nir_op_fne:      case nir_op_ine:
 550
 551          /* We assume that the limit is the "right" operand */
 552          basic_ind = get_loop_var(alu->src[0].src.ssa, state);
 553          limit = get_loop_var(alu->src[1].src.ssa, state);
 554
 555          if (basic_ind->type != basic_induction) {
 556             /* We had it the wrong way, flip things around */
 557             basic_ind = get_loop_var(alu->src[1].src.ssa, state);
 558             limit = get_loop_var(alu->src[0].src.ssa, state);
 559             limit_rhs = false;
 560          }
 561
 562          /* The comparison has to have a basic induction variable
 563           * and a constant for us to be able to find trip counts
 564           */
 565          if (basic_ind->type != basic_induction || !is_var_constant(limit)) {
 566             trip_count_known = false;
 567             continue;
 568          }
 569
 570          /* We have determined that we have the following constants:
 571           * (With the typical int i = 0; i < x; i++; as an example)
 572           *    - Upper limit.
 573           *    - Starting value
 574           *    - Step / iteration size
 575           * Thats all thats needed to calculate the trip-count
 576           */
 577
 578          nir_const_value initial_val =
 579             nir_instr_as_load_const(basic_ind->ind->def_outside_loop->
 580                                        def->parent_instr)->value;
 581
 582          nir_const_value step_val =
 583             nir_instr_as_load_const(basic_ind->ind->invariant->def->
 584                                        parent_instr)->value;
 585
 586          nir_const_value limit_val =
 587             nir_instr_as_load_const(limit->def->parent_instr)->value;
 588
 589          int iterations = calculate_iterations(&initial_val, &step_val,
 590                                                &limit_val,
 591                                                basic_ind->ind->alu_def, alu,
 592                                                limit_rhs,
 593                                                terminator->continue_from_then);
 594
 595          /* Where we not able to calculate the iteration count */
 596          if (iterations == -1) {
 597             trip_count_known = false;
 598             continue;
 599          }
 600
 601          /* If this is the first run or we have found a smaller amount of
 602           * iterations than previously (we have identified a more limiting
 603           * terminator) set the trip count and limiting terminator.
 604           */
 605          if (min_trip_count == -1 || iterations < min_trip_count) {
 606             min_trip_count = iterations;
 607             limiting_terminator = terminator;
 608          }
 609          break;
 610
 611       default:
 612          trip_count_known = false;
 613       }
 614    }
 615
 616    state->loop->info->is_trip_count_known = trip_count_known;
 617    if (min_trip_count > -1)
 618       state->loop->info->trip_count = min_trip_count;
 619    state->loop->info->limiting_terminator = limiting_terminator;
 620 }
 621
 622 static bool
 623 force_unroll_array_access(loop_info_state *state, nir_shader *ns,
 624                           nir_deref_instr *deref)
 625 {
 626    for (nir_deref_instr *d = deref; d; d = nir_deref_instr_parent(d)) {
 627       if (d->deref_type != nir_deref_type_array)
 628          continue;
 629
 630       assert(d->arr.index.is_ssa);
 631       nir_loop_variable *array_index = get_loop_var(d->arr.index.ssa, state);
 632
 633       if (array_index->type != basic_induction)
 634          continue;
 635
 636       nir_deref_instr *parent = nir_deref_instr_parent(d);
 637       assert(glsl_type_is_array(parent->type) ||
 638              glsl_type_is_matrix(parent->type));
 639       if (glsl_get_length(parent->type) == state->loop->info->trip_count) {
 640          state->loop->info->force_unroll = true;
 641          return true;
 642       }
 643
 644       if (deref->mode & state->indirect_mask) {
 645          state->loop->info->force_unroll = true;
 646          return true;
 647       }
 648    }
 649
 650    return false;
 651 }
 652
 653 static bool
 654 force_unroll_heuristics(loop_info_state *state, nir_shader *ns,
 655                         nir_block *block)
 656 {
 657    nir_foreach_instr(instr, block) {
 658       if (instr->type != nir_instr_type_intrinsic)
 659          continue;
 660
 661       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 662
 663       /* Check for arrays variably-indexed by a loop induction variable.
 664        * Unrolling the loop may convert that access into constant-indexing.
 665        */
 666       if (intrin->intrinsic == nir_intrinsic_load_deref ||
 667           intrin->intrinsic == nir_intrinsic_store_deref ||
 668           intrin->intrinsic == nir_intrinsic_copy_deref) {
 669          if (force_unroll_array_access(state, ns,
 670                                        nir_src_as_deref(intrin->src[0])))
 671             return true;
 672
 673          if (intrin->intrinsic == nir_intrinsic_copy_deref &&
 674              force_unroll_array_access(state, ns,
 675                                        nir_src_as_deref(intrin->src[1])))
 676             return true;
 677       }
 678    }
 679
 680    return false;
 681 }
 682
 683 static void
 684 get_loop_info(loop_info_state *state, nir_function_impl *impl)
 685 {
 686    /* Initialize all variables to "outside_loop". This also marks defs
 687     * invariant and constant if they are nir_instr_type_load_consts
 688     */
 689    nir_foreach_block(block, impl) {
 690       nir_foreach_instr(instr, block)
 691          nir_foreach_ssa_def(instr, initialize_ssa_def, state);
 692    }
 693
 694    /* Add all entries in the outermost part of the loop to the processing list
 695     * Mark the entries in conditionals or in nested loops accordingly
 696     */
 697    foreach_list_typed_safe(nir_cf_node, node, node, &state->loop->body) {
 698       switch (node->type) {
 699
 700       case nir_cf_node_block:
 701          init_loop_block(nir_cf_node_as_block(node), state, false);
 702          break;
 703
 704       case nir_cf_node_if:
 705          nir_foreach_block_in_cf_node(block, node)
 706             init_loop_block(block, state, true);
 707          break;
 708
 709       case nir_cf_node_loop:
 710          nir_foreach_block_in_cf_node(block, node) {
 711             init_loop_block(block, state, true);
 712          }
 713          break;
 714
 715       case nir_cf_node_function:
 716          break;
 717       }
 718    }
 719
 720    /* Try to find all simple terminators of the loop. If we can't find any,
 721     * or we find possible terminators that have side effects then bail.
 722     */
 723    if (!find_loop_terminators(state)) {
 724       list_for_each_entry_safe(nir_loop_terminator, terminator,
 725                                &state->loop->info->loop_terminator_list,
 726                                loop_terminator_link) {
 727          list_del(&terminator->loop_terminator_link);
 728          ralloc_free(terminator);
 729       }
 730       return;
 731    }
 732
 733    /* Induction analysis needs invariance information so get that first */
 734    compute_invariance_information(state);
 735
 736    /* We have invariance information so try to find induction variables */
 737    if (!compute_induction_information(state))
 738       return;
 739
 740    /* Run through each of the terminators and try to compute a trip-count */
 741    find_trip_count(state);
 742
 743    nir_shader *ns = impl->function->shader;
 744    foreach_list_typed_safe(nir_cf_node, node, node, &state->loop->body) {
 745       if (node->type == nir_cf_node_block) {
 746          if (force_unroll_heuristics(state, ns, nir_cf_node_as_block(node)))
 747             break;
 748       } else {
 749          nir_foreach_block_in_cf_node(block, node) {
 750             if (force_unroll_heuristics(state, ns, block))
 751                break;
 752          }
 753       }
 754    }
 755 }
 756
 757 static loop_info_state *
 758 initialize_loop_info_state(nir_loop *loop, void *mem_ctx,
 759                            nir_function_impl *impl)
 760 {
 761    loop_info_state *state = rzalloc(mem_ctx, loop_info_state);
 762    state->loop_vars = rzalloc_array(mem_ctx, nir_loop_variable,
 763                                     impl->ssa_alloc);
 764    state->loop = loop;
 765
 766    list_inithead(&state->process_list);
 767
 768    if (loop->info)
 769      ralloc_free(loop->info);
 770
 771    loop->info = rzalloc(loop, nir_loop_info);
 772
 773    list_inithead(&loop->info->loop_terminator_list);
 774
 775    return state;
 776 }
 777
 778 static void
 779 process_loops(nir_cf_node *cf_node, nir_variable_mode indirect_mask)
 780 {
 781    switch (cf_node->type) {
 782    case nir_cf_node_block:
 783       return;
 784    case nir_cf_node_if: {
 785       nir_if *if_stmt = nir_cf_node_as_if(cf_node);
 786       foreach_list_typed(nir_cf_node, nested_node, node, &if_stmt->then_list)
 787          process_loops(nested_node, indirect_mask);
 788       foreach_list_typed(nir_cf_node, nested_node, node, &if_stmt->else_list)
 789          process_loops(nested_node, indirect_mask);
 790       return;
 791    }
 792    case nir_cf_node_loop: {
 793       nir_loop *loop = nir_cf_node_as_loop(cf_node);
 794       foreach_list_typed(nir_cf_node, nested_node, node, &loop->body)
 795          process_loops(nested_node, indirect_mask);
 796       break;
 797    }
 798    default:
 799       unreachable("unknown cf node type");
 800    }
 801
 802    nir_loop *loop = nir_cf_node_as_loop(cf_node);
 803    nir_function_impl *impl = nir_cf_node_get_function(cf_node);
 804    void *mem_ctx = ralloc_context(NULL);
 805
 806    loop_info_state *state = initialize_loop_info_state(loop, mem_ctx, impl);
 807    state->indirect_mask = indirect_mask;
 808
 809    get_loop_info(state, impl);
 810
 811    ralloc_free(mem_ctx);
 812 }
 813
 814 void
 815 nir_loop_analyze_impl(nir_function_impl *impl,
 816                       nir_variable_mode indirect_mask)
 817 {
 818    nir_index_ssa_defs(impl);
 819    foreach_list_typed(nir_cf_node, node, node, &impl->body)
 820       process_loops(node, indirect_mask);
 821 }