src/glsl/nir/nir_opt_dead_cf.c

   1 /*
   2  * Copyright © 2014 Connor Abbott
   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  * Authors:
  24  *    Connor Abbott (cwabbott0@gmail.com)
  25  *
  26  */
  27
  28 #include "nir.h"
  29 #include "nir_control_flow.h"
  30
  31 /*
  32  * This file implements an optimization that deletes statically
  33  * unreachable/dead code. In NIR, one way this can happen if if an if
  34  * statement has a constant condition:
  35  *
  36  * if (true) {
  37  *    ...
  38  * }
  39  *
  40  * We delete the if statement and paste the contents of the always-executed
  41  * branch into the surrounding control flow, possibly removing more code if
  42  * the branch had a jump at the end.
  43  *
  44  * Another way is that control flow can end in a jump so that code after it
  45  * never gets executed. In particular, this can happen after optimizing
  46  * something like:
  47  *
  48  * if (true) {
  49  *    ...
  50  *    break;
  51  * }
  52  * ...
  53  *
  54  * We also consider the case where both branches of an if end in a jump, e.g.:
  55  *
  56  * if (...) {
  57  *    break;
  58  * } else {
  59  *    continue;
  60  * }
  61  * ...
  62  *
  63  * Finally, we also handle removing useless loops, i.e. loops with no side
  64  * effects and without any definitions that are used elsewhere. This case is a
  65  * little different from the first two in that the code is actually run (it
  66  * just never does anything), but there are similar issues with needing to
  67  * be careful with restarting after deleting the cf_node (see dead_cf_list())
  68  * so this is a convenient place to remove them.
  69  */
  70
  71 static void
  72 remove_after_cf_node(nir_cf_node *node)
  73 {
  74    nir_cf_node *end = node;
  75    while (!nir_cf_node_is_last(end))
  76       end = nir_cf_node_next(end);
  77
  78    nir_cf_list list;
  79    nir_cf_extract(&list, nir_after_cf_node(node), nir_after_cf_node(end));
  80    nir_cf_delete(&list);
  81 }
  82
  83 static void
  84 opt_constant_if(nir_if *if_stmt, bool condition)
  85 {
  86    /* First, we need to remove any phi nodes after the if by rewriting uses to
  87     * point to the correct source.
  88     */
  89    nir_block *after = nir_cf_node_as_block(nir_cf_node_next(&if_stmt->cf_node));
  90    nir_block *last_block =
  91       nir_cf_node_as_block(condition ? nir_if_last_then_node(if_stmt)
  92                                      : nir_if_last_else_node(if_stmt));
  93
  94    nir_foreach_instr_safe(after, instr) {
  95       if (instr->type != nir_instr_type_phi)
  96          break;
  97
  98       nir_phi_instr *phi = nir_instr_as_phi(instr);
  99       nir_ssa_def *def = NULL;
 100       nir_foreach_phi_src(phi, phi_src) {
 101          if (phi_src->pred != last_block)
 102             continue;
 103
 104          assert(phi_src->src.is_ssa);
 105          def = phi_src->src.ssa;
 106       }
 107
 108       assert(def);
 109       assert(phi->dest.is_ssa);
 110       nir_ssa_def_rewrite_uses(&phi->dest.ssa, nir_src_for_ssa(def));
 111       nir_instr_remove(instr);
 112    }
 113
 114    /* The control flow list we're about to paste in may include a jump at the
 115     * end, and in that case we have to delete the rest of the control flow
 116     * list after the if since it's unreachable and the validator will balk if
 117     * we don't.
 118     */
 119
 120    if (!exec_list_is_empty(&last_block->instr_list)) {
 121       nir_instr *last_instr = nir_block_last_instr(last_block);
 122       if (last_instr->type == nir_instr_type_jump)
 123          remove_after_cf_node(&if_stmt->cf_node);
 124    }
 125
 126    /* Finally, actually paste in the then or else branch and delete the if. */
 127    struct exec_list *cf_list = condition ? &if_stmt->then_list
 128                                          : &if_stmt->else_list;
 129
 130    nir_cf_list list;
 131    nir_cf_extract(&list, nir_before_cf_list(cf_list),
 132                   nir_after_cf_list(cf_list));
 133    nir_cf_reinsert(&list, nir_after_cf_node(&if_stmt->cf_node));
 134    nir_cf_node_remove(&if_stmt->cf_node);
 135 }
 136
 137 static bool
 138 block_has_no_side_effects(nir_block *block, void *state)
 139 {
 140    (void) state;
 141
 142    nir_foreach_instr(block, instr) {
 143       if (instr->type == nir_instr_type_call)
 144          return false;
 145
 146       /* Return instructions can cause us to skip over other side-effecting
 147        * instructions after the loop, so consider them to have side effects
 148        * here.
 149        */
 150
 151       if (instr->type == nir_instr_type_jump &&
 152           nir_instr_as_jump(instr)->type == nir_jump_return)
 153          return false;
 154
 155       if (instr->type != nir_instr_type_intrinsic)
 156          continue;
 157
 158       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 159       if (!nir_intrinsic_infos[intrin->intrinsic].flags &
 160           NIR_INTRINSIC_CAN_ELIMINATE)
 161          return false;
 162    }
 163
 164    return true;
 165 }
 166
 167 static bool
 168 def_not_live_out(nir_ssa_def *def, void *state)
 169 {
 170    nir_block *after = state;
 171
 172    return !BITSET_TEST(after->live_in, def->live_index);
 173 }
 174
 175 /*
 176  * Test if a loop is dead. A loop is dead if:
 177  *
 178  * 1) It has no side effects (i.e. intrinsics which could possibly affect the
 179  * state of the program aside from producing an SSA value, indicated by a lack
 180  * of NIR_INTRINSIC_CAN_ELIMINATE).
 181  *
 182  * 2) It has no phi nodes after it, since those indicate values inside the
 183  * loop being used after the loop.
 184  *
 185  * 3) If there are no phi nodes after the loop, then the only way a value
 186  * defined inside the loop can be used outside the loop is if its definition
 187  * dominates the block after the loop. If none of the definitions that
 188  * dominate the loop exit are used outside the loop, then the loop is dead
 189  * and it can be deleted.
 190  */
 191
 192 static bool
 193 loop_is_dead(nir_loop *loop)
 194 {
 195    nir_block *before = nir_cf_node_as_block(nir_cf_node_prev(&loop->cf_node));
 196    nir_block *after = nir_cf_node_as_block(nir_cf_node_next(&loop->cf_node));
 197
 198    if (!exec_list_is_empty(&after->instr_list) &&
 199        nir_block_first_instr(after)->type == nir_instr_type_phi)
 200       return false;
 201
 202    if (!nir_foreach_block_in_cf_node(&loop->cf_node, block_has_no_side_effects,
 203                                      NULL))
 204       return false;
 205
 206    for (nir_block *cur = after->imm_dom; cur != before; cur = cur->imm_dom) {
 207       nir_foreach_instr(cur, instr) {
 208          if (!nir_foreach_ssa_def(instr, def_not_live_out, after))
 209             return false;
 210       }
 211    }
 212
 213    return true;
 214 }
 215
 216 static bool
 217 dead_cf_block(nir_block *block)
 218 {
 219    nir_if *following_if = nir_block_get_following_if(block);
 220    if (following_if) {
 221      nir_const_value *const_value =
 222         nir_src_as_const_value(following_if->condition);
 223
 224      if (!const_value)
 225         return false;
 226
 227       opt_constant_if(following_if, const_value->u[0] != 0);
 228       return true;
 229    }
 230
 231    nir_loop *following_loop = nir_block_get_following_loop(block);
 232    if (!following_loop)
 233       return false;
 234
 235    if (!loop_is_dead(following_loop))
 236       return false;
 237
 238    nir_cf_node_remove(&following_loop->cf_node);
 239    return true;
 240 }
 241
 242 static bool
 243 ends_in_jump(nir_block *block)
 244 {
 245    if (exec_list_is_empty(&block->instr_list))
 246       return false;
 247
 248    nir_instr *instr = nir_block_last_instr(block);
 249    return instr->type == nir_instr_type_jump;
 250 }
 251
 252 static bool
 253 dead_cf_list(struct exec_list *list, bool *list_ends_in_jump)
 254 {
 255    bool progress = false;
 256    *list_ends_in_jump = false;
 257
 258    nir_cf_node *prev = NULL;
 259
 260    foreach_list_typed(nir_cf_node, cur, node, list) {
 261       switch (cur->type) {
 262       case nir_cf_node_block: {
 263          nir_block *block = nir_cf_node_as_block(cur);
 264          if (dead_cf_block(block)) {
 265             /* We just deleted the if or loop after this block, so we may have
 266              * deleted the block before or after it -- which one is an
 267              * implementation detail. Therefore, to recover the place we were
 268              * at, we have to use the previous cf_node.
 269              */
 270
 271             if (prev) {
 272                cur = nir_cf_node_next(prev);
 273             } else {
 274                cur = exec_node_data(nir_cf_node, exec_list_get_head(list),
 275                                     node);
 276             }
 277
 278             block = nir_cf_node_as_block(cur);
 279
 280             progress = true;
 281          }
 282
 283          if (ends_in_jump(block)) {
 284             *list_ends_in_jump = true;
 285
 286             if (!exec_node_is_tail_sentinel(cur->node.next)) {
 287                remove_after_cf_node(cur);
 288                return true;
 289             }
 290          }
 291
 292          break;
 293       }
 294
 295       case nir_cf_node_if: {
 296          nir_if *if_stmt = nir_cf_node_as_if(cur);
 297          bool then_ends_in_jump, else_ends_in_jump;
 298          progress |= dead_cf_list(&if_stmt->then_list, &then_ends_in_jump);
 299          progress |= dead_cf_list(&if_stmt->else_list, &else_ends_in_jump);
 300
 301          if (then_ends_in_jump && else_ends_in_jump) {
 302             *list_ends_in_jump = true;
 303             nir_block *next = nir_cf_node_as_block(nir_cf_node_next(cur));
 304             if (!exec_list_is_empty(&next->instr_list) ||
 305                 !exec_node_is_tail_sentinel(next->cf_node.node.next)) {
 306                remove_after_cf_node(cur);
 307                return true;
 308             }
 309          }
 310
 311          break;
 312       }
 313
 314       case nir_cf_node_loop: {
 315          nir_loop *loop = nir_cf_node_as_loop(cur);
 316          bool dummy;
 317          progress |= dead_cf_list(&loop->body, &dummy);
 318
 319          break;
 320       }
 321
 322       default:
 323          unreachable("unknown cf node type");
 324       }
 325
 326       prev = cur;
 327    }
 328
 329    return progress;
 330 }
 331
 332 static bool
 333 opt_dead_cf_impl(nir_function_impl *impl)
 334 {
 335    nir_metadata_require(impl, nir_metadata_live_variables |
 336                               nir_metadata_dominance);
 337
 338    bool dummy;
 339    bool progress = dead_cf_list(&impl->body, &dummy);
 340
 341    if (progress)
 342       nir_metadata_preserve(impl, nir_metadata_none);
 343
 344    return progress;
 345 }
 346
 347 bool
 348 nir_opt_dead_cf(nir_shader *shader)
 349 {
 350    bool progress = false;
 351
 352    nir_foreach_overload(shader, overload)
 353       if (overload->impl)
 354          progress |= opt_dead_cf_impl(overload->impl);
 355
 356    return progress;
 357 }