src/compiler/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 is when 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 = condition ? nir_if_last_then_block(if_stmt)
  91                                      : nir_if_last_else_block(if_stmt);
  92
  93    nir_foreach_instr_safe(instr, after) {
  94       if (instr->type != nir_instr_type_phi)
  95          break;
  96
  97       nir_phi_instr *phi = nir_instr_as_phi(instr);
  98       nir_ssa_def *def = NULL;
  99       nir_foreach_phi_src(phi_src, phi) {
 100          if (phi_src->pred != last_block)
 101             continue;
 102
 103          assert(phi_src->src.is_ssa);
 104          def = phi_src->src.ssa;
 105       }
 106
 107       assert(def);
 108       assert(phi->dest.is_ssa);
 109       nir_ssa_def_rewrite_uses(&phi->dest.ssa, nir_src_for_ssa(def));
 110       nir_instr_remove(instr);
 111    }
 112
 113    /* The control flow list we're about to paste in may include a jump at the
 114     * end, and in that case we have to delete the rest of the control flow
 115     * list after the if since it's unreachable and the validator will balk if
 116     * we don't.
 117     */
 118
 119    if (!exec_list_is_empty(&last_block->instr_list)) {
 120       nir_instr *last_instr = nir_block_last_instr(last_block);
 121       if (last_instr->type == nir_instr_type_jump)
 122          remove_after_cf_node(&if_stmt->cf_node);
 123    }
 124
 125    /* Finally, actually paste in the then or else branch and delete the if. */
 126    struct exec_list *cf_list = condition ? &if_stmt->then_list
 127                                          : &if_stmt->else_list;
 128
 129    nir_cf_list list;
 130    nir_cf_list_extract(&list, cf_list);
 131    nir_cf_reinsert(&list, nir_after_cf_node(&if_stmt->cf_node));
 132    nir_cf_node_remove(&if_stmt->cf_node);
 133 }
 134
 135 static bool
 136 cf_node_has_side_effects(nir_cf_node *node)
 137 {
 138    nir_foreach_block_in_cf_node(block, node) {
 139       nir_foreach_instr(instr, block) {
 140          if (instr->type == nir_instr_type_call)
 141             return true;
 142
 143          /* Return instructions can cause us to skip over other side-effecting
 144           * instructions after the loop, so consider them to have side effects
 145           * here.
 146           */
 147
 148          if (instr->type == nir_instr_type_jump &&
 149              nir_instr_as_jump(instr)->type == nir_jump_return)
 150             return true;
 151
 152          if (instr->type != nir_instr_type_intrinsic)
 153             continue;
 154
 155          nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 156          if (!(nir_intrinsic_infos[intrin->intrinsic].flags &
 157              NIR_INTRINSIC_CAN_ELIMINATE))
 158             return true;
 159       }
 160    }
 161
 162    return false;
 163 }
 164
 165 static bool
 166 def_not_live_out(nir_ssa_def *def, void *state)
 167 {
 168    nir_block *after = state;
 169
 170    return !BITSET_TEST(after->live_in, def->live_index);
 171 }
 172
 173 /*
 174  * Test if a loop is dead. A loop is dead if:
 175  *
 176  * 1) It has no side effects (i.e. intrinsics which could possibly affect the
 177  * state of the program aside from producing an SSA value, indicated by a lack
 178  * of NIR_INTRINSIC_CAN_ELIMINATE).
 179  *
 180  * 2) It has no phi nodes after it, since those indicate values inside the
 181  * loop being used after the loop.
 182  *
 183  * 3) If there are no phi nodes after the loop, then the only way a value
 184  * defined inside the loop can be used outside the loop is if its definition
 185  * dominates the block after the loop. If none of the definitions that
 186  * dominate the loop exit are used outside the loop, then the loop is dead
 187  * and it can be deleted.
 188  */
 189
 190 static bool
 191 loop_is_dead(nir_loop *loop)
 192 {
 193    nir_block *before = nir_cf_node_as_block(nir_cf_node_prev(&loop->cf_node));
 194    nir_block *after = nir_cf_node_as_block(nir_cf_node_next(&loop->cf_node));
 195
 196    if (!exec_list_is_empty(&after->instr_list) &&
 197        nir_block_first_instr(after)->type == nir_instr_type_phi)
 198       return false;
 199
 200    if (cf_node_has_side_effects(&loop->cf_node))
 201       return false;
 202
 203    nir_function_impl *impl = nir_cf_node_get_function(&loop->cf_node);
 204    nir_metadata_require(impl, nir_metadata_live_ssa_defs |
 205                               nir_metadata_dominance);
 206
 207    for (nir_block *cur = after->imm_dom; cur && cur != before;
 208         cur = cur->imm_dom) {
 209       nir_foreach_instr(instr, cur) {
 210          if (!nir_foreach_ssa_def(instr, def_not_live_out, after))
 211             return false;
 212       }
 213    }
 214
 215    return true;
 216 }
 217
 218 static bool
 219 dead_cf_block(nir_block *block)
 220 {
 221    nir_if *following_if = nir_block_get_following_if(block);
 222    if (following_if) {
 223       nir_const_value *const_value =
 224          nir_src_as_const_value(following_if->condition);
 225
 226       if (!const_value)
 227          return false;
 228
 229       opt_constant_if(following_if, const_value->u32[0] != 0);
 230       return true;
 231    }
 232
 233    nir_loop *following_loop = nir_block_get_following_loop(block);
 234    if (!following_loop)
 235       return false;
 236
 237    if (!loop_is_dead(following_loop))
 238       return false;
 239
 240    nir_cf_node_remove(&following_loop->cf_node);
 241    return true;
 242 }
 243
 244 static bool
 245 ends_in_jump(nir_block *block)
 246 {
 247    if (exec_list_is_empty(&block->instr_list))
 248       return false;
 249
 250    nir_instr *instr = nir_block_last_instr(block);
 251    return instr->type == nir_instr_type_jump;
 252 }
 253
 254 static bool
 255 dead_cf_list(struct exec_list *list, bool *list_ends_in_jump)
 256 {
 257    bool progress = false;
 258    *list_ends_in_jump = false;
 259
 260    nir_cf_node *prev = NULL;
 261
 262    foreach_list_typed(nir_cf_node, cur, node, list) {
 263       switch (cur->type) {
 264       case nir_cf_node_block: {
 265          nir_block *block = nir_cf_node_as_block(cur);
 266          if (dead_cf_block(block)) {
 267             /* We just deleted the if or loop after this block, so we may have
 268              * deleted the block before or after it -- which one is an
 269              * implementation detail. Therefore, to recover the place we were
 270              * at, we have to use the previous cf_node.
 271              */
 272
 273             if (prev) {
 274                cur = nir_cf_node_next(prev);
 275             } else {
 276                cur = exec_node_data(nir_cf_node, exec_list_get_head(list),
 277                                     node);
 278             }
 279
 280             block = nir_cf_node_as_block(cur);
 281
 282             progress = true;
 283          }
 284
 285          if (ends_in_jump(block)) {
 286             *list_ends_in_jump = true;
 287
 288             if (!exec_node_is_tail_sentinel(cur->node.next)) {
 289                remove_after_cf_node(cur);
 290                return true;
 291             }
 292          }
 293
 294          break;
 295       }
 296
 297       case nir_cf_node_if: {
 298          nir_if *if_stmt = nir_cf_node_as_if(cur);
 299          bool then_ends_in_jump, else_ends_in_jump;
 300          progress |= dead_cf_list(&if_stmt->then_list, &then_ends_in_jump);
 301          progress |= dead_cf_list(&if_stmt->else_list, &else_ends_in_jump);
 302
 303          if (then_ends_in_jump && else_ends_in_jump) {
 304             *list_ends_in_jump = true;
 305             nir_block *next = nir_cf_node_as_block(nir_cf_node_next(cur));
 306             if (!exec_list_is_empty(&next->instr_list) ||
 307                 !exec_node_is_tail_sentinel(next->cf_node.node.next)) {
 308                remove_after_cf_node(cur);
 309                return true;
 310             }
 311          }
 312
 313          break;
 314       }
 315
 316       case nir_cf_node_loop: {
 317          nir_loop *loop = nir_cf_node_as_loop(cur);
 318          bool dummy;
 319          progress |= dead_cf_list(&loop->body, &dummy);
 320
 321          break;
 322       }
 323
 324       default:
 325          unreachable("unknown cf node type");
 326       }
 327
 328       prev = cur;
 329    }
 330
 331    return progress;
 332 }
 333
 334 static bool
 335 opt_dead_cf_impl(nir_function_impl *impl)
 336 {
 337    bool dummy;
 338    bool progress = dead_cf_list(&impl->body, &dummy);
 339
 340    if (progress)
 341       nir_metadata_preserve(impl, nir_metadata_none);
 342
 343    return progress;
 344 }
 345
 346 bool
 347 nir_opt_dead_cf(nir_shader *shader)
 348 {
 349    bool progress = false;
 350
 351    nir_foreach_function(function, shader)
 352       if (function->impl)
 353          progress |= opt_dead_cf_impl(function->impl);
 354
 355    return progress;
 356 }