src/glsl/nir/nir_lower_locals_to_regs.c

   1 /*
   2  * Copyright © 2014 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 DEALINGS
  21  * IN THE SOFTWARE.
  22  *
  23  * Authors:
  24  *    Jason Ekstrand (jason@jlekstrand.net)
  25  *
  26  */
  27
  28 #include "nir.h"
  29 #include "nir_array.h"
  30
  31 struct locals_to_regs_state {
  32    nir_shader *shader;
  33    nir_function_impl *impl;
  34
  35    /* A hash table mapping derefs to registers */
  36    struct hash_table *regs_table;
  37
  38    /* A growing array of derefs that we have encountered.  There is exactly
  39     * one element of this array per element in the hash table.  This is
  40     * used to make adding register initialization code deterministic.
  41     */
  42    nir_array derefs_array;
  43 };
  44
  45 /* The following two functions implement a hash and equality check for
  46  * variable dreferences.  When the hash or equality function encounters an
  47  * array, it ignores the offset and whether it is direct or indirect
  48  * entirely.
  49  */
  50 static uint32_t
  51 hash_deref(const void *void_deref)
  52 {
  53    uint32_t hash = _mesa_fnv32_1a_offset_bias;
  54
  55    const nir_deref_var *deref_var = void_deref;
  56    hash = _mesa_fnv32_1a_accumulate(hash, deref_var->var);
  57
  58    for (const nir_deref *deref = deref_var->deref.child;
  59         deref; deref = deref->child) {
  60       if (deref->deref_type == nir_deref_type_struct) {
  61          const nir_deref_struct *deref_struct = nir_deref_as_struct(deref);
  62          hash = _mesa_fnv32_1a_accumulate(hash, deref_struct->index);
  63       }
  64    }
  65
  66    return hash;
  67 }
  68
  69 static bool
  70 derefs_equal(const void *void_a, const void *void_b)
  71 {
  72    const nir_deref_var *a_var = void_a;
  73    const nir_deref_var *b_var = void_b;
  74
  75    if (a_var->var != b_var->var)
  76       return false;
  77
  78    for (const nir_deref *a = a_var->deref.child, *b = b_var->deref.child;
  79         a != NULL; a = a->child, b = b->child) {
  80       if (a->deref_type != b->deref_type)
  81          return false;
  82
  83       if (a->deref_type == nir_deref_type_struct) {
  84          if (nir_deref_as_struct(a)->index != nir_deref_as_struct(b)->index)
  85             return false;
  86       }
  87       /* Do nothing for arrays.  They're all the same. */
  88
  89       assert((a->child == NULL) == (b->child == NULL));
  90       if((a->child == NULL) != (b->child == NULL))
  91          return false;
  92    }
  93
  94    return true;
  95 }
  96
  97 static nir_register *
  98 get_reg_for_deref(nir_deref_var *deref, struct locals_to_regs_state *state)
  99 {
 100    uint32_t hash = hash_deref(deref);
 101
 102    struct hash_entry *entry =
 103       _mesa_hash_table_search_pre_hashed(state->regs_table, hash, deref);
 104    if (entry)
 105       return entry->data;
 106
 107    unsigned array_size = 1;
 108    nir_deref *tail = &deref->deref;
 109    while (tail->child) {
 110       if (tail->child->deref_type == nir_deref_type_array)
 111          array_size *= glsl_get_length(tail->type);
 112       tail = tail->child;
 113    }
 114
 115    assert(glsl_type_is_vector(tail->type) || glsl_type_is_scalar(tail->type));
 116
 117    nir_register *reg = nir_local_reg_create(state->impl);
 118    reg->num_components = glsl_get_vector_elements(tail->type);
 119    reg->num_array_elems = array_size > 1 ? array_size : 0;
 120
 121    _mesa_hash_table_insert_pre_hashed(state->regs_table, hash, deref, reg);
 122    nir_array_add(&state->derefs_array, nir_deref_var *, deref);
 123
 124    return reg;
 125 }
 126
 127 static nir_src
 128 get_deref_reg_src(nir_deref_var *deref, nir_instr *instr,
 129                   struct locals_to_regs_state *state)
 130 {
 131    nir_src src;
 132
 133    src.is_ssa = false;
 134    src.reg.reg = get_reg_for_deref(deref, state);
 135    src.reg.base_offset = 0;
 136    src.reg.indirect = NULL;
 137
 138    /* It is possible for a user to create a shader that has an array with a
 139     * single element and then proceed to access it indirectly.  Indirectly
 140     * accessing a non-array register is not allowed in NIR.  In order to
 141     * handle this case we just convert it to a direct reference.
 142     */
 143    if (src.reg.reg->num_array_elems == 0)
 144       return src;
 145
 146    nir_deref *tail = &deref->deref;
 147    while (tail->child != NULL) {
 148       const struct glsl_type *parent_type = tail->type;
 149       tail = tail->child;
 150
 151       if (tail->deref_type != nir_deref_type_array)
 152          continue;
 153
 154       nir_deref_array *deref_array = nir_deref_as_array(tail);
 155
 156       src.reg.base_offset *= glsl_get_length(parent_type);
 157       src.reg.base_offset += deref_array->base_offset;
 158
 159       if (src.reg.indirect) {
 160          nir_load_const_instr *load_const =
 161             nir_load_const_instr_create(state->shader, 1);
 162          load_const->value.u[0] = glsl_get_length(parent_type);
 163          nir_instr_insert_before(instr, &load_const->instr);
 164
 165          nir_alu_instr *mul = nir_alu_instr_create(state->shader, nir_op_imul);
 166          mul->src[0].src = *src.reg.indirect;
 167          mul->src[1].src.is_ssa = true;
 168          mul->src[1].src.ssa = &load_const->def;
 169          mul->dest.write_mask = 1;
 170          nir_ssa_dest_init(&mul->instr, &mul->dest.dest, 1, NULL);
 171          nir_instr_insert_before(instr, &mul->instr);
 172
 173          src.reg.indirect->is_ssa = true;
 174          src.reg.indirect->ssa = &mul->dest.dest.ssa;
 175       }
 176
 177       if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
 178          if (src.reg.indirect == NULL) {
 179             src.reg.indirect = ralloc(state->shader, nir_src);
 180             nir_src_copy(src.reg.indirect, &deref_array->indirect,
 181                          state->shader);
 182          } else {
 183             nir_alu_instr *add = nir_alu_instr_create(state->shader,
 184                                                       nir_op_iadd);
 185             add->src[0].src = *src.reg.indirect;
 186             nir_src_copy(&add->src[1].src, &deref_array->indirect, add);
 187             add->dest.write_mask = 1;
 188             nir_ssa_dest_init(&add->instr, &add->dest.dest, 1, NULL);
 189             nir_instr_insert_before(instr, &add->instr);
 190
 191             src.reg.indirect->is_ssa = true;
 192             src.reg.indirect->ssa = &add->dest.dest.ssa;
 193          }
 194       }
 195    }
 196
 197    return src;
 198 }
 199
 200 static bool
 201 lower_locals_to_regs_block(nir_block *block, void *void_state)
 202 {
 203    struct locals_to_regs_state *state = void_state;
 204
 205    nir_foreach_instr_safe(block, instr) {
 206       if (instr->type != nir_instr_type_intrinsic)
 207          continue;
 208
 209       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 210
 211       switch (intrin->intrinsic) {
 212       case nir_intrinsic_load_var: {
 213          if (intrin->variables[0]->var->data.mode != nir_var_local)
 214             continue;
 215
 216          nir_alu_instr *mov = nir_alu_instr_create(state->shader, nir_op_imov);
 217          mov->src[0].src = get_deref_reg_src(intrin->variables[0],
 218                                              &intrin->instr, state);
 219          mov->dest.write_mask = (1 << intrin->num_components) - 1;
 220          if (intrin->dest.is_ssa) {
 221             nir_ssa_dest_init(&mov->instr, &mov->dest.dest,
 222                               intrin->num_components, NULL);
 223             nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
 224                                      nir_src_for_ssa(&mov->dest.dest.ssa));
 225          } else {
 226             nir_dest_copy(&mov->dest.dest, &intrin->dest, &mov->instr);
 227          }
 228          nir_instr_insert_before(&intrin->instr, &mov->instr);
 229
 230          nir_instr_remove(&intrin->instr);
 231          break;
 232       }
 233
 234       case nir_intrinsic_store_var: {
 235          if (intrin->variables[0]->var->data.mode != nir_var_local)
 236             continue;
 237
 238          nir_src reg_src = get_deref_reg_src(intrin->variables[0],
 239                                              &intrin->instr, state);
 240
 241          nir_alu_instr *mov = nir_alu_instr_create(state->shader, nir_op_imov);
 242          nir_src_copy(&mov->src[0].src, &intrin->src[0], mov);
 243          mov->dest.write_mask = (1 << intrin->num_components) - 1;
 244          mov->dest.dest.is_ssa = false;
 245          mov->dest.dest.reg.reg = reg_src.reg.reg;
 246          mov->dest.dest.reg.base_offset = reg_src.reg.base_offset;
 247          mov->dest.dest.reg.indirect = reg_src.reg.indirect;
 248
 249          nir_instr_insert_before(&intrin->instr, &mov->instr);
 250
 251          nir_instr_remove(&intrin->instr);
 252          break;
 253       }
 254
 255       case nir_intrinsic_copy_var:
 256          unreachable("There should be no copies whatsoever at this point");
 257          break;
 258
 259       default:
 260          continue;
 261       }
 262    }
 263
 264    return true;
 265 }
 266
 267 static nir_block *
 268 compute_reg_usedef_lca(nir_register *reg)
 269 {
 270    nir_block *lca = NULL;
 271
 272    list_for_each_entry(nir_dest, def_dest, &reg->defs, reg.def_link)
 273       lca = nir_dominance_lca(lca, def_dest->reg.parent_instr->block);
 274
 275    list_for_each_entry(nir_src, use_src, &reg->uses, use_link)
 276       lca = nir_dominance_lca(lca, use_src->parent_instr->block);
 277
 278    list_for_each_entry(nir_src, use_src, &reg->if_uses, use_link) {
 279       nir_cf_node *prev_node = nir_cf_node_prev(&use_src->parent_if->cf_node);
 280       assert(prev_node->type == nir_cf_node_block);
 281       lca = nir_dominance_lca(lca, nir_cf_node_as_block(prev_node));
 282    }
 283
 284    return lca;
 285 }
 286
 287 static void
 288 insert_constant_initializer(nir_deref_var *deref_head, nir_deref *deref_tail,
 289                             nir_block *block,
 290                             struct locals_to_regs_state *state)
 291 {
 292    if (deref_tail->child) {
 293       switch (deref_tail->child->deref_type) {
 294       case nir_deref_type_array: {
 295          unsigned array_elems = glsl_get_length(deref_tail->type);
 296
 297          nir_deref_array arr_deref;
 298          arr_deref.deref = *deref_tail->child;
 299          arr_deref.deref_array_type = nir_deref_array_type_direct;
 300
 301          nir_deref *old_child = deref_tail->child;
 302          deref_tail->child = &arr_deref.deref;
 303          for (unsigned i = 0; i < array_elems; i++) {
 304             arr_deref.base_offset = i;
 305             insert_constant_initializer(deref_head, &arr_deref.deref,
 306                                         block, state);
 307          }
 308          deref_tail->child = old_child;
 309          return;
 310       }
 311
 312       case nir_deref_type_struct:
 313          insert_constant_initializer(deref_head, deref_tail->child,
 314                                      block, state);
 315          return;
 316
 317       default:
 318          unreachable("Invalid deref child type");
 319       }
 320    }
 321
 322    assert(deref_tail->child == NULL);
 323
 324    nir_load_const_instr *load =
 325       nir_deref_get_const_initializer_load(state->shader, deref_head);
 326    nir_instr_insert_before_block(block, &load->instr);
 327
 328    nir_src reg_src = get_deref_reg_src(deref_head, &load->instr, state);
 329
 330    nir_alu_instr *mov = nir_alu_instr_create(state->shader, nir_op_imov);
 331    mov->src[0].src = nir_src_for_ssa(&load->def);
 332    mov->dest.write_mask = (1 << load->def.num_components) - 1;
 333    mov->dest.dest.is_ssa = false;
 334    mov->dest.dest.reg.reg = reg_src.reg.reg;
 335    mov->dest.dest.reg.base_offset = reg_src.reg.base_offset;
 336    mov->dest.dest.reg.indirect = reg_src.reg.indirect;
 337
 338    nir_instr_insert_after(&load->instr, &mov->instr);
 339 }
 340
 341 static void
 342 nir_lower_locals_to_regs_impl(nir_function_impl *impl)
 343 {
 344    struct locals_to_regs_state state;
 345
 346    state.shader = impl->overload->function->shader;
 347    state.impl = impl;
 348    state.regs_table = _mesa_hash_table_create(NULL, hash_deref, derefs_equal);
 349    nir_array_init(&state.derefs_array, NULL);
 350
 351    nir_metadata_require(impl, nir_metadata_dominance);
 352
 353    nir_foreach_block(impl, lower_locals_to_regs_block, &state);
 354
 355    nir_array_foreach(&state.derefs_array, nir_deref_var *, deref_ptr) {
 356       nir_deref_var *deref = *deref_ptr;
 357       struct hash_entry *deref_entry =
 358          _mesa_hash_table_search(state.regs_table, deref);
 359       assert(deref_entry && deref_entry->key == deref);
 360       nir_register *reg = (nir_register *)deref_entry->data;
 361
 362       if (deref->var->constant_initializer == NULL)
 363          continue;
 364
 365       nir_block *usedef_lca = compute_reg_usedef_lca(reg);
 366
 367       insert_constant_initializer(deref, &deref->deref, usedef_lca, &state);
 368    }
 369
 370    nir_metadata_preserve(impl, nir_metadata_block_index |
 371                                nir_metadata_dominance);
 372
 373    nir_array_fini(&state.derefs_array);
 374    _mesa_hash_table_destroy(state.regs_table, NULL);
 375 }
 376
 377 void
 378 nir_lower_locals_to_regs(nir_shader *shader)
 379 {
 380    nir_foreach_overload(shader, overload) {
 381       if (overload->impl)
 382          nir_lower_locals_to_regs_impl(overload->impl);
 383    }
 384 }