src/compiler/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    bool progress;
  45 };
  46
  47 /* The following two functions implement a hash and equality check for
  48  * variable dreferences.  When the hash or equality function encounters an
  49  * array, it ignores the offset and whether it is direct or indirect
  50  * entirely.
  51  */
  52 static uint32_t
  53 hash_deref(const void *void_deref)
  54 {
  55    uint32_t hash = _mesa_fnv32_1a_offset_bias;
  56
  57    const nir_deref_var *deref_var = void_deref;
  58    hash = _mesa_fnv32_1a_accumulate(hash, deref_var->var);
  59
  60    for (const nir_deref *deref = deref_var->deref.child;
  61         deref; deref = deref->child) {
  62       if (deref->deref_type == nir_deref_type_struct) {
  63          const nir_deref_struct *deref_struct = nir_deref_as_struct(deref);
  64          hash = _mesa_fnv32_1a_accumulate(hash, deref_struct->index);
  65       }
  66    }
  67
  68    return hash;
  69 }
  70
  71 static bool
  72 derefs_equal(const void *void_a, const void *void_b)
  73 {
  74    const nir_deref_var *a_var = void_a;
  75    const nir_deref_var *b_var = void_b;
  76
  77    if (a_var->var != b_var->var)
  78       return false;
  79
  80    for (const nir_deref *a = a_var->deref.child, *b = b_var->deref.child;
  81         a != NULL; a = a->child, b = b->child) {
  82       if (a->deref_type != b->deref_type)
  83          return false;
  84
  85       if (a->deref_type == nir_deref_type_struct) {
  86          if (nir_deref_as_struct(a)->index != nir_deref_as_struct(b)->index)
  87             return false;
  88       }
  89       /* Do nothing for arrays.  They're all the same. */
  90
  91       assert((a->child == NULL) == (b->child == NULL));
  92       if((a->child == NULL) != (b->child == NULL))
  93          return false;
  94    }
  95
  96    return true;
  97 }
  98
  99 static nir_register *
 100 get_reg_for_deref(nir_deref_var *deref, struct locals_to_regs_state *state)
 101 {
 102    uint32_t hash = hash_deref(deref);
 103
 104    struct hash_entry *entry =
 105       _mesa_hash_table_search_pre_hashed(state->regs_table, hash, deref);
 106    if (entry)
 107       return entry->data;
 108
 109    unsigned array_size = 1;
 110    nir_deref *tail = &deref->deref;
 111    while (tail->child) {
 112       if (tail->child->deref_type == nir_deref_type_array)
 113          array_size *= glsl_get_length(tail->type);
 114       tail = tail->child;
 115    }
 116
 117    assert(glsl_type_is_vector(tail->type) || glsl_type_is_scalar(tail->type));
 118
 119    nir_register *reg = nir_local_reg_create(state->impl);
 120    reg->num_components = glsl_get_vector_elements(tail->type);
 121    reg->num_array_elems = array_size > 1 ? array_size : 0;
 122    reg->bit_size = glsl_get_bit_size(glsl_get_base_type(tail->type));
 123
 124    _mesa_hash_table_insert_pre_hashed(state->regs_table, hash, deref, reg);
 125    nir_array_add(&state->derefs_array, nir_deref_var *, deref);
 126
 127    return reg;
 128 }
 129
 130 static nir_src
 131 get_deref_reg_src(nir_deref_var *deref, nir_instr *instr,
 132                   struct locals_to_regs_state *state)
 133 {
 134    nir_src src;
 135
 136    src.is_ssa = false;
 137    src.reg.reg = get_reg_for_deref(deref, state);
 138    src.reg.base_offset = 0;
 139    src.reg.indirect = NULL;
 140
 141    /* It is possible for a user to create a shader that has an array with a
 142     * single element and then proceed to access it indirectly.  Indirectly
 143     * accessing a non-array register is not allowed in NIR.  In order to
 144     * handle this case we just convert it to a direct reference.
 145     */
 146    if (src.reg.reg->num_array_elems == 0)
 147       return src;
 148
 149    nir_deref *tail = &deref->deref;
 150    while (tail->child != NULL) {
 151       const struct glsl_type *parent_type = tail->type;
 152       tail = tail->child;
 153
 154       if (tail->deref_type != nir_deref_type_array)
 155          continue;
 156
 157       nir_deref_array *deref_array = nir_deref_as_array(tail);
 158
 159       src.reg.base_offset *= glsl_get_length(parent_type);
 160       src.reg.base_offset += deref_array->base_offset;
 161
 162       if (src.reg.indirect) {
 163          nir_load_const_instr *load_const =
 164             nir_load_const_instr_create(state->shader, 1);
 165          load_const->value.u32[0] = glsl_get_length(parent_type);
 166          nir_instr_insert_before(instr, &load_const->instr);
 167
 168          nir_alu_instr *mul = nir_alu_instr_create(state->shader, nir_op_imul);
 169          mul->src[0].src = *src.reg.indirect;
 170          mul->src[1].src.is_ssa = true;
 171          mul->src[1].src.ssa = &load_const->def;
 172          mul->dest.write_mask = 1;
 173          nir_ssa_dest_init(&mul->instr, &mul->dest.dest, 1, 32, NULL);
 174          nir_instr_insert_before(instr, &mul->instr);
 175
 176          src.reg.indirect->is_ssa = true;
 177          src.reg.indirect->ssa = &mul->dest.dest.ssa;
 178       }
 179
 180       if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
 181          if (src.reg.indirect == NULL) {
 182             src.reg.indirect = ralloc(state->shader, nir_src);
 183             nir_src_copy(src.reg.indirect, &deref_array->indirect,
 184                          state->shader);
 185          } else {
 186             nir_alu_instr *add = nir_alu_instr_create(state->shader,
 187                                                       nir_op_iadd);
 188             add->src[0].src = *src.reg.indirect;
 189             nir_src_copy(&add->src[1].src, &deref_array->indirect, add);
 190             add->dest.write_mask = 1;
 191             nir_ssa_dest_init(&add->instr, &add->dest.dest, 1, 32, NULL);
 192             nir_instr_insert_before(instr, &add->instr);
 193
 194             src.reg.indirect->is_ssa = true;
 195             src.reg.indirect->ssa = &add->dest.dest.ssa;
 196          }
 197       }
 198    }
 199
 200    return src;
 201 }
 202
 203 static bool
 204 lower_locals_to_regs_block(nir_block *block, void *void_state)
 205 {
 206    struct locals_to_regs_state *state = void_state;
 207
 208    nir_foreach_instr_safe(block, instr) {
 209       if (instr->type != nir_instr_type_intrinsic)
 210          continue;
 211
 212       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 213
 214       switch (intrin->intrinsic) {
 215       case nir_intrinsic_load_var: {
 216          if (intrin->variables[0]->var->data.mode != nir_var_local)
 217             continue;
 218
 219          nir_alu_instr *mov = nir_alu_instr_create(state->shader, nir_op_imov);
 220          mov->src[0].src = get_deref_reg_src(intrin->variables[0],
 221                                              &intrin->instr, state);
 222          mov->dest.write_mask = (1 << intrin->num_components) - 1;
 223          if (intrin->dest.is_ssa) {
 224             nir_ssa_dest_init(&mov->instr, &mov->dest.dest,
 225                               intrin->num_components,
 226                               intrin->dest.ssa.bit_size, NULL);
 227             nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
 228                                      nir_src_for_ssa(&mov->dest.dest.ssa));
 229          } else {
 230             nir_dest_copy(&mov->dest.dest, &intrin->dest, &mov->instr);
 231          }
 232          nir_instr_insert_before(&intrin->instr, &mov->instr);
 233
 234          nir_instr_remove(&intrin->instr);
 235          state->progress = true;
 236          break;
 237       }
 238
 239       case nir_intrinsic_store_var: {
 240          if (intrin->variables[0]->var->data.mode != nir_var_local)
 241             continue;
 242
 243          nir_src reg_src = get_deref_reg_src(intrin->variables[0],
 244                                              &intrin->instr, state);
 245
 246          nir_alu_instr *mov = nir_alu_instr_create(state->shader, nir_op_imov);
 247          nir_src_copy(&mov->src[0].src, &intrin->src[0], mov);
 248          mov->dest.write_mask = nir_intrinsic_write_mask(intrin);
 249          mov->dest.dest.is_ssa = false;
 250          mov->dest.dest.reg.reg = reg_src.reg.reg;
 251          mov->dest.dest.reg.base_offset = reg_src.reg.base_offset;
 252          mov->dest.dest.reg.indirect = reg_src.reg.indirect;
 253
 254          nir_instr_insert_before(&intrin->instr, &mov->instr);
 255
 256          nir_instr_remove(&intrin->instr);
 257          state->progress = true;
 258          break;
 259       }
 260
 261       case nir_intrinsic_copy_var:
 262          unreachable("There should be no copies whatsoever at this point");
 263          break;
 264
 265       default:
 266          continue;
 267       }
 268    }
 269
 270    return true;
 271 }
 272
 273 static nir_block *
 274 compute_reg_usedef_lca(nir_register *reg)
 275 {
 276    nir_block *lca = NULL;
 277
 278    list_for_each_entry(nir_dest, def_dest, &reg->defs, reg.def_link)
 279       lca = nir_dominance_lca(lca, def_dest->reg.parent_instr->block);
 280
 281    list_for_each_entry(nir_src, use_src, &reg->uses, use_link)
 282       lca = nir_dominance_lca(lca, use_src->parent_instr->block);
 283
 284    list_for_each_entry(nir_src, use_src, &reg->if_uses, use_link) {
 285       nir_cf_node *prev_node = nir_cf_node_prev(&use_src->parent_if->cf_node);
 286       assert(prev_node->type == nir_cf_node_block);
 287       lca = nir_dominance_lca(lca, nir_cf_node_as_block(prev_node));
 288    }
 289
 290    return lca;
 291 }
 292
 293 static void
 294 insert_constant_initializer(nir_deref_var *deref_head, nir_deref *deref_tail,
 295                             nir_block *block,
 296                             struct locals_to_regs_state *state)
 297 {
 298    if (deref_tail->child) {
 299       switch (deref_tail->child->deref_type) {
 300       case nir_deref_type_array: {
 301          unsigned array_elems = glsl_get_length(deref_tail->type);
 302
 303          nir_deref_array arr_deref;
 304          arr_deref.deref = *deref_tail->child;
 305          arr_deref.deref_array_type = nir_deref_array_type_direct;
 306
 307          nir_deref *old_child = deref_tail->child;
 308          deref_tail->child = &arr_deref.deref;
 309          for (unsigned i = 0; i < array_elems; i++) {
 310             arr_deref.base_offset = i;
 311             insert_constant_initializer(deref_head, &arr_deref.deref,
 312                                         block, state);
 313          }
 314          deref_tail->child = old_child;
 315          return;
 316       }
 317
 318       case nir_deref_type_struct:
 319          insert_constant_initializer(deref_head, deref_tail->child,
 320                                      block, state);
 321          return;
 322
 323       default:
 324          unreachable("Invalid deref child type");
 325       }
 326    }
 327
 328    assert(deref_tail->child == NULL);
 329
 330    nir_load_const_instr *load =
 331       nir_deref_get_const_initializer_load(state->shader, deref_head);
 332    nir_instr_insert_before_block(block, &load->instr);
 333
 334    nir_src reg_src = get_deref_reg_src(deref_head, &load->instr, state);
 335
 336    nir_alu_instr *mov = nir_alu_instr_create(state->shader, nir_op_imov);
 337    mov->src[0].src = nir_src_for_ssa(&load->def);
 338    mov->dest.write_mask = (1 << load->def.num_components) - 1;
 339    mov->dest.dest.is_ssa = false;
 340    mov->dest.dest.reg.reg = reg_src.reg.reg;
 341    mov->dest.dest.reg.base_offset = reg_src.reg.base_offset;
 342    mov->dest.dest.reg.indirect = reg_src.reg.indirect;
 343
 344    nir_instr_insert_after(&load->instr, &mov->instr);
 345    state->progress = true;
 346 }
 347
 348 static bool
 349 nir_lower_locals_to_regs_impl(nir_function_impl *impl)
 350 {
 351    struct locals_to_regs_state state;
 352
 353    state.shader = impl->function->shader;
 354    state.impl = impl;
 355    state.progress = false;
 356    state.regs_table = _mesa_hash_table_create(NULL, hash_deref, derefs_equal);
 357    nir_array_init(&state.derefs_array, NULL);
 358
 359    nir_metadata_require(impl, nir_metadata_dominance);
 360
 361    nir_foreach_block(impl, lower_locals_to_regs_block, &state);
 362
 363    nir_array_foreach(&state.derefs_array, nir_deref_var *, deref_ptr) {
 364       nir_deref_var *deref = *deref_ptr;
 365       struct hash_entry *deref_entry =
 366          _mesa_hash_table_search(state.regs_table, deref);
 367       assert(deref_entry && deref_entry->key == deref);
 368       nir_register *reg = (nir_register *)deref_entry->data;
 369
 370       if (deref->var->constant_initializer == NULL)
 371          continue;
 372
 373       nir_block *usedef_lca = compute_reg_usedef_lca(reg);
 374
 375       insert_constant_initializer(deref, &deref->deref, usedef_lca, &state);
 376    }
 377
 378    nir_metadata_preserve(impl, nir_metadata_block_index |
 379                                nir_metadata_dominance);
 380
 381    nir_array_fini(&state.derefs_array);
 382    _mesa_hash_table_destroy(state.regs_table, NULL);
 383
 384    return state.progress;
 385 }
 386
 387 bool
 388 nir_lower_locals_to_regs(nir_shader *shader)
 389 {
 390    bool progress = false;
 391
 392    nir_foreach_function(shader, function) {
 393       if (function->impl)
 394          progress = nir_lower_locals_to_regs_impl(function->impl) || progress;
 395    }
 396
 397    return progress;
 398 }