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    assert(deref->var->constant_initializer == NULL);
 105
 106    struct hash_entry *entry =
 107       _mesa_hash_table_search_pre_hashed(state->regs_table, hash, deref);
 108    if (entry)
 109       return entry->data;
 110
 111    unsigned array_size = 1;
 112    nir_deref *tail = &deref->deref;
 113    while (tail->child) {
 114       if (tail->child->deref_type == nir_deref_type_array)
 115          array_size *= glsl_get_length(tail->type);
 116       tail = tail->child;
 117    }
 118
 119    assert(glsl_type_is_vector(tail->type) || glsl_type_is_scalar(tail->type));
 120
 121    nir_register *reg = nir_local_reg_create(state->impl);
 122    reg->num_components = glsl_get_vector_elements(tail->type);
 123    reg->num_array_elems = array_size > 1 ? array_size : 0;
 124    reg->bit_size = glsl_get_bit_size(tail->type);
 125
 126    _mesa_hash_table_insert_pre_hashed(state->regs_table, hash, deref, reg);
 127    nir_array_add(&state->derefs_array, nir_deref_var *, deref);
 128
 129    return reg;
 130 }
 131
 132 static nir_src
 133 get_deref_reg_src(nir_deref_var *deref, nir_instr *instr,
 134                   struct locals_to_regs_state *state)
 135 {
 136    nir_src src;
 137
 138    src.is_ssa = false;
 139    src.reg.reg = get_reg_for_deref(deref, state);
 140    src.reg.base_offset = 0;
 141    src.reg.indirect = NULL;
 142
 143    /* It is possible for a user to create a shader that has an array with a
 144     * single element and then proceed to access it indirectly.  Indirectly
 145     * accessing a non-array register is not allowed in NIR.  In order to
 146     * handle this case we just convert it to a direct reference.
 147     */
 148    if (src.reg.reg->num_array_elems == 0)
 149       return src;
 150
 151    nir_deref *tail = &deref->deref;
 152    while (tail->child != NULL) {
 153       const struct glsl_type *parent_type = tail->type;
 154       tail = tail->child;
 155
 156       if (tail->deref_type != nir_deref_type_array)
 157          continue;
 158
 159       nir_deref_array *deref_array = nir_deref_as_array(tail);
 160
 161       src.reg.base_offset *= glsl_get_length(parent_type);
 162       src.reg.base_offset += deref_array->base_offset;
 163
 164       if (src.reg.indirect) {
 165          nir_load_const_instr *load_const =
 166             nir_load_const_instr_create(state->shader, 1, 32);
 167          load_const->value.u32[0] = glsl_get_length(parent_type);
 168          nir_instr_insert_before(instr, &load_const->instr);
 169
 170          nir_alu_instr *mul = nir_alu_instr_create(state->shader, nir_op_imul);
 171          mul->src[0].src = *src.reg.indirect;
 172          mul->src[1].src.is_ssa = true;
 173          mul->src[1].src.ssa = &load_const->def;
 174          mul->dest.write_mask = 1;
 175          nir_ssa_dest_init(&mul->instr, &mul->dest.dest, 1, 32, NULL);
 176          nir_instr_insert_before(instr, &mul->instr);
 177
 178          src.reg.indirect->is_ssa = true;
 179          src.reg.indirect->ssa = &mul->dest.dest.ssa;
 180       }
 181
 182       if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
 183          if (src.reg.indirect == NULL) {
 184             src.reg.indirect = ralloc(state->shader, nir_src);
 185             nir_src_copy(src.reg.indirect, &deref_array->indirect,
 186                          state->shader);
 187          } else {
 188             nir_alu_instr *add = nir_alu_instr_create(state->shader,
 189                                                       nir_op_iadd);
 190             add->src[0].src = *src.reg.indirect;
 191             nir_src_copy(&add->src[1].src, &deref_array->indirect, add);
 192             add->dest.write_mask = 1;
 193             nir_ssa_dest_init(&add->instr, &add->dest.dest, 1, 32, NULL);
 194             nir_instr_insert_before(instr, &add->instr);
 195
 196             src.reg.indirect->is_ssa = true;
 197             src.reg.indirect->ssa = &add->dest.dest.ssa;
 198          }
 199       }
 200    }
 201
 202    return src;
 203 }
 204
 205 static bool
 206 lower_locals_to_regs_block(nir_block *block,
 207                            struct locals_to_regs_state *state)
 208 {
 209    nir_foreach_instr_safe(instr, block) {
 210       if (instr->type != nir_instr_type_intrinsic)
 211          continue;
 212
 213       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 214
 215       switch (intrin->intrinsic) {
 216       case nir_intrinsic_load_var: {
 217          if (intrin->variables[0]->var->data.mode != nir_var_local)
 218             continue;
 219
 220          nir_alu_instr *mov = nir_alu_instr_create(state->shader, nir_op_imov);
 221          mov->src[0].src = get_deref_reg_src(intrin->variables[0],
 222                                              &intrin->instr, state);
 223          mov->dest.write_mask = (1 << intrin->num_components) - 1;
 224          if (intrin->dest.is_ssa) {
 225             nir_ssa_dest_init(&mov->instr, &mov->dest.dest,
 226                               intrin->num_components,
 227                               intrin->dest.ssa.bit_size, NULL);
 228             nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
 229                                      nir_src_for_ssa(&mov->dest.dest.ssa));
 230          } else {
 231             nir_dest_copy(&mov->dest.dest, &intrin->dest, &mov->instr);
 232          }
 233          nir_instr_insert_before(&intrin->instr, &mov->instr);
 234
 235          nir_instr_remove(&intrin->instr);
 236          state->progress = true;
 237          break;
 238       }
 239
 240       case nir_intrinsic_store_var: {
 241          if (intrin->variables[0]->var->data.mode != nir_var_local)
 242             continue;
 243
 244          nir_src reg_src = get_deref_reg_src(intrin->variables[0],
 245                                              &intrin->instr, state);
 246
 247          nir_alu_instr *mov = nir_alu_instr_create(state->shader, nir_op_imov);
 248          nir_src_copy(&mov->src[0].src, &intrin->src[0], mov);
 249          mov->dest.write_mask = nir_intrinsic_write_mask(intrin);
 250          mov->dest.dest.is_ssa = false;
 251          mov->dest.dest.reg.reg = reg_src.reg.reg;
 252          mov->dest.dest.reg.base_offset = reg_src.reg.base_offset;
 253          mov->dest.dest.reg.indirect = reg_src.reg.indirect;
 254
 255          nir_instr_insert_before(&intrin->instr, &mov->instr);
 256
 257          nir_instr_remove(&intrin->instr);
 258          state->progress = true;
 259          break;
 260       }
 261
 262       case nir_intrinsic_copy_var:
 263          unreachable("There should be no copies whatsoever at this point");
 264          break;
 265
 266       default:
 267          continue;
 268       }
 269    }
 270
 271    return true;
 272 }
 273
 274 static bool
 275 nir_lower_locals_to_regs_impl(nir_function_impl *impl)
 276 {
 277    struct locals_to_regs_state state;
 278
 279    state.shader = impl->function->shader;
 280    state.impl = impl;
 281    state.progress = false;
 282    state.regs_table = _mesa_hash_table_create(NULL, hash_deref, derefs_equal);
 283    nir_array_init(&state.derefs_array, NULL);
 284
 285    nir_metadata_require(impl, nir_metadata_dominance);
 286
 287    nir_foreach_block(block, impl) {
 288       lower_locals_to_regs_block(block, &state);
 289    }
 290
 291    nir_metadata_preserve(impl, nir_metadata_block_index |
 292                                nir_metadata_dominance);
 293
 294    nir_array_fini(&state.derefs_array);
 295    _mesa_hash_table_destroy(state.regs_table, NULL);
 296
 297    return state.progress;
 298 }
 299
 300 bool
 301 nir_lower_locals_to_regs(nir_shader *shader)
 302 {
 303    bool progress = false;
 304
 305    nir_foreach_function(function, shader) {
 306       if (function->impl)
 307          progress = nir_lower_locals_to_regs_impl(function->impl) || progress;
 308    }
 309
 310    return progress;
 311 }