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
  30 struct locals_to_regs_state {
  31    nir_shader *shader;
  32    nir_function_impl *impl;
  33
  34    /* A hash table mapping derefs to registers */
  35    struct hash_table *regs_table;
  36
  37    bool progress;
  38 };
  39
  40 /* The following two functions implement a hash and equality check for
  41  * variable dreferences.  When the hash or equality function encounters an
  42  * array, it ignores the offset and whether it is direct or indirect
  43  * entirely.
  44  */
  45 static uint32_t
  46 hash_deref(const void *void_deref)
  47 {
  48    uint32_t hash = _mesa_fnv32_1a_offset_bias;
  49
  50    const nir_deref_var *deref_var = void_deref;
  51    hash = _mesa_fnv32_1a_accumulate(hash, deref_var->var);
  52
  53    for (const nir_deref *deref = deref_var->deref.child;
  54         deref; deref = deref->child) {
  55       if (deref->deref_type == nir_deref_type_struct) {
  56          const nir_deref_struct *deref_struct = nir_deref_as_struct(deref);
  57          hash = _mesa_fnv32_1a_accumulate(hash, deref_struct->index);
  58       }
  59    }
  60
  61    return hash;
  62 }
  63
  64 static bool
  65 derefs_equal(const void *void_a, const void *void_b)
  66 {
  67    const nir_deref_var *a_var = void_a;
  68    const nir_deref_var *b_var = void_b;
  69
  70    if (a_var->var != b_var->var)
  71       return false;
  72
  73    for (const nir_deref *a = a_var->deref.child, *b = b_var->deref.child;
  74         a != NULL; a = a->child, b = b->child) {
  75       if (a->deref_type != b->deref_type)
  76          return false;
  77
  78       if (a->deref_type == nir_deref_type_struct) {
  79          if (nir_deref_as_struct(a)->index != nir_deref_as_struct(b)->index)
  80             return false;
  81       }
  82       /* Do nothing for arrays.  They're all the same. */
  83
  84       assert((a->child == NULL) == (b->child == NULL));
  85       if((a->child == NULL) != (b->child == NULL))
  86          return false;
  87    }
  88
  89    return true;
  90 }
  91
  92 static nir_register *
  93 get_reg_for_deref(nir_deref_var *deref, struct locals_to_regs_state *state)
  94 {
  95    uint32_t hash = hash_deref(deref);
  96
  97    assert(deref->var->constant_initializer == NULL);
  98
  99    struct hash_entry *entry =
 100       _mesa_hash_table_search_pre_hashed(state->regs_table, hash, deref);
 101    if (entry)
 102       return entry->data;
 103
 104    unsigned array_size = 1;
 105    nir_deref *tail = &deref->deref;
 106    while (tail->child) {
 107       if (tail->child->deref_type == nir_deref_type_array)
 108          array_size *= glsl_get_length(tail->type);
 109       tail = tail->child;
 110    }
 111
 112    assert(glsl_type_is_vector(tail->type) || glsl_type_is_scalar(tail->type));
 113
 114    nir_register *reg = nir_local_reg_create(state->impl);
 115    reg->num_components = glsl_get_vector_elements(tail->type);
 116    reg->num_array_elems = array_size > 1 ? array_size : 0;
 117    reg->bit_size = glsl_get_bit_size(tail->type);
 118
 119    _mesa_hash_table_insert_pre_hashed(state->regs_table, hash, deref, reg);
 120
 121    return reg;
 122 }
 123
 124 static nir_src
 125 get_deref_reg_src(nir_deref_var *deref, nir_instr *instr,
 126                   struct locals_to_regs_state *state)
 127 {
 128    nir_src src;
 129
 130    src.is_ssa = false;
 131    src.reg.reg = get_reg_for_deref(deref, state);
 132    src.reg.base_offset = 0;
 133    src.reg.indirect = NULL;
 134
 135    /* It is possible for a user to create a shader that has an array with a
 136     * single element and then proceed to access it indirectly.  Indirectly
 137     * accessing a non-array register is not allowed in NIR.  In order to
 138     * handle this case we just convert it to a direct reference.
 139     */
 140    if (src.reg.reg->num_array_elems == 0)
 141       return src;
 142
 143    nir_deref *tail = &deref->deref;
 144    while (tail->child != NULL) {
 145       const struct glsl_type *parent_type = tail->type;
 146       tail = tail->child;
 147
 148       if (tail->deref_type != nir_deref_type_array)
 149          continue;
 150
 151       nir_deref_array *deref_array = nir_deref_as_array(tail);
 152
 153       src.reg.base_offset *= glsl_get_length(parent_type);
 154       src.reg.base_offset += deref_array->base_offset;
 155
 156       if (src.reg.indirect) {
 157          nir_load_const_instr *load_const =
 158             nir_load_const_instr_create(state->shader, 1, 32);
 159          load_const->value.u32[0] = glsl_get_length(parent_type);
 160          nir_instr_insert_before(instr, &load_const->instr);
 161
 162          nir_alu_instr *mul = nir_alu_instr_create(state->shader, nir_op_imul);
 163          mul->src[0].src = *src.reg.indirect;
 164          mul->src[1].src.is_ssa = true;
 165          mul->src[1].src.ssa = &load_const->def;
 166          mul->dest.write_mask = 1;
 167          nir_ssa_dest_init(&mul->instr, &mul->dest.dest, 1, 32, NULL);
 168          nir_instr_insert_before(instr, &mul->instr);
 169
 170          src.reg.indirect->is_ssa = true;
 171          src.reg.indirect->ssa = &mul->dest.dest.ssa;
 172       }
 173
 174       if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
 175          if (src.reg.indirect == NULL) {
 176             src.reg.indirect = ralloc(state->shader, nir_src);
 177             nir_src_copy(src.reg.indirect, &deref_array->indirect,
 178                          state->shader);
 179          } else {
 180             nir_alu_instr *add = nir_alu_instr_create(state->shader,
 181                                                       nir_op_iadd);
 182             add->src[0].src = *src.reg.indirect;
 183             nir_src_copy(&add->src[1].src, &deref_array->indirect, add);
 184             add->dest.write_mask = 1;
 185             nir_ssa_dest_init(&add->instr, &add->dest.dest, 1, 32, NULL);
 186             nir_instr_insert_before(instr, &add->instr);
 187
 188             src.reg.indirect->is_ssa = true;
 189             src.reg.indirect->ssa = &add->dest.dest.ssa;
 190          }
 191       }
 192    }
 193
 194    return src;
 195 }
 196
 197 static bool
 198 lower_locals_to_regs_block(nir_block *block,
 199                            struct locals_to_regs_state *state)
 200 {
 201    nir_foreach_instr_safe(instr, block) {
 202       if (instr->type != nir_instr_type_intrinsic)
 203          continue;
 204
 205       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 206
 207       switch (intrin->intrinsic) {
 208       case nir_intrinsic_load_var: {
 209          if (intrin->variables[0]->var->data.mode != nir_var_local)
 210             continue;
 211
 212          nir_alu_instr *mov = nir_alu_instr_create(state->shader, nir_op_imov);
 213          mov->src[0].src = get_deref_reg_src(intrin->variables[0],
 214                                              &intrin->instr, state);
 215          mov->dest.write_mask = (1 << intrin->num_components) - 1;
 216          if (intrin->dest.is_ssa) {
 217             nir_ssa_dest_init(&mov->instr, &mov->dest.dest,
 218                               intrin->num_components,
 219                               intrin->dest.ssa.bit_size, NULL);
 220             nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
 221                                      nir_src_for_ssa(&mov->dest.dest.ssa));
 222          } else {
 223             nir_dest_copy(&mov->dest.dest, &intrin->dest, &mov->instr);
 224          }
 225          nir_instr_insert_before(&intrin->instr, &mov->instr);
 226
 227          nir_instr_remove(&intrin->instr);
 228          state->progress = true;
 229          break;
 230       }
 231
 232       case nir_intrinsic_store_var: {
 233          if (intrin->variables[0]->var->data.mode != nir_var_local)
 234             continue;
 235
 236          nir_src reg_src = get_deref_reg_src(intrin->variables[0],
 237                                              &intrin->instr, state);
 238
 239          nir_alu_instr *mov = nir_alu_instr_create(state->shader, nir_op_imov);
 240          nir_src_copy(&mov->src[0].src, &intrin->src[0], mov);
 241          mov->dest.write_mask = nir_intrinsic_write_mask(intrin);
 242          mov->dest.dest.is_ssa = false;
 243          mov->dest.dest.reg.reg = reg_src.reg.reg;
 244          mov->dest.dest.reg.base_offset = reg_src.reg.base_offset;
 245          mov->dest.dest.reg.indirect = reg_src.reg.indirect;
 246
 247          nir_instr_insert_before(&intrin->instr, &mov->instr);
 248
 249          nir_instr_remove(&intrin->instr);
 250          state->progress = true;
 251          break;
 252       }
 253
 254       case nir_intrinsic_copy_var:
 255          unreachable("There should be no copies whatsoever at this point");
 256          break;
 257
 258       default:
 259          continue;
 260       }
 261    }
 262
 263    return true;
 264 }
 265
 266 static bool
 267 nir_lower_locals_to_regs_impl(nir_function_impl *impl)
 268 {
 269    struct locals_to_regs_state state;
 270
 271    state.shader = impl->function->shader;
 272    state.impl = impl;
 273    state.progress = false;
 274    state.regs_table = _mesa_hash_table_create(NULL, hash_deref, derefs_equal);
 275
 276    nir_metadata_require(impl, nir_metadata_dominance);
 277
 278    nir_foreach_block(block, impl) {
 279       lower_locals_to_regs_block(block, &state);
 280    }
 281
 282    nir_metadata_preserve(impl, nir_metadata_block_index |
 283                                nir_metadata_dominance);
 284
 285    _mesa_hash_table_destroy(state.regs_table, NULL);
 286
 287    return state.progress;
 288 }
 289
 290 bool
 291 nir_lower_locals_to_regs(nir_shader *shader)
 292 {
 293    bool progress = false;
 294
 295    nir_foreach_function(function, shader) {
 296       if (function->impl)
 297          progress = nir_lower_locals_to_regs_impl(function->impl) || progress;
 298    }
 299
 300    return progress;
 301 }