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_builder.h"
  30
  31 struct locals_to_regs_state {
  32    nir_builder builder;
  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    for (const nir_deref_instr *deref = void_deref; deref;
  51         deref = nir_deref_instr_parent(deref)) {
  52       switch (deref->deref_type) {
  53       case nir_deref_type_var:
  54          return _mesa_fnv32_1a_accumulate(hash, deref->var);
  55
  56       case nir_deref_type_array:
  57          continue; /* Do nothing */
  58
  59       case nir_deref_type_struct:
  60          hash = _mesa_fnv32_1a_accumulate(hash, deref->strct.index);
  61          continue;
  62
  63       default:
  64          unreachable("Invalid deref type");
  65       }
  66    }
  67
  68    unreachable("We should have hit a variable dereference");
  69 }
  70
  71 static bool
  72 derefs_equal(const void *void_a, const void *void_b)
  73 {
  74    for (const nir_deref_instr *a = void_a, *b = void_b; a || b;
  75         a = nir_deref_instr_parent(a), b = nir_deref_instr_parent(b)) {
  76       if (a->deref_type != b->deref_type)
  77          return false;
  78
  79       switch (a->deref_type) {
  80       case nir_deref_type_var:
  81          return a->var == b->var;
  82
  83       case nir_deref_type_array:
  84          continue; /* Do nothing */
  85
  86       case nir_deref_type_struct:
  87          if (a->strct.index != b->strct.index)
  88             return false;
  89          continue;
  90
  91       default:
  92          unreachable("Invalid deref type");
  93       }
  94    }
  95
  96    unreachable("We should have hit a variable dereference");
  97 }
  98
  99 static nir_register *
 100 get_reg_for_deref(nir_deref_instr *deref, struct locals_to_regs_state *state)
 101 {
 102    uint32_t hash = hash_deref(deref);
 103
 104    assert(nir_deref_instr_get_variable(deref)->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    for (nir_deref_instr *d = deref; d; d = nir_deref_instr_parent(d)) {
 113       if (d->deref_type == nir_deref_type_array)
 114          array_size *= glsl_get_length(nir_deref_instr_parent(d)->type);
 115    }
 116
 117    assert(glsl_type_is_vector_or_scalar(deref->type));
 118
 119    nir_register *reg = nir_local_reg_create(state->builder.impl);
 120    reg->num_components = glsl_get_vector_elements(deref->type);
 121    reg->num_array_elems = array_size > 1 ? array_size : 0;
 122    reg->bit_size = glsl_get_bit_size(deref->type);
 123
 124    _mesa_hash_table_insert_pre_hashed(state->regs_table, hash, deref, reg);
 125
 126    return reg;
 127 }
 128
 129 static nir_src
 130 get_deref_reg_src(nir_deref_instr *deref, struct locals_to_regs_state *state)
 131 {
 132    nir_builder *b = &state->builder;
 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    unsigned inner_array_size = 1;
 150    for (const nir_deref_instr *d = deref; d; d = nir_deref_instr_parent(d)) {
 151       if (d->deref_type != nir_deref_type_array)
 152          continue;
 153
 154       if (nir_src_is_const(d->arr.index) && !src.reg.indirect) {
 155          src.reg.base_offset += nir_src_as_uint(d->arr.index) *
 156                                 inner_array_size;
 157       } else {
 158          if (src.reg.indirect) {
 159             assert(src.reg.base_offset == 0);
 160          } else {
 161             src.reg.indirect = ralloc(b->shader, nir_src);
 162             *src.reg.indirect =
 163                nir_src_for_ssa(nir_imm_int(b, src.reg.base_offset));
 164             src.reg.base_offset = 0;
 165          }
 166
 167          assert(src.reg.indirect->is_ssa);
 168          src.reg.indirect->ssa =
 169             nir_iadd(b, src.reg.indirect->ssa,
 170                         nir_imul(b, nir_ssa_for_src(b, d->arr.index, 1),
 171                                     nir_imm_int(b, inner_array_size)));
 172       }
 173
 174       inner_array_size *= glsl_get_length(nir_deref_instr_parent(d)->type);
 175    }
 176
 177    return src;
 178 }
 179
 180 static bool
 181 lower_locals_to_regs_block(nir_block *block,
 182                            struct locals_to_regs_state *state)
 183 {
 184    nir_builder *b = &state->builder;
 185
 186    nir_foreach_instr_safe(instr, block) {
 187       if (instr->type != nir_instr_type_intrinsic)
 188          continue;
 189
 190       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 191
 192       switch (intrin->intrinsic) {
 193       case nir_intrinsic_load_deref: {
 194          nir_deref_instr *deref = nir_src_as_deref(intrin->src[0]);
 195          if (deref->mode != nir_var_local)
 196             continue;
 197
 198          b->cursor = nir_before_instr(&intrin->instr);
 199
 200          nir_alu_instr *mov = nir_alu_instr_create(b->shader, nir_op_imov);
 201          mov->src[0].src = get_deref_reg_src(deref, state);
 202          mov->dest.write_mask = (1 << intrin->num_components) - 1;
 203          if (intrin->dest.is_ssa) {
 204             nir_ssa_dest_init(&mov->instr, &mov->dest.dest,
 205                               intrin->num_components,
 206                               intrin->dest.ssa.bit_size, NULL);
 207             nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
 208                                      nir_src_for_ssa(&mov->dest.dest.ssa));
 209          } else {
 210             nir_dest_copy(&mov->dest.dest, &intrin->dest, &mov->instr);
 211          }
 212          nir_builder_instr_insert(b, &mov->instr);
 213
 214          nir_instr_remove(&intrin->instr);
 215          state->progress = true;
 216          break;
 217       }
 218
 219       case nir_intrinsic_store_deref: {
 220          nir_deref_instr *deref = nir_src_as_deref(intrin->src[0]);
 221          if (deref->mode != nir_var_local)
 222             continue;
 223
 224          b->cursor = nir_before_instr(&intrin->instr);
 225
 226          nir_src reg_src = get_deref_reg_src(deref, state);
 227
 228          nir_alu_instr *mov = nir_alu_instr_create(b->shader, nir_op_imov);
 229          nir_src_copy(&mov->src[0].src, &intrin->src[1], mov);
 230          mov->dest.write_mask = nir_intrinsic_write_mask(intrin);
 231          mov->dest.dest.is_ssa = false;
 232          mov->dest.dest.reg.reg = reg_src.reg.reg;
 233          mov->dest.dest.reg.base_offset = reg_src.reg.base_offset;
 234          mov->dest.dest.reg.indirect = reg_src.reg.indirect;
 235
 236          nir_builder_instr_insert(b, &mov->instr);
 237
 238          nir_instr_remove(&intrin->instr);
 239          state->progress = true;
 240          break;
 241       }
 242
 243       case nir_intrinsic_copy_deref:
 244          unreachable("There should be no copies whatsoever at this point");
 245          break;
 246
 247       default:
 248          continue;
 249       }
 250    }
 251
 252    return true;
 253 }
 254
 255 static bool
 256 nir_lower_locals_to_regs_impl(nir_function_impl *impl)
 257 {
 258    struct locals_to_regs_state state;
 259
 260    nir_builder_init(&state.builder, impl);
 261    state.progress = false;
 262    state.regs_table = _mesa_hash_table_create(NULL, hash_deref, derefs_equal);
 263
 264    nir_metadata_require(impl, nir_metadata_dominance);
 265
 266    nir_foreach_block(block, impl) {
 267       lower_locals_to_regs_block(block, &state);
 268    }
 269
 270    nir_metadata_preserve(impl, nir_metadata_block_index |
 271                                nir_metadata_dominance);
 272
 273    _mesa_hash_table_destroy(state.regs_table, NULL);
 274
 275    return state.progress;
 276 }
 277
 278 bool
 279 nir_lower_locals_to_regs(nir_shader *shader)
 280 {
 281    bool progress = false;
 282
 283    nir_foreach_function(function, shader) {
 284       if (function->impl)
 285          progress = nir_lower_locals_to_regs_impl(function->impl) || progress;
 286    }
 287
 288    return progress;
 289 }