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
  30 struct locals_to_regs_state {
  31    void *mem_ctx;
  32    nir_function_impl *impl;
  33
  34    /* A hash table mapping derefs to registers */
  35    struct hash_table *regs_table;
  36 };
  37
  38 /* The following two functions implement a hash and equality check for
  39  * variable dreferences.  When the hash or equality function encounters an
  40  * array, it ignores the offset and whether it is direct or indirect
  41  * entirely.
  42  */
  43 static uint32_t
  44 hash_deref(const void *void_deref)
  45 {
  46    uint32_t hash = _mesa_fnv32_1a_offset_bias;
  47
  48    const nir_deref_var *deref_var = void_deref;
  49    hash = _mesa_fnv32_1a_accumulate(hash, deref_var->var);
  50
  51    for (const nir_deref *deref = deref_var->deref.child;
  52         deref; deref = deref->child) {
  53       if (deref->deref_type == nir_deref_type_struct) {
  54          const nir_deref_struct *deref_struct = nir_deref_as_struct(deref);
  55          hash = _mesa_fnv32_1a_accumulate(hash, deref_struct->index);
  56       }
  57    }
  58
  59    return hash;
  60 }
  61
  62 static bool
  63 derefs_equal(const void *void_a, const void *void_b)
  64 {
  65    const nir_deref_var *a_var = void_a;
  66    const nir_deref_var *b_var = void_b;
  67
  68    if (a_var->var != b_var->var)
  69       return false;
  70
  71    for (const nir_deref *a = a_var->deref.child, *b = b_var->deref.child;
  72         a != NULL; a = a->child, b = b->child) {
  73       if (a->deref_type != b->deref_type)
  74          return false;
  75
  76       if (a->deref_type == nir_deref_type_struct) {
  77          if (nir_deref_as_struct(a)->index != nir_deref_as_struct(b)->index)
  78             return false;
  79       }
  80       /* Do nothing for arrays.  They're all the same. */
  81
  82       assert((a->child == NULL) == (b->child == NULL));
  83       if((a->child == NULL) != (b->child == NULL))
  84          return false;
  85    }
  86
  87    return true;
  88 }
  89
  90 static nir_register *
  91 get_reg_for_deref(nir_deref_var *deref, struct locals_to_regs_state *state)
  92 {
  93    uint32_t hash = hash_deref(deref);
  94
  95    struct hash_entry *entry =
  96       _mesa_hash_table_search_pre_hashed(state->regs_table, hash, deref);
  97    if (entry)
  98       return entry->data;
  99
 100    unsigned array_size = 1;
 101    nir_deref *tail = &deref->deref;
 102    while (tail->child) {
 103       if (tail->child->deref_type == nir_deref_type_array) {
 104          /* Multiply by the parent's type. */
 105          if (glsl_type_is_matrix(tail->type)) {
 106             array_size *= glsl_get_matrix_columns(tail->type);
 107          } else {
 108             assert(glsl_get_length(tail->type) > 0);
 109             array_size *= glsl_get_length(tail->type);
 110          }
 111       }
 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
 123    return reg;
 124 }
 125
 126 static nir_src
 127 get_deref_reg_src(nir_deref_var *deref, nir_instr *instr,
 128                   struct locals_to_regs_state *state)
 129 {
 130    nir_src src;
 131
 132    src.is_ssa = false;
 133    src.reg.reg = get_reg_for_deref(deref, state);
 134    src.reg.base_offset = 0;
 135    src.reg.indirect = NULL;
 136
 137    nir_deref *tail = &deref->deref;
 138    while (tail->child != NULL) {
 139       const struct glsl_type *parent_type = tail->type;
 140       tail = tail->child;
 141
 142       if (tail->deref_type != nir_deref_type_array)
 143          continue;
 144
 145       nir_deref_array *deref_array = nir_deref_as_array(tail);
 146
 147       src.reg.base_offset *= glsl_get_length(parent_type);
 148       src.reg.base_offset += deref_array->base_offset;
 149
 150       if (src.reg.indirect) {
 151          nir_load_const_instr *load_const =
 152             nir_load_const_instr_create(state->mem_ctx, 1);
 153          load_const->value.u[0] = glsl_get_length(parent_type);
 154          nir_instr_insert_before(instr, &load_const->instr);
 155
 156          nir_alu_instr *mul = nir_alu_instr_create(state->mem_ctx, nir_op_imul);
 157          mul->src[0].src = *src.reg.indirect;
 158          mul->src[1].src.is_ssa = true;
 159          mul->src[1].src.ssa = &load_const->def;
 160          mul->dest.write_mask = 1;
 161          nir_ssa_dest_init(&mul->instr, &mul->dest.dest, 1, NULL);
 162          nir_instr_insert_before(instr, &mul->instr);
 163
 164          src.reg.indirect->is_ssa = true;
 165          src.reg.indirect->ssa = &mul->dest.dest.ssa;
 166       }
 167
 168       if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
 169          if (src.reg.indirect == NULL) {
 170             src.reg.indirect = ralloc(state->mem_ctx, nir_src);
 171             nir_src_copy(src.reg.indirect, &deref_array->indirect,
 172                          state->mem_ctx);
 173          } else {
 174             nir_alu_instr *add = nir_alu_instr_create(state->mem_ctx,
 175                                                       nir_op_iadd);
 176             add->src[0].src = *src.reg.indirect;
 177             nir_src_copy(&add->src[1].src, &deref_array->indirect,
 178                          state->mem_ctx);
 179             add->dest.write_mask = 1;
 180             nir_ssa_dest_init(&add->instr, &add->dest.dest, 1, NULL);
 181             nir_instr_insert_before(instr, &add->instr);
 182
 183             src.reg.indirect->is_ssa = true;
 184             src.reg.indirect->ssa = &add->dest.dest.ssa;
 185          }
 186       }
 187    }
 188
 189    return src;
 190 }
 191
 192 static bool
 193 lower_locals_to_regs_block(nir_block *block, void *void_state)
 194 {
 195    struct locals_to_regs_state *state = void_state;
 196
 197    nir_foreach_instr_safe(block, instr) {
 198       if (instr->type != nir_instr_type_intrinsic)
 199          continue;
 200
 201       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 202
 203       switch (intrin->intrinsic) {
 204       case nir_intrinsic_load_var: {
 205          if (intrin->variables[0]->var->data.mode != nir_var_local)
 206             continue;
 207
 208          nir_alu_instr *mov = nir_alu_instr_create(state->mem_ctx, nir_op_imov);
 209          mov->src[0].src = get_deref_reg_src(intrin->variables[0],
 210                                              &intrin->instr, state);
 211          mov->dest.write_mask = (1 << intrin->num_components) - 1;
 212          if (intrin->dest.is_ssa) {
 213             nir_ssa_dest_init(&mov->instr, &mov->dest.dest,
 214                               intrin->num_components, NULL);
 215             nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
 216                                      nir_src_for_ssa(&mov->dest.dest.ssa),
 217                                      state->mem_ctx);
 218          } else {
 219             nir_dest_copy(&mov->dest.dest, &intrin->dest, state->mem_ctx);
 220          }
 221          nir_instr_insert_before(&intrin->instr, &mov->instr);
 222
 223          nir_instr_remove(&intrin->instr);
 224          break;
 225       }
 226
 227       case nir_intrinsic_store_var: {
 228          if (intrin->variables[0]->var->data.mode != nir_var_local)
 229             continue;
 230
 231          nir_src reg_src = get_deref_reg_src(intrin->variables[0],
 232                                              &intrin->instr, state);
 233
 234          nir_alu_instr *mov = nir_alu_instr_create(state->mem_ctx, nir_op_imov);
 235          nir_src_copy(&mov->src[0].src, &intrin->src[0], state->mem_ctx);
 236          mov->dest.write_mask = (1 << intrin->num_components) - 1;
 237          mov->dest.dest.is_ssa = false;
 238          mov->dest.dest.reg.reg = reg_src.reg.reg;
 239          mov->dest.dest.reg.base_offset = reg_src.reg.base_offset;
 240          mov->dest.dest.reg.indirect = reg_src.reg.indirect;
 241
 242          nir_instr_insert_before(&intrin->instr, &mov->instr);
 243
 244          nir_instr_remove(&intrin->instr);
 245          break;
 246       }
 247
 248       case nir_intrinsic_copy_var:
 249          unreachable("There should be no copies whatsoever at this point");
 250          break;
 251
 252       default:
 253          continue;
 254       }
 255    }
 256
 257    return true;
 258 }
 259
 260 static void
 261 nir_lower_locals_to_regs_impl(nir_function_impl *impl)
 262 {
 263    struct locals_to_regs_state state;
 264
 265    state.mem_ctx = ralloc_parent(impl);
 266    state.impl = impl;
 267    state.regs_table = _mesa_hash_table_create(NULL, hash_deref, derefs_equal);
 268
 269    nir_foreach_block(impl, lower_locals_to_regs_block, &state);
 270
 271    nir_metadata_preserve(impl, nir_metadata_block_index |
 272                                nir_metadata_dominance);
 273
 274    _mesa_hash_table_destroy(state.regs_table, NULL);
 275 }
 276
 277 void
 278 nir_lower_locals_to_regs(nir_shader *shader)
 279 {
 280    nir_foreach_overload(shader, overload) {
 281       if (overload->impl)
 282          nir_lower_locals_to_regs_impl(overload->impl);
 283    }
 284 }