src/glsl/opt_structure_splitting.cpp

   1 /*
   2  * Copyright © 2010 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
  21  * DEALINGS IN THE SOFTWARE.
  22  */
  23
  24 /**
  25  * \file opt_structure_splitting.cpp
  26  *
  27  * If a structure is only ever referenced by its components, then
  28  * split those components out to individual variables so they can be
  29  * handled normally by other optimization passes.
  30  *
  31  * This skips structures like uniforms, which need to be accessible as
  32  * structures for their access by the GL.
  33  */
  34
  35 #include "ir.h"
  36 #include "ir_visitor.h"
  37 #include "ir_rvalue_visitor.h"
  38 #include "glsl_types.h"
  39
  40 namespace {
  41
  42 static bool debug = false;
  43
  44 class variable_entry : public exec_node
  45 {
  46 public:
  47    variable_entry(ir_variable *var)
  48    {
  49       this->var = var;
  50       this->whole_structure_access = 0;
  51       this->declaration = false;
  52       this->components = NULL;
  53       this->mem_ctx = NULL;
  54    }
  55
  56    ir_variable *var; /* The key: the variable's pointer. */
  57
  58    /** Number of times the variable is referenced, including assignments. */
  59    unsigned whole_structure_access;
  60
  61    /* If the variable had a decl we can work with in the instruction
  62     * stream.  We can't do splitting on function arguments, which
  63     * don't get this variable set.
  64     */
  65    bool declaration;
  66
  67    ir_variable **components;
  68
  69    /** ralloc_parent(this->var) -- the shader's ralloc context. */
  70    void *mem_ctx;
  71 };
  72
  73
  74 class ir_structure_reference_visitor : public ir_hierarchical_visitor {
  75 public:
  76    ir_structure_reference_visitor(void)
  77    {
  78       this->mem_ctx = ralloc_context(NULL);
  79       this->variable_list.make_empty();
  80    }
  81
  82    ~ir_structure_reference_visitor(void)
  83    {
  84       ralloc_free(mem_ctx);
  85    }
  86
  87    virtual ir_visitor_status visit(ir_variable *);
  88    virtual ir_visitor_status visit(ir_dereference_variable *);
  89    virtual ir_visitor_status visit_enter(ir_dereference_record *);
  90    virtual ir_visitor_status visit_enter(ir_assignment *);
  91    virtual ir_visitor_status visit_enter(ir_function_signature *);
  92
  93    variable_entry *get_variable_entry(ir_variable *var);
  94
  95    /* List of variable_entry */
  96    exec_list variable_list;
  97
  98    void *mem_ctx;
  99 };
 100
 101 variable_entry *
 102 ir_structure_reference_visitor::get_variable_entry(ir_variable *var)
 103 {
 104    assert(var);
 105
 106    if (!var->type->is_record() || var->data.mode == ir_var_uniform
 107        || var->data.mode == ir_var_shader_in || var->data.mode == ir_var_shader_out)
 108       return NULL;
 109
 110    foreach_in_list(variable_entry, entry, &this->variable_list) {
 111       if (entry->var == var)
 112          return entry;
 113    }
 114
 115    variable_entry *entry = new(mem_ctx) variable_entry(var);
 116    this->variable_list.push_tail(entry);
 117    return entry;
 118 }
 119
 120
 121 ir_visitor_status
 122 ir_structure_reference_visitor::visit(ir_variable *ir)
 123 {
 124    variable_entry *entry = this->get_variable_entry(ir);
 125
 126    if (entry)
 127       entry->declaration = true;
 128
 129    return visit_continue;
 130 }
 131
 132 ir_visitor_status
 133 ir_structure_reference_visitor::visit(ir_dereference_variable *ir)
 134 {
 135    ir_variable *const var = ir->variable_referenced();
 136    variable_entry *entry = this->get_variable_entry(var);
 137
 138    if (entry)
 139       entry->whole_structure_access++;
 140
 141    return visit_continue;
 142 }
 143
 144 ir_visitor_status
 145 ir_structure_reference_visitor::visit_enter(ir_dereference_record *ir)
 146 {
 147    (void) ir;
 148    /* Don't descend into the ir_dereference_variable below. */
 149    return visit_continue_with_parent;
 150 }
 151
 152 ir_visitor_status
 153 ir_structure_reference_visitor::visit_enter(ir_assignment *ir)
 154 {
 155    /* If there are no structure references yet, no need to bother with
 156     * processing the expression tree.
 157     */
 158    if (this->variable_list.is_empty())
 159       return visit_continue_with_parent;
 160
 161    if (ir->lhs->as_dereference_variable() &&
 162        ir->rhs->as_dereference_variable() &&
 163        !ir->condition) {
 164       /* We'll split copies of a structure to copies of components, so don't
 165        * descend to the ir_dereference_variables.
 166        */
 167       return visit_continue_with_parent;
 168    }
 169    return visit_continue;
 170 }
 171
 172 ir_visitor_status
 173 ir_structure_reference_visitor::visit_enter(ir_function_signature *ir)
 174 {
 175    /* We don't have logic for structure-splitting function arguments,
 176     * so just look at the body instructions and not the parameter
 177     * declarations.
 178     */
 179    visit_list_elements(this, &ir->body);
 180    return visit_continue_with_parent;
 181 }
 182
 183 class ir_structure_splitting_visitor : public ir_rvalue_visitor {
 184 public:
 185    ir_structure_splitting_visitor(exec_list *vars)
 186    {
 187       this->variable_list = vars;
 188    }
 189
 190    virtual ~ir_structure_splitting_visitor()
 191    {
 192    }
 193
 194    virtual ir_visitor_status visit_leave(ir_assignment *);
 195
 196    void split_deref(ir_dereference **deref);
 197    void handle_rvalue(ir_rvalue **rvalue);
 198    variable_entry *get_splitting_entry(ir_variable *var);
 199
 200    exec_list *variable_list;
 201 };
 202
 203 variable_entry *
 204 ir_structure_splitting_visitor::get_splitting_entry(ir_variable *var)
 205 {
 206    assert(var);
 207
 208    if (!var->type->is_record())
 209       return NULL;
 210
 211    foreach_in_list(variable_entry, entry, this->variable_list) {
 212       if (entry->var == var) {
 213          return entry;
 214       }
 215    }
 216
 217    return NULL;
 218 }
 219
 220 void
 221 ir_structure_splitting_visitor::split_deref(ir_dereference **deref)
 222 {
 223    if ((*deref)->ir_type != ir_type_dereference_record)
 224       return;
 225
 226    ir_dereference_record *deref_record = (ir_dereference_record *)*deref;
 227    ir_dereference_variable *deref_var = deref_record->record->as_dereference_variable();
 228    if (!deref_var)
 229       return;
 230
 231    variable_entry *entry = get_splitting_entry(deref_var->var);
 232    if (!entry)
 233       return;
 234
 235    unsigned int i;
 236    for (i = 0; i < entry->var->type->length; i++) {
 237       if (strcmp(deref_record->field,
 238                  entry->var->type->fields.structure[i].name) == 0)
 239          break;
 240    }
 241    assert(i != entry->var->type->length);
 242
 243    *deref = new(entry->mem_ctx) ir_dereference_variable(entry->components[i]);
 244 }
 245
 246 void
 247 ir_structure_splitting_visitor::handle_rvalue(ir_rvalue **rvalue)
 248 {
 249    if (!*rvalue)
 250       return;
 251
 252    ir_dereference *deref = (*rvalue)->as_dereference();
 253
 254    if (!deref)
 255       return;
 256
 257    split_deref(&deref);
 258    *rvalue = deref;
 259 }
 260
 261 ir_visitor_status
 262 ir_structure_splitting_visitor::visit_leave(ir_assignment *ir)
 263 {
 264    ir_dereference_variable *lhs_deref = ir->lhs->as_dereference_variable();
 265    ir_dereference_variable *rhs_deref = ir->rhs->as_dereference_variable();
 266    variable_entry *lhs_entry = lhs_deref ? get_splitting_entry(lhs_deref->var) : NULL;
 267    variable_entry *rhs_entry = rhs_deref ? get_splitting_entry(rhs_deref->var) : NULL;
 268    const glsl_type *type = ir->rhs->type;
 269
 270    if ((lhs_entry || rhs_entry) && !ir->condition) {
 271       for (unsigned int i = 0; i < type->length; i++) {
 272          ir_dereference *new_lhs, *new_rhs;
 273          void *mem_ctx = lhs_entry ? lhs_entry->mem_ctx : rhs_entry->mem_ctx;
 274
 275          if (lhs_entry) {
 276             new_lhs = new(mem_ctx) ir_dereference_variable(lhs_entry->components[i]);
 277          } else {
 278             new_lhs = new(mem_ctx)
 279                ir_dereference_record(ir->lhs->clone(mem_ctx, NULL),
 280                                      type->fields.structure[i].name);
 281          }
 282
 283          if (rhs_entry) {
 284             new_rhs = new(mem_ctx) ir_dereference_variable(rhs_entry->components[i]);
 285          } else {
 286             new_rhs = new(mem_ctx)
 287                ir_dereference_record(ir->rhs->clone(mem_ctx, NULL),
 288                                      type->fields.structure[i].name);
 289          }
 290
 291          ir->insert_before(new(mem_ctx) ir_assignment(new_lhs,
 292                                                       new_rhs,
 293                                                       NULL));
 294       }
 295       ir->remove();
 296    } else {
 297       handle_rvalue(&ir->rhs);
 298       split_deref(&ir->lhs);
 299    }
 300
 301    handle_rvalue(&ir->condition);
 302
 303    return visit_continue;
 304 }
 305
 306 } /* unnamed namespace */
 307
 308 bool
 309 do_structure_splitting(exec_list *instructions)
 310 {
 311    ir_structure_reference_visitor refs;
 312
 313    visit_list_elements(&refs, instructions);
 314
 315    /* Trim out variables we can't split. */
 316    foreach_in_list_safe(variable_entry, entry, &refs.variable_list) {
 317       if (debug) {
 318          printf("structure %s@%p: decl %d, whole_access %d\n",
 319                 entry->var->name, (void *) entry->var, entry->declaration,
 320                 entry->whole_structure_access);
 321       }
 322
 323       if (!entry->declaration || entry->whole_structure_access) {
 324          entry->remove();
 325       }
 326    }
 327
 328    if (refs.variable_list.is_empty())
 329       return false;
 330
 331    void *mem_ctx = ralloc_context(NULL);
 332
 333    /* Replace the decls of the structures to be split with their split
 334     * components.
 335     */
 336    foreach_in_list_safe(variable_entry, entry, &refs.variable_list) {
 337       const struct glsl_type *type = entry->var->type;
 338
 339       entry->mem_ctx = ralloc_parent(entry->var);
 340
 341       entry->components = ralloc_array(mem_ctx,
 342                                        ir_variable *,
 343                                        type->length);
 344
 345       for (unsigned int i = 0; i < entry->var->type->length; i++) {
 346          const char *name = ralloc_asprintf(mem_ctx, "%s_%s",
 347                                             entry->var->name,
 348                                             type->fields.structure[i].name);
 349
 350          entry->components[i] =
 351             new(entry->mem_ctx) ir_variable(type->fields.structure[i].type,
 352                                             name,
 353                                             ir_var_temporary);
 354          entry->var->insert_before(entry->components[i]);
 355       }
 356
 357       entry->var->remove();
 358    }
 359
 360    ir_structure_splitting_visitor split(&refs.variable_list);
 361    visit_list_elements(&split, instructions);
 362
 363    ralloc_free(mem_ctx);
 364
 365    return true;
 366 }