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_print_visitor.h"
  38 #include "ir_rvalue_visitor.h"
  39 #include "glsl_types.h"
  40
  41 static bool debug = false;
  42
  43 // XXX using variable_entry2 here to avoid collision (MSVC multiply-defined
  44 // function) with the variable_entry class seen in ir_variable_refcount.h
  45 // Perhaps we can use the one in ir_variable_refcount.h and make this class
  46 // here go away?
  47 class variable_entry2 : public exec_node
  48 {
  49 public:
  50    variable_entry2(ir_variable *var)
  51    {
  52       this->var = var;
  53       this->whole_structure_access = 0;
  54       this->declaration = false;
  55       this->components = NULL;
  56       this->mem_ctx = NULL;
  57    }
  58
  59    ir_variable *var; /* The key: the variable's pointer. */
  60
  61    /** Number of times the variable is referenced, including assignments. */
  62    unsigned whole_structure_access;
  63
  64    bool declaration; /* If the variable had a decl in the instruction stream */
  65
  66    ir_variable **components;
  67
  68    /** ralloc_parent(this->var) -- the shader's ralloc context. */
  69    void *mem_ctx;
  70 };
  71
  72
  73 class ir_structure_reference_visitor : public ir_hierarchical_visitor {
  74 public:
  75    ir_structure_reference_visitor(void)
  76    {
  77       this->mem_ctx = ralloc_context(NULL);
  78       this->variable_list.make_empty();
  79    }
  80
  81    ~ir_structure_reference_visitor(void)
  82    {
  83       ralloc_free(mem_ctx);
  84    }
  85
  86    virtual ir_visitor_status visit(ir_variable *);
  87    virtual ir_visitor_status visit(ir_dereference_variable *);
  88    virtual ir_visitor_status visit_enter(ir_dereference_record *);
  89    virtual ir_visitor_status visit_enter(ir_assignment *);
  90    virtual ir_visitor_status visit_enter(ir_function_signature *);
  91
  92    variable_entry2 *get_variable_entry2(ir_variable *var);
  93
  94    /* List of variable_entry */
  95    exec_list variable_list;
  96
  97    void *mem_ctx;
  98 };
  99
 100 variable_entry2 *
 101 ir_structure_reference_visitor::get_variable_entry2(ir_variable *var)
 102 {
 103    assert(var);
 104
 105    if (!var->type->is_record() || var->mode == ir_var_uniform)
 106       return NULL;
 107
 108    foreach_iter(exec_list_iterator, iter, this->variable_list) {
 109       variable_entry2 *entry = (variable_entry2 *)iter.get();
 110       if (entry->var == var)
 111          return entry;
 112    }
 113
 114    variable_entry2 *entry = new(mem_ctx) variable_entry2(var);
 115    this->variable_list.push_tail(entry);
 116    return entry;
 117 }
 118
 119
 120 ir_visitor_status
 121 ir_structure_reference_visitor::visit(ir_variable *ir)
 122 {
 123    variable_entry2 *entry = this->get_variable_entry2(ir);
 124
 125    if (entry)
 126       entry->declaration = true;
 127
 128    return visit_continue;
 129 }
 130
 131 ir_visitor_status
 132 ir_structure_reference_visitor::visit(ir_dereference_variable *ir)
 133 {
 134    ir_variable *const var = ir->variable_referenced();
 135    variable_entry2 *entry = this->get_variable_entry2(var);
 136
 137    if (entry)
 138       entry->whole_structure_access++;
 139
 140    return visit_continue;
 141 }
 142
 143 ir_visitor_status
 144 ir_structure_reference_visitor::visit_enter(ir_dereference_record *ir)
 145 {
 146    (void) ir;
 147    /* Don't descend into the ir_dereference_variable below. */
 148    return visit_continue_with_parent;
 149 }
 150
 151 ir_visitor_status
 152 ir_structure_reference_visitor::visit_enter(ir_assignment *ir)
 153 {
 154    /* If there are no structure references yet, no need to bother with
 155     * processing the expression tree.
 156     */
 157    if (this->variable_list.is_empty())
 158       return visit_continue_with_parent;
 159
 160    if (ir->lhs->as_dereference_variable() &&
 161        ir->rhs->as_dereference_variable() &&
 162        !ir->condition) {
 163       /* We'll split copies of a structure to copies of components, so don't
 164        * descend to the ir_dereference_variables.
 165        */
 166       return visit_continue_with_parent;
 167    }
 168    return visit_continue;
 169 }
 170
 171 ir_visitor_status
 172 ir_structure_reference_visitor::visit_enter(ir_function_signature *ir)
 173 {
 174    /* We don't want to descend into the function parameters and
 175     * dead-code eliminate them, so just accept the body here.
 176     */
 177    visit_list_elements(this, &ir->body);
 178    return visit_continue_with_parent;
 179 }
 180
 181 class ir_structure_splitting_visitor : public ir_rvalue_visitor {
 182 public:
 183    ir_structure_splitting_visitor(exec_list *vars)
 184    {
 185       this->variable_list = vars;
 186    }
 187
 188    virtual ~ir_structure_splitting_visitor()
 189    {
 190    }
 191
 192    virtual ir_visitor_status visit_leave(ir_assignment *);
 193
 194    void split_deref(ir_dereference **deref);
 195    void handle_rvalue(ir_rvalue **rvalue);
 196    variable_entry2 *get_splitting_entry(ir_variable *var);
 197
 198    exec_list *variable_list;
 199    void *mem_ctx;
 200 };
 201
 202 variable_entry2 *
 203 ir_structure_splitting_visitor::get_splitting_entry(ir_variable *var)
 204 {
 205    assert(var);
 206
 207    if (!var->type->is_record())
 208       return NULL;
 209
 210    foreach_iter(exec_list_iterator, iter, *this->variable_list) {
 211       variable_entry2 *entry = (variable_entry2 *)iter.get();
 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_entry2 *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_entry2 *lhs_entry = lhs_deref ? get_splitting_entry(lhs_deref->var) : NULL;
 267    variable_entry2 *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 bool
 307 do_structure_splitting(exec_list *instructions)
 308 {
 309    ir_structure_reference_visitor refs;
 310
 311    visit_list_elements(&refs, instructions);
 312
 313    /* Trim out variables we can't split. */
 314    foreach_iter(exec_list_iterator, iter, refs.variable_list) {
 315       variable_entry2 *entry = (variable_entry2 *)iter.get();
 316
 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_iter(exec_list_iterator, iter, refs.variable_list) {
 337       variable_entry2 *entry = (variable_entry2 *)iter.get();
 338       const struct glsl_type *type = entry->var->type;
 339
 340       entry->mem_ctx = ralloc_parent(entry->var);
 341
 342       entry->components = ralloc_array(mem_ctx,
 343                                        ir_variable *,
 344                                        type->length);
 345
 346       for (unsigned int i = 0; i < entry->var->type->length; i++) {
 347          const char *name = ralloc_asprintf(mem_ctx, "%s_%s",
 348                                             entry->var->name,
 349                                             type->fields.structure[i].name);
 350
 351          entry->components[i] =
 352             new(entry->mem_ctx) ir_variable(type->fields.structure[i].type,
 353                                             name,
 354                                             ir_var_temporary);
 355          entry->var->insert_before(entry->components[i]);
 356       }
 357
 358       entry->var->remove();
 359    }
 360
 361    ir_structure_splitting_visitor split(&refs.variable_list);
 362    visit_list_elements(&split, instructions);
 363
 364    ralloc_free(mem_ctx);
 365
 366    return true;
 367 }