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 namespace {
  42
  43 static bool debug = false;
  44
  45 // XXX using variable_entry2 here to avoid collision (MSVC multiply-defined
  46 // function) with the variable_entry class seen in ir_variable_refcount.h
  47 // Perhaps we can use the one in ir_variable_refcount.h and make this class
  48 // here go away?
  49 class variable_entry2 : public exec_node
  50 {
  51 public:
  52    variable_entry2(ir_variable *var)
  53    {
  54       this->var = var;
  55       this->whole_structure_access = 0;
  56       this->declaration = false;
  57       this->components = NULL;
  58       this->mem_ctx = NULL;
  59    }
  60
  61    ir_variable *var; /* The key: the variable's pointer. */
  62
  63    /** Number of times the variable is referenced, including assignments. */
  64    unsigned whole_structure_access;
  65
  66    /* If the variable had a decl we can work with in the instruction
  67     * stream.  We can't do splitting on function arguments, which
  68     * don't get this variable set.
  69     */
  70    bool declaration;
  71
  72    ir_variable **components;
  73
  74    /** ralloc_parent(this->var) -- the shader's ralloc context. */
  75    void *mem_ctx;
  76 };
  77
  78
  79 class ir_structure_reference_visitor : public ir_hierarchical_visitor {
  80 public:
  81    ir_structure_reference_visitor(void)
  82    {
  83       this->mem_ctx = ralloc_context(NULL);
  84       this->variable_list.make_empty();
  85    }
  86
  87    ~ir_structure_reference_visitor(void)
  88    {
  89       ralloc_free(mem_ctx);
  90    }
  91
  92    virtual ir_visitor_status visit(ir_variable *);
  93    virtual ir_visitor_status visit(ir_dereference_variable *);
  94    virtual ir_visitor_status visit_enter(ir_dereference_record *);
  95    virtual ir_visitor_status visit_enter(ir_assignment *);
  96    virtual ir_visitor_status visit_enter(ir_function_signature *);
  97
  98    variable_entry2 *get_variable_entry2(ir_variable *var);
  99
 100    /* List of variable_entry */
 101    exec_list variable_list;
 102
 103    void *mem_ctx;
 104 };
 105
 106 variable_entry2 *
 107 ir_structure_reference_visitor::get_variable_entry2(ir_variable *var)
 108 {
 109    assert(var);
 110
 111    if (!var->type->is_record() || var->mode == ir_var_uniform)
 112       return NULL;
 113
 114    foreach_iter(exec_list_iterator, iter, this->variable_list) {
 115       variable_entry2 *entry = (variable_entry2 *)iter.get();
 116       if (entry->var == var)
 117          return entry;
 118    }
 119
 120    variable_entry2 *entry = new(mem_ctx) variable_entry2(var);
 121    this->variable_list.push_tail(entry);
 122    return entry;
 123 }
 124
 125
 126 ir_visitor_status
 127 ir_structure_reference_visitor::visit(ir_variable *ir)
 128 {
 129    variable_entry2 *entry = this->get_variable_entry2(ir);
 130
 131    if (entry)
 132       entry->declaration = true;
 133
 134    return visit_continue;
 135 }
 136
 137 ir_visitor_status
 138 ir_structure_reference_visitor::visit(ir_dereference_variable *ir)
 139 {
 140    ir_variable *const var = ir->variable_referenced();
 141    variable_entry2 *entry = this->get_variable_entry2(var);
 142
 143    if (entry)
 144       entry->whole_structure_access++;
 145
 146    return visit_continue;
 147 }
 148
 149 ir_visitor_status
 150 ir_structure_reference_visitor::visit_enter(ir_dereference_record *ir)
 151 {
 152    (void) ir;
 153    /* Don't descend into the ir_dereference_variable below. */
 154    return visit_continue_with_parent;
 155 }
 156
 157 ir_visitor_status
 158 ir_structure_reference_visitor::visit_enter(ir_assignment *ir)
 159 {
 160    /* If there are no structure references yet, no need to bother with
 161     * processing the expression tree.
 162     */
 163    if (this->variable_list.is_empty())
 164       return visit_continue_with_parent;
 165
 166    if (ir->lhs->as_dereference_variable() &&
 167        ir->rhs->as_dereference_variable() &&
 168        !ir->condition) {
 169       /* We'll split copies of a structure to copies of components, so don't
 170        * descend to the ir_dereference_variables.
 171        */
 172       return visit_continue_with_parent;
 173    }
 174    return visit_continue;
 175 }
 176
 177 ir_visitor_status
 178 ir_structure_reference_visitor::visit_enter(ir_function_signature *ir)
 179 {
 180    /* We don't have logic for structure-splitting function arguments,
 181     * so just look at the body instructions and not the parameter
 182     * declarations.
 183     */
 184    visit_list_elements(this, &ir->body);
 185    return visit_continue_with_parent;
 186 }
 187
 188 class ir_structure_splitting_visitor : public ir_rvalue_visitor {
 189 public:
 190    ir_structure_splitting_visitor(exec_list *vars)
 191    {
 192       this->variable_list = vars;
 193    }
 194
 195    virtual ~ir_structure_splitting_visitor()
 196    {
 197    }
 198
 199    virtual ir_visitor_status visit_leave(ir_assignment *);
 200
 201    void split_deref(ir_dereference **deref);
 202    void handle_rvalue(ir_rvalue **rvalue);
 203    variable_entry2 *get_splitting_entry(ir_variable *var);
 204
 205    exec_list *variable_list;
 206 };
 207
 208 variable_entry2 *
 209 ir_structure_splitting_visitor::get_splitting_entry(ir_variable *var)
 210 {
 211    assert(var);
 212
 213    if (!var->type->is_record())
 214       return NULL;
 215
 216    foreach_iter(exec_list_iterator, iter, *this->variable_list) {
 217       variable_entry2 *entry = (variable_entry2 *)iter.get();
 218       if (entry->var == var) {
 219          return entry;
 220       }
 221    }
 222
 223    return NULL;
 224 }
 225
 226 void
 227 ir_structure_splitting_visitor::split_deref(ir_dereference **deref)
 228 {
 229    if ((*deref)->ir_type != ir_type_dereference_record)
 230       return;
 231
 232    ir_dereference_record *deref_record = (ir_dereference_record *)*deref;
 233    ir_dereference_variable *deref_var = deref_record->record->as_dereference_variable();
 234    if (!deref_var)
 235       return;
 236
 237    variable_entry2 *entry = get_splitting_entry(deref_var->var);
 238    if (!entry)
 239       return;
 240
 241    unsigned int i;
 242    for (i = 0; i < entry->var->type->length; i++) {
 243       if (strcmp(deref_record->field,
 244                  entry->var->type->fields.structure[i].name) == 0)
 245          break;
 246    }
 247    assert(i != entry->var->type->length);
 248
 249    *deref = new(entry->mem_ctx) ir_dereference_variable(entry->components[i]);
 250 }
 251
 252 void
 253 ir_structure_splitting_visitor::handle_rvalue(ir_rvalue **rvalue)
 254 {
 255    if (!*rvalue)
 256       return;
 257
 258    ir_dereference *deref = (*rvalue)->as_dereference();
 259
 260    if (!deref)
 261       return;
 262
 263    split_deref(&deref);
 264    *rvalue = deref;
 265 }
 266
 267 ir_visitor_status
 268 ir_structure_splitting_visitor::visit_leave(ir_assignment *ir)
 269 {
 270    ir_dereference_variable *lhs_deref = ir->lhs->as_dereference_variable();
 271    ir_dereference_variable *rhs_deref = ir->rhs->as_dereference_variable();
 272    variable_entry2 *lhs_entry = lhs_deref ? get_splitting_entry(lhs_deref->var) : NULL;
 273    variable_entry2 *rhs_entry = rhs_deref ? get_splitting_entry(rhs_deref->var) : NULL;
 274    const glsl_type *type = ir->rhs->type;
 275
 276    if ((lhs_entry || rhs_entry) && !ir->condition) {
 277       for (unsigned int i = 0; i < type->length; i++) {
 278          ir_dereference *new_lhs, *new_rhs;
 279          void *mem_ctx = lhs_entry ? lhs_entry->mem_ctx : rhs_entry->mem_ctx;
 280
 281          if (lhs_entry) {
 282             new_lhs = new(mem_ctx) ir_dereference_variable(lhs_entry->components[i]);
 283          } else {
 284             new_lhs = new(mem_ctx)
 285                ir_dereference_record(ir->lhs->clone(mem_ctx, NULL),
 286                                      type->fields.structure[i].name);
 287          }
 288
 289          if (rhs_entry) {
 290             new_rhs = new(mem_ctx) ir_dereference_variable(rhs_entry->components[i]);
 291          } else {
 292             new_rhs = new(mem_ctx)
 293                ir_dereference_record(ir->rhs->clone(mem_ctx, NULL),
 294                                      type->fields.structure[i].name);
 295          }
 296
 297          ir->insert_before(new(mem_ctx) ir_assignment(new_lhs,
 298                                                       new_rhs,
 299                                                       NULL));
 300       }
 301       ir->remove();
 302    } else {
 303       handle_rvalue(&ir->rhs);
 304       split_deref(&ir->lhs);
 305    }
 306
 307    handle_rvalue(&ir->condition);
 308
 309    return visit_continue;
 310 }
 311
 312 } /* unnamed namespace */
 313
 314 bool
 315 do_structure_splitting(exec_list *instructions)
 316 {
 317    ir_structure_reference_visitor refs;
 318
 319    visit_list_elements(&refs, instructions);
 320
 321    /* Trim out variables we can't split. */
 322    foreach_iter(exec_list_iterator, iter, refs.variable_list) {
 323       variable_entry2 *entry = (variable_entry2 *)iter.get();
 324
 325       if (debug) {
 326          printf("structure %s@%p: decl %d, whole_access %d\n",
 327                 entry->var->name, (void *) entry->var, entry->declaration,
 328                 entry->whole_structure_access);
 329       }
 330
 331       if (!entry->declaration || entry->whole_structure_access) {
 332          entry->remove();
 333       }
 334    }
 335
 336    if (refs.variable_list.is_empty())
 337       return false;
 338
 339    void *mem_ctx = ralloc_context(NULL);
 340
 341    /* Replace the decls of the structures to be split with their split
 342     * components.
 343     */
 344    foreach_iter(exec_list_iterator, iter, refs.variable_list) {
 345       variable_entry2 *entry = (variable_entry2 *)iter.get();
 346       const struct glsl_type *type = entry->var->type;
 347
 348       entry->mem_ctx = ralloc_parent(entry->var);
 349
 350       entry->components = ralloc_array(mem_ctx,
 351                                        ir_variable *,
 352                                        type->length);
 353
 354       for (unsigned int i = 0; i < entry->var->type->length; i++) {
 355          const char *name = ralloc_asprintf(mem_ctx, "%s_%s",
 356                                             entry->var->name,
 357                                             type->fields.structure[i].name);
 358
 359          entry->components[i] =
 360             new(entry->mem_ctx) ir_variable(type->fields.structure[i].type,
 361                                             name,
 362                                             ir_var_temporary);
 363          entry->var->insert_before(entry->components[i]);
 364       }
 365
 366       entry->var->remove();
 367    }
 368
 369    ir_structure_splitting_visitor split(&refs.variable_list);
 370    visit_list_elements(&split, instructions);
 371
 372    ralloc_free(mem_ctx);
 373
 374    return true;
 375 }