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