src/glsl/opt_array_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_array_splitting.cpp
  26  *
  27  * If an array is always dereferenced with a constant index, then
  28  * split it apart into its elements, making it more amenable to other
  29  * optimization passes.
  30  *
  31  * This skips uniform/varying arrays, which would need careful
  32  * handling due to their ir->location fields tying them to the GL API
  33  * and other shader stages.
  34  */
  35
  36 #include "ir.h"
  37 #include "ir_visitor.h"
  38 #include "ir_rvalue_visitor.h"
  39 #include "ir_print_visitor.h"
  40 #include "glsl_types.h"
  41
  42 static bool debug = false;
  43
  44 namespace opt_array_splitting {
  45
  46 class variable_entry : public exec_node
  47 {
  48 public:
  49    variable_entry(ir_variable *var)
  50    {
  51       this->var = var;
  52       this->split = true;
  53       this->declaration = false;
  54       this->components = NULL;
  55       this->mem_ctx = NULL;
  56       if (var->type->is_array())
  57          this->size = var->type->length;
  58       else
  59          this->size = var->type->matrix_columns;
  60    }
  61
  62    ir_variable *var; /* The key: the variable's pointer. */
  63    unsigned size; /* array length or matrix columns */
  64
  65    /** Whether this array should be split or not. */
  66    bool split;
  67
  68    bool declaration; /* If the variable had a decl in the instruction stream */
  69
  70    ir_variable **components;
  71
  72    /** ralloc_parent(this->var) -- the shader's talloc context. */
  73    void *mem_ctx;
  74 };
  75
  76 } /* namespace */
  77 using namespace opt_array_splitting;
  78
  79 /**
  80  * This class does a walk over the tree, coming up with the set of
  81  * variables that could be split by looking to see if they are arrays
  82  * that are only ever constant-index dereferenced.
  83  */
  84 class ir_array_reference_visitor : public ir_hierarchical_visitor {
  85 public:
  86    ir_array_reference_visitor(void)
  87    {
  88       this->mem_ctx = ralloc_context(NULL);
  89       this->variable_list.make_empty();
  90    }
  91
  92    ~ir_array_reference_visitor(void)
  93    {
  94       ralloc_free(mem_ctx);
  95    }
  96
  97    bool get_split_list(exec_list *instructions, bool linked);
  98
  99    virtual ir_visitor_status visit(ir_variable *);
 100    virtual ir_visitor_status visit(ir_dereference_variable *);
 101    virtual ir_visitor_status visit_enter(ir_dereference_array *);
 102
 103    variable_entry *get_variable_entry(ir_variable *var);
 104
 105    /* List of variable_entry */
 106    exec_list variable_list;
 107
 108    void *mem_ctx;
 109 };
 110
 111 variable_entry *
 112 ir_array_reference_visitor::get_variable_entry(ir_variable *var)
 113 {
 114    assert(var);
 115
 116    if (var->mode != ir_var_auto &&
 117        var->mode != ir_var_temporary)
 118       return NULL;
 119
 120    if (!(var->type->is_array() || var->type->is_matrix()))
 121       return NULL;
 122
 123    /* If the array hasn't been sized yet, we can't split it.  After
 124     * linking, this should be resolved.
 125     */
 126    if (var->type->is_array() && var->type->length == 0)
 127       return NULL;
 128
 129    foreach_iter(exec_list_iterator, iter, this->variable_list) {
 130       variable_entry *entry = (variable_entry *)iter.get();
 131       if (entry->var == var)
 132          return entry;
 133    }
 134
 135    variable_entry *entry = new(mem_ctx) variable_entry(var);
 136    this->variable_list.push_tail(entry);
 137    return entry;
 138 }
 139
 140
 141 ir_visitor_status
 142 ir_array_reference_visitor::visit(ir_variable *ir)
 143 {
 144    variable_entry *entry = this->get_variable_entry(ir);
 145
 146    if (entry)
 147       entry->declaration = true;
 148
 149    return visit_continue;
 150 }
 151
 152 ir_visitor_status
 153 ir_array_reference_visitor::visit(ir_dereference_variable *ir)
 154 {
 155    variable_entry *entry = this->get_variable_entry(ir->var);
 156
 157    /* If we made it to here without seeing an ir_dereference_array,
 158     * then the dereference of this array didn't have a constant index
 159     * (see the visit_continue_with_parent below), so we can't split
 160     * the variable.
 161     */
 162    if (entry)
 163       entry->split = false;
 164
 165    return visit_continue;
 166 }
 167
 168 ir_visitor_status
 169 ir_array_reference_visitor::visit_enter(ir_dereference_array *ir)
 170 {
 171    ir_dereference_variable *deref = ir->array->as_dereference_variable();
 172    if (!deref)
 173       return visit_continue;
 174
 175    variable_entry *entry = this->get_variable_entry(deref->var);
 176
 177    /* If the access to the array has a variable index, we wouldn't
 178     * know which split variable this dereference should go to.
 179     */
 180    if (entry && !ir->array_index->as_constant())
 181       entry->split = false;
 182
 183    return visit_continue_with_parent;
 184 }
 185
 186 bool
 187 ir_array_reference_visitor::get_split_list(exec_list *instructions,
 188                                            bool linked)
 189 {
 190    visit_list_elements(this, instructions);
 191
 192    /* If the shaders aren't linked yet, we can't mess with global
 193     * declarations, which need to be matched by name across shaders.
 194     */
 195    if (!linked) {
 196       foreach_list(node, instructions) {
 197          ir_variable *var = ((ir_instruction *)node)->as_variable();
 198          if (var) {
 199             variable_entry *entry = get_variable_entry(var);
 200             if (entry)
 201                entry->remove();
 202          }
 203       }
 204    }
 205
 206    /* Trim out variables we found that we can't split. */
 207    foreach_iter(exec_list_iterator, iter, variable_list) {
 208       variable_entry *entry = (variable_entry *)iter.get();
 209
 210       if (debug) {
 211          printf("array %s@%p: decl %d, split %d\n",
 212                 entry->var->name, (void *) entry->var, entry->declaration,
 213                 entry->split);
 214       }
 215
 216       if (!(entry->declaration && entry->split)) {
 217          entry->remove();
 218       }
 219    }
 220
 221    return !variable_list.is_empty();
 222 }
 223
 224 /**
 225  * This class rewrites the dereferences of arrays that have been split
 226  * to use the newly created ir_variables for each component.
 227  */
 228 class ir_array_splitting_visitor : public ir_rvalue_visitor {
 229 public:
 230    ir_array_splitting_visitor(exec_list *vars)
 231    {
 232       this->variable_list = vars;
 233    }
 234
 235    virtual ~ir_array_splitting_visitor()
 236    {
 237    }
 238
 239    virtual ir_visitor_status visit_leave(ir_assignment *);
 240
 241    void split_deref(ir_dereference **deref);
 242    void handle_rvalue(ir_rvalue **rvalue);
 243    variable_entry *get_splitting_entry(ir_variable *var);
 244
 245    exec_list *variable_list;
 246 };
 247
 248 variable_entry *
 249 ir_array_splitting_visitor::get_splitting_entry(ir_variable *var)
 250 {
 251    assert(var);
 252
 253    foreach_iter(exec_list_iterator, iter, *this->variable_list) {
 254       variable_entry *entry = (variable_entry *)iter.get();
 255       if (entry->var == var) {
 256          return entry;
 257       }
 258    }
 259
 260    return NULL;
 261 }
 262
 263 void
 264 ir_array_splitting_visitor::split_deref(ir_dereference **deref)
 265 {
 266    ir_dereference_array *deref_array = (*deref)->as_dereference_array();
 267    if (!deref_array)
 268       return;
 269
 270    ir_dereference_variable *deref_var = deref_array->array->as_dereference_variable();
 271    if (!deref_var)
 272       return;
 273    ir_variable *var = deref_var->var;
 274
 275    variable_entry *entry = get_splitting_entry(var);
 276    if (!entry)
 277       return;
 278
 279    ir_constant *constant = deref_array->array_index->as_constant();
 280    assert(constant);
 281
 282    if (constant->value.i[0] < (int)entry->size) {
 283       *deref = new(entry->mem_ctx)
 284          ir_dereference_variable(entry->components[constant->value.i[0]]);
 285    } else {
 286       /* There was a constant array access beyond the end of the
 287        * array.  This might have happened due to constant folding
 288        * after the initial parse.  This produces an undefined value,
 289        * but shouldn't crash.  Just give them an uninitialized
 290        * variable.
 291        */
 292       ir_variable *temp = new(entry->mem_ctx) ir_variable(deref_array->type,
 293                                                           "undef",
 294                                                           ir_var_temporary);
 295       entry->components[0]->insert_before(temp);
 296       *deref = new(entry->mem_ctx) ir_dereference_variable(temp);
 297    }
 298 }
 299
 300 void
 301 ir_array_splitting_visitor::handle_rvalue(ir_rvalue **rvalue)
 302 {
 303    if (!*rvalue)
 304       return;
 305
 306    ir_dereference *deref = (*rvalue)->as_dereference();
 307
 308    if (!deref)
 309       return;
 310
 311    split_deref(&deref);
 312    *rvalue = deref;
 313 }
 314
 315 ir_visitor_status
 316 ir_array_splitting_visitor::visit_leave(ir_assignment *ir)
 317 {
 318    /* The normal rvalue visitor skips the LHS of assignments, but we
 319     * need to process those just the same.
 320     */
 321    ir_rvalue *lhs = ir->lhs;
 322
 323    handle_rvalue(&lhs);
 324    ir->lhs = lhs->as_dereference();
 325
 326    ir->lhs->accept(this);
 327
 328    handle_rvalue(&ir->rhs);
 329    ir->rhs->accept(this);
 330
 331    if (ir->condition) {
 332       handle_rvalue(&ir->condition);
 333       ir->condition->accept(this);
 334    }
 335
 336    return visit_continue;
 337 }
 338
 339 bool
 340 optimize_split_arrays(exec_list *instructions, bool linked)
 341 {
 342    ir_array_reference_visitor refs;
 343    if (!refs.get_split_list(instructions, linked))
 344       return false;
 345
 346    void *mem_ctx = ralloc_context(NULL);
 347
 348    /* Replace the decls of the arrays to be split with their split
 349     * components.
 350     */
 351    foreach_iter(exec_list_iterator, iter, refs.variable_list) {
 352       variable_entry *entry = (variable_entry *)iter.get();
 353       const struct glsl_type *type = entry->var->type;
 354       const struct glsl_type *subtype;
 355
 356       if (type->is_matrix())
 357          subtype = type->column_type();
 358       else
 359          subtype = type->fields.array;
 360
 361       entry->mem_ctx = ralloc_parent(entry->var);
 362
 363       entry->components = ralloc_array(mem_ctx,
 364                                        ir_variable *,
 365                                        entry->size);
 366
 367       for (unsigned int i = 0; i < entry->size; i++) {
 368          const char *name = ralloc_asprintf(mem_ctx, "%s_%d",
 369                                             entry->var->name, i);
 370
 371          entry->components[i] =
 372             new(entry->mem_ctx) ir_variable(subtype, name, ir_var_temporary);
 373          entry->var->insert_before(entry->components[i]);
 374       }
 375
 376       entry->var->remove();
 377    }
 378
 379    ir_array_splitting_visitor split(&refs.variable_list);
 380    visit_list_elements(&split, instructions);
 381
 382    if (debug)
 383       _mesa_print_ir(instructions, NULL);
 384
 385    ralloc_free(mem_ctx);
 386
 387    return true;
 388
 389 }