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 "glsl_types.h"
  40
  41 static bool debug = false;
  42
  43 namespace {
  44
  45 namespace opt_array_splitting {
  46
  47 class variable_entry : public exec_node
  48 {
  49 public:
  50    variable_entry(ir_variable *var)
  51    {
  52       this->var = var;
  53       this->split = true;
  54       this->declaration = false;
  55       this->components = NULL;
  56       this->mem_ctx = NULL;
  57       if (var->type->is_array())
  58          this->size = var->type->length;
  59       else
  60          this->size = var->type->matrix_columns;
  61    }
  62
  63    ir_variable *var; /* The key: the variable's pointer. */
  64    unsigned size; /* array length or matrix columns */
  65
  66    /** Whether this array should be split or not. */
  67    bool split;
  68
  69    /* If the variable had a decl we can work with in the instruction
  70     * stream.  We can't do splitting on function arguments, which
  71     * don't get this variable set.
  72     */
  73    bool declaration;
  74
  75    ir_variable **components;
  76
  77    /** ralloc_parent(this->var) -- the shader's talloc context. */
  78    void *mem_ctx;
  79 };
  80
  81 } /* namespace */
  82
  83 using namespace opt_array_splitting;
  84
  85 /**
  86  * This class does a walk over the tree, coming up with the set of
  87  * variables that could be split by looking to see if they are arrays
  88  * that are only ever constant-index dereferenced.
  89  */
  90 class ir_array_reference_visitor : public ir_hierarchical_visitor {
  91 public:
  92    ir_array_reference_visitor(void)
  93    {
  94       this->mem_ctx = ralloc_context(NULL);
  95       this->variable_list.make_empty();
  96    }
  97
  98    ~ir_array_reference_visitor(void)
  99    {
 100       ralloc_free(mem_ctx);
 101    }
 102
 103    bool get_split_list(exec_list *instructions, bool linked);
 104
 105    virtual ir_visitor_status visit(ir_variable *);
 106    virtual ir_visitor_status visit(ir_dereference_variable *);
 107    virtual ir_visitor_status visit_enter(ir_dereference_array *);
 108    virtual ir_visitor_status visit_enter(ir_function_signature *);
 109
 110    variable_entry *get_variable_entry(ir_variable *var);
 111
 112    /* List of variable_entry */
 113    exec_list variable_list;
 114
 115    void *mem_ctx;
 116 };
 117
 118 } /* namespace */
 119
 120 variable_entry *
 121 ir_array_reference_visitor::get_variable_entry(ir_variable *var)
 122 {
 123    assert(var);
 124
 125    if (var->data.mode != ir_var_auto &&
 126        var->data.mode != ir_var_temporary)
 127       return NULL;
 128
 129    if (!(var->type->is_array() || var->type->is_matrix()))
 130       return NULL;
 131
 132    /* If the array hasn't been sized yet, we can't split it.  After
 133     * linking, this should be resolved.
 134     */
 135    if (var->type->is_unsized_array())
 136       return NULL;
 137
 138    foreach_in_list(variable_entry, entry, &this->variable_list) {
 139       if (entry->var == var)
 140          return entry;
 141    }
 142
 143    variable_entry *entry = new(mem_ctx) variable_entry(var);
 144    this->variable_list.push_tail(entry);
 145    return entry;
 146 }
 147
 148
 149 ir_visitor_status
 150 ir_array_reference_visitor::visit(ir_variable *ir)
 151 {
 152    variable_entry *entry = this->get_variable_entry(ir);
 153
 154    if (entry)
 155       entry->declaration = true;
 156
 157    return visit_continue;
 158 }
 159
 160 ir_visitor_status
 161 ir_array_reference_visitor::visit(ir_dereference_variable *ir)
 162 {
 163    variable_entry *entry = this->get_variable_entry(ir->var);
 164
 165    /* If we made it to here without seeing an ir_dereference_array,
 166     * then the dereference of this array didn't have a constant index
 167     * (see the visit_continue_with_parent below), so we can't split
 168     * the variable.
 169     */
 170    if (entry)
 171       entry->split = false;
 172
 173    return visit_continue;
 174 }
 175
 176 ir_visitor_status
 177 ir_array_reference_visitor::visit_enter(ir_dereference_array *ir)
 178 {
 179    ir_dereference_variable *deref = ir->array->as_dereference_variable();
 180    if (!deref)
 181       return visit_continue;
 182
 183    variable_entry *entry = this->get_variable_entry(deref->var);
 184
 185    /* If the access to the array has a variable index, we wouldn't
 186     * know which split variable this dereference should go to.
 187     */
 188    if (entry && !ir->array_index->as_constant())
 189       entry->split = false;
 190
 191    return visit_continue_with_parent;
 192 }
 193
 194 ir_visitor_status
 195 ir_array_reference_visitor::visit_enter(ir_function_signature *ir)
 196 {
 197    /* We don't have logic for array-splitting function arguments,
 198     * so just look at the body instructions and not the parameter
 199     * declarations.
 200     */
 201    visit_list_elements(this, &ir->body);
 202    return visit_continue_with_parent;
 203 }
 204
 205 bool
 206 ir_array_reference_visitor::get_split_list(exec_list *instructions,
 207                                            bool linked)
 208 {
 209    visit_list_elements(this, instructions);
 210
 211    /* If the shaders aren't linked yet, we can't mess with global
 212     * declarations, which need to be matched by name across shaders.
 213     */
 214    if (!linked) {
 215       foreach_in_list(ir_instruction, node, instructions) {
 216          ir_variable *var = node->as_variable();
 217          if (var) {
 218             variable_entry *entry = get_variable_entry(var);
 219             if (entry)
 220                entry->remove();
 221          }
 222       }
 223    }
 224
 225    /* Trim out variables we found that we can't split. */
 226    foreach_list_safe(n, &variable_list) {
 227       variable_entry *entry = (variable_entry *) n;
 228
 229       if (debug) {
 230          printf("array %s@%p: decl %d, split %d\n",
 231                 entry->var->name, (void *) entry->var, entry->declaration,
 232                 entry->split);
 233       }
 234
 235       if (!(entry->declaration && entry->split)) {
 236          entry->remove();
 237       }
 238    }
 239
 240    return !variable_list.is_empty();
 241 }
 242
 243 /**
 244  * This class rewrites the dereferences of arrays that have been split
 245  * to use the newly created ir_variables for each component.
 246  */
 247 class ir_array_splitting_visitor : public ir_rvalue_visitor {
 248 public:
 249    ir_array_splitting_visitor(exec_list *vars)
 250    {
 251       this->variable_list = vars;
 252    }
 253
 254    virtual ~ir_array_splitting_visitor()
 255    {
 256    }
 257
 258    virtual ir_visitor_status visit_leave(ir_assignment *);
 259
 260    void split_deref(ir_dereference **deref);
 261    void handle_rvalue(ir_rvalue **rvalue);
 262    variable_entry *get_splitting_entry(ir_variable *var);
 263
 264    exec_list *variable_list;
 265 };
 266
 267 variable_entry *
 268 ir_array_splitting_visitor::get_splitting_entry(ir_variable *var)
 269 {
 270    assert(var);
 271
 272    foreach_in_list(variable_entry, entry, this->variable_list) {
 273       if (entry->var == var) {
 274          return entry;
 275       }
 276    }
 277
 278    return NULL;
 279 }
 280
 281 void
 282 ir_array_splitting_visitor::split_deref(ir_dereference **deref)
 283 {
 284    ir_dereference_array *deref_array = (*deref)->as_dereference_array();
 285    if (!deref_array)
 286       return;
 287
 288    ir_dereference_variable *deref_var = deref_array->array->as_dereference_variable();
 289    if (!deref_var)
 290       return;
 291    ir_variable *var = deref_var->var;
 292
 293    variable_entry *entry = get_splitting_entry(var);
 294    if (!entry)
 295       return;
 296
 297    ir_constant *constant = deref_array->array_index->as_constant();
 298    assert(constant);
 299
 300    if (constant->value.i[0] < (int)entry->size) {
 301       *deref = new(entry->mem_ctx)
 302          ir_dereference_variable(entry->components[constant->value.i[0]]);
 303    } else {
 304       /* There was a constant array access beyond the end of the
 305        * array.  This might have happened due to constant folding
 306        * after the initial parse.  This produces an undefined value,
 307        * but shouldn't crash.  Just give them an uninitialized
 308        * variable.
 309        */
 310       ir_variable *temp = new(entry->mem_ctx) ir_variable(deref_array->type,
 311                                                           "undef",
 312                                                           ir_var_temporary);
 313       entry->components[0]->insert_before(temp);
 314       *deref = new(entry->mem_ctx) ir_dereference_variable(temp);
 315    }
 316 }
 317
 318 void
 319 ir_array_splitting_visitor::handle_rvalue(ir_rvalue **rvalue)
 320 {
 321    if (!*rvalue)
 322       return;
 323
 324    ir_dereference *deref = (*rvalue)->as_dereference();
 325
 326    if (!deref)
 327       return;
 328
 329    split_deref(&deref);
 330    *rvalue = deref;
 331 }
 332
 333 ir_visitor_status
 334 ir_array_splitting_visitor::visit_leave(ir_assignment *ir)
 335 {
 336    /* The normal rvalue visitor skips the LHS of assignments, but we
 337     * need to process those just the same.
 338     */
 339    ir_rvalue *lhs = ir->lhs;
 340
 341    handle_rvalue(&lhs);
 342    ir->lhs = lhs->as_dereference();
 343
 344    ir->lhs->accept(this);
 345
 346    handle_rvalue(&ir->rhs);
 347    ir->rhs->accept(this);
 348
 349    if (ir->condition) {
 350       handle_rvalue(&ir->condition);
 351       ir->condition->accept(this);
 352    }
 353
 354    return visit_continue;
 355 }
 356
 357 bool
 358 optimize_split_arrays(exec_list *instructions, bool linked)
 359 {
 360    ir_array_reference_visitor refs;
 361    if (!refs.get_split_list(instructions, linked))
 362       return false;
 363
 364    void *mem_ctx = ralloc_context(NULL);
 365
 366    /* Replace the decls of the arrays to be split with their split
 367     * components.
 368     */
 369    foreach_in_list(variable_entry, entry, &refs.variable_list) {
 370       const struct glsl_type *type = entry->var->type;
 371       const struct glsl_type *subtype;
 372
 373       if (type->is_matrix())
 374          subtype = type->column_type();
 375       else
 376          subtype = type->fields.array;
 377
 378       entry->mem_ctx = ralloc_parent(entry->var);
 379
 380       entry->components = ralloc_array(mem_ctx,
 381                                        ir_variable *,
 382                                        entry->size);
 383
 384       for (unsigned int i = 0; i < entry->size; i++) {
 385          const char *name = ralloc_asprintf(mem_ctx, "%s_%d",
 386                                             entry->var->name, i);
 387
 388          entry->components[i] =
 389             new(entry->mem_ctx) ir_variable(subtype, name, ir_var_temporary);
 390          entry->var->insert_before(entry->components[i]);
 391       }
 392
 393       entry->var->remove();
 394    }
 395
 396    ir_array_splitting_visitor split(&refs.variable_list);
 397    visit_list_elements(&split, instructions);
 398
 399    if (debug)
 400       _mesa_print_ir(stdout, instructions, NULL);
 401
 402    ralloc_free(mem_ctx);
 403
 404    return true;
 405
 406 }