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_list(n, &this->variable_list) {
 139       variable_entry *entry = (variable_entry *) n;
 140       if (entry->var == var)
 141          return entry;
 142    }
 143
 144    variable_entry *entry = new(mem_ctx) variable_entry(var);
 145    this->variable_list.push_tail(entry);
 146    return entry;
 147 }
 148
 149
 150 ir_visitor_status
 151 ir_array_reference_visitor::visit(ir_variable *ir)
 152 {
 153    variable_entry *entry = this->get_variable_entry(ir);
 154
 155    if (entry)
 156       entry->declaration = true;
 157
 158    return visit_continue;
 159 }
 160
 161 ir_visitor_status
 162 ir_array_reference_visitor::visit(ir_dereference_variable *ir)
 163 {
 164    variable_entry *entry = this->get_variable_entry(ir->var);
 165
 166    /* If we made it to here without seeing an ir_dereference_array,
 167     * then the dereference of this array didn't have a constant index
 168     * (see the visit_continue_with_parent below), so we can't split
 169     * the variable.
 170     */
 171    if (entry)
 172       entry->split = false;
 173
 174    return visit_continue;
 175 }
 176
 177 ir_visitor_status
 178 ir_array_reference_visitor::visit_enter(ir_dereference_array *ir)
 179 {
 180    ir_dereference_variable *deref = ir->array->as_dereference_variable();
 181    if (!deref)
 182       return visit_continue;
 183
 184    variable_entry *entry = this->get_variable_entry(deref->var);
 185
 186    /* If the access to the array has a variable index, we wouldn't
 187     * know which split variable this dereference should go to.
 188     */
 189    if (entry && !ir->array_index->as_constant())
 190       entry->split = false;
 191
 192    return visit_continue_with_parent;
 193 }
 194
 195 ir_visitor_status
 196 ir_array_reference_visitor::visit_enter(ir_function_signature *ir)
 197 {
 198    /* We don't have logic for array-splitting function arguments,
 199     * so just look at the body instructions and not the parameter
 200     * declarations.
 201     */
 202    visit_list_elements(this, &ir->body);
 203    return visit_continue_with_parent;
 204 }
 205
 206 bool
 207 ir_array_reference_visitor::get_split_list(exec_list *instructions,
 208                                            bool linked)
 209 {
 210    visit_list_elements(this, instructions);
 211
 212    /* If the shaders aren't linked yet, we can't mess with global
 213     * declarations, which need to be matched by name across shaders.
 214     */
 215    if (!linked) {
 216       foreach_list(node, instructions) {
 217          ir_variable *var = ((ir_instruction *)node)->as_variable();
 218          if (var) {
 219             variable_entry *entry = get_variable_entry(var);
 220             if (entry)
 221                entry->remove();
 222          }
 223       }
 224    }
 225
 226    /* Trim out variables we found that we can't split. */
 227    foreach_list_safe(n, &variable_list) {
 228       variable_entry *entry = (variable_entry *) n;
 229
 230       if (debug) {
 231          printf("array %s@%p: decl %d, split %d\n",
 232                 entry->var->name, (void *) entry->var, entry->declaration,
 233                 entry->split);
 234       }
 235
 236       if (!(entry->declaration && entry->split)) {
 237          entry->remove();
 238       }
 239    }
 240
 241    return !variable_list.is_empty();
 242 }
 243
 244 /**
 245  * This class rewrites the dereferences of arrays that have been split
 246  * to use the newly created ir_variables for each component.
 247  */
 248 class ir_array_splitting_visitor : public ir_rvalue_visitor {
 249 public:
 250    ir_array_splitting_visitor(exec_list *vars)
 251    {
 252       this->variable_list = vars;
 253    }
 254
 255    virtual ~ir_array_splitting_visitor()
 256    {
 257    }
 258
 259    virtual ir_visitor_status visit_leave(ir_assignment *);
 260
 261    void split_deref(ir_dereference **deref);
 262    void handle_rvalue(ir_rvalue **rvalue);
 263    variable_entry *get_splitting_entry(ir_variable *var);
 264
 265    exec_list *variable_list;
 266 };
 267
 268 variable_entry *
 269 ir_array_splitting_visitor::get_splitting_entry(ir_variable *var)
 270 {
 271    assert(var);
 272
 273    foreach_list(n, this->variable_list) {
 274       variable_entry *entry = (variable_entry *) n;
 275       if (entry->var == var) {
 276          return entry;
 277       }
 278    }
 279
 280    return NULL;
 281 }
 282
 283 void
 284 ir_array_splitting_visitor::split_deref(ir_dereference **deref)
 285 {
 286    ir_dereference_array *deref_array = (*deref)->as_dereference_array();
 287    if (!deref_array)
 288       return;
 289
 290    ir_dereference_variable *deref_var = deref_array->array->as_dereference_variable();
 291    if (!deref_var)
 292       return;
 293    ir_variable *var = deref_var->var;
 294
 295    variable_entry *entry = get_splitting_entry(var);
 296    if (!entry)
 297       return;
 298
 299    ir_constant *constant = deref_array->array_index->as_constant();
 300    assert(constant);
 301
 302    if (constant->value.i[0] < (int)entry->size) {
 303       *deref = new(entry->mem_ctx)
 304          ir_dereference_variable(entry->components[constant->value.i[0]]);
 305    } else {
 306       /* There was a constant array access beyond the end of the
 307        * array.  This might have happened due to constant folding
 308        * after the initial parse.  This produces an undefined value,
 309        * but shouldn't crash.  Just give them an uninitialized
 310        * variable.
 311        */
 312       ir_variable *temp = new(entry->mem_ctx) ir_variable(deref_array->type,
 313                                                           "undef",
 314                                                           ir_var_temporary);
 315       entry->components[0]->insert_before(temp);
 316       *deref = new(entry->mem_ctx) ir_dereference_variable(temp);
 317    }
 318 }
 319
 320 void
 321 ir_array_splitting_visitor::handle_rvalue(ir_rvalue **rvalue)
 322 {
 323    if (!*rvalue)
 324       return;
 325
 326    ir_dereference *deref = (*rvalue)->as_dereference();
 327
 328    if (!deref)
 329       return;
 330
 331    split_deref(&deref);
 332    *rvalue = deref;
 333 }
 334
 335 ir_visitor_status
 336 ir_array_splitting_visitor::visit_leave(ir_assignment *ir)
 337 {
 338    /* The normal rvalue visitor skips the LHS of assignments, but we
 339     * need to process those just the same.
 340     */
 341    ir_rvalue *lhs = ir->lhs;
 342
 343    handle_rvalue(&lhs);
 344    ir->lhs = lhs->as_dereference();
 345
 346    ir->lhs->accept(this);
 347
 348    handle_rvalue(&ir->rhs);
 349    ir->rhs->accept(this);
 350
 351    if (ir->condition) {
 352       handle_rvalue(&ir->condition);
 353       ir->condition->accept(this);
 354    }
 355
 356    return visit_continue;
 357 }
 358
 359 bool
 360 optimize_split_arrays(exec_list *instructions, bool linked)
 361 {
 362    ir_array_reference_visitor refs;
 363    if (!refs.get_split_list(instructions, linked))
 364       return false;
 365
 366    void *mem_ctx = ralloc_context(NULL);
 367
 368    /* Replace the decls of the arrays to be split with their split
 369     * components.
 370     */
 371    foreach_list(n, &refs.variable_list) {
 372       variable_entry *entry = (variable_entry *) n;
 373       const struct glsl_type *type = entry->var->type;
 374       const struct glsl_type *subtype;
 375
 376       if (type->is_matrix())
 377          subtype = type->column_type();
 378       else
 379          subtype = type->fields.array;
 380
 381       entry->mem_ctx = ralloc_parent(entry->var);
 382
 383       entry->components = ralloc_array(mem_ctx,
 384                                        ir_variable *,
 385                                        entry->size);
 386
 387       for (unsigned int i = 0; i < entry->size; i++) {
 388          const char *name = ralloc_asprintf(mem_ctx, "%s_%d",
 389                                             entry->var->name, i);
 390
 391          entry->components[i] =
 392             new(entry->mem_ctx) ir_variable(subtype, name, ir_var_temporary);
 393          entry->var->insert_before(entry->components[i]);
 394       }
 395
 396       entry->var->remove();
 397    }
 398
 399    ir_array_splitting_visitor split(&refs.variable_list);
 400    visit_list_elements(&split, instructions);
 401
 402    if (debug)
 403       _mesa_print_ir(stdout, instructions, NULL);
 404
 405    ralloc_free(mem_ctx);
 406
 407    return true;
 408
 409 }