src/compiler/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 "compiler/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 (!ir->array_index->as_constant()) {
 189       if (entry)
 190          entry->split = false;
 191       /* This variable indexing could come from a different array dereference
 192        * that also has variable indexing, that is, something like a[b[a[b[0]]]].
 193        * If we return visit_continue_with_parent here for the first appearence
 194        * of a, then we can miss that b also has indirect indexing (if this is
 195        * the only place in the program where such indirect indexing into b
 196        * happens), so keep going.
 197        */
 198       return visit_continue;
 199    }
 200
 201    /* If the index is also array dereference, visit index. */
 202    if (ir->array_index->as_dereference_array())
 203       visit_enter(ir->array_index->as_dereference_array());
 204
 205    return visit_continue_with_parent;
 206 }
 207
 208 ir_visitor_status
 209 ir_array_reference_visitor::visit_enter(ir_function_signature *ir)
 210 {
 211    /* We don't have logic for array-splitting function arguments,
 212     * so just look at the body instructions and not the parameter
 213     * declarations.
 214     */
 215    visit_list_elements(this, &ir->body);
 216    return visit_continue_with_parent;
 217 }
 218
 219 bool
 220 ir_array_reference_visitor::get_split_list(exec_list *instructions,
 221                                            bool linked)
 222 {
 223    visit_list_elements(this, instructions);
 224
 225    /* If the shaders aren't linked yet, we can't mess with global
 226     * declarations, which need to be matched by name across shaders.
 227     */
 228    if (!linked) {
 229       foreach_in_list(ir_instruction, node, instructions) {
 230          ir_variable *var = node->as_variable();
 231          if (var) {
 232             variable_entry *entry = get_variable_entry(var);
 233             if (entry)
 234                entry->remove();
 235          }
 236       }
 237    }
 238
 239    /* Trim out variables we found that we can't split. */
 240    foreach_in_list_safe(variable_entry, entry, &variable_list) {
 241       if (debug) {
 242          printf("array %s@%p: decl %d, split %d\n",
 243                 entry->var->name, (void *) entry->var, entry->declaration,
 244                 entry->split);
 245       }
 246
 247       if (!(entry->declaration && entry->split)) {
 248          entry->remove();
 249       }
 250    }
 251
 252    return !variable_list.is_empty();
 253 }
 254
 255 /**
 256  * This class rewrites the dereferences of arrays that have been split
 257  * to use the newly created ir_variables for each component.
 258  */
 259 class ir_array_splitting_visitor : public ir_rvalue_visitor {
 260 public:
 261    ir_array_splitting_visitor(exec_list *vars)
 262    {
 263       this->variable_list = vars;
 264    }
 265
 266    virtual ~ir_array_splitting_visitor()
 267    {
 268    }
 269
 270    virtual ir_visitor_status visit_leave(ir_assignment *);
 271
 272    void split_deref(ir_dereference **deref);
 273    void handle_rvalue(ir_rvalue **rvalue);
 274    variable_entry *get_splitting_entry(ir_variable *var);
 275
 276    exec_list *variable_list;
 277 };
 278
 279 variable_entry *
 280 ir_array_splitting_visitor::get_splitting_entry(ir_variable *var)
 281 {
 282    assert(var);
 283
 284    foreach_in_list(variable_entry, entry, this->variable_list) {
 285       if (entry->var == var) {
 286          return entry;
 287       }
 288    }
 289
 290    return NULL;
 291 }
 292
 293 void
 294 ir_array_splitting_visitor::split_deref(ir_dereference **deref)
 295 {
 296    ir_dereference_array *deref_array = (*deref)->as_dereference_array();
 297    if (!deref_array)
 298       return;
 299
 300    ir_dereference_variable *deref_var = deref_array->array->as_dereference_variable();
 301    if (!deref_var)
 302       return;
 303    ir_variable *var = deref_var->var;
 304
 305    variable_entry *entry = get_splitting_entry(var);
 306    if (!entry)
 307       return;
 308
 309    ir_constant *constant = deref_array->array_index->as_constant();
 310    assert(constant);
 311
 312    if (constant->value.i[0] >= 0 && constant->value.i[0] < (int)entry->size) {
 313       *deref = new(entry->mem_ctx)
 314                ir_dereference_variable(entry->components[constant->value.i[0]]);
 315    } else {
 316       /* There was a constant array access beyond the end of the
 317        * array.  This might have happened due to constant folding
 318        * after the initial parse.  This produces an undefined value,
 319        * but shouldn't crash.  Just give them an uninitialized
 320        * variable.
 321        */
 322       ir_variable *temp = new(entry->mem_ctx) ir_variable(deref_array->type,
 323                                                           "undef",
 324                                                           ir_var_temporary);
 325       entry->components[0]->insert_before(temp);
 326       *deref = new(entry->mem_ctx) ir_dereference_variable(temp);
 327    }
 328 }
 329
 330 void
 331 ir_array_splitting_visitor::handle_rvalue(ir_rvalue **rvalue)
 332 {
 333    if (!*rvalue)
 334       return;
 335
 336    ir_dereference *deref = (*rvalue)->as_dereference();
 337
 338    if (!deref)
 339       return;
 340
 341    split_deref(&deref);
 342    *rvalue = deref;
 343 }
 344
 345 ir_visitor_status
 346 ir_array_splitting_visitor::visit_leave(ir_assignment *ir)
 347 {
 348    /* The normal rvalue visitor skips the LHS of assignments, but we
 349     * need to process those just the same.
 350     */
 351    ir_rvalue *lhs = ir->lhs;
 352
 353    handle_rvalue(&lhs);
 354    ir->lhs = lhs->as_dereference();
 355
 356    ir->lhs->accept(this);
 357
 358    handle_rvalue(&ir->rhs);
 359    ir->rhs->accept(this);
 360
 361    if (ir->condition) {
 362       handle_rvalue(&ir->condition);
 363       ir->condition->accept(this);
 364    }
 365
 366    return visit_continue;
 367 }
 368
 369 bool
 370 optimize_split_arrays(exec_list *instructions, bool linked)
 371 {
 372    ir_array_reference_visitor refs;
 373    if (!refs.get_split_list(instructions, linked))
 374       return false;
 375
 376    void *mem_ctx = ralloc_context(NULL);
 377
 378    /* Replace the decls of the arrays to be split with their split
 379     * components.
 380     */
 381    foreach_in_list(variable_entry, entry, &refs.variable_list) {
 382       const struct glsl_type *type = entry->var->type;
 383       const struct glsl_type *subtype;
 384
 385       if (type->is_matrix())
 386          subtype = type->column_type();
 387       else
 388          subtype = type->fields.array;
 389
 390       entry->mem_ctx = ralloc_parent(entry->var);
 391
 392       entry->components = ralloc_array(mem_ctx, ir_variable *, entry->size);
 393
 394       for (unsigned int i = 0; i < entry->size; i++) {
 395          const char *name = ralloc_asprintf(mem_ctx, "%s_%d",
 396                                             entry->var->name, i);
 397
 398          entry->components[i] =
 399             new(entry->mem_ctx) ir_variable(subtype, name, ir_var_temporary);
 400          entry->var->insert_before(entry->components[i]);
 401       }
 402
 403       entry->var->remove();
 404    }
 405
 406    ir_array_splitting_visitor split(&refs.variable_list);
 407    visit_list_elements(&split, instructions);
 408
 409    if (debug)
 410       _mesa_print_ir(stdout, instructions, NULL);
 411
 412    ralloc_free(mem_ctx);
 413
 414    return true;
 415
 416 }