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->whole_array_access = 0;
  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    /** Number of times the variable is referenced, including assignments. */
  66    unsigned whole_array_access;
  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, then the dereference of this array didn't
 158     * have a constant index (see the visit_continue_with_parent
 159     * below), so we can't split the variable.
 160     */
 161    if (entry)
 162       entry->whole_array_access++;
 163
 164    return visit_continue;
 165 }
 166
 167 ir_visitor_status
 168 ir_array_reference_visitor::visit_enter(ir_dereference_array *ir)
 169 {
 170    ir_dereference_variable *deref = ir->array->as_dereference_variable();
 171    if (!deref)
 172       return visit_continue;
 173
 174    variable_entry *entry = this->get_variable_entry(deref->var);
 175
 176    if (entry && !ir->array_index->as_constant())
 177       entry->whole_array_access++;
 178
 179    return visit_continue_with_parent;
 180 }
 181
 182 bool
 183 ir_array_reference_visitor::get_split_list(exec_list *instructions,
 184                                            bool linked)
 185 {
 186    visit_list_elements(this, instructions);
 187
 188    /* If the shaders aren't linked yet, we can't mess with global
 189     * declarations, which need to be matched by name across shaders.
 190     */
 191    if (!linked) {
 192       foreach_list(node, instructions) {
 193          ir_variable *var = ((ir_instruction *)node)->as_variable();
 194          if (var) {
 195             variable_entry *entry = get_variable_entry(var);
 196             if (entry)
 197                entry->remove();
 198          }
 199       }
 200    }
 201
 202    /* Trim out variables we found that we can't split. */
 203    foreach_iter(exec_list_iterator, iter, variable_list) {
 204       variable_entry *entry = (variable_entry *)iter.get();
 205
 206       if (debug) {
 207          printf("array %s@%p: decl %d, whole_access %d\n",
 208                 entry->var->name, (void *) entry->var, entry->declaration,
 209                 entry->whole_array_access);
 210       }
 211
 212       if (!entry->declaration || entry->whole_array_access) {
 213          entry->remove();
 214       }
 215    }
 216
 217    return !variable_list.is_empty();
 218 }
 219
 220 /** This is the class that does the actual work of splitting. */
 221 class ir_array_splitting_visitor : public ir_rvalue_visitor {
 222 public:
 223    ir_array_splitting_visitor(exec_list *vars)
 224    {
 225       this->variable_list = vars;
 226    }
 227
 228    virtual ~ir_array_splitting_visitor()
 229    {
 230    }
 231
 232    virtual ir_visitor_status visit_leave(ir_assignment *);
 233
 234    void split_deref(ir_dereference **deref);
 235    void handle_rvalue(ir_rvalue **rvalue);
 236    variable_entry *get_splitting_entry(ir_variable *var);
 237
 238    exec_list *variable_list;
 239 };
 240
 241 variable_entry *
 242 ir_array_splitting_visitor::get_splitting_entry(ir_variable *var)
 243 {
 244    assert(var);
 245
 246    foreach_iter(exec_list_iterator, iter, *this->variable_list) {
 247       variable_entry *entry = (variable_entry *)iter.get();
 248       if (entry->var == var) {
 249          return entry;
 250       }
 251    }
 252
 253    return NULL;
 254 }
 255
 256 void
 257 ir_array_splitting_visitor::split_deref(ir_dereference **deref)
 258 {
 259    ir_dereference_array *deref_array = (*deref)->as_dereference_array();
 260    if (!deref_array)
 261       return;
 262
 263    ir_dereference_variable *deref_var = deref_array->array->as_dereference_variable();
 264    if (!deref_var)
 265       return;
 266    ir_variable *var = deref_var->var;
 267
 268    variable_entry *entry = get_splitting_entry(var);
 269    if (!entry)
 270       return;
 271
 272    ir_constant *constant = deref_array->array_index->as_constant();
 273    assert(constant);
 274
 275    if (constant->value.i[0] < (int)entry->size) {
 276       *deref = new(entry->mem_ctx)
 277          ir_dereference_variable(entry->components[constant->value.i[0]]);
 278    } else {
 279       /* There was a constant array access beyond the end of the
 280        * array.  This might have happened due to constant folding
 281        * after the initial parse.  This produces an undefined value,
 282        * but shouldn't crash.  Just give them an uninitialized
 283        * variable.
 284        */
 285       ir_variable *temp = new(entry->mem_ctx) ir_variable(deref_array->type,
 286                                                           "undef",
 287                                                           ir_var_temporary);
 288       entry->components[0]->insert_before(temp);
 289       *deref = new(entry->mem_ctx) ir_dereference_variable(temp);
 290    }
 291 }
 292
 293 void
 294 ir_array_splitting_visitor::handle_rvalue(ir_rvalue **rvalue)
 295 {
 296    if (!*rvalue)
 297       return;
 298
 299    ir_dereference *deref = (*rvalue)->as_dereference();
 300
 301    if (!deref)
 302       return;
 303
 304    split_deref(&deref);
 305    *rvalue = deref;
 306 }
 307
 308 ir_visitor_status
 309 ir_array_splitting_visitor::visit_leave(ir_assignment *ir)
 310 {
 311    /* The normal rvalue visitor skips the LHS of assignments, but we
 312     * need to process those just the same.
 313     */
 314    ir_rvalue *lhs = ir->lhs;
 315
 316    handle_rvalue(&lhs);
 317    ir->lhs = lhs->as_dereference();
 318
 319    ir->lhs->accept(this);
 320
 321    handle_rvalue(&ir->rhs);
 322    ir->rhs->accept(this);
 323
 324    if (ir->condition) {
 325       handle_rvalue(&ir->condition);
 326       ir->condition->accept(this);
 327    }
 328
 329    return visit_continue;
 330 }
 331
 332 bool
 333 optimize_split_arrays(exec_list *instructions, bool linked)
 334 {
 335    ir_array_reference_visitor refs;
 336    if (!refs.get_split_list(instructions, linked))
 337       return false;
 338
 339    void *mem_ctx = ralloc_context(NULL);
 340
 341    /* Replace the decls of the arrays to be split with their split
 342     * components.
 343     */
 344    foreach_iter(exec_list_iterator, iter, refs.variable_list) {
 345       variable_entry *entry = (variable_entry *)iter.get();
 346       const struct glsl_type *type = entry->var->type;
 347       const struct glsl_type *subtype;
 348
 349       if (type->is_matrix())
 350          subtype = type->column_type();
 351       else
 352          subtype = type->fields.array;
 353
 354       entry->mem_ctx = ralloc_parent(entry->var);
 355
 356       entry->components = ralloc_array(mem_ctx,
 357                                        ir_variable *,
 358                                        entry->size);
 359
 360       for (unsigned int i = 0; i < entry->size; i++) {
 361          const char *name = ralloc_asprintf(mem_ctx, "%s_%d",
 362                                             entry->var->name, i);
 363
 364          entry->components[i] =
 365             new(entry->mem_ctx) ir_variable(subtype, name, ir_var_temporary);
 366          entry->var->insert_before(entry->components[i]);
 367       }
 368
 369       entry->var->remove();
 370    }
 371
 372    ir_array_splitting_visitor split(&refs.variable_list);
 373    visit_list_elements(&split, instructions);
 374
 375    if (debug)
 376       _mesa_print_ir(instructions, NULL);
 377
 378    ralloc_free(mem_ctx);
 379
 380    return true;
 381
 382 }