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