ir_function_inlining.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 ir_function_inlining.cpp
  26  *
  27  * Replaces calls to functions with the body of the function.
  28  */
  29
  30 #define NULL 0
  31 #include "ir.h"
  32 #include "ir_visitor.h"
  33 #include "ir_function_inlining.h"
  34 #include "ir_expression_flattening.h"
  35 #include "glsl_types.h"
  36
  37 class variable_remap : public exec_node {
  38 public:
  39    variable_remap(const ir_variable *old_var, ir_variable *new_var)
  40       : old_var(old_var), new_var(new_var)
  41    {
  42       /* empty */
  43    }
  44    const ir_variable *old_var;
  45    ir_variable *new_var;
  46 };
  47
  48 class ir_function_cloning_visitor : public ir_visitor {
  49 public:
  50    ir_function_cloning_visitor(ir_variable *retval)
  51       : retval(retval)
  52    {
  53       /* empty */
  54    }
  55
  56    virtual ~ir_function_cloning_visitor()
  57    {
  58       /* empty */
  59    }
  60
  61    void remap_variable(const ir_variable *old_var, ir_variable *new_var) {
  62       variable_remap *remap = new variable_remap(old_var, new_var);
  63       this->remap_list.push_tail(remap);
  64    }
  65
  66    ir_variable *get_remapped_variable(ir_variable *var) {
  67       foreach_iter(exec_list_iterator, iter, this->remap_list) {
  68          variable_remap *remap = (variable_remap *)iter.get();
  69
  70          if (var == remap->old_var)
  71             return remap->new_var;
  72       }
  73
  74       /* Not a reapped variable, so a global scoped reference, for example. */
  75       return var;
  76    }
  77
  78    /* List of variable_remap for mapping from original function body variables
  79     * to inlined function body variables.
  80     */
  81    exec_list remap_list;
  82
  83    /* Return value for the inlined function. */
  84    ir_variable *retval;
  85
  86    /**
  87     * \name Visit methods
  88     *
  89     * As typical for the visitor pattern, there must be one \c visit method for
  90     * each concrete subclass of \c ir_instruction.  Virtual base classes within
  91     * the hierarchy should not have \c visit methods.
  92     */
  93    /*@{*/
  94    virtual void visit(ir_variable *);
  95    virtual void visit(ir_loop *);
  96    virtual void visit(ir_loop_jump *);
  97    virtual void visit(ir_function_signature *);
  98    virtual void visit(ir_function *);
  99    virtual void visit(ir_expression *);
 100    virtual void visit(ir_swizzle *);
 101    virtual void visit(ir_dereference *);
 102    virtual void visit(ir_assignment *);
 103    virtual void visit(ir_constant *);
 104    virtual void visit(ir_call *);
 105    virtual void visit(ir_return *);
 106    virtual void visit(ir_if *);
 107    /*@}*/
 108
 109    ir_instruction *result;
 110 };
 111
 112 void
 113 ir_function_cloning_visitor::visit(ir_variable *ir)
 114 {
 115    ir_variable *new_var = ir->clone();
 116
 117    this->result = new_var;
 118
 119    this->remap_variable(ir, new_var);
 120 }
 121
 122 void
 123 ir_function_cloning_visitor::visit(ir_loop *ir)
 124 {
 125    (void)ir;
 126    this->result = NULL;
 127 }
 128
 129 void
 130 ir_function_cloning_visitor::visit(ir_loop_jump *ir)
 131 {
 132    (void) ir;
 133    this->result = NULL;
 134 }
 135
 136
 137 void
 138 ir_function_cloning_visitor::visit(ir_function_signature *ir)
 139 {
 140    (void)ir;
 141    this->result = NULL;
 142 }
 143
 144
 145 void
 146 ir_function_cloning_visitor::visit(ir_function *ir)
 147 {
 148    (void) ir;
 149    this->result = NULL;
 150 }
 151
 152 void
 153 ir_function_cloning_visitor::visit(ir_expression *ir)
 154 {
 155    unsigned int operand;
 156    ir_rvalue *op[2] = {NULL, NULL};
 157
 158    for (operand = 0; operand < ir->get_num_operands(); operand++) {
 159       ir->operands[operand]->accept(this);
 160       op[operand] = this->result->as_rvalue();
 161       assert(op[operand]);
 162    }
 163
 164    this->result = new ir_expression(ir->operation, ir->type, op[0], op[1]);
 165 }
 166
 167
 168 void
 169 ir_function_cloning_visitor::visit(ir_swizzle *ir)
 170 {
 171    ir->val->accept(this);
 172
 173    this->result = new ir_swizzle(this->result->as_rvalue(), ir->mask);
 174 }
 175
 176 void
 177 ir_function_cloning_visitor::visit(ir_dereference *ir)
 178 {
 179    if (ir->mode == ir_dereference::ir_reference_variable) {
 180       ir_variable *old_var = ir->var->as_variable();
 181
 182       /* If it's a deref of a real variable, then we need to remap it if
 183        * it was local to the function.
 184        */
 185       if (old_var) {
 186          ir_variable *new_var;
 187
 188          new_var = this->get_remapped_variable(old_var);
 189
 190          this->result = new ir_dereference(new_var);
 191       } else {
 192          ir->var->accept(this);
 193
 194          this->result = new ir_dereference(this->result);
 195       }
 196    } else if (ir->mode == ir_dereference::ir_reference_array) {
 197       ir_instruction *variable;
 198       ir_rvalue *index;
 199
 200       ir->var->accept(this);
 201       variable = this->result;
 202
 203       ir->selector.array_index->accept(this);
 204       index = this->result->as_rvalue();
 205
 206       this->result = new ir_dereference(variable, index);
 207    } else {
 208       assert(ir->mode == ir_dereference::ir_reference_record);
 209       /* FINISHME: inlining of structure references */
 210       assert(0);
 211    }
 212 }
 213
 214 void
 215 ir_function_cloning_visitor::visit(ir_assignment *ir)
 216 {
 217    ir_rvalue *lhs, *rhs, *condition = NULL;
 218
 219    ir->lhs->accept(this);
 220    lhs = this->result->as_rvalue();
 221
 222    ir->rhs->accept(this);
 223    rhs = this->result->as_rvalue();
 224
 225    if (ir->condition) {
 226       ir->condition->accept(this);
 227       condition = this->result->as_rvalue();
 228    }
 229
 230    this->result = new ir_assignment(lhs, rhs, condition);
 231 }
 232
 233
 234 void
 235 ir_function_cloning_visitor::visit(ir_constant *ir)
 236 {
 237    this->result = ir->clone();
 238 }
 239
 240
 241 void
 242 ir_function_cloning_visitor::visit(ir_call *ir)
 243 {
 244    exec_list parameters;
 245
 246    foreach_iter(exec_list_iterator, iter, *ir) {
 247       ir_rvalue *param = (ir_rvalue *)iter.get();
 248
 249       param->accept(this);
 250       parameters.push_tail(this->result);
 251    }
 252
 253    this->result = new ir_call(ir->get_callee(), &parameters);
 254 }
 255
 256
 257 void
 258 ir_function_cloning_visitor::visit(ir_return *ir)
 259 {
 260    ir_rvalue *rval;
 261
 262    assert(this->retval);
 263
 264    rval = ir->get_value();
 265    rval->accept(this);
 266    rval = this->result->as_rvalue();
 267    assert(rval);
 268
 269    result = new ir_assignment(new ir_dereference(this->retval),
 270                               ir->get_value(), NULL);
 271 }
 272
 273
 274 void
 275 ir_function_cloning_visitor::visit(ir_if *ir)
 276 {
 277    (void) ir;
 278    result = NULL;
 279 }
 280
 281 bool
 282 can_inline(ir_call *call)
 283 {
 284    bool found_return = false;
 285
 286    /* FINISHME: Right now we only allow a single statement that is a return.
 287     */
 288    foreach_iter(exec_list_iterator, iter, call->get_callee()->body) {
 289       ir_instruction *ir = (ir_instruction *)iter.get();
 290       if (ir->get_next()->get_next() != NULL)
 291          return false;
 292
 293       if (!ir->as_return())
 294          return false;
 295
 296       found_return = true;
 297    }
 298
 299    return found_return;
 300 }
 301
 302 bool
 303 automatic_inlining_predicate(ir_instruction *ir)
 304 {
 305    ir_call *call = ir->as_call();
 306
 307    if (call && can_inline(call))
 308       return true;
 309
 310    return false;
 311 }
 312
 313 bool
 314 do_function_inlining(exec_list *instructions)
 315 {
 316    bool progress = false;
 317
 318    do_expression_flattening(instructions, automatic_inlining_predicate);
 319
 320    foreach_iter(exec_list_iterator, iter, *instructions) {
 321       ir_instruction *ir = (ir_instruction *)iter.get();
 322       ir_assignment *assign = ir->as_assignment();
 323       ir_call *call;
 324
 325       if (assign) {
 326          call = assign->rhs->as_call();
 327          if (!call || !can_inline(call))
 328             continue;
 329
 330          /* generates the parameter setup, function body, and returns the return
 331           * value of the function
 332           */
 333          ir_rvalue *rhs = call->generate_inline(ir);
 334          assert(rhs);
 335
 336          assign->rhs = rhs;
 337          progress = true;
 338       } else if ((call = ir->as_call()) && can_inline(call)) {
 339          (void)call->generate_inline(ir);
 340          ir->remove();
 341          progress = true;
 342       } else {
 343          ir_function_inlining_visitor v;
 344          ir->accept(&v);
 345       }
 346    }
 347
 348    return progress;
 349 }
 350
 351 ir_rvalue *
 352 ir_call::generate_inline(ir_instruction *next_ir)
 353 {
 354    ir_variable **parameters;
 355    int num_parameters;
 356    int i;
 357    ir_variable *retval = NULL;
 358
 359    num_parameters = 0;
 360    foreach_iter(exec_list_iterator, iter_sig, this->callee->parameters)
 361       num_parameters++;
 362
 363    parameters = new ir_variable *[num_parameters];
 364
 365    /* Generate storage for the return value. */
 366    if (this->callee->return_type) {
 367       retval = new ir_variable(this->callee->return_type, "__retval");
 368       next_ir->insert_before(retval);
 369    }
 370
 371    ir_function_cloning_visitor v = ir_function_cloning_visitor(retval);
 372
 373    /* Generate the declarations for the parameters to our inlined code,
 374     * and set up the mapping of real function body variables to ours.
 375     */
 376    i = 0;
 377    exec_list_iterator sig_param_iter = this->callee->parameters.iterator();
 378    exec_list_iterator param_iter = this->actual_parameters.iterator();
 379    for (i = 0; i < num_parameters; i++) {
 380       const ir_variable *const sig_param = (ir_variable *) sig_param_iter.get();
 381       ir_rvalue *param = (ir_rvalue *) param_iter.get();
 382
 383       /* Generate a new variable for the parameter. */
 384       parameters[i] = sig_param->clone();
 385       next_ir->insert_before(parameters[i]);
 386
 387       v.remap_variable(sig_param, parameters[i]);
 388
 389       /* Move the actual param into our param variable if it's an 'in' type. */
 390       if (parameters[i]->mode == ir_var_in ||
 391           parameters[i]->mode == ir_var_inout) {
 392          ir_assignment *assign;
 393
 394          assign = new ir_assignment(new ir_dereference(parameters[i]),
 395                                     param, NULL);
 396          next_ir->insert_before(assign);
 397       }
 398
 399       sig_param_iter.next();
 400       param_iter.next();
 401    }
 402
 403    /* Generate the inlined body of the function. */
 404    foreach_iter(exec_list_iterator, iter, callee->body) {
 405       ir_instruction *ir = (ir_instruction *)iter.get();
 406
 407       ir->accept(&v);
 408       assert(v.result);
 409       next_ir->insert_before(v.result);
 410    }
 411
 412    /* Generate the declarations for the parameters to our inlined code,
 413     * and set up the mapping of real function body variables to ours.
 414     */
 415    i = 0;
 416    param_iter = this->actual_parameters.iterator();
 417    for (i = 0; i < num_parameters; i++) {
 418       ir_instruction *const param = (ir_instruction *) param_iter.get();
 419
 420       /* Move the actual param into our param variable if it's an 'in' type. */
 421       if (parameters[i]->mode == ir_var_out ||
 422           parameters[i]->mode == ir_var_inout) {
 423          ir_assignment *assign;
 424
 425          assign = new ir_assignment(param->as_rvalue(),
 426                                     new ir_dereference(parameters[i]),
 427                                     NULL);
 428          next_ir->insert_before(assign);
 429       }
 430
 431       param_iter.next();
 432    }
 433
 434    delete(parameters);
 435
 436    if (retval)
 437       return new ir_dereference(retval);
 438    else
 439       return NULL;
 440 }
 441
 442 void
 443 ir_function_inlining_visitor::visit(ir_variable *ir)
 444 {
 445    (void) ir;
 446 }
 447
 448
 449 void
 450 ir_function_inlining_visitor::visit(ir_loop *ir)
 451 {
 452    do_function_inlining(&ir->body_instructions);
 453 }
 454
 455 void
 456 ir_function_inlining_visitor::visit(ir_loop_jump *ir)
 457 {
 458    (void) ir;
 459 }
 460
 461
 462 void
 463 ir_function_inlining_visitor::visit(ir_function_signature *ir)
 464 {
 465    do_function_inlining(&ir->body);
 466 }
 467
 468
 469 void
 470 ir_function_inlining_visitor::visit(ir_function *ir)
 471 {
 472    foreach_iter(exec_list_iterator, iter, *ir) {
 473       ir_function_signature *const sig = (ir_function_signature *) iter.get();
 474       sig->accept(this);
 475    }
 476 }
 477
 478 void
 479 ir_function_inlining_visitor::visit(ir_expression *ir)
 480 {
 481    unsigned int operand;
 482
 483    for (operand = 0; operand < ir->get_num_operands(); operand++) {
 484       ir->operands[operand]->accept(this);
 485    }
 486 }
 487
 488
 489 void
 490 ir_function_inlining_visitor::visit(ir_swizzle *ir)
 491 {
 492    ir->val->accept(this);
 493 }
 494
 495
 496 void
 497 ir_function_inlining_visitor::visit(ir_dereference *ir)
 498 {
 499    if (ir->mode == ir_dereference::ir_reference_array) {
 500       ir->selector.array_index->accept(this);
 501    }
 502    ir->var->accept(this);
 503 }
 504
 505 void
 506 ir_function_inlining_visitor::visit(ir_assignment *ir)
 507 {
 508    ir->rhs->accept(this);
 509 }
 510
 511
 512 void
 513 ir_function_inlining_visitor::visit(ir_constant *ir)
 514 {
 515    (void) ir;
 516 }
 517
 518
 519 void
 520 ir_function_inlining_visitor::visit(ir_call *ir)
 521 {
 522    (void) ir;
 523 }
 524
 525
 526 void
 527 ir_function_inlining_visitor::visit(ir_return *ir)
 528 {
 529    (void) ir;
 530 }
 531
 532
 533 void
 534 ir_function_inlining_visitor::visit(ir_if *ir)
 535 {
 536    ir->condition->accept(this);
 537
 538    do_function_inlining(&ir->then_instructions);
 539    do_function_inlining(&ir->else_instructions);
 540 }