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    /* FINISHME: Implement loop cloning. */
 126    assert(0);
 127
 128    (void)ir;
 129    this->result = NULL;
 130 }
 131
 132 void
 133 ir_function_cloning_visitor::visit(ir_loop_jump *ir)
 134 {
 135    /* FINISHME: Implement loop cloning. */
 136    assert(0);
 137
 138    (void) ir;
 139    this->result = NULL;
 140 }
 141
 142
 143 void
 144 ir_function_cloning_visitor::visit(ir_function_signature *ir)
 145 {
 146    assert(0);
 147    (void)ir;
 148    this->result = NULL;
 149 }
 150
 151
 152 void
 153 ir_function_cloning_visitor::visit(ir_function *ir)
 154 {
 155    assert(0);
 156    (void) ir;
 157    this->result = NULL;
 158 }
 159
 160 void
 161 ir_function_cloning_visitor::visit(ir_expression *ir)
 162 {
 163    unsigned int operand;
 164    ir_rvalue *op[2] = {NULL, NULL};
 165
 166    for (operand = 0; operand < ir->get_num_operands(); operand++) {
 167       ir->operands[operand]->accept(this);
 168       op[operand] = this->result->as_rvalue();
 169       assert(op[operand]);
 170    }
 171
 172    this->result = new ir_expression(ir->operation, ir->type, op[0], op[1]);
 173 }
 174
 175
 176 void
 177 ir_function_cloning_visitor::visit(ir_swizzle *ir)
 178 {
 179    ir->val->accept(this);
 180
 181    this->result = new ir_swizzle(this->result->as_rvalue(), ir->mask);
 182 }
 183
 184 void
 185 ir_function_cloning_visitor::visit(ir_dereference *ir)
 186 {
 187    ir_variable *old_var = ir->var->as_variable();
 188    ir_instruction *var;
 189
 190    if (old_var)
 191       var = this->get_remapped_variable(old_var);
 192    else {
 193       ir->var->accept(this);
 194       var = this->result;
 195    }
 196
 197    if (ir->mode == ir_dereference::ir_reference_variable) {
 198       this->result = new ir_dereference(var);
 199    } else if (ir->mode == ir_dereference::ir_reference_array) {
 200       ir_rvalue *index;
 201
 202       ir->selector.array_index->accept(this);
 203       index = this->result->as_rvalue();
 204
 205       this->result = new ir_dereference(var, index);
 206    } else {
 207       assert(ir->mode == ir_dereference::ir_reference_record);
 208       this->result = new ir_dereference(var, strdup(ir->selector.field));
 209    }
 210 }
 211
 212 void
 213 ir_function_cloning_visitor::visit(ir_assignment *ir)
 214 {
 215    ir_rvalue *lhs, *rhs, *condition = NULL;
 216
 217    ir->lhs->accept(this);
 218    lhs = this->result->as_rvalue();
 219
 220    ir->rhs->accept(this);
 221    rhs = this->result->as_rvalue();
 222
 223    if (ir->condition) {
 224       ir->condition->accept(this);
 225       condition = this->result->as_rvalue();
 226    }
 227
 228    this->result = new ir_assignment(lhs, rhs, condition);
 229 }
 230
 231
 232 void
 233 ir_function_cloning_visitor::visit(ir_constant *ir)
 234 {
 235    this->result = ir->clone();
 236 }
 237
 238
 239 void
 240 ir_function_cloning_visitor::visit(ir_call *ir)
 241 {
 242    exec_list parameters;
 243
 244    foreach_iter(exec_list_iterator, iter, *ir) {
 245       ir_rvalue *param = (ir_rvalue *)iter.get();
 246
 247       param->accept(this);
 248       parameters.push_tail(this->result);
 249    }
 250
 251    this->result = new ir_call(ir->get_callee(), &parameters);
 252 }
 253
 254
 255 void
 256 ir_function_cloning_visitor::visit(ir_return *ir)
 257 {
 258    ir_rvalue *rval;
 259
 260    assert(this->retval);
 261
 262    rval = ir->get_value();
 263    rval->accept(this);
 264    rval = this->result->as_rvalue();
 265    assert(rval);
 266
 267    result = new ir_assignment(new ir_dereference(this->retval), rval, NULL);
 268 }
 269
 270
 271 void
 272 ir_function_cloning_visitor::visit(ir_if *ir)
 273 {
 274    /* FINISHME: Implement if cloning. */
 275    assert(0);
 276
 277    (void) ir;
 278    result = NULL;
 279 }
 280
 281 bool
 282 automatic_inlining_predicate(ir_instruction *ir)
 283 {
 284    ir_call *call = ir->as_call();
 285
 286    if (call && can_inline(call))
 287       return true;
 288
 289    return false;
 290 }
 291
 292 bool
 293 do_function_inlining(exec_list *instructions)
 294 {
 295    bool progress = false;
 296
 297    do_expression_flattening(instructions, automatic_inlining_predicate);
 298
 299    foreach_iter(exec_list_iterator, iter, *instructions) {
 300       ir_instruction *ir = (ir_instruction *)iter.get();
 301       ir_assignment *assign = ir->as_assignment();
 302       ir_call *call;
 303
 304       if (assign) {
 305          call = assign->rhs->as_call();
 306          if (!call || !can_inline(call))
 307             continue;
 308
 309          /* generates the parameter setup, function body, and returns the return
 310           * value of the function
 311           */
 312          ir_rvalue *rhs = call->generate_inline(ir);
 313          assert(rhs);
 314
 315          assign->rhs = rhs;
 316          progress = true;
 317       } else if ((call = ir->as_call()) && can_inline(call)) {
 318          (void)call->generate_inline(ir);
 319          ir->remove();
 320          progress = true;
 321       } else {
 322          ir_function_inlining_visitor v;
 323          ir->accept(&v);
 324       }
 325    }
 326
 327    return progress;
 328 }
 329
 330 ir_rvalue *
 331 ir_call::generate_inline(ir_instruction *next_ir)
 332 {
 333    ir_variable **parameters;
 334    int num_parameters;
 335    int i;
 336    ir_variable *retval = NULL;
 337
 338    num_parameters = 0;
 339    foreach_iter(exec_list_iterator, iter_sig, this->callee->parameters)
 340       num_parameters++;
 341
 342    parameters = new ir_variable *[num_parameters];
 343
 344    /* Generate storage for the return value. */
 345    if (this->callee->return_type) {
 346       retval = new ir_variable(this->callee->return_type, "__retval");
 347       next_ir->insert_before(retval);
 348    }
 349
 350    ir_function_cloning_visitor v = ir_function_cloning_visitor(retval);
 351
 352    /* Generate the declarations for the parameters to our inlined code,
 353     * and set up the mapping of real function body variables to ours.
 354     */
 355    i = 0;
 356    exec_list_iterator sig_param_iter = this->callee->parameters.iterator();
 357    exec_list_iterator param_iter = this->actual_parameters.iterator();
 358    for (i = 0; i < num_parameters; i++) {
 359       const ir_variable *const sig_param = (ir_variable *) sig_param_iter.get();
 360       ir_rvalue *param = (ir_rvalue *) param_iter.get();
 361
 362       /* Generate a new variable for the parameter. */
 363       parameters[i] = sig_param->clone();
 364       next_ir->insert_before(parameters[i]);
 365
 366       v.remap_variable(sig_param, parameters[i]);
 367
 368       /* Move the actual param into our param variable if it's an 'in' type. */
 369       if (parameters[i]->mode == ir_var_in ||
 370           parameters[i]->mode == ir_var_inout) {
 371          ir_assignment *assign;
 372
 373          assign = new ir_assignment(new ir_dereference(parameters[i]),
 374                                     param, NULL);
 375          next_ir->insert_before(assign);
 376       }
 377
 378       sig_param_iter.next();
 379       param_iter.next();
 380    }
 381
 382    /* Generate the inlined body of the function. */
 383    foreach_iter(exec_list_iterator, iter, callee->body) {
 384       ir_instruction *ir = (ir_instruction *)iter.get();
 385
 386       ir->accept(&v);
 387       assert(v.result);
 388       next_ir->insert_before(v.result);
 389    }
 390
 391    /* Copy back the value of any 'out' parameters from the function body
 392     * variables to our own.
 393     */
 394    i = 0;
 395    param_iter = this->actual_parameters.iterator();
 396    for (i = 0; i < num_parameters; i++) {
 397       ir_instruction *const param = (ir_instruction *) param_iter.get();
 398
 399       /* Move our param variable into the actual param if it's an 'out' type. */
 400       if (parameters[i]->mode == ir_var_out ||
 401           parameters[i]->mode == ir_var_inout) {
 402          ir_assignment *assign;
 403
 404          assign = new ir_assignment(param->as_rvalue(),
 405                                     new ir_dereference(parameters[i]),
 406                                     NULL);
 407          next_ir->insert_before(assign);
 408       }
 409
 410       param_iter.next();
 411    }
 412
 413    delete(parameters);
 414
 415    if (retval)
 416       return new ir_dereference(retval);
 417    else
 418       return NULL;
 419 }
 420
 421 void
 422 ir_function_inlining_visitor::visit(ir_variable *ir)
 423 {
 424    (void) ir;
 425 }
 426
 427
 428 void
 429 ir_function_inlining_visitor::visit(ir_loop *ir)
 430 {
 431    do_function_inlining(&ir->body_instructions);
 432 }
 433
 434 void
 435 ir_function_inlining_visitor::visit(ir_loop_jump *ir)
 436 {
 437    (void) ir;
 438 }
 439
 440
 441 void
 442 ir_function_inlining_visitor::visit(ir_function_signature *ir)
 443 {
 444    do_function_inlining(&ir->body);
 445 }
 446
 447
 448 void
 449 ir_function_inlining_visitor::visit(ir_function *ir)
 450 {
 451    foreach_iter(exec_list_iterator, iter, *ir) {
 452       ir_function_signature *const sig = (ir_function_signature *) iter.get();
 453       sig->accept(this);
 454    }
 455 }
 456
 457 void
 458 ir_function_inlining_visitor::visit(ir_expression *ir)
 459 {
 460    unsigned int operand;
 461
 462    for (operand = 0; operand < ir->get_num_operands(); operand++) {
 463       ir->operands[operand]->accept(this);
 464    }
 465 }
 466
 467
 468 void
 469 ir_function_inlining_visitor::visit(ir_swizzle *ir)
 470 {
 471    ir->val->accept(this);
 472 }
 473
 474
 475 void
 476 ir_function_inlining_visitor::visit(ir_dereference *ir)
 477 {
 478    if (ir->mode == ir_dereference::ir_reference_array) {
 479       ir->selector.array_index->accept(this);
 480    }
 481    ir->var->accept(this);
 482 }
 483
 484 void
 485 ir_function_inlining_visitor::visit(ir_assignment *ir)
 486 {
 487    ir->rhs->accept(this);
 488 }
 489
 490
 491 void
 492 ir_function_inlining_visitor::visit(ir_constant *ir)
 493 {
 494    (void) ir;
 495 }
 496
 497
 498 void
 499 ir_function_inlining_visitor::visit(ir_call *ir)
 500 {
 501    (void) ir;
 502 }
 503
 504
 505 void
 506 ir_function_inlining_visitor::visit(ir_return *ir)
 507 {
 508    (void) ir;
 509 }
 510
 511
 512 void
 513 ir_function_inlining_visitor::visit(ir_if *ir)
 514 {
 515    ir->condition->accept(this);
 516
 517    do_function_inlining(&ir->then_instructions);
 518    do_function_inlining(&ir->else_instructions);
 519 }