src/compiler/glsl/opt_constant_propagation.cpp

   1 /*
   2  * Copyright © 2010 Intel Corporation
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * constant 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, constant, 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 constantright 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 CONSTANTRIGHT 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_constant_propagation.cpp
  26  *
  27  * Tracks assignments of constants to channels of variables, and
  28  * usage of those constant channels with direct usage of the constants.
  29  *
  30  * This can lead to constant folding and algebraic optimizations in
  31  * those later expressions, while causing no increase in instruction
  32  * count (due to constants being generally free to load from a
  33  * constant push buffer or as instruction immediate values) and
  34  * possibly reducing register pressure.
  35  */
  36
  37 #include "ir.h"
  38 #include "ir_visitor.h"
  39 #include "ir_rvalue_visitor.h"
  40 #include "ir_basic_block.h"
  41 #include "ir_optimization.h"
  42 #include "compiler/glsl_types.h"
  43 #include "util/hash_table.h"
  44
  45 namespace {
  46
  47 class acp_entry : public exec_node
  48 {
  49 public:
  50    /* override operator new from exec_node */
  51    DECLARE_LINEAR_ZALLOC_CXX_OPERATORS(acp_entry)
  52
  53    acp_entry(ir_variable *var, unsigned write_mask, ir_constant *constant)
  54    {
  55       assert(var);
  56       assert(constant);
  57       this->var = var;
  58       this->write_mask = write_mask;
  59       this->constant = constant;
  60       this->initial_values = write_mask;
  61    }
  62
  63    acp_entry(const acp_entry *src)
  64    {
  65       this->var = src->var;
  66       this->write_mask = src->write_mask;
  67       this->constant = src->constant;
  68       this->initial_values = src->initial_values;
  69    }
  70
  71    ir_variable *var;
  72    ir_constant *constant;
  73    unsigned write_mask;
  74
  75    /** Mask of values initially available in the constant. */
  76    unsigned initial_values;
  77 };
  78
  79
  80 class kill_entry : public exec_node
  81 {
  82 public:
  83    /* override operator new from exec_node */
  84    DECLARE_LINEAR_ZALLOC_CXX_OPERATORS(kill_entry)
  85
  86    kill_entry(ir_variable *var, unsigned write_mask)
  87    {
  88       assert(var);
  89       this->var = var;
  90       this->write_mask = write_mask;
  91    }
  92
  93    ir_variable *var;
  94    unsigned write_mask;
  95 };
  96
  97 class ir_constant_propagation_visitor : public ir_rvalue_visitor {
  98 public:
  99    ir_constant_propagation_visitor()
 100    {
 101       progress = false;
 102       killed_all = false;
 103       mem_ctx = ralloc_context(0);
 104       this->lin_ctx = linear_alloc_parent(this->mem_ctx, 0);
 105       this->acp = new(mem_ctx) exec_list;
 106       this->kills = _mesa_hash_table_create(mem_ctx, _mesa_hash_pointer,
 107                                             _mesa_key_pointer_equal);
 108    }
 109    ~ir_constant_propagation_visitor()
 110    {
 111       ralloc_free(mem_ctx);
 112    }
 113
 114    virtual ir_visitor_status visit_enter(class ir_loop *);
 115    virtual ir_visitor_status visit_enter(class ir_function_signature *);
 116    virtual ir_visitor_status visit_enter(class ir_function *);
 117    virtual ir_visitor_status visit_leave(class ir_assignment *);
 118    virtual ir_visitor_status visit_enter(class ir_call *);
 119    virtual ir_visitor_status visit_enter(class ir_if *);
 120
 121    void add_constant(ir_assignment *ir);
 122    void constant_folding(ir_rvalue **rvalue);
 123    void constant_propagation(ir_rvalue **rvalue);
 124    void kill(ir_variable *ir, unsigned write_mask);
 125    void handle_if_block(exec_list *instructions);
 126    void handle_rvalue(ir_rvalue **rvalue);
 127
 128    /** List of acp_entry: The available constants to propagate */
 129    exec_list *acp;
 130
 131    /**
 132     * Hash table of kill_entry: The masks of variables whose values were
 133     * killed in this block.
 134     */
 135    hash_table *kills;
 136
 137    bool progress;
 138
 139    bool killed_all;
 140
 141    void *mem_ctx;
 142    void *lin_ctx;
 143 };
 144
 145
 146 void
 147 ir_constant_propagation_visitor::constant_folding(ir_rvalue **rvalue)
 148 {
 149    if (this->in_assignee || *rvalue == NULL)
 150       return;
 151
 152    if (ir_constant_fold(rvalue))
 153       this->progress = true;
 154
 155    ir_dereference_variable *var_ref = (*rvalue)->as_dereference_variable();
 156    if (var_ref && !var_ref->type->is_array()) {
 157       ir_constant *constant =
 158          var_ref->constant_expression_value(ralloc_parent(var_ref));
 159       if (constant) {
 160          *rvalue = constant;
 161          this->progress = true;
 162       }
 163    }
 164 }
 165
 166 void
 167 ir_constant_propagation_visitor::constant_propagation(ir_rvalue **rvalue) {
 168
 169    if (this->in_assignee || !*rvalue)
 170       return;
 171
 172    const glsl_type *type = (*rvalue)->type;
 173    if (!type->is_scalar() && !type->is_vector())
 174       return;
 175
 176    ir_swizzle *swiz = NULL;
 177    ir_dereference_variable *deref = (*rvalue)->as_dereference_variable();
 178    if (!deref) {
 179       swiz = (*rvalue)->as_swizzle();
 180       if (!swiz)
 181          return;
 182
 183       deref = swiz->val->as_dereference_variable();
 184       if (!deref)
 185          return;
 186    }
 187
 188    ir_constant_data data;
 189    memset(&data, 0, sizeof(data));
 190
 191    for (unsigned int i = 0; i < type->components(); i++) {
 192       int channel;
 193       acp_entry *found = NULL;
 194
 195       if (swiz) {
 196          switch (i) {
 197          case 0: channel = swiz->mask.x; break;
 198          case 1: channel = swiz->mask.y; break;
 199          case 2: channel = swiz->mask.z; break;
 200          case 3: channel = swiz->mask.w; break;
 201          default: assert(!"shouldn't be reached"); channel = 0; break;
 202          }
 203       } else {
 204          channel = i;
 205       }
 206
 207       foreach_in_list(acp_entry, entry, this->acp) {
 208          if (entry->var == deref->var && entry->write_mask & (1 << channel)) {
 209             found = entry;
 210             break;
 211          }
 212       }
 213
 214       if (!found)
 215          return;
 216
 217       int rhs_channel = 0;
 218       for (int j = 0; j < 4; j++) {
 219          if (j == channel)
 220             break;
 221          if (found->initial_values & (1 << j))
 222             rhs_channel++;
 223       }
 224
 225       switch (type->base_type) {
 226       case GLSL_TYPE_FLOAT:
 227          data.f[i] = found->constant->value.f[rhs_channel];
 228          break;
 229       case GLSL_TYPE_DOUBLE:
 230          data.d[i] = found->constant->value.d[rhs_channel];
 231          break;
 232       case GLSL_TYPE_INT:
 233          data.i[i] = found->constant->value.i[rhs_channel];
 234          break;
 235       case GLSL_TYPE_UINT:
 236          data.u[i] = found->constant->value.u[rhs_channel];
 237          break;
 238       case GLSL_TYPE_BOOL:
 239          data.b[i] = found->constant->value.b[rhs_channel];
 240          break;
 241       case GLSL_TYPE_UINT64:
 242          data.u64[i] = found->constant->value.u64[rhs_channel];
 243          break;
 244       case GLSL_TYPE_INT64:
 245          data.i64[i] = found->constant->value.i64[rhs_channel];
 246          break;
 247       default:
 248          assert(!"not reached");
 249          break;
 250       }
 251    }
 252
 253    *rvalue = new(ralloc_parent(deref)) ir_constant(type, &data);
 254    this->progress = true;
 255 }
 256
 257 void
 258 ir_constant_propagation_visitor::handle_rvalue(ir_rvalue **rvalue)
 259 {
 260    constant_propagation(rvalue);
 261    constant_folding(rvalue);
 262 }
 263
 264 ir_visitor_status
 265 ir_constant_propagation_visitor::visit_enter(ir_function_signature *ir)
 266 {
 267    /* Treat entry into a function signature as a completely separate
 268     * block.  Any instructions at global scope will be shuffled into
 269     * main() at link time, so they're irrelevant to us.
 270     */
 271    exec_list *orig_acp = this->acp;
 272    hash_table *orig_kills = this->kills;
 273    bool orig_killed_all = this->killed_all;
 274
 275    this->acp = new(mem_ctx) exec_list;
 276    this->kills = _mesa_hash_table_create(mem_ctx, _mesa_hash_pointer,
 277                                          _mesa_key_pointer_equal);
 278    this->killed_all = false;
 279
 280    visit_list_elements(this, &ir->body);
 281
 282    this->kills = orig_kills;
 283    this->acp = orig_acp;
 284    this->killed_all = orig_killed_all;
 285
 286    return visit_continue_with_parent;
 287 }
 288
 289 ir_visitor_status
 290 ir_constant_propagation_visitor::visit_leave(ir_assignment *ir)
 291 {
 292   constant_folding(&ir->rhs);
 293
 294    if (this->in_assignee)
 295       return visit_continue;
 296
 297    unsigned kill_mask = ir->write_mask;
 298    if (ir->lhs->as_dereference_array()) {
 299       /* The LHS of the assignment uses an array indexing operator (e.g. v[i]
 300        * = ...;).  Since we only try to constant propagate vectors and
 301        * scalars, this means that either (a) array indexing is being used to
 302        * select a vector component, or (b) the variable in question is neither
 303        * a scalar or a vector, so we don't care about it.  In the former case,
 304        * we want to kill the whole vector, since in general we can't predict
 305        * which vector component will be selected by array indexing.  In the
 306        * latter case, it doesn't matter what we do, so go ahead and kill the
 307        * whole variable anyway.
 308        *
 309        * Note that if the array index is constant (e.g. v[2] = ...;), we could
 310        * in principle be smarter, but we don't need to, because a future
 311        * optimization pass will convert it to a simple assignment with the
 312        * correct mask.
 313        */
 314       kill_mask = ~0;
 315    }
 316    kill(ir->lhs->variable_referenced(), kill_mask);
 317
 318    add_constant(ir);
 319
 320    return visit_continue;
 321 }
 322
 323 ir_visitor_status
 324 ir_constant_propagation_visitor::visit_enter(ir_function *ir)
 325 {
 326    (void) ir;
 327    return visit_continue;
 328 }
 329
 330 ir_visitor_status
 331 ir_constant_propagation_visitor::visit_enter(ir_call *ir)
 332 {
 333    /* Do constant propagation on call parameters, but skip any out params */
 334    foreach_two_lists(formal_node, &ir->callee->parameters,
 335                      actual_node, &ir->actual_parameters) {
 336       ir_variable *sig_param = (ir_variable *) formal_node;
 337       ir_rvalue *param = (ir_rvalue *) actual_node;
 338       if (sig_param->data.mode != ir_var_function_out
 339           && sig_param->data.mode != ir_var_function_inout) {
 340          ir_rvalue *new_param = param;
 341          handle_rvalue(&new_param);
 342          if (new_param != param)
 343             param->replace_with(new_param);
 344          else
 345             param->accept(this);
 346       }
 347    }
 348
 349    /* Since we're unlinked, we don't (necssarily) know the side effects of
 350     * this call.  So kill all copies.
 351     */
 352    acp->make_empty();
 353    this->killed_all = true;
 354
 355    return visit_continue_with_parent;
 356 }
 357
 358 void
 359 ir_constant_propagation_visitor::handle_if_block(exec_list *instructions)
 360 {
 361    exec_list *orig_acp = this->acp;
 362    hash_table *orig_kills = this->kills;
 363    bool orig_killed_all = this->killed_all;
 364
 365    this->acp = new(mem_ctx) exec_list;
 366    this->kills = _mesa_hash_table_create(mem_ctx, _mesa_hash_pointer,
 367                                          _mesa_key_pointer_equal);
 368    this->killed_all = false;
 369
 370    /* Populate the initial acp with a constant of the original */
 371    foreach_in_list(acp_entry, a, orig_acp) {
 372       this->acp->push_tail(new(this->lin_ctx) acp_entry(a));
 373    }
 374
 375    visit_list_elements(this, instructions);
 376
 377    if (this->killed_all) {
 378       orig_acp->make_empty();
 379    }
 380
 381    hash_table *new_kills = this->kills;
 382    this->kills = orig_kills;
 383    this->acp = orig_acp;
 384    this->killed_all = this->killed_all || orig_killed_all;
 385
 386    hash_entry *htk;
 387    hash_table_foreach(new_kills, htk) {
 388       kill_entry *k = (kill_entry *) htk->data;
 389       kill(k->var, k->write_mask);
 390    }
 391 }
 392
 393 ir_visitor_status
 394 ir_constant_propagation_visitor::visit_enter(ir_if *ir)
 395 {
 396    ir->condition->accept(this);
 397    handle_rvalue(&ir->condition);
 398
 399    handle_if_block(&ir->then_instructions);
 400    handle_if_block(&ir->else_instructions);
 401
 402    /* handle_if_block() already descended into the children. */
 403    return visit_continue_with_parent;
 404 }
 405
 406 ir_visitor_status
 407 ir_constant_propagation_visitor::visit_enter(ir_loop *ir)
 408 {
 409    exec_list *orig_acp = this->acp;
 410    hash_table *orig_kills = this->kills;
 411    bool orig_killed_all = this->killed_all;
 412
 413    /* FINISHME: For now, the initial acp for loops is totally empty.
 414     * We could go through once, then go through again with the acp
 415     * cloned minus the killed entries after the first run through.
 416     */
 417    this->acp = new(mem_ctx) exec_list;
 418    this->kills = _mesa_hash_table_create(mem_ctx, _mesa_hash_pointer,
 419                                          _mesa_key_pointer_equal);
 420    this->killed_all = false;
 421
 422    visit_list_elements(this, &ir->body_instructions);
 423
 424    if (this->killed_all) {
 425       orig_acp->make_empty();
 426    }
 427
 428    hash_table *new_kills = this->kills;
 429    this->kills = orig_kills;
 430    this->acp = orig_acp;
 431    this->killed_all = this->killed_all || orig_killed_all;
 432
 433    hash_entry *htk;
 434    hash_table_foreach(new_kills, htk) {
 435       kill_entry *k = (kill_entry *) htk->data;
 436       kill(k->var, k->write_mask);
 437    }
 438
 439    /* already descended into the children. */
 440    return visit_continue_with_parent;
 441 }
 442
 443 void
 444 ir_constant_propagation_visitor::kill(ir_variable *var, unsigned write_mask)
 445 {
 446    assert(var != NULL);
 447
 448    /* We don't track non-vectors. */
 449    if (!var->type->is_vector() && !var->type->is_scalar())
 450       return;
 451
 452    /* Remove any entries currently in the ACP for this kill. */
 453    foreach_in_list_safe(acp_entry, entry, this->acp) {
 454       if (entry->var == var) {
 455          entry->write_mask &= ~write_mask;
 456          if (entry->write_mask == 0)
 457             entry->remove();
 458       }
 459    }
 460
 461    /* Add this writemask of the variable to the hash table of killed
 462     * variables in this block.
 463     */
 464    hash_entry *kill_hash_entry = _mesa_hash_table_search(this->kills, var);
 465    if (kill_hash_entry) {
 466       kill_entry *entry = (kill_entry *) kill_hash_entry->data;
 467       entry->write_mask |= write_mask;
 468       return;
 469    }
 470    /* Not already in the hash table.  Make new entry. */
 471    _mesa_hash_table_insert(this->kills, var,
 472                            new(this->lin_ctx) kill_entry(var, write_mask));
 473 }
 474
 475 /**
 476  * Adds an entry to the available constant list if it's a plain assignment
 477  * of a variable to a variable.
 478  */
 479 void
 480 ir_constant_propagation_visitor::add_constant(ir_assignment *ir)
 481 {
 482    acp_entry *entry;
 483
 484    if (ir->condition)
 485       return;
 486
 487    if (!ir->write_mask)
 488       return;
 489
 490    ir_dereference_variable *deref = ir->lhs->as_dereference_variable();
 491    ir_constant *constant = ir->rhs->as_constant();
 492
 493    if (!deref || !constant)
 494       return;
 495
 496    /* Only do constant propagation on vectors.  Constant matrices,
 497     * arrays, or structures would require more work elsewhere.
 498     */
 499    if (!deref->var->type->is_vector() && !deref->var->type->is_scalar())
 500       return;
 501
 502    /* We can't do copy propagation on buffer variables, since the underlying
 503     * memory storage is shared across multiple threads we can't be sure that
 504     * the variable value isn't modified between this assignment and the next
 505     * instruction where its value is read.
 506     */
 507    if (deref->var->data.mode == ir_var_shader_storage ||
 508        deref->var->data.mode == ir_var_shader_shared)
 509       return;
 510
 511    entry = new(this->lin_ctx) acp_entry(deref->var, ir->write_mask, constant);
 512    this->acp->push_tail(entry);
 513 }
 514
 515 } /* unnamed namespace */
 516
 517 /**
 518  * Does a constant propagation pass on the code present in the instruction stream.
 519  */
 520 bool
 521 do_constant_propagation(exec_list *instructions)
 522 {
 523    ir_constant_propagation_visitor v;
 524
 525    visit_list_elements(&v, instructions);
 526
 527    return v.progress;
 528 }