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