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