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 "glsl_types.h"
  43
  44 namespace {
  45
  46 class acp_entry : public exec_node
  47 {
  48 public:
  49    acp_entry(ir_variable *var, unsigned write_mask, ir_constant *constant)
  50    {
  51       assert(var);
  52       assert(constant);
  53       this->var = var;
  54       this->write_mask = write_mask;
  55       this->constant = constant;
  56       this->initial_values = write_mask;
  57    }
  58
  59    acp_entry(const acp_entry *src)
  60    {
  61       this->var = src->var;
  62       this->write_mask = src->write_mask;
  63       this->constant = src->constant;
  64       this->initial_values = src->initial_values;
  65    }
  66
  67    ir_variable *var;
  68    ir_constant *constant;
  69    unsigned write_mask;
  70
  71    /** Mask of values initially available in the constant. */
  72    unsigned initial_values;
  73 };
  74
  75
  76 class kill_entry : public exec_node
  77 {
  78 public:
  79    kill_entry(ir_variable *var, unsigned write_mask)
  80    {
  81       assert(var);
  82       this->var = var;
  83       this->write_mask = write_mask;
  84    }
  85
  86    ir_variable *var;
  87    unsigned write_mask;
  88 };
  89
  90 class ir_constant_propagation_visitor : public ir_rvalue_visitor {
  91 public:
  92    ir_constant_propagation_visitor()
  93    {
  94       progress = false;
  95       killed_all = false;
  96       mem_ctx = ralloc_context(0);
  97       this->acp = new(mem_ctx) exec_list;
  98       this->kills = new(mem_ctx) exec_list;
  99    }
 100    ~ir_constant_propagation_visitor()
 101    {
 102       ralloc_free(mem_ctx);
 103    }
 104
 105    virtual ir_visitor_status visit_enter(class ir_loop *);
 106    virtual ir_visitor_status visit_enter(class ir_function_signature *);
 107    virtual ir_visitor_status visit_enter(class ir_function *);
 108    virtual ir_visitor_status visit_leave(class ir_assignment *);
 109    virtual ir_visitor_status visit_enter(class ir_call *);
 110    virtual ir_visitor_status visit_enter(class ir_if *);
 111
 112    void add_constant(ir_assignment *ir);
 113    void constant_folding(ir_rvalue **rvalue);
 114    void constant_propagation(ir_rvalue **rvalue);
 115    void kill(ir_variable *ir, unsigned write_mask);
 116    void handle_if_block(exec_list *instructions);
 117    void handle_rvalue(ir_rvalue **rvalue);
 118
 119    /** List of acp_entry: The available constants to propagate */
 120    exec_list *acp;
 121
 122    /**
 123     * List of kill_entry: The masks of variables whose values were
 124     * killed in this block.
 125     */
 126    exec_list *kills;
 127
 128    bool progress;
 129
 130    bool killed_all;
 131
 132    void *mem_ctx;
 133 };
 134
 135
 136 void
 137 ir_constant_propagation_visitor::constant_folding(ir_rvalue **rvalue) {
 138
 139    if (*rvalue == NULL || (*rvalue)->ir_type == ir_type_constant)
 140       return;
 141
 142    /* Note that we visit rvalues one leaving.  So if an expression has a
 143     * non-constant operand, no need to go looking down it to find if it's
 144     * constant.  This cuts the time of this pass down drastically.
 145     */
 146    ir_expression *expr = (*rvalue)->as_expression();
 147    if (expr) {
 148       for (unsigned int i = 0; i < expr->get_num_operands(); i++) {
 149          if (!expr->operands[i]->as_constant())
 150             return;
 151       }
 152    }
 153
 154    /* Ditto for swizzles. */
 155    ir_swizzle *swiz = (*rvalue)->as_swizzle();
 156    if (swiz && !swiz->val->as_constant())
 157       return;
 158
 159    ir_constant *constant = (*rvalue)->constant_expression_value();
 160    if (constant) {
 161       *rvalue = constant;
 162       this->progress = true;
 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       default:
 242          assert(!"not reached");
 243          break;
 244       }
 245    }
 246
 247    *rvalue = new(ralloc_parent(deref)) ir_constant(type, &data);
 248    this->progress = true;
 249 }
 250
 251 void
 252 ir_constant_propagation_visitor::handle_rvalue(ir_rvalue **rvalue)
 253 {
 254    constant_propagation(rvalue);
 255    constant_folding(rvalue);
 256 }
 257
 258 ir_visitor_status
 259 ir_constant_propagation_visitor::visit_enter(ir_function_signature *ir)
 260 {
 261    /* Treat entry into a function signature as a completely separate
 262     * block.  Any instructions at global scope will be shuffled into
 263     * main() at link time, so they're irrelevant to us.
 264     */
 265    exec_list *orig_acp = this->acp;
 266    exec_list *orig_kills = this->kills;
 267    bool orig_killed_all = this->killed_all;
 268
 269    this->acp = new(mem_ctx) exec_list;
 270    this->kills = new(mem_ctx) exec_list;
 271    this->killed_all = false;
 272
 273    visit_list_elements(this, &ir->body);
 274
 275    this->kills = orig_kills;
 276    this->acp = orig_acp;
 277    this->killed_all = orig_killed_all;
 278
 279    return visit_continue_with_parent;
 280 }
 281
 282 ir_visitor_status
 283 ir_constant_propagation_visitor::visit_leave(ir_assignment *ir)
 284 {
 285   constant_folding(&ir->rhs);
 286
 287    if (this->in_assignee)
 288       return visit_continue;
 289
 290    unsigned kill_mask = ir->write_mask;
 291    if (ir->lhs->as_dereference_array()) {
 292       /* The LHS of the assignment uses an array indexing operator (e.g. v[i]
 293        * = ...;).  Since we only try to constant propagate vectors and
 294        * scalars, this means that either (a) array indexing is being used to
 295        * select a vector component, or (b) the variable in question is neither
 296        * a scalar or a vector, so we don't care about it.  In the former case,
 297        * we want to kill the whole vector, since in general we can't predict
 298        * which vector component will be selected by array indexing.  In the
 299        * latter case, it doesn't matter what we do, so go ahead and kill the
 300        * whole variable anyway.
 301        *
 302        * Note that if the array index is constant (e.g. v[2] = ...;), we could
 303        * in principle be smarter, but we don't need to, because a future
 304        * optimization pass will convert it to a simple assignment with the
 305        * correct mask.
 306        */
 307       kill_mask = ~0;
 308    }
 309    kill(ir->lhs->variable_referenced(), kill_mask);
 310
 311    add_constant(ir);
 312
 313    return visit_continue;
 314 }
 315
 316 ir_visitor_status
 317 ir_constant_propagation_visitor::visit_enter(ir_function *ir)
 318 {
 319    (void) ir;
 320    return visit_continue;
 321 }
 322
 323 ir_visitor_status
 324 ir_constant_propagation_visitor::visit_enter(ir_call *ir)
 325 {
 326    /* Do constant propagation on call parameters, but skip any out params */
 327    foreach_two_lists(formal_node, &ir->callee->parameters,
 328                      actual_node, &ir->actual_parameters) {
 329       ir_variable *sig_param = (ir_variable *) formal_node;
 330       ir_rvalue *param = (ir_rvalue *) actual_node;
 331       if (sig_param->data.mode != ir_var_function_out
 332           && sig_param->data.mode != ir_var_function_inout) {
 333          ir_rvalue *new_param = param;
 334          handle_rvalue(&new_param);
 335          if (new_param != param)
 336             param->replace_with(new_param);
 337          else
 338             param->accept(this);
 339       }
 340    }
 341
 342    /* Since we're unlinked, we don't (necssarily) know the side effects of
 343     * this call.  So kill all copies.
 344     */
 345    acp->make_empty();
 346    this->killed_all = true;
 347
 348    return visit_continue_with_parent;
 349 }
 350
 351 void
 352 ir_constant_propagation_visitor::handle_if_block(exec_list *instructions)
 353 {
 354    exec_list *orig_acp = this->acp;
 355    exec_list *orig_kills = this->kills;
 356    bool orig_killed_all = this->killed_all;
 357
 358    this->acp = new(mem_ctx) exec_list;
 359    this->kills = new(mem_ctx) exec_list;
 360    this->killed_all = false;
 361
 362    /* Populate the initial acp with a constant of the original */
 363    foreach_in_list(acp_entry, a, orig_acp) {
 364       this->acp->push_tail(new(this->mem_ctx) acp_entry(a));
 365    }
 366
 367    visit_list_elements(this, instructions);
 368
 369    if (this->killed_all) {
 370       orig_acp->make_empty();
 371    }
 372
 373    exec_list *new_kills = this->kills;
 374    this->kills = orig_kills;
 375    this->acp = orig_acp;
 376    this->killed_all = this->killed_all || orig_killed_all;
 377
 378    foreach_in_list(kill_entry, k, new_kills) {
 379       kill(k->var, k->write_mask);
 380    }
 381 }
 382
 383 ir_visitor_status
 384 ir_constant_propagation_visitor::visit_enter(ir_if *ir)
 385 {
 386    ir->condition->accept(this);
 387    handle_rvalue(&ir->condition);
 388
 389    handle_if_block(&ir->then_instructions);
 390    handle_if_block(&ir->else_instructions);
 391
 392    /* handle_if_block() already descended into the children. */
 393    return visit_continue_with_parent;
 394 }
 395
 396 ir_visitor_status
 397 ir_constant_propagation_visitor::visit_enter(ir_loop *ir)
 398 {
 399    exec_list *orig_acp = this->acp;
 400    exec_list *orig_kills = this->kills;
 401    bool orig_killed_all = this->killed_all;
 402
 403    /* FINISHME: For now, the initial acp for loops is totally empty.
 404     * We could go through once, then go through again with the acp
 405     * cloned minus the killed entries after the first run through.
 406     */
 407    this->acp = new(mem_ctx) exec_list;
 408    this->kills = new(mem_ctx) exec_list;
 409    this->killed_all = false;
 410
 411    visit_list_elements(this, &ir->body_instructions);
 412
 413    if (this->killed_all) {
 414       orig_acp->make_empty();
 415    }
 416
 417    exec_list *new_kills = this->kills;
 418    this->kills = orig_kills;
 419    this->acp = orig_acp;
 420    this->killed_all = this->killed_all || orig_killed_all;
 421
 422    foreach_in_list(kill_entry, k, new_kills) {
 423       kill(k->var, k->write_mask);
 424    }
 425
 426    /* already descended into the children. */
 427    return visit_continue_with_parent;
 428 }
 429
 430 void
 431 ir_constant_propagation_visitor::kill(ir_variable *var, unsigned write_mask)
 432 {
 433    assert(var != NULL);
 434
 435    /* We don't track non-vectors. */
 436    if (!var->type->is_vector() && !var->type->is_scalar())
 437       return;
 438
 439    /* Remove any entries currently in the ACP for this kill. */
 440    foreach_in_list_safe(acp_entry, entry, this->acp) {
 441       if (entry->var == var) {
 442          entry->write_mask &= ~write_mask;
 443          if (entry->write_mask == 0)
 444             entry->remove();
 445       }
 446    }
 447
 448    /* Add this writemask of the variable to the list of killed
 449     * variables in this block.
 450     */
 451    foreach_in_list(kill_entry, entry, this->kills) {
 452       if (entry->var == var) {
 453          entry->write_mask |= write_mask;
 454          return;
 455       }
 456    }
 457    /* Not already in the list.  Make new entry. */
 458    this->kills->push_tail(new(this->mem_ctx) kill_entry(var, write_mask));
 459 }
 460
 461 /**
 462  * Adds an entry to the available constant list if it's a plain assignment
 463  * of a variable to a variable.
 464  */
 465 void
 466 ir_constant_propagation_visitor::add_constant(ir_assignment *ir)
 467 {
 468    acp_entry *entry;
 469
 470    if (ir->condition)
 471       return;
 472
 473    if (!ir->write_mask)
 474       return;
 475
 476    ir_dereference_variable *deref = ir->lhs->as_dereference_variable();
 477    ir_constant *constant = ir->rhs->as_constant();
 478
 479    if (!deref || !constant)
 480       return;
 481
 482    /* Only do constant propagation on vectors.  Constant matrices,
 483     * arrays, or structures would require more work elsewhere.
 484     */
 485    if (!deref->var->type->is_vector() && !deref->var->type->is_scalar())
 486       return;
 487
 488    /* We can't do copy propagation on buffer variables, since the underlying
 489     * memory storage is shared across multiple threads we can't be sure that
 490     * the variable value isn't modified between this assignment and the next
 491     * instruction where its value is read.
 492     */
 493    if (deref->var->data.mode == ir_var_shader_storage)
 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 }