reactos/dll/opengl/mesa/src/glsl/opt_copy_propagation_elements.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 opt_copy_propagation_elements.cpp
  26  *
  27  * Replaces usage of recently-copied components of variables with the
  28  * previous copy of the variable.
  29  *
  30  * This pass can be compared with opt_copy_propagation, which operands
  31  * on arbitrary whole-variable copies.  However, in order to handle
  32  * the copy propagation of swizzled variables or writemasked writes,
  33  * we want to track things on a channel-wise basis.  I found that
  34  * trying to mix the swizzled/writemasked support here with the
  35  * whole-variable stuff in opt_copy_propagation.cpp just made a mess,
  36  * so this is separate despite the ACP handling being somewhat
  37  * similar.
  38  *
  39  * This should reduce the number of MOV instructions in the generated
  40  * programs unless copy propagation is also done on the LIR, and may
  41  * help anyway by triggering other optimizations that live in the HIR.
  42  */
  43
  44 #include "ir.h"
  45 #include "ir_rvalue_visitor.h"
  46 #include "ir_basic_block.h"
  47 #include "ir_optimization.h"
  48 #include "glsl_types.h"
  49
  50 static bool debug = false;
  51
  52 class acp_entry : public exec_node
  53 {
  54 public:
  55    acp_entry(ir_variable *lhs, ir_variable *rhs, int write_mask, int swizzle[4])
  56    {
  57       this->lhs = lhs;
  58       this->rhs = rhs;
  59       this->write_mask = write_mask;
  60       memcpy(this->swizzle, swizzle, sizeof(this->swizzle));
  61    }
  62
  63    acp_entry(acp_entry *a)
  64    {
  65       this->lhs = a->lhs;
  66       this->rhs = a->rhs;
  67       this->write_mask = a->write_mask;
  68       memcpy(this->swizzle, a->swizzle, sizeof(this->swizzle));
  69    }
  70
  71    ir_variable *lhs;
  72    ir_variable *rhs;
  73    unsigned int write_mask;
  74    int swizzle[4];
  75 };
  76
  77
  78 class kill_entry : public exec_node
  79 {
  80 public:
  81    kill_entry(ir_variable *var, int write_mask)
  82    {
  83       this->var = var;
  84       this->write_mask = write_mask;
  85    }
  86
  87    ir_variable *var;
  88    unsigned int write_mask;
  89 };
  90
  91 class ir_copy_propagation_elements_visitor : public ir_rvalue_visitor {
  92 public:
  93    ir_copy_propagation_elements_visitor()
  94    {
  95       this->progress = false;
  96       this->mem_ctx = ralloc_context(NULL);
  97       this->shader_mem_ctx = NULL;
  98       this->acp = new(mem_ctx) exec_list;
  99       this->kills = new(mem_ctx) exec_list;
 100    }
 101    ~ir_copy_propagation_elements_visitor()
 102    {
 103       ralloc_free(mem_ctx);
 104    }
 105
 106    virtual ir_visitor_status visit_enter(class ir_loop *);
 107    virtual ir_visitor_status visit_enter(class ir_function_signature *);
 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    virtual ir_visitor_status visit_leave(class ir_swizzle *);
 112
 113    void handle_rvalue(ir_rvalue **rvalue);
 114
 115    void add_copy(ir_assignment *ir);
 116    void kill(kill_entry *k);
 117    void handle_if_block(exec_list *instructions);
 118
 119    /** List of acp_entry: The available copies to propagate */
 120    exec_list *acp;
 121    /**
 122     * List of kill_entry: The variables whose values were killed in this
 123     * block.
 124     */
 125    exec_list *kills;
 126
 127    bool progress;
 128
 129    bool killed_all;
 130
 131    /* Context for our local data structures. */
 132    void *mem_ctx;
 133    /* Context for allocating new shader nodes. */
 134    void *shader_mem_ctx;
 135 };
 136
 137 ir_visitor_status
 138 ir_copy_propagation_elements_visitor::visit_enter(ir_function_signature *ir)
 139 {
 140    /* Treat entry into a function signature as a completely separate
 141     * block.  Any instructions at global scope will be shuffled into
 142     * main() at link time, so they're irrelevant to us.
 143     */
 144    exec_list *orig_acp = this->acp;
 145    exec_list *orig_kills = this->kills;
 146    bool orig_killed_all = this->killed_all;
 147
 148    this->acp = new(mem_ctx) exec_list;
 149    this->kills = new(mem_ctx) exec_list;
 150    this->killed_all = false;
 151
 152    visit_list_elements(this, &ir->body);
 153
 154    this->kills = orig_kills;
 155    this->acp = orig_acp;
 156    this->killed_all = orig_killed_all;
 157
 158    return visit_continue_with_parent;
 159 }
 160
 161 ir_visitor_status
 162 ir_copy_propagation_elements_visitor::visit_leave(ir_assignment *ir)
 163 {
 164    ir_dereference_variable *lhs = ir->lhs->as_dereference_variable();
 165    ir_variable *var = ir->lhs->variable_referenced();
 166
 167    if (var->type->is_scalar() || var->type->is_vector()) {
 168       kill_entry *k;
 169
 170       if (lhs)
 171          k = new(mem_ctx) kill_entry(var, ir->write_mask);
 172       else
 173          k = new(mem_ctx) kill_entry(var, ~0);
 174
 175       kill(k);
 176    }
 177
 178    add_copy(ir);
 179
 180    return visit_continue;
 181 }
 182
 183 ir_visitor_status
 184 ir_copy_propagation_elements_visitor::visit_leave(ir_swizzle *ir)
 185 {
 186    /* Don't visit the values of swizzles since they are handled while
 187     * visiting the swizzle itself.
 188     */
 189    return visit_continue;
 190 }
 191
 192 /**
 193  * Replaces dereferences of ACP RHS variables with ACP LHS variables.
 194  *
 195  * This is where the actual copy propagation occurs.  Note that the
 196  * rewriting of ir_dereference means that the ir_dereference instance
 197  * must not be shared by multiple IR operations!
 198  */
 199 void
 200 ir_copy_propagation_elements_visitor::handle_rvalue(ir_rvalue **ir)
 201 {
 202    int swizzle_chan[4];
 203    ir_dereference_variable *deref_var;
 204    ir_variable *source[4] = {NULL, NULL, NULL, NULL};
 205    int source_chan[4];
 206    int chans;
 207
 208    if (!*ir)
 209       return;
 210
 211    ir_swizzle *swizzle = (*ir)->as_swizzle();
 212    if (swizzle) {
 213       deref_var = swizzle->val->as_dereference_variable();
 214       if (!deref_var)
 215          return;
 216
 217       swizzle_chan[0] = swizzle->mask.x;
 218       swizzle_chan[1] = swizzle->mask.y;
 219       swizzle_chan[2] = swizzle->mask.z;
 220       swizzle_chan[3] = swizzle->mask.w;
 221       chans = swizzle->type->vector_elements;
 222    } else {
 223       deref_var = (*ir)->as_dereference_variable();
 224       if (!deref_var)
 225          return;
 226
 227       swizzle_chan[0] = 0;
 228       swizzle_chan[1] = 1;
 229       swizzle_chan[2] = 2;
 230       swizzle_chan[3] = 3;
 231       chans = deref_var->type->vector_elements;
 232    }
 233
 234    if (this->in_assignee)
 235       return;
 236
 237    ir_variable *var = deref_var->var;
 238
 239    /* Try to find ACP entries covering swizzle_chan[], hoping they're
 240     * the same source variable.
 241     */
 242    foreach_iter(exec_list_iterator, iter, *this->acp) {
 243       acp_entry *entry = (acp_entry *)iter.get();
 244
 245       if (var == entry->lhs) {
 246          for (int c = 0; c < chans; c++) {
 247             if (entry->write_mask & (1 << swizzle_chan[c])) {
 248                source[c] = entry->rhs;
 249                source_chan[c] = entry->swizzle[swizzle_chan[c]];
 250             }
 251          }
 252       }
 253    }
 254
 255    /* Make sure all channels are copying from the same source variable. */
 256    if (!source[0])
 257       return;
 258    for (int c = 1; c < chans; c++) {
 259       if (source[c] != source[0])
 260          return;
 261    }
 262
 263    if (!shader_mem_ctx)
 264       shader_mem_ctx = ralloc_parent(deref_var);
 265
 266    if (debug) {
 267       printf("Copy propagation from:\n");
 268       (*ir)->print();
 269    }
 270
 271    deref_var = new(shader_mem_ctx) ir_dereference_variable(source[0]);
 272    *ir = new(shader_mem_ctx) ir_swizzle(deref_var,
 273                                         source_chan[0],
 274                                         source_chan[1],
 275                                         source_chan[2],
 276                                         source_chan[3],
 277                                         chans);
 278
 279    if (debug) {
 280       printf("to:\n");
 281       (*ir)->print();
 282       printf("\n");
 283    }
 284 }
 285
 286
 287 ir_visitor_status
 288 ir_copy_propagation_elements_visitor::visit_enter(ir_call *ir)
 289 {
 290    /* Do copy propagation on call parameters, but skip any out params */
 291    exec_list_iterator sig_param_iter = ir->get_callee()->parameters.iterator();
 292    foreach_iter(exec_list_iterator, iter, ir->actual_parameters) {
 293       ir_variable *sig_param = (ir_variable *)sig_param_iter.get();
 294       ir_instruction *ir = (ir_instruction *)iter.get();
 295       if (sig_param->mode != ir_var_out && sig_param->mode != ir_var_inout) {
 296          ir->accept(this);
 297       }
 298       sig_param_iter.next();
 299    }
 300
 301    /* Since we're unlinked, we don't (necessarily) know the side effects of
 302     * this call.  So kill all copies.
 303     */
 304    acp->make_empty();
 305    this->killed_all = true;
 306
 307    return visit_continue_with_parent;
 308 }
 309
 310 void
 311 ir_copy_propagation_elements_visitor::handle_if_block(exec_list *instructions)
 312 {
 313    exec_list *orig_acp = this->acp;
 314    exec_list *orig_kills = this->kills;
 315    bool orig_killed_all = this->killed_all;
 316
 317    this->acp = new(mem_ctx) exec_list;
 318    this->kills = new(mem_ctx) exec_list;
 319    this->killed_all = false;
 320
 321    /* Populate the initial acp with a copy of the original */
 322    foreach_iter(exec_list_iterator, iter, *orig_acp) {
 323       acp_entry *a = (acp_entry *)iter.get();
 324       this->acp->push_tail(new(this->mem_ctx) acp_entry(a));
 325    }
 326
 327    visit_list_elements(this, instructions);
 328
 329    if (this->killed_all) {
 330       orig_acp->make_empty();
 331    }
 332
 333    exec_list *new_kills = this->kills;
 334    this->kills = orig_kills;
 335    this->acp = orig_acp;
 336    this->killed_all = this->killed_all || orig_killed_all;
 337
 338    /* Move the new kills into the parent block's list, removing them
 339     * from the parent's ACP list in the process.
 340     */
 341    foreach_list_safe(node, new_kills) {
 342       kill_entry *k = (kill_entry *)node;
 343       kill(k);
 344    }
 345 }
 346
 347 ir_visitor_status
 348 ir_copy_propagation_elements_visitor::visit_enter(ir_if *ir)
 349 {
 350    ir->condition->accept(this);
 351
 352    handle_if_block(&ir->then_instructions);
 353    handle_if_block(&ir->else_instructions);
 354
 355    /* handle_if_block() already descended into the children. */
 356    return visit_continue_with_parent;
 357 }
 358
 359 ir_visitor_status
 360 ir_copy_propagation_elements_visitor::visit_enter(ir_loop *ir)
 361 {
 362    exec_list *orig_acp = this->acp;
 363    exec_list *orig_kills = this->kills;
 364    bool orig_killed_all = this->killed_all;
 365
 366    /* FINISHME: For now, the initial acp for loops is totally empty.
 367     * We could go through once, then go through again with the acp
 368     * cloned minus the killed entries after the first run through.
 369     */
 370    this->acp = new(mem_ctx) exec_list;
 371    this->kills = new(mem_ctx) exec_list;
 372    this->killed_all = false;
 373
 374    visit_list_elements(this, &ir->body_instructions);
 375
 376    if (this->killed_all) {
 377       orig_acp->make_empty();
 378    }
 379
 380    exec_list *new_kills = this->kills;
 381    this->kills = orig_kills;
 382    this->acp = orig_acp;
 383    this->killed_all = this->killed_all || orig_killed_all;
 384
 385    foreach_list_safe(node, new_kills) {
 386       kill_entry *k = (kill_entry *)node;
 387       kill(k);
 388    }
 389
 390    /* already descended into the children. */
 391    return visit_continue_with_parent;
 392 }
 393
 394 /* Remove any entries currently in the ACP for this kill. */
 395 void
 396 ir_copy_propagation_elements_visitor::kill(kill_entry *k)
 397 {
 398    foreach_list_safe(node, acp) {
 399       acp_entry *entry = (acp_entry *)node;
 400
 401       if (entry->lhs == k->var) {
 402          entry->write_mask = entry->write_mask & ~k->write_mask;
 403          if (entry->write_mask == 0) {
 404             entry->remove();
 405             continue;
 406          }
 407       }
 408       if (entry->rhs == k->var) {
 409          entry->remove();
 410       }
 411    }
 412
 413    /* If we were on a list, remove ourselves before inserting */
 414    if (k->next)
 415       k->remove();
 416
 417    this->kills->push_tail(k);
 418 }
 419
 420 /**
 421  * Adds directly-copied channels between vector variables to the available
 422  * copy propagation list.
 423  */
 424 void
 425 ir_copy_propagation_elements_visitor::add_copy(ir_assignment *ir)
 426 {
 427    acp_entry *entry;
 428    int orig_swizzle[4] = {0, 1, 2, 3};
 429    int swizzle[4];
 430
 431    if (ir->condition)
 432       return;
 433
 434    ir_dereference_variable *lhs = ir->lhs->as_dereference_variable();
 435    if (!lhs || !(lhs->type->is_scalar() || lhs->type->is_vector()))
 436       return;
 437
 438    ir_dereference_variable *rhs = ir->rhs->as_dereference_variable();
 439    if (!rhs) {
 440       ir_swizzle *swiz = ir->rhs->as_swizzle();
 441       if (!swiz)
 442          return;
 443
 444       rhs = swiz->val->as_dereference_variable();
 445       if (!rhs)
 446          return;
 447
 448       orig_swizzle[0] = swiz->mask.x;
 449       orig_swizzle[1] = swiz->mask.y;
 450       orig_swizzle[2] = swiz->mask.z;
 451       orig_swizzle[3] = swiz->mask.w;
 452    }
 453
 454    /* Move the swizzle channels out to the positions they match in the
 455     * destination.  We don't want to have to rewrite the swizzle[]
 456     * array every time we clear a bit of the write_mask.
 457     */
 458    int j = 0;
 459    for (int i = 0; i < 4; i++) {
 460       if (ir->write_mask & (1 << i))
 461          swizzle[i] = orig_swizzle[j++];
 462    }
 463
 464    int write_mask = ir->write_mask;
 465    if (lhs->var == rhs->var) {
 466       /* If this is a copy from the variable to itself, then we need
 467        * to be sure not to include the updated channels from this
 468        * instruction in the set of new source channels to be
 469        * copy-propagated from.
 470        */
 471       for (int i = 0; i < 4; i++) {
 472          if (ir->write_mask & (1 << orig_swizzle[i]))
 473             write_mask &= ~(1 << i);
 474       }
 475    }
 476
 477    entry = new(this->mem_ctx) acp_entry(lhs->var, rhs->var, write_mask,
 478                                         swizzle);
 479    this->acp->push_tail(entry);
 480 }
 481
 482 bool
 483 do_copy_propagation_elements(exec_list *instructions)
 484 {
 485    ir_copy_propagation_elements_visitor v;
 486
 487    visit_list_elements(&v, instructions);
 488
 489    return v.progress;
 490 }