src/compiler/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 "compiler/glsl_types.h"
  49
  50 static bool debug = false;
  51
  52 namespace {
  53
  54 class acp_entry : public exec_node
  55 {
  56 public:
  57    acp_entry(ir_variable *lhs, ir_variable *rhs, int write_mask, int swizzle[4])
  58    {
  59       this->lhs = lhs;
  60       this->rhs = rhs;
  61       this->write_mask = write_mask;
  62       memcpy(this->swizzle, swizzle, sizeof(this->swizzle));
  63    }
  64
  65    acp_entry(acp_entry *a)
  66    {
  67       this->lhs = a->lhs;
  68       this->rhs = a->rhs;
  69       this->write_mask = a->write_mask;
  70       memcpy(this->swizzle, a->swizzle, sizeof(this->swizzle));
  71    }
  72
  73    ir_variable *lhs;
  74    ir_variable *rhs;
  75    unsigned int write_mask;
  76    int swizzle[4];
  77 };
  78
  79
  80 class kill_entry : public exec_node
  81 {
  82 public:
  83    kill_entry(ir_variable *var, int write_mask)
  84    {
  85       this->var = var;
  86       this->write_mask = write_mask;
  87    }
  88
  89    ir_variable *var;
  90    unsigned int write_mask;
  91 };
  92
  93 class ir_copy_propagation_elements_visitor : public ir_rvalue_visitor {
  94 public:
  95    ir_copy_propagation_elements_visitor()
  96    {
  97       this->progress = false;
  98       this->killed_all = false;
  99       this->mem_ctx = ralloc_context(NULL);
 100       this->shader_mem_ctx = NULL;
 101       this->acp = new(mem_ctx) exec_list;
 102       this->kills = new(mem_ctx) exec_list;
 103    }
 104    ~ir_copy_propagation_elements_visitor()
 105    {
 106       ralloc_free(mem_ctx);
 107    }
 108
 109    void handle_loop(ir_loop *, bool keep_acp);
 110    virtual ir_visitor_status visit_enter(class ir_loop *);
 111    virtual ir_visitor_status visit_enter(class ir_function_signature *);
 112    virtual ir_visitor_status visit_leave(class ir_assignment *);
 113    virtual ir_visitor_status visit_enter(class ir_call *);
 114    virtual ir_visitor_status visit_enter(class ir_if *);
 115    virtual ir_visitor_status visit_leave(class ir_swizzle *);
 116
 117    void handle_rvalue(ir_rvalue **rvalue);
 118
 119    void add_copy(ir_assignment *ir);
 120    void kill(kill_entry *k);
 121    void handle_if_block(exec_list *instructions);
 122
 123    /** List of acp_entry: The available copies to propagate */
 124    exec_list *acp;
 125    /**
 126     * List of kill_entry: The variables whose values were killed in this
 127     * block.
 128     */
 129    exec_list *kills;
 130
 131    bool progress;
 132
 133    bool killed_all;
 134
 135    /* Context for our local data structures. */
 136    void *mem_ctx;
 137    /* Context for allocating new shader nodes. */
 138    void *shader_mem_ctx;
 139 };
 140
 141 } /* unnamed namespace */
 142
 143 ir_visitor_status
 144 ir_copy_propagation_elements_visitor::visit_enter(ir_function_signature *ir)
 145 {
 146    /* Treat entry into a function signature as a completely separate
 147     * block.  Any instructions at global scope will be shuffled into
 148     * main() at link time, so they're irrelevant to us.
 149     */
 150    exec_list *orig_acp = this->acp;
 151    exec_list *orig_kills = this->kills;
 152    bool orig_killed_all = this->killed_all;
 153
 154    this->acp = new(mem_ctx) exec_list;
 155    this->kills = new(mem_ctx) exec_list;
 156    this->killed_all = false;
 157
 158    visit_list_elements(this, &ir->body);
 159
 160    ralloc_free(this->acp);
 161    ralloc_free(this->kills);
 162
 163    this->kills = orig_kills;
 164    this->acp = orig_acp;
 165    this->killed_all = orig_killed_all;
 166
 167    return visit_continue_with_parent;
 168 }
 169
 170 ir_visitor_status
 171 ir_copy_propagation_elements_visitor::visit_leave(ir_assignment *ir)
 172 {
 173    ir_dereference_variable *lhs = ir->lhs->as_dereference_variable();
 174    ir_variable *var = ir->lhs->variable_referenced();
 175
 176    if (var->type->is_scalar() || var->type->is_vector()) {
 177       kill_entry *k;
 178
 179       if (lhs)
 180          k = new(this->kills) kill_entry(var, ir->write_mask);
 181       else
 182          k = new(this->kills) kill_entry(var, ~0);
 183
 184       kill(k);
 185    }
 186
 187    add_copy(ir);
 188
 189    return visit_continue;
 190 }
 191
 192 ir_visitor_status
 193 ir_copy_propagation_elements_visitor::visit_leave(ir_swizzle *)
 194 {
 195    /* Don't visit the values of swizzles since they are handled while
 196     * visiting the swizzle itself.
 197     */
 198    return visit_continue;
 199 }
 200
 201 /**
 202  * Replaces dereferences of ACP RHS variables with ACP LHS variables.
 203  *
 204  * This is where the actual copy propagation occurs.  Note that the
 205  * rewriting of ir_dereference means that the ir_dereference instance
 206  * must not be shared by multiple IR operations!
 207  */
 208 void
 209 ir_copy_propagation_elements_visitor::handle_rvalue(ir_rvalue **ir)
 210 {
 211    int swizzle_chan[4];
 212    ir_dereference_variable *deref_var;
 213    ir_variable *source[4] = {NULL, NULL, NULL, NULL};
 214    int source_chan[4] = {0, 0, 0, 0};
 215    int chans;
 216    bool noop_swizzle = true;
 217
 218    if (!*ir)
 219       return;
 220
 221    ir_swizzle *swizzle = (*ir)->as_swizzle();
 222    if (swizzle) {
 223       deref_var = swizzle->val->as_dereference_variable();
 224       if (!deref_var)
 225          return;
 226
 227       swizzle_chan[0] = swizzle->mask.x;
 228       swizzle_chan[1] = swizzle->mask.y;
 229       swizzle_chan[2] = swizzle->mask.z;
 230       swizzle_chan[3] = swizzle->mask.w;
 231       chans = swizzle->type->vector_elements;
 232    } else {
 233       deref_var = (*ir)->as_dereference_variable();
 234       if (!deref_var)
 235          return;
 236
 237       swizzle_chan[0] = 0;
 238       swizzle_chan[1] = 1;
 239       swizzle_chan[2] = 2;
 240       swizzle_chan[3] = 3;
 241       chans = deref_var->type->vector_elements;
 242    }
 243
 244    if (this->in_assignee)
 245       return;
 246
 247    ir_variable *var = deref_var->var;
 248
 249    /* Try to find ACP entries covering swizzle_chan[], hoping they're
 250     * the same source variable.
 251     */
 252    foreach_in_list(acp_entry, entry, this->acp) {
 253       if (var == entry->lhs) {
 254          for (int c = 0; c < chans; c++) {
 255             if (entry->write_mask & (1 << swizzle_chan[c])) {
 256                source[c] = entry->rhs;
 257                source_chan[c] = entry->swizzle[swizzle_chan[c]];
 258
 259                if (source_chan[c] != swizzle_chan[c])
 260                   noop_swizzle = false;
 261             }
 262          }
 263       }
 264    }
 265
 266    /* Make sure all channels are copying from the same source variable. */
 267    if (!source[0])
 268       return;
 269    for (int c = 1; c < chans; c++) {
 270       if (source[c] != source[0])
 271          return;
 272    }
 273
 274    if (!shader_mem_ctx)
 275       shader_mem_ctx = ralloc_parent(deref_var);
 276
 277    /* Don't pointlessly replace the rvalue with itself (or a noop swizzle
 278     * of itself, which would just be deleted by opt_noop_swizzle).
 279     */
 280    if (source[0] == var && noop_swizzle)
 281       return;
 282
 283    if (debug) {
 284       printf("Copy propagation from:\n");
 285       (*ir)->print();
 286    }
 287
 288    deref_var = new(shader_mem_ctx) ir_dereference_variable(source[0]);
 289    *ir = new(shader_mem_ctx) ir_swizzle(deref_var,
 290                                         source_chan[0],
 291                                         source_chan[1],
 292                                         source_chan[2],
 293                                         source_chan[3],
 294                                         chans);
 295    progress = true;
 296
 297    if (debug) {
 298       printf("to:\n");
 299       (*ir)->print();
 300       printf("\n");
 301    }
 302 }
 303
 304
 305 ir_visitor_status
 306 ir_copy_propagation_elements_visitor::visit_enter(ir_call *ir)
 307 {
 308    /* Do copy propagation on call parameters, but skip any out params */
 309    foreach_two_lists(formal_node, &ir->callee->parameters,
 310                      actual_node, &ir->actual_parameters) {
 311       ir_variable *sig_param = (ir_variable *) formal_node;
 312       ir_rvalue *ir = (ir_rvalue *) actual_node;
 313       if (sig_param->data.mode != ir_var_function_out
 314           && sig_param->data.mode != ir_var_function_inout) {
 315          ir->accept(this);
 316       }
 317    }
 318
 319    /* Since we're unlinked, we don't (necessarily) know the side effects of
 320     * this call.  So kill all copies.
 321     */
 322    acp->make_empty();
 323    this->killed_all = true;
 324
 325    return visit_continue_with_parent;
 326 }
 327
 328 void
 329 ir_copy_propagation_elements_visitor::handle_if_block(exec_list *instructions)
 330 {
 331    exec_list *orig_acp = this->acp;
 332    exec_list *orig_kills = this->kills;
 333    bool orig_killed_all = this->killed_all;
 334
 335    this->acp = new(mem_ctx) exec_list;
 336    this->kills = new(mem_ctx) exec_list;
 337    this->killed_all = false;
 338
 339    /* Populate the initial acp with a copy of the original */
 340    foreach_in_list(acp_entry, a, orig_acp) {
 341       this->acp->push_tail(new(this->acp) acp_entry(a));
 342    }
 343
 344    visit_list_elements(this, instructions);
 345
 346    if (this->killed_all) {
 347       orig_acp->make_empty();
 348    }
 349
 350    exec_list *new_kills = this->kills;
 351    this->kills = orig_kills;
 352    ralloc_free(this->acp);
 353    this->acp = orig_acp;
 354    this->killed_all = this->killed_all || orig_killed_all;
 355
 356    /* Move the new kills into the parent block's list, removing them
 357     * from the parent's ACP list in the process.
 358     */
 359    foreach_in_list_safe(kill_entry, k, new_kills) {
 360       kill(k);
 361    }
 362
 363    ralloc_free(new_kills);
 364 }
 365
 366 ir_visitor_status
 367 ir_copy_propagation_elements_visitor::visit_enter(ir_if *ir)
 368 {
 369    ir->condition->accept(this);
 370
 371    handle_if_block(&ir->then_instructions);
 372    handle_if_block(&ir->else_instructions);
 373
 374    /* handle_if_block() already descended into the children. */
 375    return visit_continue_with_parent;
 376 }
 377
 378 void
 379 ir_copy_propagation_elements_visitor::handle_loop(ir_loop *ir, bool keep_acp)
 380 {
 381    exec_list *orig_acp = this->acp;
 382    exec_list *orig_kills = this->kills;
 383    bool orig_killed_all = this->killed_all;
 384
 385    /* FINISHME: For now, the initial acp for loops is totally empty.
 386     * We could go through once, then go through again with the acp
 387     * cloned minus the killed entries after the first run through.
 388     */
 389    this->acp = new(mem_ctx) exec_list;
 390    this->kills = new(mem_ctx) exec_list;
 391    this->killed_all = false;
 392
 393    if (keep_acp) {
 394       /* Populate the initial acp with a copy of the original */
 395       foreach_in_list(acp_entry, a, orig_acp) {
 396          this->acp->push_tail(new(this->acp) acp_entry(a));
 397       }
 398    }
 399
 400    visit_list_elements(this, &ir->body_instructions);
 401
 402    if (this->killed_all) {
 403       orig_acp->make_empty();
 404    }
 405
 406    exec_list *new_kills = this->kills;
 407    this->kills = orig_kills;
 408    ralloc_free(this->acp);
 409    this->acp = orig_acp;
 410    this->killed_all = this->killed_all || orig_killed_all;
 411
 412    foreach_in_list_safe(kill_entry, k, new_kills) {
 413       kill(k);
 414    }
 415
 416    ralloc_free(new_kills);
 417 }
 418
 419 ir_visitor_status
 420 ir_copy_propagation_elements_visitor::visit_enter(ir_loop *ir)
 421 {
 422    handle_loop(ir, false);
 423    handle_loop(ir, true);
 424
 425    /* already descended into the children. */
 426    return visit_continue_with_parent;
 427 }
 428
 429 /* Remove any entries currently in the ACP for this kill. */
 430 void
 431 ir_copy_propagation_elements_visitor::kill(kill_entry *k)
 432 {
 433    foreach_in_list_safe(acp_entry, entry, acp) {
 434       if (entry->lhs == k->var) {
 435          entry->write_mask = entry->write_mask & ~k->write_mask;
 436          if (entry->write_mask == 0) {
 437             entry->remove();
 438             continue;
 439          }
 440       }
 441       if (entry->rhs == k->var) {
 442          entry->remove();
 443       }
 444    }
 445
 446    /* If we were on a list, remove ourselves before inserting */
 447    if (k->next)
 448       k->remove();
 449
 450    ralloc_steal(this->kills, k);
 451    this->kills->push_tail(k);
 452 }
 453
 454 /**
 455  * Adds directly-copied channels between vector variables to the available
 456  * copy propagation list.
 457  */
 458 void
 459 ir_copy_propagation_elements_visitor::add_copy(ir_assignment *ir)
 460 {
 461    acp_entry *entry;
 462    int orig_swizzle[4] = {0, 1, 2, 3};
 463    int swizzle[4];
 464
 465    if (ir->condition)
 466       return;
 467
 468    ir_dereference_variable *lhs = ir->lhs->as_dereference_variable();
 469    if (!lhs || !(lhs->type->is_scalar() || lhs->type->is_vector()))
 470       return;
 471
 472    ir_dereference_variable *rhs = ir->rhs->as_dereference_variable();
 473    if (!rhs) {
 474       ir_swizzle *swiz = ir->rhs->as_swizzle();
 475       if (!swiz)
 476          return;
 477
 478       rhs = swiz->val->as_dereference_variable();
 479       if (!rhs)
 480          return;
 481
 482       orig_swizzle[0] = swiz->mask.x;
 483       orig_swizzle[1] = swiz->mask.y;
 484       orig_swizzle[2] = swiz->mask.z;
 485       orig_swizzle[3] = swiz->mask.w;
 486    }
 487
 488    /* Move the swizzle channels out to the positions they match in the
 489     * destination.  We don't want to have to rewrite the swizzle[]
 490     * array every time we clear a bit of the write_mask.
 491     */
 492    int j = 0;
 493    for (int i = 0; i < 4; i++) {
 494       if (ir->write_mask & (1 << i))
 495          swizzle[i] = orig_swizzle[j++];
 496    }
 497
 498    int write_mask = ir->write_mask;
 499    if (lhs->var == rhs->var) {
 500       /* If this is a copy from the variable to itself, then we need
 501        * to be sure not to include the updated channels from this
 502        * instruction in the set of new source channels to be
 503        * copy-propagated from.
 504        */
 505       for (int i = 0; i < 4; i++) {
 506          if (ir->write_mask & (1 << orig_swizzle[i]))
 507             write_mask &= ~(1 << i);
 508       }
 509    }
 510
 511    if (lhs->var->data.precise != rhs->var->data.precise)
 512       return;
 513
 514    entry = new(this->mem_ctx) acp_entry(lhs->var, rhs->var, write_mask,
 515                                         swizzle);
 516    this->acp->push_tail(entry);
 517 }
 518
 519 bool
 520 do_copy_propagation_elements(exec_list *instructions)
 521 {
 522    ir_copy_propagation_elements_visitor v;
 523
 524    visit_list_elements(&v, instructions);
 525
 526    return v.progress;
 527 }