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 #include "util/hash_table.h"
  50
  51 static bool debug = false;
  52
  53 namespace {
  54
  55 class acp_entry;
  56
  57 /* Class that refers to acp_entry in another exec_list. Used
  58  * when making removals based on rhs.
  59  */
  60 class acp_ref : public exec_node
  61 {
  62 public:
  63    acp_ref(acp_entry *e)
  64    {
  65       entry = e;
  66    }
  67    acp_entry *entry;
  68 };
  69
  70 class acp_entry : public exec_node
  71 {
  72 public:
  73    /* override operator new from exec_node */
  74    DECLARE_LINEAR_ZALLOC_CXX_OPERATORS(acp_entry)
  75
  76    acp_entry(ir_variable *lhs, ir_variable *rhs, int write_mask, int swizzle[4])
  77       : rhs_node(this)
  78    {
  79       this->lhs = lhs;
  80       this->rhs = rhs;
  81       this->write_mask = write_mask;
  82       memcpy(this->swizzle, swizzle, sizeof(this->swizzle));
  83    }
  84
  85    ir_variable *lhs;
  86    ir_variable *rhs;
  87    unsigned int write_mask;
  88    int swizzle[4];
  89    acp_ref rhs_node;
  90 };
  91
  92
  93 class kill_entry : public exec_node
  94 {
  95 public:
  96    /* override operator new from exec_node */
  97    DECLARE_LINEAR_ZALLOC_CXX_OPERATORS(kill_entry)
  98
  99    kill_entry(ir_variable *var, int write_mask)
 100    {
 101       this->var = var;
 102       this->write_mask = write_mask;
 103    }
 104
 105    ir_variable *var;
 106    unsigned int write_mask;
 107 };
 108
 109 class ir_copy_propagation_elements_visitor : public ir_rvalue_visitor {
 110 public:
 111    ir_copy_propagation_elements_visitor()
 112    {
 113       this->progress = false;
 114       this->killed_all = false;
 115       this->mem_ctx = ralloc_context(NULL);
 116       this->lin_ctx = linear_alloc_parent(this->mem_ctx, 0);
 117       this->shader_mem_ctx = NULL;
 118       this->kills = new(mem_ctx) exec_list;
 119
 120       create_acp();
 121    }
 122    ~ir_copy_propagation_elements_visitor()
 123    {
 124       ralloc_free(mem_ctx);
 125    }
 126
 127    void clone_acp(hash_table *lhs, hash_table *rhs)
 128    {
 129       lhs_ht = _mesa_hash_table_clone(lhs, mem_ctx);
 130       rhs_ht = _mesa_hash_table_clone(rhs, mem_ctx);
 131    }
 132
 133    void create_acp()
 134    {
 135       lhs_ht = _mesa_hash_table_create(mem_ctx, _mesa_hash_pointer,
 136                                        _mesa_key_pointer_equal);
 137       rhs_ht = _mesa_hash_table_create(mem_ctx, _mesa_hash_pointer,
 138                                        _mesa_key_pointer_equal);
 139    }
 140
 141    void destroy_acp()
 142    {
 143       _mesa_hash_table_destroy(lhs_ht, NULL);
 144       _mesa_hash_table_destroy(rhs_ht, NULL);
 145    }
 146
 147    void handle_loop(ir_loop *, bool keep_acp);
 148    virtual ir_visitor_status visit_enter(class ir_loop *);
 149    virtual ir_visitor_status visit_enter(class ir_function_signature *);
 150    virtual ir_visitor_status visit_leave(class ir_assignment *);
 151    virtual ir_visitor_status visit_enter(class ir_call *);
 152    virtual ir_visitor_status visit_enter(class ir_if *);
 153    virtual ir_visitor_status visit_leave(class ir_swizzle *);
 154
 155    void handle_rvalue(ir_rvalue **rvalue);
 156
 157    void add_copy(ir_assignment *ir);
 158    void kill(kill_entry *k);
 159    void handle_if_block(exec_list *instructions);
 160
 161    /** Hash of acp_entry: The available copies to propagate */
 162    hash_table *lhs_ht;
 163    hash_table *rhs_ht;
 164
 165    /**
 166     * List of kill_entry: The variables whose values were killed in this
 167     * block.
 168     */
 169    exec_list *kills;
 170
 171    bool progress;
 172
 173    bool killed_all;
 174
 175    /* Context for our local data structures. */
 176    void *mem_ctx;
 177    void *lin_ctx;
 178    /* Context for allocating new shader nodes. */
 179    void *shader_mem_ctx;
 180 };
 181
 182 } /* unnamed namespace */
 183
 184 ir_visitor_status
 185 ir_copy_propagation_elements_visitor::visit_enter(ir_function_signature *ir)
 186 {
 187    /* Treat entry into a function signature as a completely separate
 188     * block.  Any instructions at global scope will be shuffled into
 189     * main() at link time, so they're irrelevant to us.
 190     */
 191    exec_list *orig_kills = this->kills;
 192    bool orig_killed_all = this->killed_all;
 193
 194    hash_table *orig_lhs_ht = lhs_ht;
 195    hash_table *orig_rhs_ht = rhs_ht;
 196
 197    this->kills = new(mem_ctx) exec_list;
 198    this->killed_all = false;
 199
 200    create_acp();
 201
 202    visit_list_elements(this, &ir->body);
 203
 204    ralloc_free(this->kills);
 205
 206    destroy_acp();
 207
 208    this->kills = orig_kills;
 209    this->killed_all = orig_killed_all;
 210
 211    lhs_ht = orig_lhs_ht;
 212    rhs_ht = orig_rhs_ht;
 213
 214    return visit_continue_with_parent;
 215 }
 216
 217 ir_visitor_status
 218 ir_copy_propagation_elements_visitor::visit_leave(ir_assignment *ir)
 219 {
 220    ir_dereference_variable *lhs = ir->lhs->as_dereference_variable();
 221    ir_variable *var = ir->lhs->variable_referenced();
 222
 223    if (var->type->is_scalar() || var->type->is_vector()) {
 224       kill_entry *k;
 225
 226       if (lhs)
 227          k = new(this->lin_ctx) kill_entry(var, ir->write_mask);
 228       else
 229          k = new(this->lin_ctx) kill_entry(var, ~0);
 230
 231       kill(k);
 232    }
 233
 234    add_copy(ir);
 235
 236    return visit_continue;
 237 }
 238
 239 ir_visitor_status
 240 ir_copy_propagation_elements_visitor::visit_leave(ir_swizzle *)
 241 {
 242    /* Don't visit the values of swizzles since they are handled while
 243     * visiting the swizzle itself.
 244     */
 245    return visit_continue;
 246 }
 247
 248 /**
 249  * Replaces dereferences of ACP RHS variables with ACP LHS variables.
 250  *
 251  * This is where the actual copy propagation occurs.  Note that the
 252  * rewriting of ir_dereference means that the ir_dereference instance
 253  * must not be shared by multiple IR operations!
 254  */
 255 void
 256 ir_copy_propagation_elements_visitor::handle_rvalue(ir_rvalue **ir)
 257 {
 258    int swizzle_chan[4];
 259    ir_dereference_variable *deref_var;
 260    ir_variable *source[4] = {NULL, NULL, NULL, NULL};
 261    int source_chan[4] = {0, 0, 0, 0};
 262    int chans;
 263    bool noop_swizzle = true;
 264
 265    if (!*ir)
 266       return;
 267
 268    ir_swizzle *swizzle = (*ir)->as_swizzle();
 269    if (swizzle) {
 270       deref_var = swizzle->val->as_dereference_variable();
 271       if (!deref_var)
 272          return;
 273
 274       swizzle_chan[0] = swizzle->mask.x;
 275       swizzle_chan[1] = swizzle->mask.y;
 276       swizzle_chan[2] = swizzle->mask.z;
 277       swizzle_chan[3] = swizzle->mask.w;
 278       chans = swizzle->type->vector_elements;
 279    } else {
 280       deref_var = (*ir)->as_dereference_variable();
 281       if (!deref_var)
 282          return;
 283
 284       swizzle_chan[0] = 0;
 285       swizzle_chan[1] = 1;
 286       swizzle_chan[2] = 2;
 287       swizzle_chan[3] = 3;
 288       chans = deref_var->type->vector_elements;
 289    }
 290
 291    if (this->in_assignee)
 292       return;
 293
 294    ir_variable *var = deref_var->var;
 295
 296    /* Try to find ACP entries covering swizzle_chan[], hoping they're
 297     * the same source variable.
 298     */
 299    hash_entry *ht_entry = _mesa_hash_table_search(lhs_ht, var);
 300    if (ht_entry) {
 301       exec_list *ht_list = (exec_list *) ht_entry->data;
 302       foreach_in_list(acp_entry, entry, ht_list) {
 303          for (int c = 0; c < chans; c++) {
 304             if (entry->write_mask & (1 << swizzle_chan[c])) {
 305                source[c] = entry->rhs;
 306                source_chan[c] = entry->swizzle[swizzle_chan[c]];
 307
 308                if (source_chan[c] != swizzle_chan[c])
 309                   noop_swizzle = false;
 310             }
 311          }
 312       }
 313    }
 314
 315    /* Make sure all channels are copying from the same source variable. */
 316    if (!source[0])
 317       return;
 318    for (int c = 1; c < chans; c++) {
 319       if (source[c] != source[0])
 320          return;
 321    }
 322
 323    if (!shader_mem_ctx)
 324       shader_mem_ctx = ralloc_parent(deref_var);
 325
 326    /* Don't pointlessly replace the rvalue with itself (or a noop swizzle
 327     * of itself, which would just be deleted by opt_noop_swizzle).
 328     */
 329    if (source[0] == var && noop_swizzle)
 330       return;
 331
 332    if (debug) {
 333       printf("Copy propagation from:\n");
 334       (*ir)->print();
 335    }
 336
 337    deref_var = new(shader_mem_ctx) ir_dereference_variable(source[0]);
 338    *ir = new(shader_mem_ctx) ir_swizzle(deref_var,
 339                                         source_chan[0],
 340                                         source_chan[1],
 341                                         source_chan[2],
 342                                         source_chan[3],
 343                                         chans);
 344    progress = true;
 345
 346    if (debug) {
 347       printf("to:\n");
 348       (*ir)->print();
 349       printf("\n");
 350    }
 351 }
 352
 353
 354 ir_visitor_status
 355 ir_copy_propagation_elements_visitor::visit_enter(ir_call *ir)
 356 {
 357    /* Do copy propagation on call parameters, but skip any out params */
 358    foreach_two_lists(formal_node, &ir->callee->parameters,
 359                      actual_node, &ir->actual_parameters) {
 360       ir_variable *sig_param = (ir_variable *) formal_node;
 361       ir_rvalue *ir = (ir_rvalue *) actual_node;
 362       if (sig_param->data.mode != ir_var_function_out
 363           && sig_param->data.mode != ir_var_function_inout) {
 364          ir->accept(this);
 365       }
 366    }
 367
 368    /* Since we're unlinked, we don't (necessarily) know the side effects of
 369     * this call.  So kill all copies.
 370     */
 371    _mesa_hash_table_clear(lhs_ht, NULL);
 372    _mesa_hash_table_clear(rhs_ht, NULL);
 373
 374    this->killed_all = true;
 375
 376    return visit_continue_with_parent;
 377 }
 378
 379 void
 380 ir_copy_propagation_elements_visitor::handle_if_block(exec_list *instructions)
 381 {
 382    exec_list *orig_kills = this->kills;
 383    bool orig_killed_all = this->killed_all;
 384
 385    hash_table *orig_lhs_ht = lhs_ht;
 386    hash_table *orig_rhs_ht = rhs_ht;
 387
 388    this->kills = new(mem_ctx) exec_list;
 389    this->killed_all = false;
 390
 391    /* Populate the initial acp with a copy of the original */
 392    clone_acp(orig_lhs_ht, orig_rhs_ht);
 393
 394    visit_list_elements(this, instructions);
 395
 396    if (this->killed_all) {
 397       _mesa_hash_table_clear(orig_lhs_ht, NULL);
 398       _mesa_hash_table_clear(orig_rhs_ht, NULL);
 399    }
 400
 401    exec_list *new_kills = this->kills;
 402    this->kills = orig_kills;
 403    this->killed_all = this->killed_all || orig_killed_all;
 404
 405    destroy_acp();
 406
 407    lhs_ht = orig_lhs_ht;
 408    rhs_ht = orig_rhs_ht;
 409
 410    /* Move the new kills into the parent block's list, removing them
 411     * from the parent's ACP list in the process.
 412     */
 413    foreach_in_list_safe(kill_entry, k, new_kills) {
 414       kill(k);
 415    }
 416
 417    ralloc_free(new_kills);
 418 }
 419
 420 ir_visitor_status
 421 ir_copy_propagation_elements_visitor::visit_enter(ir_if *ir)
 422 {
 423    ir->condition->accept(this);
 424
 425    handle_if_block(&ir->then_instructions);
 426    handle_if_block(&ir->else_instructions);
 427
 428    /* handle_if_block() already descended into the children. */
 429    return visit_continue_with_parent;
 430 }
 431
 432 void
 433 ir_copy_propagation_elements_visitor::handle_loop(ir_loop *ir, bool keep_acp)
 434 {
 435    exec_list *orig_kills = this->kills;
 436    bool orig_killed_all = this->killed_all;
 437
 438    hash_table *orig_lhs_ht = lhs_ht;
 439    hash_table *orig_rhs_ht = rhs_ht;
 440
 441    /* FINISHME: For now, the initial acp for loops is totally empty.
 442     * We could go through once, then go through again with the acp
 443     * cloned minus the killed entries after the first run through.
 444     */
 445    this->kills = new(mem_ctx) exec_list;
 446    this->killed_all = false;
 447
 448    if (keep_acp) {
 449       /* Populate the initial acp with a copy of the original */
 450       clone_acp(orig_lhs_ht, orig_rhs_ht);
 451    } else {
 452       create_acp();
 453    }
 454
 455    visit_list_elements(this, &ir->body_instructions);
 456
 457    if (this->killed_all) {
 458       _mesa_hash_table_clear(orig_lhs_ht, NULL);
 459       _mesa_hash_table_clear(orig_rhs_ht, NULL);
 460    }
 461
 462    exec_list *new_kills = this->kills;
 463    this->kills = orig_kills;
 464    this->killed_all = this->killed_all || orig_killed_all;
 465
 466    destroy_acp();
 467
 468    lhs_ht = orig_lhs_ht;
 469    rhs_ht = orig_rhs_ht;
 470
 471    foreach_in_list_safe(kill_entry, k, new_kills) {
 472       kill(k);
 473    }
 474
 475    ralloc_free(new_kills);
 476 }
 477
 478 ir_visitor_status
 479 ir_copy_propagation_elements_visitor::visit_enter(ir_loop *ir)
 480 {
 481    handle_loop(ir, false);
 482    handle_loop(ir, true);
 483
 484    /* already descended into the children. */
 485    return visit_continue_with_parent;
 486 }
 487
 488 /* Remove any entries currently in the ACP for this kill. */
 489 void
 490 ir_copy_propagation_elements_visitor::kill(kill_entry *k)
 491 {
 492    /* removal of lhs entries */
 493    hash_entry *ht_entry = _mesa_hash_table_search(lhs_ht, k->var);
 494    if (ht_entry) {
 495       exec_list *lhs_list = (exec_list *) ht_entry->data;
 496       foreach_in_list_safe(acp_entry, entry, lhs_list) {
 497          entry->write_mask = entry->write_mask & ~k->write_mask;
 498          if (entry->write_mask == 0) {
 499             entry->remove();
 500             continue;
 501          }
 502       }
 503    }
 504
 505    /* removal of rhs entries */
 506    ht_entry = _mesa_hash_table_search(rhs_ht, k->var);
 507    if (ht_entry) {
 508       exec_list *rhs_list = (exec_list *) ht_entry->data;
 509       acp_ref *ref;
 510
 511       while ((ref = (acp_ref *) rhs_list->pop_head()) != NULL) {
 512          acp_entry *entry = ref->entry;
 513
 514          /* If entry is still in a list (not already removed by lhs entry
 515           * removal above), remove it.
 516           */
 517          if (entry->prev || entry->next)
 518             entry->remove();
 519       }
 520    }
 521
 522    /* If we were on a list, remove ourselves before inserting */
 523    if (k->next)
 524       k->remove();
 525
 526    this->kills->push_tail(k);
 527 }
 528
 529 /**
 530  * Adds directly-copied channels between vector variables to the available
 531  * copy propagation list.
 532  */
 533 void
 534 ir_copy_propagation_elements_visitor::add_copy(ir_assignment *ir)
 535 {
 536    acp_entry *entry;
 537    int orig_swizzle[4] = {0, 1, 2, 3};
 538    int swizzle[4];
 539
 540    if (ir->condition)
 541       return;
 542
 543    ir_dereference_variable *lhs = ir->lhs->as_dereference_variable();
 544    if (!lhs || !(lhs->type->is_scalar() || lhs->type->is_vector()))
 545       return;
 546
 547    if (lhs->var->data.mode == ir_var_shader_storage ||
 548        lhs->var->data.mode == ir_var_shader_shared)
 549       return;
 550
 551    ir_dereference_variable *rhs = ir->rhs->as_dereference_variable();
 552    if (!rhs) {
 553       ir_swizzle *swiz = ir->rhs->as_swizzle();
 554       if (!swiz)
 555          return;
 556
 557       rhs = swiz->val->as_dereference_variable();
 558       if (!rhs)
 559          return;
 560
 561       orig_swizzle[0] = swiz->mask.x;
 562       orig_swizzle[1] = swiz->mask.y;
 563       orig_swizzle[2] = swiz->mask.z;
 564       orig_swizzle[3] = swiz->mask.w;
 565    }
 566
 567    if (rhs->var->data.mode == ir_var_shader_storage ||
 568        rhs->var->data.mode == ir_var_shader_shared)
 569       return;
 570
 571    /* Move the swizzle channels out to the positions they match in the
 572     * destination.  We don't want to have to rewrite the swizzle[]
 573     * array every time we clear a bit of the write_mask.
 574     */
 575    int j = 0;
 576    for (int i = 0; i < 4; i++) {
 577       if (ir->write_mask & (1 << i))
 578          swizzle[i] = orig_swizzle[j++];
 579    }
 580
 581    int write_mask = ir->write_mask;
 582    if (lhs->var == rhs->var) {
 583       /* If this is a copy from the variable to itself, then we need
 584        * to be sure not to include the updated channels from this
 585        * instruction in the set of new source channels to be
 586        * copy-propagated from.
 587        */
 588       for (int i = 0; i < 4; i++) {
 589          if (ir->write_mask & (1 << orig_swizzle[i]))
 590             write_mask &= ~(1 << i);
 591       }
 592    }
 593
 594    if (lhs->var->data.precise != rhs->var->data.precise)
 595       return;
 596
 597    entry = new(this->lin_ctx) acp_entry(lhs->var, rhs->var, write_mask,
 598                                         swizzle);
 599
 600    /* lhs hash, hash of lhs -> acp_entry lists */
 601    hash_entry *ht_entry = _mesa_hash_table_search(lhs_ht, lhs->var);
 602    if (ht_entry) {
 603       exec_list *lhs_list = (exec_list *) ht_entry->data;
 604       lhs_list->push_tail(entry);
 605    } else {
 606       exec_list *lhs_list = new(mem_ctx) exec_list;
 607       lhs_list->push_tail(entry);
 608       _mesa_hash_table_insert(lhs_ht, lhs->var, lhs_list);
 609    }
 610
 611    /* rhs hash, hash of rhs -> acp_entry pointers to lhs lists */
 612    ht_entry = _mesa_hash_table_search(rhs_ht, rhs->var);
 613    if (ht_entry) {
 614       exec_list *rhs_list = (exec_list *) ht_entry->data;
 615       rhs_list->push_tail(&entry->rhs_node);
 616    } else {
 617       exec_list *rhs_list = new(mem_ctx) exec_list;
 618       rhs_list->push_tail(&entry->rhs_node);
 619       _mesa_hash_table_insert(rhs_ht, rhs->var, rhs_list);
 620    }
 621 }
 622
 623 bool
 624 do_copy_propagation_elements(exec_list *instructions)
 625 {
 626    ir_copy_propagation_elements_visitor v;
 627
 628    visit_list_elements(&v, instructions);
 629
 630    return v.progress;
 631 }