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