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