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];
 211    int chans;
 212
 213    if (!*ir)
 214       return;
 215
 216    ir_swizzle *swizzle = (*ir)->as_swizzle();
 217    if (swizzle) {
 218       deref_var = swizzle->val->as_dereference_variable();
 219       if (!deref_var)
 220          return;
 221
 222       swizzle_chan[0] = swizzle->mask.x;
 223       swizzle_chan[1] = swizzle->mask.y;
 224       swizzle_chan[2] = swizzle->mask.z;
 225       swizzle_chan[3] = swizzle->mask.w;
 226       chans = swizzle->type->vector_elements;
 227    } else {
 228       deref_var = (*ir)->as_dereference_variable();
 229       if (!deref_var)
 230          return;
 231
 232       swizzle_chan[0] = 0;
 233       swizzle_chan[1] = 1;
 234       swizzle_chan[2] = 2;
 235       swizzle_chan[3] = 3;
 236       chans = deref_var->type->vector_elements;
 237    }
 238
 239    if (this->in_assignee)
 240       return;
 241
 242    ir_variable *var = deref_var->var;
 243
 244    /* Try to find ACP entries covering swizzle_chan[], hoping they're
 245     * the same source variable.
 246     */
 247    foreach_in_list(acp_entry, entry, this->acp) {
 248       if (var == entry->lhs) {
 249          for (int c = 0; c < chans; c++) {
 250             if (entry->write_mask & (1 << swizzle_chan[c])) {
 251                source[c] = entry->rhs;
 252                source_chan[c] = entry->swizzle[swizzle_chan[c]];
 253             }
 254          }
 255       }
 256    }
 257
 258    /* Make sure all channels are copying from the same source variable. */
 259    if (!source[0])
 260       return;
 261    for (int c = 1; c < chans; c++) {
 262       if (source[c] != source[0])
 263          return;
 264    }
 265
 266    if (!shader_mem_ctx)
 267       shader_mem_ctx = ralloc_parent(deref_var);
 268
 269    if (debug) {
 270       printf("Copy propagation from:\n");
 271       (*ir)->print();
 272    }
 273
 274    deref_var = new(shader_mem_ctx) ir_dereference_variable(source[0]);
 275    *ir = new(shader_mem_ctx) ir_swizzle(deref_var,
 276                                         source_chan[0],
 277                                         source_chan[1],
 278                                         source_chan[2],
 279                                         source_chan[3],
 280                                         chans);
 281
 282    if (debug) {
 283       printf("to:\n");
 284       (*ir)->print();
 285       printf("\n");
 286    }
 287 }
 288
 289
 290 ir_visitor_status
 291 ir_copy_propagation_elements_visitor::visit_enter(ir_call *ir)
 292 {
 293    /* Do copy propagation on call parameters, but skip any out params */
 294    foreach_two_lists(formal_node, &ir->callee->parameters,
 295                      actual_node, &ir->actual_parameters) {
 296       ir_variable *sig_param = (ir_variable *) formal_node;
 297       ir_rvalue *ir = (ir_rvalue *) actual_node;
 298       if (sig_param->data.mode != ir_var_function_out
 299           && sig_param->data.mode != ir_var_function_inout) {
 300          ir->accept(this);
 301       }
 302    }
 303
 304    /* Since we're unlinked, we don't (necessarily) know the side effects of
 305     * this call.  So kill all copies.
 306     */
 307    acp->make_empty();
 308    this->killed_all = true;
 309
 310    return visit_continue_with_parent;
 311 }
 312
 313 void
 314 ir_copy_propagation_elements_visitor::handle_if_block(exec_list *instructions)
 315 {
 316    exec_list *orig_acp = this->acp;
 317    exec_list *orig_kills = this->kills;
 318    bool orig_killed_all = this->killed_all;
 319
 320    this->acp = new(mem_ctx) exec_list;
 321    this->kills = new(mem_ctx) exec_list;
 322    this->killed_all = false;
 323
 324    /* Populate the initial acp with a copy of the original */
 325    foreach_in_list(acp_entry, a, orig_acp) {
 326       this->acp->push_tail(new(this->mem_ctx) acp_entry(a));
 327    }
 328
 329    visit_list_elements(this, instructions);
 330
 331    if (this->killed_all) {
 332       orig_acp->make_empty();
 333    }
 334
 335    exec_list *new_kills = this->kills;
 336    this->kills = orig_kills;
 337    this->acp = orig_acp;
 338    this->killed_all = this->killed_all || orig_killed_all;
 339
 340    /* Move the new kills into the parent block's list, removing them
 341     * from the parent's ACP list in the process.
 342     */
 343    foreach_list_safe(node, new_kills) {
 344       kill_entry *k = (kill_entry *)node;
 345       kill(k);
 346    }
 347 }
 348
 349 ir_visitor_status
 350 ir_copy_propagation_elements_visitor::visit_enter(ir_if *ir)
 351 {
 352    ir->condition->accept(this);
 353
 354    handle_if_block(&ir->then_instructions);
 355    handle_if_block(&ir->else_instructions);
 356
 357    /* handle_if_block() already descended into the children. */
 358    return visit_continue_with_parent;
 359 }
 360
 361 ir_visitor_status
 362 ir_copy_propagation_elements_visitor::visit_enter(ir_loop *ir)
 363 {
 364    exec_list *orig_acp = this->acp;
 365    exec_list *orig_kills = this->kills;
 366    bool orig_killed_all = this->killed_all;
 367
 368    /* FINISHME: For now, the initial acp for loops is totally empty.
 369     * We could go through once, then go through again with the acp
 370     * cloned minus the killed entries after the first run through.
 371     */
 372    this->acp = new(mem_ctx) exec_list;
 373    this->kills = new(mem_ctx) exec_list;
 374    this->killed_all = false;
 375
 376    visit_list_elements(this, &ir->body_instructions);
 377
 378    if (this->killed_all) {
 379       orig_acp->make_empty();
 380    }
 381
 382    exec_list *new_kills = this->kills;
 383    this->kills = orig_kills;
 384    this->acp = orig_acp;
 385    this->killed_all = this->killed_all || orig_killed_all;
 386
 387    foreach_list_safe(node, new_kills) {
 388       kill_entry *k = (kill_entry *)node;
 389       kill(k);
 390    }
 391
 392    /* already descended into the children. */
 393    return visit_continue_with_parent;
 394 }
 395
 396 /* Remove any entries currently in the ACP for this kill. */
 397 void
 398 ir_copy_propagation_elements_visitor::kill(kill_entry *k)
 399 {
 400    foreach_list_safe(node, acp) {
 401       acp_entry *entry = (acp_entry *)node;
 402
 403       if (entry->lhs == k->var) {
 404          entry->write_mask = entry->write_mask & ~k->write_mask;
 405          if (entry->write_mask == 0) {
 406             entry->remove();
 407             continue;
 408          }
 409       }
 410       if (entry->rhs == k->var) {
 411          entry->remove();
 412       }
 413    }
 414
 415    /* If we were on a list, remove ourselves before inserting */
 416    if (k->next)
 417       k->remove();
 418
 419    this->kills->push_tail(k);
 420 }
 421
 422 /**
 423  * Adds directly-copied channels between vector variables to the available
 424  * copy propagation list.
 425  */
 426 void
 427 ir_copy_propagation_elements_visitor::add_copy(ir_assignment *ir)
 428 {
 429    acp_entry *entry;
 430    int orig_swizzle[4] = {0, 1, 2, 3};
 431    int swizzle[4];
 432
 433    if (ir->condition)
 434       return;
 435
 436    ir_dereference_variable *lhs = ir->lhs->as_dereference_variable();
 437    if (!lhs || !(lhs->type->is_scalar() || lhs->type->is_vector()))
 438       return;
 439
 440    ir_dereference_variable *rhs = ir->rhs->as_dereference_variable();
 441    if (!rhs) {
 442       ir_swizzle *swiz = ir->rhs->as_swizzle();
 443       if (!swiz)
 444          return;
 445
 446       rhs = swiz->val->as_dereference_variable();
 447       if (!rhs)
 448          return;
 449
 450       orig_swizzle[0] = swiz->mask.x;
 451       orig_swizzle[1] = swiz->mask.y;
 452       orig_swizzle[2] = swiz->mask.z;
 453       orig_swizzle[3] = swiz->mask.w;
 454    }
 455
 456    /* Move the swizzle channels out to the positions they match in the
 457     * destination.  We don't want to have to rewrite the swizzle[]
 458     * array every time we clear a bit of the write_mask.
 459     */
 460    int j = 0;
 461    for (int i = 0; i < 4; i++) {
 462       if (ir->write_mask & (1 << i))
 463          swizzle[i] = orig_swizzle[j++];
 464    }
 465
 466    int write_mask = ir->write_mask;
 467    if (lhs->var == rhs->var) {
 468       /* If this is a copy from the variable to itself, then we need
 469        * to be sure not to include the updated channels from this
 470        * instruction in the set of new source channels to be
 471        * copy-propagated from.
 472        */
 473       for (int i = 0; i < 4; i++) {
 474          if (ir->write_mask & (1 << orig_swizzle[i]))
 475             write_mask &= ~(1 << i);
 476       }
 477    }
 478
 479    entry = new(this->mem_ctx) acp_entry(lhs->var, rhs->var, write_mask,
 480                                         swizzle);
 481    this->acp->push_tail(entry);
 482 }
 483
 484 bool
 485 do_copy_propagation_elements(exec_list *instructions)
 486 {
 487    ir_copy_propagation_elements_visitor v;
 488
 489    visit_list_elements(&v, instructions);
 490
 491    return v.progress;
 492 }