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