src/compiler/glsl/opt_copy_propagation.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.cpp
  26  *
  27  * Moves usage of recently-copied variables to the previous copy of
  28  * the variable.
  29  *
  30  * This should reduce the number of MOV instructions in the generated
  31  * programs unless copy propagation is also done on the LIR, and may
  32  * help anyway by triggering other optimizations that live in the HIR.
  33  */
  34
  35 #include "ir.h"
  36 #include "ir_visitor.h"
  37 #include "ir_basic_block.h"
  38 #include "ir_optimization.h"
  39 #include "compiler/glsl_types.h"
  40
  41 namespace {
  42
  43 class acp_entry : public exec_node
  44 {
  45 public:
  46    acp_entry(ir_variable *lhs, ir_variable *rhs)
  47    {
  48       assert(lhs);
  49       assert(rhs);
  50       this->lhs = lhs;
  51       this->rhs = rhs;
  52    }
  53
  54    ir_variable *lhs;
  55    ir_variable *rhs;
  56 };
  57
  58
  59 class kill_entry : public exec_node
  60 {
  61 public:
  62    kill_entry(ir_variable *var)
  63    {
  64       assert(var);
  65       this->var = var;
  66    }
  67
  68    ir_variable *var;
  69 };
  70
  71 class ir_copy_propagation_visitor : public ir_hierarchical_visitor {
  72 public:
  73    ir_copy_propagation_visitor()
  74    {
  75       progress = false;
  76       mem_ctx = ralloc_context(0);
  77       this->acp = new(mem_ctx) exec_list;
  78       this->kills = new(mem_ctx) exec_list;
  79       killed_all = false;
  80    }
  81    ~ir_copy_propagation_visitor()
  82    {
  83       ralloc_free(mem_ctx);
  84    }
  85
  86    virtual ir_visitor_status visit(class ir_dereference_variable *);
  87    void handle_loop(class ir_loop *, bool keep_acp);
  88    virtual ir_visitor_status visit_enter(class ir_loop *);
  89    virtual ir_visitor_status visit_enter(class ir_function_signature *);
  90    virtual ir_visitor_status visit_enter(class ir_function *);
  91    virtual ir_visitor_status visit_leave(class ir_assignment *);
  92    virtual ir_visitor_status visit_enter(class ir_call *);
  93    virtual ir_visitor_status visit_enter(class ir_if *);
  94
  95    void add_copy(ir_assignment *ir);
  96    void kill(ir_variable *ir);
  97    void handle_if_block(exec_list *instructions);
  98
  99    /** List of acp_entry: The available copies to propagate */
 100    exec_list *acp;
 101    /**
 102     * List of kill_entry: The variables whose values were killed in this
 103     * block.
 104     */
 105    exec_list *kills;
 106
 107    bool progress;
 108
 109    bool killed_all;
 110
 111    void *mem_ctx;
 112 };
 113
 114 } /* unnamed namespace */
 115
 116 ir_visitor_status
 117 ir_copy_propagation_visitor::visit_enter(ir_function_signature *ir)
 118 {
 119    /* Treat entry into a function signature as a completely separate
 120     * block.  Any instructions at global scope will be shuffled into
 121     * main() at link time, so they're irrelevant to us.
 122     */
 123    exec_list *orig_acp = this->acp;
 124    exec_list *orig_kills = this->kills;
 125    bool orig_killed_all = this->killed_all;
 126
 127    this->acp = new(mem_ctx) exec_list;
 128    this->kills = new(mem_ctx) exec_list;
 129    this->killed_all = false;
 130
 131    visit_list_elements(this, &ir->body);
 132
 133    ralloc_free(this->acp);
 134    ralloc_free(this->kills);
 135
 136    this->kills = orig_kills;
 137    this->acp = orig_acp;
 138    this->killed_all = orig_killed_all;
 139
 140    return visit_continue_with_parent;
 141 }
 142
 143 ir_visitor_status
 144 ir_copy_propagation_visitor::visit_leave(ir_assignment *ir)
 145 {
 146    kill(ir->lhs->variable_referenced());
 147
 148    add_copy(ir);
 149
 150    return visit_continue;
 151 }
 152
 153 ir_visitor_status
 154 ir_copy_propagation_visitor::visit_enter(ir_function *ir)
 155 {
 156    (void) ir;
 157    return visit_continue;
 158 }
 159
 160 /**
 161  * Replaces dereferences of ACP RHS variables with ACP LHS variables.
 162  *
 163  * This is where the actual copy propagation occurs.  Note that the
 164  * rewriting of ir_dereference means that the ir_dereference instance
 165  * must not be shared by multiple IR operations!
 166  */
 167 ir_visitor_status
 168 ir_copy_propagation_visitor::visit(ir_dereference_variable *ir)
 169 {
 170    if (this->in_assignee)
 171       return visit_continue;
 172
 173    ir_variable *var = ir->var;
 174
 175    foreach_in_list(acp_entry, entry, this->acp) {
 176       if (var == entry->lhs) {
 177          ir->var = entry->rhs;
 178          this->progress = true;
 179          break;
 180       }
 181    }
 182
 183    return visit_continue;
 184 }
 185
 186
 187 ir_visitor_status
 188 ir_copy_propagation_visitor::visit_enter(ir_call *ir)
 189 {
 190    /* Do copy propagation on call parameters, but skip any out params */
 191    foreach_two_lists(formal_node, &ir->callee->parameters,
 192                      actual_node, &ir->actual_parameters) {
 193       ir_variable *sig_param = (ir_variable *) formal_node;
 194       ir_rvalue *ir = (ir_rvalue *) actual_node;
 195       if (sig_param->data.mode != ir_var_function_out
 196           && sig_param->data.mode != ir_var_function_inout) {
 197          ir->accept(this);
 198       }
 199    }
 200
 201    /* Since we're unlinked, we don't (necessarily) know the side effects of
 202     * this call.  So kill all copies.
 203     */
 204    acp->make_empty();
 205    this->killed_all = true;
 206
 207    return visit_continue_with_parent;
 208 }
 209
 210 void
 211 ir_copy_propagation_visitor::handle_if_block(exec_list *instructions)
 212 {
 213    exec_list *orig_acp = this->acp;
 214    exec_list *orig_kills = this->kills;
 215    bool orig_killed_all = this->killed_all;
 216
 217    this->acp = new(mem_ctx) exec_list;
 218    this->kills = new(mem_ctx) exec_list;
 219    this->killed_all = false;
 220
 221    /* Populate the initial acp with a copy of the original */
 222    foreach_in_list(acp_entry, a, orig_acp) {
 223       this->acp->push_tail(new(this->acp) acp_entry(a->lhs, a->rhs));
 224    }
 225
 226    visit_list_elements(this, instructions);
 227
 228    if (this->killed_all) {
 229       orig_acp->make_empty();
 230    }
 231
 232    exec_list *new_kills = this->kills;
 233    this->kills = orig_kills;
 234    ralloc_free(this->acp);
 235    this->acp = orig_acp;
 236    this->killed_all = this->killed_all || orig_killed_all;
 237
 238    foreach_in_list(kill_entry, k, new_kills) {
 239       kill(k->var);
 240    }
 241
 242    ralloc_free(new_kills);
 243 }
 244
 245 ir_visitor_status
 246 ir_copy_propagation_visitor::visit_enter(ir_if *ir)
 247 {
 248    ir->condition->accept(this);
 249
 250    handle_if_block(&ir->then_instructions);
 251    handle_if_block(&ir->else_instructions);
 252
 253    /* handle_if_block() already descended into the children. */
 254    return visit_continue_with_parent;
 255 }
 256
 257 void
 258 ir_copy_propagation_visitor::handle_loop(ir_loop *ir, bool keep_acp)
 259 {
 260    exec_list *orig_acp = this->acp;
 261    exec_list *orig_kills = this->kills;
 262    bool orig_killed_all = this->killed_all;
 263
 264    this->acp = new(mem_ctx) exec_list;
 265    this->kills = new(mem_ctx) exec_list;
 266    this->killed_all = false;
 267
 268    if (keep_acp) {
 269       /* Populate the initial acp with a copy of the original */
 270       foreach_in_list(acp_entry, a, orig_acp) {
 271          this->acp->push_tail(new(this->acp) acp_entry(a->lhs, a->rhs));
 272       }
 273    }
 274
 275    visit_list_elements(this, &ir->body_instructions);
 276
 277    if (this->killed_all) {
 278       orig_acp->make_empty();
 279    }
 280
 281    exec_list *new_kills = this->kills;
 282    this->kills = orig_kills;
 283    ralloc_free(this->acp);
 284    this->acp = orig_acp;
 285    this->killed_all = this->killed_all || orig_killed_all;
 286
 287    foreach_in_list(kill_entry, k, new_kills) {
 288       kill(k->var);
 289    }
 290
 291    ralloc_free(new_kills);
 292 }
 293
 294 ir_visitor_status
 295 ir_copy_propagation_visitor::visit_enter(ir_loop *ir)
 296 {
 297    /* Make a conservative first pass over the loop with an empty ACP set.
 298     * This also removes any killed entries from the original ACP set.
 299     */
 300    handle_loop(ir, false);
 301
 302    /* Then, run it again with the real ACP set, minus any killed entries.
 303     * This takes care of propagating values from before the loop into it.
 304     */
 305    handle_loop(ir, true);
 306
 307    /* already descended into the children. */
 308    return visit_continue_with_parent;
 309 }
 310
 311 void
 312 ir_copy_propagation_visitor::kill(ir_variable *var)
 313 {
 314    assert(var != NULL);
 315
 316    /* Remove any entries currently in the ACP for this kill. */
 317    foreach_in_list_safe(acp_entry, entry, acp) {
 318       if (entry->lhs == var || entry->rhs == var) {
 319          entry->remove();
 320       }
 321    }
 322
 323    /* Add the LHS variable to the list of killed variables in this block.
 324     */
 325    this->kills->push_tail(new(this->kills) kill_entry(var));
 326 }
 327
 328 /**
 329  * Adds an entry to the available copy list if it's a plain assignment
 330  * of a variable to a variable.
 331  */
 332 void
 333 ir_copy_propagation_visitor::add_copy(ir_assignment *ir)
 334 {
 335    acp_entry *entry;
 336
 337    if (ir->condition)
 338       return;
 339
 340    ir_variable *lhs_var = ir->whole_variable_written();
 341    ir_variable *rhs_var = ir->rhs->whole_variable_referenced();
 342
 343    if ((lhs_var != NULL) && (rhs_var != NULL)) {
 344       if (lhs_var == rhs_var) {
 345          /* This is a dumb assignment, but we've conveniently noticed
 346           * it here.  Removing it now would mess up the loop iteration
 347           * calling us.  Just flag it to not execute, and someone else
 348           * will clean up the mess.
 349           */
 350          ir->condition = new(ralloc_parent(ir)) ir_constant(false);
 351          this->progress = true;
 352       } else if (lhs_var->data.mode != ir_var_shader_storage &&
 353                  lhs_var->data.mode != ir_var_shader_shared &&
 354                  lhs_var->data.precise == rhs_var->data.precise) {
 355          entry = new(this->acp) acp_entry(lhs_var, rhs_var);
 356          this->acp->push_tail(entry);
 357       }
 358    }
 359 }
 360
 361 /**
 362  * Does a copy propagation pass on the code present in the instruction stream.
 363  */
 364 bool
 365 do_copy_propagation(exec_list *instructions)
 366 {
 367    ir_copy_propagation_visitor v;
 368
 369    visit_list_elements(&v, instructions);
 370
 371    return v.progress;
 372 }