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