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