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