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 #include "util/set.h"
  42
  43 namespace {
  44
  45 class ir_copy_propagation_visitor : public ir_hierarchical_visitor {
  46 public:
  47    ir_copy_propagation_visitor()
  48    {
  49       progress = false;
  50       mem_ctx = ralloc_context(0);
  51       lin_ctx = linear_alloc_parent(mem_ctx, 0);
  52       acp = _mesa_hash_table_create(mem_ctx, _mesa_hash_pointer,
  53                                     _mesa_key_pointer_equal);
  54       kills = _mesa_set_create(mem_ctx, _mesa_hash_pointer,
  55                                _mesa_key_pointer_equal);
  56       killed_all = false;
  57    }
  58    ~ir_copy_propagation_visitor()
  59    {
  60       ralloc_free(mem_ctx);
  61    }
  62
  63    virtual ir_visitor_status visit(class ir_dereference_variable *);
  64    void handle_loop(class ir_loop *, bool keep_acp);
  65    virtual ir_visitor_status visit_enter(class ir_loop *);
  66    virtual ir_visitor_status visit_enter(class ir_function_signature *);
  67    virtual ir_visitor_status visit_enter(class ir_function *);
  68    virtual ir_visitor_status visit_leave(class ir_assignment *);
  69    virtual ir_visitor_status visit_enter(class ir_call *);
  70    virtual ir_visitor_status visit_enter(class ir_if *);
  71
  72    void add_copy(ir_assignment *ir);
  73    void kill(ir_variable *ir);
  74    void handle_if_block(exec_list *instructions);
  75
  76    /** Hash of lhs->rhs: The available copies to propagate */
  77    hash_table *acp;
  78
  79    /**
  80     * Set of ir_variables: Whose values were killed in this block.
  81     */
  82    set *kills;
  83
  84    bool progress;
  85
  86    bool killed_all;
  87
  88    void *mem_ctx;
  89    void *lin_ctx;
  90 };
  91
  92 } /* unnamed namespace */
  93
  94 ir_visitor_status
  95 ir_copy_propagation_visitor::visit_enter(ir_function_signature *ir)
  96 {
  97    /* Treat entry into a function signature as a completely separate
  98     * block.  Any instructions at global scope will be shuffled into
  99     * main() at link time, so they're irrelevant to us.
 100     */
 101    hash_table *orig_acp = this->acp;
 102    set *orig_kills = this->kills;
 103    bool orig_killed_all = this->killed_all;
 104
 105    acp = _mesa_hash_table_create(NULL, _mesa_hash_pointer,
 106                                  _mesa_key_pointer_equal);
 107    kills = _mesa_set_create(NULL, _mesa_hash_pointer,
 108                             _mesa_key_pointer_equal);
 109    this->killed_all = false;
 110
 111    visit_list_elements(this, &ir->body);
 112
 113    _mesa_hash_table_destroy(acp, NULL);
 114    _mesa_set_destroy(kills, NULL);
 115
 116    this->kills = orig_kills;
 117    this->acp = orig_acp;
 118    this->killed_all = orig_killed_all;
 119
 120    return visit_continue_with_parent;
 121 }
 122
 123 ir_visitor_status
 124 ir_copy_propagation_visitor::visit_leave(ir_assignment *ir)
 125 {
 126    kill(ir->lhs->variable_referenced());
 127
 128    add_copy(ir);
 129
 130    return visit_continue;
 131 }
 132
 133 ir_visitor_status
 134 ir_copy_propagation_visitor::visit_enter(ir_function *ir)
 135 {
 136    (void) ir;
 137    return visit_continue;
 138 }
 139
 140 /**
 141  * Replaces dereferences of ACP RHS variables with ACP LHS variables.
 142  *
 143  * This is where the actual copy propagation occurs.  Note that the
 144  * rewriting of ir_dereference means that the ir_dereference instance
 145  * must not be shared by multiple IR operations!
 146  */
 147 ir_visitor_status
 148 ir_copy_propagation_visitor::visit(ir_dereference_variable *ir)
 149 {
 150    if (this->in_assignee)
 151       return visit_continue;
 152
 153    struct hash_entry *entry = _mesa_hash_table_search(acp, ir->var);
 154    if (entry) {
 155       ir->var = (ir_variable *) entry->data;
 156       progress = true;
 157    }
 158
 159    return visit_continue;
 160 }
 161
 162
 163 ir_visitor_status
 164 ir_copy_propagation_visitor::visit_enter(ir_call *ir)
 165 {
 166    /* Do copy propagation on call parameters, but skip any out params */
 167    foreach_two_lists(formal_node, &ir->callee->parameters,
 168                      actual_node, &ir->actual_parameters) {
 169       ir_variable *sig_param = (ir_variable *) formal_node;
 170       ir_rvalue *ir = (ir_rvalue *) actual_node;
 171       if (sig_param->data.mode != ir_var_function_out
 172           && sig_param->data.mode != ir_var_function_inout) {
 173          ir->accept(this);
 174       }
 175    }
 176
 177    /* Since this pass can run when unlinked, we don't (necessarily) know
 178     * the side effects of calls.  (When linked, most calls are inlined
 179     * anyway, so it doesn't matter much.)
 180     *
 181     * One place where this does matter is IR intrinsics.  They're never
 182     * inlined.  We also know what they do - while some have side effects
 183     * (such as image writes), none edit random global variables.  So we
 184     * can assume they're side-effect free (other than the return value
 185     * and out parameters).
 186     */
 187    if (!ir->callee->is_intrinsic()) {
 188       _mesa_hash_table_clear(acp, NULL);
 189       this->killed_all = true;
 190    } else {
 191       if (ir->return_deref)
 192          kill(ir->return_deref->var);
 193
 194       foreach_two_lists(formal_node, &ir->callee->parameters,
 195                         actual_node, &ir->actual_parameters) {
 196          ir_variable *sig_param = (ir_variable *) formal_node;
 197          if (sig_param->data.mode == ir_var_function_out ||
 198              sig_param->data.mode == ir_var_function_inout) {
 199             ir_rvalue *ir = (ir_rvalue *) actual_node;
 200             ir_variable *var = ir->variable_referenced();
 201             kill(var);
 202          }
 203       }
 204    }
 205
 206    return visit_continue_with_parent;
 207 }
 208
 209 void
 210 ir_copy_propagation_visitor::handle_if_block(exec_list *instructions)
 211 {
 212    hash_table *orig_acp = this->acp;
 213    set *orig_kills = this->kills;
 214    bool orig_killed_all = this->killed_all;
 215
 216    kills = _mesa_set_create(NULL, _mesa_hash_pointer,
 217                             _mesa_key_pointer_equal);
 218    this->killed_all = false;
 219
 220    /* Populate the initial acp with a copy of the original */
 221    acp = _mesa_hash_table_clone(orig_acp, NULL);
 222
 223    visit_list_elements(this, instructions);
 224
 225    if (this->killed_all) {
 226       _mesa_hash_table_clear(orig_acp, NULL);
 227    }
 228
 229    set *new_kills = this->kills;
 230    this->kills = orig_kills;
 231    _mesa_hash_table_destroy(acp, NULL);
 232    this->acp = orig_acp;
 233    this->killed_all = this->killed_all || orig_killed_all;
 234
 235    struct set_entry *s_entry;
 236    set_foreach(new_kills, s_entry) {
 237       kill((ir_variable *) s_entry->key);
 238    }
 239
 240    _mesa_set_destroy(new_kills, NULL);
 241 }
 242
 243 ir_visitor_status
 244 ir_copy_propagation_visitor::visit_enter(ir_if *ir)
 245 {
 246    ir->condition->accept(this);
 247
 248    handle_if_block(&ir->then_instructions);
 249    handle_if_block(&ir->else_instructions);
 250
 251    /* handle_if_block() already descended into the children. */
 252    return visit_continue_with_parent;
 253 }
 254
 255 void
 256 ir_copy_propagation_visitor::handle_loop(ir_loop *ir, bool keep_acp)
 257 {
 258    hash_table *orig_acp = this->acp;
 259    set *orig_kills = this->kills;
 260    bool orig_killed_all = this->killed_all;
 261
 262    kills = _mesa_set_create(NULL, _mesa_hash_pointer,
 263                             _mesa_key_pointer_equal);
 264    this->killed_all = false;
 265
 266    if (keep_acp) {
 267       acp = _mesa_hash_table_clone(orig_acp, NULL);
 268    } else {
 269       acp = _mesa_hash_table_create(NULL, _mesa_hash_pointer,
 270                                     _mesa_key_pointer_equal);
 271    }
 272
 273    visit_list_elements(this, &ir->body_instructions);
 274
 275    if (this->killed_all) {
 276       _mesa_hash_table_clear(orig_acp, NULL);
 277    }
 278
 279    set *new_kills = this->kills;
 280    this->kills = orig_kills;
 281    _mesa_hash_table_destroy(acp, NULL);
 282    this->acp = orig_acp;
 283    this->killed_all = this->killed_all || orig_killed_all;
 284
 285    struct set_entry *entry;
 286    set_foreach(new_kills, entry) {
 287       kill((ir_variable *) entry->key);
 288    }
 289
 290    _mesa_set_destroy(new_kills, NULL);
 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    struct hash_entry *entry = _mesa_hash_table_search(acp, var);
 317    if (entry) {
 318       _mesa_hash_table_remove(acp, entry);
 319    }
 320
 321    hash_table_foreach(acp, entry) {
 322       if (var == (ir_variable *) entry->data) {
 323          _mesa_hash_table_remove(acp, entry);
 324       }
 325    }
 326
 327    /* Add the LHS variable to the set of killed variables in this block. */
 328    _mesa_set_add(kills, var);
 329 }
 330
 331 /**
 332  * Adds an entry to the available copy list if it's a plain assignment
 333  * of a variable to a variable.
 334  */
 335 void
 336 ir_copy_propagation_visitor::add_copy(ir_assignment *ir)
 337 {
 338    if (ir->condition)
 339       return;
 340
 341    ir_variable *lhs_var = ir->whole_variable_written();
 342    ir_variable *rhs_var = ir->rhs->whole_variable_referenced();
 343
 344    /* Don't try to remove a dumb assignment of a variable to itself.  Removing
 345     * it now would mess up the loop iteration calling us.
 346     */
 347    if (lhs_var != NULL && rhs_var != NULL && lhs_var != rhs_var) {
 348       if (lhs_var->data.mode != ir_var_shader_storage &&
 349           lhs_var->data.mode != ir_var_shader_shared &&
 350           rhs_var->data.mode != ir_var_shader_storage &&
 351           rhs_var->data.mode != ir_var_shader_shared &&
 352           lhs_var->data.precise == rhs_var->data.precise) {
 353          _mesa_hash_table_insert(acp, lhs_var, rhs_var);
 354       }
 355    }
 356 }
 357
 358 /**
 359  * Does a copy propagation pass on the code present in the instruction stream.
 360  */
 361 bool
 362 do_copy_propagation(exec_list *instructions)
 363 {
 364    ir_copy_propagation_visitor v;
 365
 366    visit_list_elements(&v, instructions);
 367
 368    return v.progress;
 369 }