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