src/glsl/opt_dead_code_local.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_dead_code_local.cpp
  26  *
  27  * Eliminates local dead assignments from the code.
  28  *
  29  * This operates on basic blocks, tracking assignments and finding if
  30  * they're used before the variable is completely reassigned.
  31  *
  32  * Compare this to ir_dead_code.cpp, which operates globally looking
  33  * for assignments to variables that are never read.
  34  */
  35
  36 #include "ir.h"
  37 #include "ir_basic_block.h"
  38 #include "ir_optimization.h"
  39 #include "glsl_types.h"
  40
  41 static bool debug = false;
  42
  43 class assignment_entry : public exec_node
  44 {
  45 public:
  46    assignment_entry(ir_variable *lhs, ir_assignment *ir)
  47    {
  48       assert(lhs);
  49       assert(ir);
  50       this->lhs = lhs;
  51       this->ir = ir;
  52       this->available = ir->write_mask;
  53    }
  54
  55    ir_variable *lhs;
  56    ir_assignment *ir;
  57
  58    /* bitmask of xyzw channels written that haven't been used so far. */
  59    int available;
  60 };
  61
  62 class kill_for_derefs_visitor : public ir_hierarchical_visitor {
  63 public:
  64    kill_for_derefs_visitor(exec_list *assignments)
  65    {
  66       this->assignments = assignments;
  67    }
  68
  69    void kill_channels(ir_variable *const var, int used)
  70    {
  71       foreach_iter(exec_list_iterator, iter, *this->assignments) {
  72          assignment_entry *entry = (assignment_entry *)iter.get();
  73
  74          if (entry->lhs == var) {
  75             if (var->type->is_scalar() || var->type->is_vector()) {
  76                if (debug)
  77                   printf("kill %s (0x%01x - 0x%01x)\n", entry->lhs->name,
  78                          entry->available, used);
  79                entry->available &= ~used;
  80                if (!entry->available)
  81                   entry->remove();
  82             } else {
  83                if (debug)
  84                   printf("kill %s\n", entry->lhs->name);
  85                entry->remove();
  86             }
  87          }
  88       }
  89    }
  90
  91    virtual ir_visitor_status visit(ir_dereference_variable *ir)
  92    {
  93       kill_channels(ir->var, ~0);
  94
  95       return visit_continue;
  96    }
  97
  98    virtual ir_visitor_status visit(ir_swizzle *ir)
  99    {
 100       ir_dereference_variable *deref = ir->val->as_dereference_variable();
 101       if (!deref)
 102          return visit_continue;
 103
 104       int used = 0;
 105       used |= 1 << ir->mask.x;
 106       used |= 1 << ir->mask.y;
 107       used |= 1 << ir->mask.z;
 108       used |= 1 << ir->mask.w;
 109
 110       kill_channels(deref->var, used);
 111
 112       return visit_continue_with_parent;
 113    }
 114
 115 private:
 116    exec_list *assignments;
 117 };
 118
 119 class array_index_visit : public ir_hierarchical_visitor {
 120 public:
 121    array_index_visit(ir_hierarchical_visitor *v)
 122    {
 123       this->visitor = v;
 124    }
 125
 126    virtual ir_visitor_status visit_enter(class ir_dereference_array *ir)
 127    {
 128       ir->array_index->accept(visitor);
 129       return visit_continue;
 130    }
 131
 132    static void run(ir_instruction *ir, ir_hierarchical_visitor *v)
 133    {
 134       array_index_visit top_visit(v);
 135       ir->accept(& top_visit);
 136    }
 137
 138    ir_hierarchical_visitor *visitor;
 139 };
 140
 141
 142 /**
 143  * Adds an entry to the available copy list if it's a plain assignment
 144  * of a variable to a variable.
 145  */
 146 static bool
 147 process_assignment(void *ctx, ir_assignment *ir, exec_list *assignments)
 148 {
 149    ir_variable *var = NULL;
 150    bool progress = false;
 151    kill_for_derefs_visitor v(assignments);
 152
 153    /* Kill assignment entries for things used to produce this assignment. */
 154    ir->rhs->accept(&v);
 155    if (ir->condition) {
 156       ir->condition->accept(&v);
 157    }
 158
 159    /* Kill assignment enties used as array indices.
 160     */
 161    array_index_visit::run(ir->lhs, &v);
 162    var = ir->lhs->variable_referenced();
 163    assert(var);
 164
 165    /* Now, check if we did a whole-variable assignment. */
 166    if (!ir->condition) {
 167       ir_dereference_variable *deref_var = ir->lhs->as_dereference_variable();
 168
 169       /* If it's a vector type, we can do per-channel elimination of
 170        * use of the RHS.
 171        */
 172       if (deref_var && (deref_var->var->type->is_scalar() ||
 173                         deref_var->var->type->is_vector())) {
 174
 175          if (debug)
 176             printf("looking for %s.0x%01x to remove\n", var->name,
 177                    ir->write_mask);
 178
 179          foreach_iter(exec_list_iterator, iter, *assignments) {
 180             assignment_entry *entry = (assignment_entry *)iter.get();
 181
 182             if (entry->lhs != var)
 183                continue;
 184
 185             int remove = entry->available & ir->write_mask;
 186             if (debug) {
 187                printf("%s 0x%01x - 0x%01x = 0x%01x\n",
 188                       var->name,
 189                       entry->ir->write_mask,
 190                       remove, entry->ir->write_mask & ~remove);
 191             }
 192             if (remove) {
 193                progress = true;
 194
 195                if (debug) {
 196                   printf("rewriting:\n  ");
 197                   entry->ir->print();
 198                   printf("\n");
 199                }
 200
 201                entry->ir->write_mask &= ~remove;
 202                entry->available &= ~remove;
 203                if (entry->ir->write_mask == 0) {
 204                   /* Delete the dead assignment. */
 205                   entry->ir->remove();
 206                   entry->remove();
 207                } else {
 208                   void *mem_ctx = ralloc_parent(entry->ir);
 209                   /* Reswizzle the RHS arguments according to the new
 210                    * write_mask.
 211                    */
 212                   unsigned components[4];
 213                   unsigned channels = 0;
 214                   unsigned next = 0;
 215
 216                   for (int i = 0; i < 4; i++) {
 217                      if ((entry->ir->write_mask | remove) & (1 << i)) {
 218                         if (!(remove & (1 << i)))
 219                            components[channels++] = next;
 220                         next++;
 221                      }
 222                   }
 223
 224                   entry->ir->rhs = new(mem_ctx) ir_swizzle(entry->ir->rhs,
 225                                                            components,
 226                                                            channels);
 227                   if (debug) {
 228                      printf("to:\n  ");
 229                      entry->ir->print();
 230                      printf("\n");
 231                   }
 232                }
 233             }
 234          }
 235       } else if (ir->whole_variable_written() != NULL) {
 236          /* We did a whole-variable assignment.  So, any instruction in
 237           * the assignment list with the same LHS is dead.
 238           */
 239          if (debug)
 240             printf("looking for %s to remove\n", var->name);
 241          foreach_iter(exec_list_iterator, iter, *assignments) {
 242             assignment_entry *entry = (assignment_entry *)iter.get();
 243
 244             if (entry->lhs == var) {
 245                if (debug)
 246                   printf("removing %s\n", var->name);
 247                entry->ir->remove();
 248                entry->remove();
 249                progress = true;
 250             }
 251          }
 252       }
 253    }
 254
 255    /* Add this instruction to the assignment list available to be removed. */
 256    assignment_entry *entry = new(ctx) assignment_entry(var, ir);
 257    assignments->push_tail(entry);
 258
 259    if (debug) {
 260       printf("add %s\n", var->name);
 261
 262       printf("current entries\n");
 263       foreach_iter(exec_list_iterator, iter, *assignments) {
 264          assignment_entry *entry = (assignment_entry *)iter.get();
 265
 266          printf("    %s (0x%01x)\n", entry->lhs->name, entry->available);
 267       }
 268    }
 269
 270    return progress;
 271 }
 272
 273 static void
 274 dead_code_local_basic_block(ir_instruction *first,
 275                              ir_instruction *last,
 276                              void *data)
 277 {
 278    ir_instruction *ir, *ir_next;
 279    /* List of avaialble_copy */
 280    exec_list assignments;
 281    bool *out_progress = (bool *)data;
 282    bool progress = false;
 283
 284    void *ctx = ralloc_context(NULL);
 285    /* Safe looping, since process_assignment */
 286    for (ir = first, ir_next = (ir_instruction *)first->next;;
 287         ir = ir_next, ir_next = (ir_instruction *)ir->next) {
 288       ir_assignment *ir_assign = ir->as_assignment();
 289
 290       if (debug) {
 291          ir->print();
 292          printf("\n");
 293       }
 294
 295       if (ir_assign) {
 296          progress = process_assignment(ctx, ir_assign, &assignments) || progress;
 297       } else {
 298          kill_for_derefs_visitor kill(&assignments);
 299          ir->accept(&kill);
 300       }
 301
 302       if (ir == last)
 303          break;
 304    }
 305    *out_progress = progress;
 306    ralloc_free(ctx);
 307 }
 308
 309 /**
 310  * Does a copy propagation pass on the code present in the instruction stream.
 311  */
 312 bool
 313 do_dead_code_local(exec_list *instructions)
 314 {
 315    bool progress = false;
 316
 317    call_for_basic_blocks(instructions, dead_code_local_basic_block, &progress);
 318
 319    return progress;
 320 }