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 namespace {
  44
  45 class assignment_entry : public exec_node
  46 {
  47 public:
  48    assignment_entry(ir_variable *lhs, ir_assignment *ir)
  49    {
  50       assert(lhs);
  51       assert(ir);
  52       this->lhs = lhs;
  53       this->ir = ir;
  54       this->available = ir->write_mask;
  55    }
  56
  57    ir_variable *lhs;
  58    ir_assignment *ir;
  59
  60    /* bitmask of xyzw channels written that haven't been used so far. */
  61    int available;
  62 };
  63
  64 class kill_for_derefs_visitor : public ir_hierarchical_visitor {
  65 public:
  66    kill_for_derefs_visitor(exec_list *assignments)
  67    {
  68       this->assignments = assignments;
  69    }
  70
  71    void kill_channels(ir_variable *const var, int used)
  72    {
  73       foreach_iter(exec_list_iterator, iter, *this->assignments) {
  74          assignment_entry *entry = (assignment_entry *)iter.get();
  75
  76          if (entry->lhs == var) {
  77             if (var->type->is_scalar() || var->type->is_vector()) {
  78                if (debug)
  79                   printf("kill %s (0x%01x - 0x%01x)\n", entry->lhs->name,
  80                          entry->available, used);
  81                entry->available &= ~used;
  82                if (!entry->available)
  83                   entry->remove();
  84             } else {
  85                if (debug)
  86                   printf("kill %s\n", entry->lhs->name);
  87                entry->remove();
  88             }
  89          }
  90       }
  91    }
  92
  93    virtual ir_visitor_status visit(ir_dereference_variable *ir)
  94    {
  95       kill_channels(ir->var, ~0);
  96
  97       return visit_continue;
  98    }
  99
 100    virtual ir_visitor_status visit(ir_swizzle *ir)
 101    {
 102       ir_dereference_variable *deref = ir->val->as_dereference_variable();
 103       if (!deref)
 104          return visit_continue;
 105
 106       int used = 0;
 107       used |= 1 << ir->mask.x;
 108       used |= 1 << ir->mask.y;
 109       used |= 1 << ir->mask.z;
 110       used |= 1 << ir->mask.w;
 111
 112       kill_channels(deref->var, used);
 113
 114       return visit_continue_with_parent;
 115    }
 116
 117    virtual ir_visitor_status visit(ir_emit_vertex *ir)
 118    {
 119       /* For the purpose of dead code elimination, emitting a vertex counts as
 120        * "reading" all of the currently assigned output variables.
 121        */
 122       foreach_iter(exec_list_iterator, iter, *this->assignments) {
 123          assignment_entry *entry = (assignment_entry *)iter.get();
 124          if (entry->lhs->mode == ir_var_shader_out) {
 125             if (debug)
 126                printf("kill %s\n", entry->lhs->name);
 127             entry->remove();
 128          }
 129       }
 130
 131       return visit_continue;
 132    }
 133
 134 private:
 135    exec_list *assignments;
 136 };
 137
 138 class array_index_visit : public ir_hierarchical_visitor {
 139 public:
 140    array_index_visit(ir_hierarchical_visitor *v)
 141    {
 142       this->visitor = v;
 143    }
 144
 145    virtual ir_visitor_status visit_enter(class ir_dereference_array *ir)
 146    {
 147       ir->array_index->accept(visitor);
 148       return visit_continue;
 149    }
 150
 151    static void run(ir_instruction *ir, ir_hierarchical_visitor *v)
 152    {
 153       array_index_visit top_visit(v);
 154       ir->accept(& top_visit);
 155    }
 156
 157    ir_hierarchical_visitor *visitor;
 158 };
 159
 160 } /* unnamed namespace */
 161
 162 /**
 163  * Adds an entry to the available copy list if it's a plain assignment
 164  * of a variable to a variable.
 165  */
 166 static bool
 167 process_assignment(void *ctx, ir_assignment *ir, exec_list *assignments)
 168 {
 169    ir_variable *var = NULL;
 170    bool progress = false;
 171    kill_for_derefs_visitor v(assignments);
 172
 173    /* Kill assignment entries for things used to produce this assignment. */
 174    ir->rhs->accept(&v);
 175    if (ir->condition) {
 176       ir->condition->accept(&v);
 177    }
 178
 179    /* Kill assignment enties used as array indices.
 180     */
 181    array_index_visit::run(ir->lhs, &v);
 182    var = ir->lhs->variable_referenced();
 183    assert(var);
 184
 185    /* Now, check if we did a whole-variable assignment. */
 186    if (!ir->condition) {
 187       ir_dereference_variable *deref_var = ir->lhs->as_dereference_variable();
 188
 189       /* If it's a vector type, we can do per-channel elimination of
 190        * use of the RHS.
 191        */
 192       if (deref_var && (deref_var->var->type->is_scalar() ||
 193                         deref_var->var->type->is_vector())) {
 194
 195          if (debug)
 196             printf("looking for %s.0x%01x to remove\n", var->name,
 197                    ir->write_mask);
 198
 199          foreach_iter(exec_list_iterator, iter, *assignments) {
 200             assignment_entry *entry = (assignment_entry *)iter.get();
 201
 202             if (entry->lhs != var)
 203                continue;
 204
 205             int remove = entry->available & ir->write_mask;
 206             if (debug) {
 207                printf("%s 0x%01x - 0x%01x = 0x%01x\n",
 208                       var->name,
 209                       entry->ir->write_mask,
 210                       remove, entry->ir->write_mask & ~remove);
 211             }
 212             if (remove) {
 213                progress = true;
 214
 215                if (debug) {
 216                   printf("rewriting:\n  ");
 217                   entry->ir->print();
 218                   printf("\n");
 219                }
 220
 221                entry->ir->write_mask &= ~remove;
 222                entry->available &= ~remove;
 223                if (entry->ir->write_mask == 0) {
 224                   /* Delete the dead assignment. */
 225                   entry->ir->remove();
 226                   entry->remove();
 227                } else {
 228                   void *mem_ctx = ralloc_parent(entry->ir);
 229                   /* Reswizzle the RHS arguments according to the new
 230                    * write_mask.
 231                    */
 232                   unsigned components[4];
 233                   unsigned channels = 0;
 234                   unsigned next = 0;
 235
 236                   for (int i = 0; i < 4; i++) {
 237                      if ((entry->ir->write_mask | remove) & (1 << i)) {
 238                         if (!(remove & (1 << i)))
 239                            components[channels++] = next;
 240                         next++;
 241                      }
 242                   }
 243
 244                   entry->ir->rhs = new(mem_ctx) ir_swizzle(entry->ir->rhs,
 245                                                            components,
 246                                                            channels);
 247                   if (debug) {
 248                      printf("to:\n  ");
 249                      entry->ir->print();
 250                      printf("\n");
 251                   }
 252                }
 253             }
 254          }
 255       } else if (ir->whole_variable_written() != NULL) {
 256          /* We did a whole-variable assignment.  So, any instruction in
 257           * the assignment list with the same LHS is dead.
 258           */
 259          if (debug)
 260             printf("looking for %s to remove\n", var->name);
 261          foreach_iter(exec_list_iterator, iter, *assignments) {
 262             assignment_entry *entry = (assignment_entry *)iter.get();
 263
 264             if (entry->lhs == var) {
 265                if (debug)
 266                   printf("removing %s\n", var->name);
 267                entry->ir->remove();
 268                entry->remove();
 269                progress = true;
 270             }
 271          }
 272       }
 273    }
 274
 275    /* Add this instruction to the assignment list available to be removed. */
 276    assignment_entry *entry = new(ctx) assignment_entry(var, ir);
 277    assignments->push_tail(entry);
 278
 279    if (debug) {
 280       printf("add %s\n", var->name);
 281
 282       printf("current entries\n");
 283       foreach_iter(exec_list_iterator, iter, *assignments) {
 284          assignment_entry *entry = (assignment_entry *)iter.get();
 285
 286          printf("    %s (0x%01x)\n", entry->lhs->name, entry->available);
 287       }
 288    }
 289
 290    return progress;
 291 }
 292
 293 static void
 294 dead_code_local_basic_block(ir_instruction *first,
 295                              ir_instruction *last,
 296                              void *data)
 297 {
 298    ir_instruction *ir, *ir_next;
 299    /* List of avaialble_copy */
 300    exec_list assignments;
 301    bool *out_progress = (bool *)data;
 302    bool progress = false;
 303
 304    void *ctx = ralloc_context(NULL);
 305    /* Safe looping, since process_assignment */
 306    for (ir = first, ir_next = (ir_instruction *)first->next;;
 307         ir = ir_next, ir_next = (ir_instruction *)ir->next) {
 308       ir_assignment *ir_assign = ir->as_assignment();
 309
 310       if (debug) {
 311          ir->print();
 312          printf("\n");
 313       }
 314
 315       if (ir_assign) {
 316          progress = process_assignment(ctx, ir_assign, &assignments) || progress;
 317       } else {
 318          kill_for_derefs_visitor kill(&assignments);
 319          ir->accept(&kill);
 320       }
 321
 322       if (ir == last)
 323          break;
 324    }
 325    *out_progress = progress;
 326    ralloc_free(ctx);
 327 }
 328
 329 /**
 330  * Does a copy propagation pass on the code present in the instruction stream.
 331  */
 332 bool
 333 do_dead_code_local(exec_list *instructions)
 334 {
 335    bool progress = false;
 336
 337    call_for_basic_blocks(instructions, dead_code_local_basic_block, &progress);
 338
 339    return progress;
 340 }