src/glsl/ir_constant_propagation.cpp

   1 /*
   2  * Constantright © 2010 Intel Corporation
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * constant 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, constant, 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 constantright 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 CONSTANTRIGHT 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_constant_propagation.cpp
  26  *
  27  * Tracks assignments of constants to channels of variables, and
  28  * usage of those constant channels with direct usage of the constants.
  29  *
  30  * This can lead to constant folding and algebraic optimizations in
  31  * those later expressions, while causing no increase in instruction
  32  * count (due to constants being generally free to load from a
  33  * constant push buffer or as instruction immediate values) and
  34  * possibly reducing register pressure.
  35  */
  36
  37 #include "ir.h"
  38 #include "ir_visitor.h"
  39 #include "ir_rvalue_visitor.h"
  40 #include "ir_basic_block.h"
  41 #include "ir_optimization.h"
  42 #include "glsl_types.h"
  43
  44 class acp_entry : public exec_node
  45 {
  46 public:
  47    acp_entry(ir_variable *var, unsigned write_mask, ir_constant *constant)
  48    {
  49       assert(var);
  50       assert(constant);
  51       this->var = var;
  52       this->write_mask = write_mask;
  53       this->constant = constant;
  54    }
  55
  56    ir_variable *var;
  57    ir_constant *constant;
  58    unsigned write_mask;
  59 };
  60
  61
  62 class kill_entry : public exec_node
  63 {
  64 public:
  65    kill_entry(ir_variable *var, unsigned write_mask)
  66    {
  67       assert(var);
  68       this->var = var;
  69       this->write_mask = write_mask;
  70    }
  71
  72    ir_variable *var;
  73    unsigned write_mask;
  74 };
  75
  76 class ir_constant_propagation_visitor : public ir_rvalue_visitor {
  77 public:
  78    ir_constant_propagation_visitor()
  79    {
  80       progress = false;
  81       mem_ctx = talloc_new(0);
  82       this->acp = new(mem_ctx) exec_list;
  83       this->kills = new(mem_ctx) exec_list;
  84    }
  85    ~ir_constant_propagation_visitor()
  86    {
  87       talloc_free(mem_ctx);
  88    }
  89
  90    virtual ir_visitor_status visit_enter(class ir_loop *);
  91    virtual ir_visitor_status visit_enter(class ir_function_signature *);
  92    virtual ir_visitor_status visit_enter(class ir_function *);
  93    virtual ir_visitor_status visit_enter(class ir_assignment *);
  94    virtual ir_visitor_status visit_enter(class ir_call *);
  95    virtual ir_visitor_status visit_enter(class ir_if *);
  96
  97    void add_constant(ir_assignment *ir);
  98    void kill(ir_variable *ir, unsigned write_mask);
  99    void handle_if_block(exec_list *instructions);
 100    void handle_rvalue(ir_rvalue **rvalue);
 101
 102    /** List of acp_entry: The available constants to propagate */
 103    exec_list *acp;
 104
 105    /**
 106     * List of kill_entry: The masks of variables whose values were
 107     * killed in this block.
 108     */
 109    exec_list *kills;
 110
 111    bool progress;
 112
 113    bool killed_all;
 114
 115    void *mem_ctx;
 116 };
 117
 118
 119 void
 120 ir_constant_propagation_visitor::handle_rvalue(ir_rvalue **rvalue)
 121 {
 122    if (!*rvalue)
 123       return;
 124
 125    const glsl_type *type = (*rvalue)->type;
 126    if (!type->is_scalar() && !type->is_vector())
 127       return;
 128
 129    ir_swizzle *swiz = NULL;
 130    ir_dereference_variable *deref = (*rvalue)->as_dereference_variable();
 131    if (!deref) {
 132       swiz = (*rvalue)->as_swizzle();
 133       if (!swiz)
 134          return;
 135
 136       deref = swiz->val->as_dereference_variable();
 137       if (!deref)
 138          return;
 139    }
 140
 141    ir_constant_data data;
 142    memset(&data, 0, sizeof(data));
 143
 144    for (unsigned int i = 0; i < type->components(); i++) {
 145       int channel;
 146       acp_entry *found = NULL;
 147
 148       if (swiz) {
 149          switch (i) {
 150          case 0: channel = swiz->mask.x; break;
 151          case 1: channel = swiz->mask.y; break;
 152          case 2: channel = swiz->mask.z; break;
 153          case 3: channel = swiz->mask.w; break;
 154          default: assert(!"shouldn't be reached"); channel = 0; break;
 155          }
 156       } else {
 157          channel = i;
 158       }
 159
 160       foreach_iter(exec_list_iterator, iter, *this->acp) {
 161          acp_entry *entry = (acp_entry *)iter.get();
 162          if (entry->var == deref->var && entry->write_mask & (1 << channel)) {
 163             found = entry;
 164             break;
 165          }
 166       }
 167
 168       if (!found)
 169          return;
 170
 171       switch (type->base_type) {
 172       case GLSL_TYPE_FLOAT:
 173          data.f[i] = found->constant->value.f[channel];
 174          break;
 175       case GLSL_TYPE_INT:
 176          data.i[i] = found->constant->value.i[channel];
 177          break;
 178       case GLSL_TYPE_UINT:
 179          data.u[i] = found->constant->value.u[channel];
 180          break;
 181       case GLSL_TYPE_BOOL:
 182          data.b[i] = found->constant->value.b[channel];
 183          break;
 184       default:
 185          assert(!"not reached");
 186          break;
 187       }
 188    }
 189
 190    *rvalue = new(talloc_parent(deref)) ir_constant(type, &data);
 191    this->progress = true;
 192 }
 193
 194 ir_visitor_status
 195 ir_constant_propagation_visitor::visit_enter(ir_function_signature *ir)
 196 {
 197    /* Treat entry into a function signature as a completely separate
 198     * block.  Any instructions at global scope will be shuffled into
 199     * main() at link time, so they're irrelevant to us.
 200     */
 201    exec_list *orig_acp = this->acp;
 202    exec_list *orig_kills = this->kills;
 203    bool orig_killed_all = this->killed_all;
 204
 205    this->acp = new(mem_ctx) exec_list;
 206    this->kills = new(mem_ctx) exec_list;
 207    this->killed_all = false;
 208
 209    visit_list_elements(this, &ir->body);
 210
 211    this->kills = orig_kills;
 212    this->acp = orig_acp;
 213    this->killed_all = orig_killed_all;
 214
 215    return visit_continue_with_parent;
 216 }
 217
 218 ir_visitor_status
 219 ir_constant_propagation_visitor::visit_enter(ir_assignment *ir)
 220 {
 221    /* Inline accepting children, skipping the LHS. */
 222    ir->rhs->accept(this);
 223    handle_rvalue(&ir->rhs);
 224
 225    if (ir->condition) {
 226       ir->condition->accept(this);
 227       handle_rvalue(&ir->condition);
 228    }
 229
 230    kill(ir->lhs->variable_referenced(), ir->write_mask);
 231
 232    add_constant(ir);
 233
 234    return visit_continue_with_parent;
 235 }
 236
 237 ir_visitor_status
 238 ir_constant_propagation_visitor::visit_enter(ir_function *ir)
 239 {
 240    (void) ir;
 241    return visit_continue;
 242 }
 243
 244 ir_visitor_status
 245 ir_constant_propagation_visitor::visit_enter(ir_call *ir)
 246 {
 247    /* Do constant propagation on call parameters, but skip any out params */
 248    exec_list_iterator sig_param_iter = ir->get_callee()->parameters.iterator();
 249    foreach_iter(exec_list_iterator, iter, ir->actual_parameters) {
 250       ir_variable *sig_param = (ir_variable *)sig_param_iter.get();
 251       ir_rvalue *param = (ir_rvalue *)iter.get();
 252       if (sig_param->mode != ir_var_out && sig_param->mode != ir_var_inout) {
 253          ir_rvalue *new_param = param;
 254          handle_rvalue(&new_param);
 255          if (new_param != param)
 256             param->replace_with(new_param);
 257          else
 258             param->accept(this);
 259       }
 260       sig_param_iter.next();
 261    }
 262
 263    /* Since we're unlinked, we don't (necssarily) know the side effects of
 264     * this call.  So kill all copies.
 265     */
 266    acp->make_empty();
 267    this->killed_all = true;
 268
 269    return visit_continue_with_parent;
 270 }
 271
 272 void
 273 ir_constant_propagation_visitor::handle_if_block(exec_list *instructions)
 274 {
 275    exec_list *orig_acp = this->acp;
 276    exec_list *orig_kills = this->kills;
 277    bool orig_killed_all = this->killed_all;
 278
 279    this->acp = new(mem_ctx) exec_list;
 280    this->kills = new(mem_ctx) exec_list;
 281    this->killed_all = false;
 282
 283    /* Populate the initial acp with a constant of the original */
 284    foreach_iter(exec_list_iterator, iter, *orig_acp) {
 285       acp_entry *a = (acp_entry *)iter.get();
 286       this->acp->push_tail(new(this->mem_ctx) acp_entry(a->var, a->write_mask,
 287                                                         a->constant));
 288    }
 289
 290    visit_list_elements(this, instructions);
 291
 292    if (this->killed_all) {
 293       orig_acp->make_empty();
 294    }
 295
 296    exec_list *new_kills = this->kills;
 297    this->kills = orig_kills;
 298    this->acp = orig_acp;
 299    this->killed_all = this->killed_all || orig_killed_all;
 300
 301    foreach_iter(exec_list_iterator, iter, *new_kills) {
 302       kill_entry *k = (kill_entry *)iter.get();
 303       kill(k->var, k->write_mask);
 304    }
 305 }
 306
 307 ir_visitor_status
 308 ir_constant_propagation_visitor::visit_enter(ir_if *ir)
 309 {
 310    ir->condition->accept(this);
 311    handle_rvalue(&ir->condition);
 312
 313    handle_if_block(&ir->then_instructions);
 314    handle_if_block(&ir->else_instructions);
 315
 316    /* handle_if_block() already descended into the children. */
 317    return visit_continue_with_parent;
 318 }
 319
 320 ir_visitor_status
 321 ir_constant_propagation_visitor::visit_enter(ir_loop *ir)
 322 {
 323    exec_list *orig_acp = this->acp;
 324    exec_list *orig_kills = this->kills;
 325    bool orig_killed_all = this->killed_all;
 326
 327    /* FINISHME: For now, the initial acp for loops is totally empty.
 328     * We could go through once, then go through again with the acp
 329     * cloned minus the killed entries after the first run through.
 330     */
 331    this->acp = new(mem_ctx) exec_list;
 332    this->kills = new(mem_ctx) exec_list;
 333    this->killed_all = false;
 334
 335    visit_list_elements(this, &ir->body_instructions);
 336
 337    if (this->killed_all) {
 338       orig_acp->make_empty();
 339    }
 340
 341    exec_list *new_kills = this->kills;
 342    this->kills = orig_kills;
 343    this->acp = orig_acp;
 344    this->killed_all = this->killed_all || orig_killed_all;
 345
 346    foreach_iter(exec_list_iterator, iter, *new_kills) {
 347       kill_entry *k = (kill_entry *)iter.get();
 348       kill(k->var, k->write_mask);
 349    }
 350
 351    /* already descended into the children. */
 352    return visit_continue_with_parent;
 353 }
 354
 355 void
 356 ir_constant_propagation_visitor::kill(ir_variable *var, unsigned write_mask)
 357 {
 358    assert(var != NULL);
 359
 360    /* We don't track non-vectors. */
 361    if (!var->type->is_vector() && !var->type->is_scalar())
 362       return;
 363
 364    /* Remove any entries currently in the ACP for this kill. */
 365    foreach_iter(exec_list_iterator, iter, *this->acp) {
 366       acp_entry *entry = (acp_entry *)iter.get();
 367
 368       if (entry->var == var) {
 369          entry->write_mask &= ~write_mask;
 370          if (entry->write_mask == 0)
 371             entry->remove();
 372       }
 373    }
 374
 375    /* Add this writemask of the variable to the list of killed
 376     * variables in this block.
 377     */
 378    foreach_iter(exec_list_iterator, iter, *this->kills) {
 379       kill_entry *entry = (kill_entry *)iter.get();
 380
 381       if (entry->var == var) {
 382          entry->write_mask |= write_mask;
 383          return;
 384       }
 385    }
 386    /* Not already in the list.  Make new entry. */
 387    this->kills->push_tail(new(this->mem_ctx) kill_entry(var, write_mask));
 388 }
 389
 390 /**
 391  * Adds an entry to the available constant list if it's a plain assignment
 392  * of a variable to a variable.
 393  */
 394 void
 395 ir_constant_propagation_visitor::add_constant(ir_assignment *ir)
 396 {
 397    acp_entry *entry;
 398
 399    if (ir->condition) {
 400       ir_constant *condition = ir->condition->as_constant();
 401       if (!condition || !condition->value.b[0])
 402          return;
 403    }
 404
 405    if (!ir->write_mask)
 406       return;
 407
 408    ir_dereference_variable *deref = ir->lhs->as_dereference_variable();
 409    ir_constant *constant = ir->rhs->as_constant();
 410
 411    if (!deref || !constant)
 412       return;
 413
 414    /* Only do constant propagation on vectors.  Constant matrices,
 415     * arrays, or structures would require more work elsewhere.
 416     */
 417    if (!deref->var->type->is_vector() && !deref->var->type->is_scalar())
 418       return;
 419
 420    entry = new(this->mem_ctx) acp_entry(deref->var, ir->write_mask, constant);
 421    this->acp->push_tail(entry);
 422 }
 423
 424 /**
 425  * Does a constant propagation pass on the code present in the instruction stream.
 426  */
 427 bool
 428 do_constant_propagation(exec_list *instructions)
 429 {
 430    ir_constant_propagation_visitor v;
 431
 432    visit_list_elements(&v, instructions);
 433
 434    return v.progress;
 435 }