src/glsl/opt_function_inlining.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_function_inlining.cpp
  26  *
  27  * Replaces calls to functions with the body of the function.
  28  */
  29
  30 #include <inttypes.h>
  31 #include "ir.h"
  32 #include "ir_visitor.h"
  33 #include "ir_function_inlining.h"
  34 #include "ir_expression_flattening.h"
  35 #include "glsl_types.h"
  36 #include "program/hash_table.h"
  37
  38 static void
  39 do_sampler_replacement(exec_list *instructions,
  40                        ir_variable *sampler,
  41                        ir_dereference *deref);
  42
  43 class ir_function_inlining_visitor : public ir_hierarchical_visitor {
  44 public:
  45    ir_function_inlining_visitor()
  46    {
  47       progress = false;
  48    }
  49
  50    virtual ~ir_function_inlining_visitor()
  51    {
  52       /* empty */
  53    }
  54
  55    virtual ir_visitor_status visit_enter(ir_expression *);
  56    virtual ir_visitor_status visit_enter(ir_call *);
  57    virtual ir_visitor_status visit_enter(ir_assignment *);
  58    virtual ir_visitor_status visit_enter(ir_return *);
  59    virtual ir_visitor_status visit_enter(ir_texture *);
  60    virtual ir_visitor_status visit_enter(ir_swizzle *);
  61
  62    bool progress;
  63 };
  64
  65
  66 bool
  67 automatic_inlining_predicate(ir_instruction *ir)
  68 {
  69    ir_call *call = ir->as_call();
  70
  71    if (call && can_inline(call))
  72       return true;
  73
  74    return false;
  75 }
  76
  77 bool
  78 do_function_inlining(exec_list *instructions)
  79 {
  80    ir_function_inlining_visitor v;
  81
  82    do_expression_flattening(instructions, automatic_inlining_predicate);
  83
  84    v.run(instructions);
  85
  86    return v.progress;
  87 }
  88
  89 static void
  90 replace_return_with_assignment(ir_instruction *ir, void *data)
  91 {
  92    void *ctx = ralloc_parent(ir);
  93    ir_variable *retval = (ir_variable *)data;
  94    ir_return *ret = ir->as_return();
  95
  96    if (ret) {
  97       if (ret->value) {
  98          ir_rvalue *lhs = new(ctx) ir_dereference_variable(retval);
  99          ret->replace_with(new(ctx) ir_assignment(lhs, ret->value, NULL));
 100       } else {
 101          /* un-valued return has to be the last return, or we shouldn't
 102           * have reached here. (see can_inline()).
 103           */
 104          assert(ret->next->is_tail_sentinel());
 105          ret->remove();
 106       }
 107    }
 108 }
 109
 110 ir_rvalue *
 111 ir_call::generate_inline(ir_instruction *next_ir)
 112 {
 113    void *ctx = ralloc_parent(this);
 114    ir_variable **parameters;
 115    int num_parameters;
 116    int i;
 117    ir_variable *retval = NULL;
 118    struct hash_table *ht;
 119
 120    ht = hash_table_ctor(0, hash_table_pointer_hash, hash_table_pointer_compare);
 121
 122    num_parameters = 0;
 123    foreach_iter(exec_list_iterator, iter_sig, this->callee->parameters)
 124       num_parameters++;
 125
 126    parameters = new ir_variable *[num_parameters];
 127
 128    /* Generate storage for the return value. */
 129    if (!this->callee->return_type->is_void()) {
 130       retval = new(ctx) ir_variable(this->callee->return_type, "_ret_val",
 131                                     ir_var_auto);
 132       next_ir->insert_before(retval);
 133    }
 134
 135    /* Generate the declarations for the parameters to our inlined code,
 136     * and set up the mapping of real function body variables to ours.
 137     */
 138    i = 0;
 139    exec_list_iterator sig_param_iter = this->callee->parameters.iterator();
 140    exec_list_iterator param_iter = this->actual_parameters.iterator();
 141    for (i = 0; i < num_parameters; i++) {
 142       ir_variable *sig_param = (ir_variable *) sig_param_iter.get();
 143       ir_rvalue *param = (ir_rvalue *) param_iter.get();
 144
 145       /* Generate a new variable for the parameter. */
 146       if (sig_param->type->base_type == GLSL_TYPE_SAMPLER) {
 147          /* For samplers, we want the inlined sampler references
 148           * referencing the passed in sampler variable, since that
 149           * will have the location information, which an assignment of
 150           * a sampler wouldn't.  Fix it up below.
 151           */
 152          parameters[i] = NULL;
 153       } else {
 154          parameters[i] = sig_param->clone(ctx, ht);
 155          parameters[i]->mode = ir_var_auto;
 156
 157          /* Remove the read-only decoration becuase we're going to write
 158           * directly to this variable.  If the cloned variable is left
 159           * read-only and the inlined function is inside a loop, the loop
 160           * analysis code will get confused.
 161           */
 162          parameters[i]->read_only = false;
 163          next_ir->insert_before(parameters[i]);
 164       }
 165
 166       /* Move the actual param into our param variable if it's an 'in' type. */
 167       if (parameters[i] && (sig_param->mode == ir_var_in ||
 168                             sig_param->mode == ir_var_const_in ||
 169                             sig_param->mode == ir_var_inout)) {
 170          ir_assignment *assign;
 171
 172          assign = new(ctx) ir_assignment(new(ctx) ir_dereference_variable(parameters[i]),
 173                                          param, NULL);
 174          next_ir->insert_before(assign);
 175       }
 176
 177       sig_param_iter.next();
 178       param_iter.next();
 179    }
 180
 181    exec_list new_instructions;
 182
 183    /* Generate the inlined body of the function to a new list */
 184    foreach_iter(exec_list_iterator, iter, callee->body) {
 185       ir_instruction *ir = (ir_instruction *)iter.get();
 186       ir_instruction *new_ir = ir->clone(ctx, ht);
 187
 188       new_instructions.push_tail(new_ir);
 189       visit_tree(new_ir, replace_return_with_assignment, retval);
 190    }
 191
 192    /* If any samplers were passed in, replace any deref of the sampler
 193     * with a deref of the sampler argument.
 194     */
 195    param_iter = this->actual_parameters.iterator();
 196    sig_param_iter = this->callee->parameters.iterator();
 197    for (i = 0; i < num_parameters; i++) {
 198       ir_instruction *const param = (ir_instruction *) param_iter.get();
 199       ir_variable *sig_param = (ir_variable *) sig_param_iter.get();
 200
 201       if (sig_param->type->base_type == GLSL_TYPE_SAMPLER) {
 202          ir_dereference *deref = param->as_dereference();
 203
 204          assert(deref);
 205          do_sampler_replacement(&new_instructions, sig_param, deref);
 206       }
 207       param_iter.next();
 208       sig_param_iter.next();
 209    }
 210
 211    /* Now push those new instructions in. */
 212    foreach_iter(exec_list_iterator, iter, new_instructions) {
 213       ir_instruction *ir = (ir_instruction *)iter.get();
 214       next_ir->insert_before(ir);
 215    }
 216
 217    /* Copy back the value of any 'out' parameters from the function body
 218     * variables to our own.
 219     */
 220    i = 0;
 221    param_iter = this->actual_parameters.iterator();
 222    sig_param_iter = this->callee->parameters.iterator();
 223    for (i = 0; i < num_parameters; i++) {
 224       ir_instruction *const param = (ir_instruction *) param_iter.get();
 225       const ir_variable *const sig_param = (ir_variable *) sig_param_iter.get();
 226
 227       /* Move our param variable into the actual param if it's an 'out' type. */
 228       if (parameters[i] && (sig_param->mode == ir_var_out ||
 229                             sig_param->mode == ir_var_inout)) {
 230          ir_assignment *assign;
 231
 232          assign = new(ctx) ir_assignment(param->clone(ctx, NULL)->as_rvalue(),
 233                                          new(ctx) ir_dereference_variable(parameters[i]),
 234                                          NULL);
 235          next_ir->insert_before(assign);
 236       }
 237
 238       param_iter.next();
 239       sig_param_iter.next();
 240    }
 241
 242    delete [] parameters;
 243
 244    hash_table_dtor(ht);
 245
 246    if (retval)
 247       return new(ctx) ir_dereference_variable(retval);
 248    else
 249       return NULL;
 250 }
 251
 252
 253 ir_visitor_status
 254 ir_function_inlining_visitor::visit_enter(ir_expression *ir)
 255 {
 256    (void) ir;
 257    return visit_continue_with_parent;
 258 }
 259
 260
 261 ir_visitor_status
 262 ir_function_inlining_visitor::visit_enter(ir_return *ir)
 263 {
 264    (void) ir;
 265    return visit_continue_with_parent;
 266 }
 267
 268
 269 ir_visitor_status
 270 ir_function_inlining_visitor::visit_enter(ir_texture *ir)
 271 {
 272    (void) ir;
 273    return visit_continue_with_parent;
 274 }
 275
 276
 277 ir_visitor_status
 278 ir_function_inlining_visitor::visit_enter(ir_swizzle *ir)
 279 {
 280    (void) ir;
 281    return visit_continue_with_parent;
 282 }
 283
 284
 285 ir_visitor_status
 286 ir_function_inlining_visitor::visit_enter(ir_call *ir)
 287 {
 288    if (can_inline(ir)) {
 289       /* If the call was part of some tree, then it should have been
 290        * flattened out or we shouldn't have seen it because of a
 291        * visit_continue_with_parent in this visitor.
 292        */
 293       assert(ir == base_ir);
 294
 295       (void) ir->generate_inline(ir);
 296       ir->remove();
 297       this->progress = true;
 298    }
 299
 300    return visit_continue;
 301 }
 302
 303
 304 ir_visitor_status
 305 ir_function_inlining_visitor::visit_enter(ir_assignment *ir)
 306 {
 307    ir_call *call = ir->rhs->as_call();
 308    if (!call || !can_inline(call))
 309       return visit_continue;
 310
 311    /* generates the parameter setup, function body, and returns the return
 312     * value of the function
 313     */
 314    ir_rvalue *rhs = call->generate_inline(ir);
 315    assert(rhs);
 316
 317    ir->rhs = rhs;
 318    this->progress = true;
 319
 320    return visit_continue;
 321 }
 322
 323 /**
 324  * Replaces references to the "sampler" variable with a clone of "deref."
 325  *
 326  * From the spec, samplers can appear in the tree as function
 327  * (non-out) parameters and as the result of array indexing and
 328  * structure field selection.  In our builtin implementation, they
 329  * also appear in the sampler field of an ir_tex instruction.
 330  */
 331
 332 class ir_sampler_replacement_visitor : public ir_hierarchical_visitor {
 333 public:
 334    ir_sampler_replacement_visitor(ir_variable *sampler, ir_dereference *deref)
 335    {
 336       this->sampler = sampler;
 337       this->deref = deref;
 338    }
 339
 340    virtual ~ir_sampler_replacement_visitor()
 341    {
 342    }
 343
 344    virtual ir_visitor_status visit_leave(ir_call *);
 345    virtual ir_visitor_status visit_leave(ir_dereference_array *);
 346    virtual ir_visitor_status visit_leave(ir_dereference_record *);
 347    virtual ir_visitor_status visit_leave(ir_texture *);
 348
 349    void replace_deref(ir_dereference **deref);
 350    void replace_rvalue(ir_rvalue **rvalue);
 351
 352    ir_variable *sampler;
 353    ir_dereference *deref;
 354 };
 355
 356 void
 357 ir_sampler_replacement_visitor::replace_deref(ir_dereference **deref)
 358 {
 359    ir_dereference_variable *deref_var = (*deref)->as_dereference_variable();
 360    if (deref_var && deref_var->var == this->sampler) {
 361       *deref = this->deref->clone(ralloc_parent(*deref), NULL);
 362    }
 363 }
 364
 365 void
 366 ir_sampler_replacement_visitor::replace_rvalue(ir_rvalue **rvalue)
 367 {
 368    if (!*rvalue)
 369       return;
 370
 371    ir_dereference *deref = (*rvalue)->as_dereference();
 372
 373    if (!deref)
 374       return;
 375
 376    replace_deref(&deref);
 377    *rvalue = deref;
 378 }
 379
 380 ir_visitor_status
 381 ir_sampler_replacement_visitor::visit_leave(ir_texture *ir)
 382 {
 383    replace_deref(&ir->sampler);
 384
 385    return visit_continue;
 386 }
 387
 388 ir_visitor_status
 389 ir_sampler_replacement_visitor::visit_leave(ir_dereference_array *ir)
 390 {
 391    replace_rvalue(&ir->array);
 392    return visit_continue;
 393 }
 394
 395 ir_visitor_status
 396 ir_sampler_replacement_visitor::visit_leave(ir_dereference_record *ir)
 397 {
 398    replace_rvalue(&ir->record);
 399    return visit_continue;
 400 }
 401
 402 ir_visitor_status
 403 ir_sampler_replacement_visitor::visit_leave(ir_call *ir)
 404 {
 405    foreach_iter(exec_list_iterator, iter, *ir) {
 406       ir_rvalue *param = (ir_rvalue *)iter.get();
 407       ir_rvalue *new_param = param;
 408       replace_rvalue(&new_param);
 409
 410       if (new_param != param) {
 411          param->replace_with(new_param);
 412       }
 413    }
 414    return visit_continue;
 415 }
 416
 417 static void
 418 do_sampler_replacement(exec_list *instructions,
 419                        ir_variable *sampler,
 420                        ir_dereference *deref)
 421 {
 422    ir_sampler_replacement_visitor v(sampler, deref);
 423
 424    visit_list_elements(&v, instructions);
 425 }