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