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 = talloc_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 = talloc_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) {
 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_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    foreach_iter(exec_list_iterator, iter, new_instructions) {
 212       ir_instruction *ir = (ir_instruction *)iter.get();
 213       next_ir->insert_before(ir);
 214    }
 215
 216    /* Copy back the value of any 'out' parameters from the function body
 217     * variables to our own.
 218     */
 219    i = 0;
 220    param_iter = this->actual_parameters.iterator();
 221    sig_param_iter = this->callee->parameters.iterator();
 222    for (i = 0; i < num_parameters; i++) {
 223       ir_instruction *const param = (ir_instruction *) param_iter.get();
 224       const ir_variable *const sig_param = (ir_variable *) sig_param_iter.get();
 225
 226       /* Move our param variable into the actual param if it's an 'out' type. */
 227       if (parameters[i] && (sig_param->mode == ir_var_out ||
 228                             sig_param->mode == ir_var_inout)) {
 229          ir_assignment *assign;
 230
 231          assign = new(ctx) ir_assignment(param->clone(ctx, NULL)->as_rvalue(),
 232                                          new(ctx) ir_dereference_variable(parameters[i]),
 233                                          NULL);
 234          next_ir->insert_before(assign);
 235       }
 236
 237       param_iter.next();
 238       sig_param_iter.next();
 239    }
 240
 241    delete [] parameters;
 242
 243    hash_table_dtor(ht);
 244
 245    if (retval)
 246       return new(ctx) ir_dereference_variable(retval);
 247    else
 248       return NULL;
 249 }
 250
 251
 252 ir_visitor_status
 253 ir_function_inlining_visitor::visit_enter(ir_expression *ir)
 254 {
 255    (void) ir;
 256    return visit_continue_with_parent;
 257 }
 258
 259
 260 ir_visitor_status
 261 ir_function_inlining_visitor::visit_enter(ir_return *ir)
 262 {
 263    (void) ir;
 264    return visit_continue_with_parent;
 265 }
 266
 267
 268 ir_visitor_status
 269 ir_function_inlining_visitor::visit_enter(ir_texture *ir)
 270 {
 271    (void) ir;
 272    return visit_continue_with_parent;
 273 }
 274
 275
 276 ir_visitor_status
 277 ir_function_inlining_visitor::visit_enter(ir_swizzle *ir)
 278 {
 279    (void) ir;
 280    return visit_continue_with_parent;
 281 }
 282
 283
 284 ir_visitor_status
 285 ir_function_inlining_visitor::visit_enter(ir_call *ir)
 286 {
 287    if (can_inline(ir)) {
 288       /* If the call was part of some tree, then it should have been
 289        * flattened out or we shouldn't have seen it because of a
 290        * visit_continue_with_parent in this visitor.
 291        */
 292       assert(ir == base_ir);
 293
 294       (void) ir->generate_inline(ir);
 295       ir->remove();
 296       this->progress = true;
 297    }
 298
 299    return visit_continue;
 300 }
 301
 302
 303 ir_visitor_status
 304 ir_function_inlining_visitor::visit_enter(ir_assignment *ir)
 305 {
 306    ir_call *call = ir->rhs->as_call();
 307    if (!call || !can_inline(call))
 308       return visit_continue;
 309
 310    /* generates the parameter setup, function body, and returns the return
 311     * value of the function
 312     */
 313    ir_rvalue *rhs = call->generate_inline(ir);
 314    assert(rhs);
 315
 316    ir->rhs = rhs;
 317    this->progress = true;
 318
 319    return visit_continue;
 320 }
 321
 322 /**
 323  * Replaces references to the "sampler" variable with a clone of "deref."
 324  *
 325  * From the spec, samplers can appear in the tree as function
 326  * (non-out) parameters and as the result of array indexing and
 327  * structure field selection.  In our builtin implementation, they
 328  * also appear in the sampler field of an ir_tex instruction.
 329  */
 330
 331 class ir_sampler_replacement_visitor : public ir_hierarchical_visitor {
 332 public:
 333    ir_sampler_replacement_visitor(ir_variable *sampler, ir_dereference *deref)
 334    {
 335       this->sampler = sampler;
 336       this->deref = deref;
 337    }
 338
 339    virtual ~ir_sampler_replacement_visitor()
 340    {
 341    }
 342
 343    virtual ir_visitor_status visit_leave(ir_call *);
 344    virtual ir_visitor_status visit_leave(ir_dereference_array *);
 345    virtual ir_visitor_status visit_leave(ir_dereference_record *);
 346    virtual ir_visitor_status visit_leave(ir_texture *);
 347
 348    void replace_deref(ir_dereference **deref);
 349    void replace_rvalue(ir_rvalue **rvalue);
 350
 351    ir_variable *sampler;
 352    ir_dereference *deref;
 353 };
 354
 355 void
 356 ir_sampler_replacement_visitor::replace_deref(ir_dereference **deref)
 357 {
 358    ir_dereference_variable *deref_var = (*deref)->as_dereference_variable();
 359    if (deref_var && deref_var->var == this->sampler) {
 360       *deref = this->deref->clone(talloc_parent(*deref), NULL);
 361    }
 362 }
 363
 364 void
 365 ir_sampler_replacement_visitor::replace_rvalue(ir_rvalue **rvalue)
 366 {
 367    if (!*rvalue)
 368       return;
 369
 370    ir_dereference *deref = (*rvalue)->as_dereference();
 371
 372    if (!deref)
 373       return;
 374
 375    replace_deref(&deref);
 376    *rvalue = deref;
 377 }
 378
 379 ir_visitor_status
 380 ir_sampler_replacement_visitor::visit_leave(ir_texture *ir)
 381 {
 382    replace_deref(&ir->sampler);
 383
 384    return visit_continue;
 385 }
 386
 387 ir_visitor_status
 388 ir_sampler_replacement_visitor::visit_leave(ir_dereference_array *ir)
 389 {
 390    replace_rvalue(&ir->array);
 391    return visit_continue;
 392 }
 393
 394 ir_visitor_status
 395 ir_sampler_replacement_visitor::visit_leave(ir_dereference_record *ir)
 396 {
 397    replace_rvalue(&ir->record);
 398    return visit_continue;
 399 }
 400
 401 ir_visitor_status
 402 ir_sampler_replacement_visitor::visit_leave(ir_call *ir)
 403 {
 404    foreach_iter(exec_list_iterator, iter, *ir) {
 405       ir_rvalue *param = (ir_rvalue *)iter.get();
 406       ir_rvalue *new_param = param;
 407       replace_rvalue(&new_param);
 408
 409       if (new_param != param) {
 410          param->replace_with(new_param);
 411       }
 412    }
 413    return visit_continue;
 414 }
 415
 416 static void
 417 do_sampler_replacement(exec_list *instructions,
 418                        ir_variable *sampler,
 419                        ir_dereference *deref)
 420 {
 421    ir_sampler_replacement_visitor v(sampler, deref);
 422
 423    visit_list_elements(&v, instructions);
 424 }