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_variable_replacement(exec_list *instructions,
  39                         ir_variable *orig,
  40                         ir_dereference *repl);
  41
  42 namespace {
  43
  44 class ir_function_inlining_visitor : public ir_hierarchical_visitor {
  45 public:
  46    ir_function_inlining_visitor()
  47    {
  48       progress = false;
  49    }
  50
  51    virtual ~ir_function_inlining_visitor()
  52    {
  53       /* empty */
  54    }
  55
  56    virtual ir_visitor_status visit_enter(ir_expression *);
  57    virtual ir_visitor_status visit_enter(ir_call *);
  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 } /* unnamed namespace */
  66
  67 bool
  68 do_function_inlining(exec_list *instructions)
  69 {
  70    ir_function_inlining_visitor v;
  71
  72    v.run(instructions);
  73
  74    return v.progress;
  75 }
  76
  77 static void
  78 replace_return_with_assignment(ir_instruction *ir, void *data)
  79 {
  80    void *ctx = ralloc_parent(ir);
  81    ir_dereference *orig_deref = (ir_dereference *) data;
  82    ir_return *ret = ir->as_return();
  83
  84    if (ret) {
  85       if (ret->value) {
  86          ir_rvalue *lhs = orig_deref->clone(ctx, NULL);
  87          ret->replace_with(new(ctx) ir_assignment(lhs, ret->value, NULL));
  88       } else {
  89          /* un-valued return has to be the last return, or we shouldn't
  90           * have reached here. (see can_inline()).
  91           */
  92          assert(ret->next->is_tail_sentinel());
  93          ret->remove();
  94       }
  95    }
  96 }
  97
  98 void
  99 ir_call::generate_inline(ir_instruction *next_ir)
 100 {
 101    void *ctx = ralloc_parent(this);
 102    ir_variable **parameters;
 103    int num_parameters;
 104    int i;
 105    struct hash_table *ht;
 106
 107    ht = hash_table_ctor(0, hash_table_pointer_hash, hash_table_pointer_compare);
 108
 109    num_parameters = 0;
 110    foreach_list(n, &this->callee->parameters)
 111       num_parameters++;
 112
 113    parameters = new ir_variable *[num_parameters];
 114
 115    /* Generate the declarations for the parameters to our inlined code,
 116     * and set up the mapping of real function body variables to ours.
 117     */
 118    i = 0;
 119    foreach_two_lists(formal_node, &this->callee->parameters,
 120                      actual_node, &this->actual_parameters) {
 121       ir_variable *sig_param = (ir_variable *) formal_node;
 122       ir_rvalue *param = (ir_rvalue *) actual_node;
 123
 124       /* Generate a new variable for the parameter. */
 125       if (sig_param->type->contains_opaque()) {
 126          /* For opaque types, we want the inlined variable references
 127           * referencing the passed in variable, since that will have
 128           * the location information, which an assignment of an opaque
 129           * variable wouldn't.  Fix it up below.
 130           */
 131          parameters[i] = NULL;
 132       } else {
 133          parameters[i] = sig_param->clone(ctx, ht);
 134          parameters[i]->data.mode = ir_var_auto;
 135
 136          /* Remove the read-only decoration becuase we're going to write
 137           * directly to this variable.  If the cloned variable is left
 138           * read-only and the inlined function is inside a loop, the loop
 139           * analysis code will get confused.
 140           */
 141          parameters[i]->data.read_only = false;
 142          next_ir->insert_before(parameters[i]);
 143       }
 144
 145       /* Move the actual param into our param variable if it's an 'in' type. */
 146       if (parameters[i] && (sig_param->data.mode == ir_var_function_in ||
 147                             sig_param->data.mode == ir_var_const_in ||
 148                             sig_param->data.mode == ir_var_function_inout)) {
 149          ir_assignment *assign;
 150
 151          assign = new(ctx) ir_assignment(new(ctx) ir_dereference_variable(parameters[i]),
 152                                          param, NULL);
 153          next_ir->insert_before(assign);
 154       }
 155
 156       ++i;
 157    }
 158
 159    exec_list new_instructions;
 160
 161    /* Generate the inlined body of the function to a new list */
 162    foreach_list(n, &callee->body) {
 163       ir_instruction *ir = (ir_instruction *) n;
 164       ir_instruction *new_ir = ir->clone(ctx, ht);
 165
 166       new_instructions.push_tail(new_ir);
 167       visit_tree(new_ir, replace_return_with_assignment, this->return_deref);
 168    }
 169
 170    /* If any opaque types were passed in, replace any deref of the
 171     * opaque variable with a deref of the argument.
 172     */
 173    foreach_two_lists(formal_node, &this->callee->parameters,
 174                      actual_node, &this->actual_parameters) {
 175       ir_rvalue *const param = (ir_rvalue *) actual_node;
 176       ir_variable *sig_param = (ir_variable *) formal_node;
 177
 178       if (sig_param->type->contains_opaque()) {
 179          ir_dereference *deref = param->as_dereference();
 180
 181          assert(deref);
 182          do_variable_replacement(&new_instructions, sig_param, deref);
 183       }
 184    }
 185
 186    /* Now push those new instructions in. */
 187    next_ir->insert_before(&new_instructions);
 188
 189    /* Copy back the value of any 'out' parameters from the function body
 190     * variables to our own.
 191     */
 192    i = 0;
 193    foreach_two_lists(formal_node, &this->callee->parameters,
 194                      actual_node, &this->actual_parameters) {
 195       ir_rvalue *const param = (ir_rvalue *) actual_node;
 196       const ir_variable *const sig_param = (ir_variable *) formal_node;
 197
 198       /* Move our param variable into the actual param if it's an 'out' type. */
 199       if (parameters[i] && (sig_param->data.mode == ir_var_function_out ||
 200                             sig_param->data.mode == ir_var_function_inout)) {
 201          ir_assignment *assign;
 202
 203          assign = new(ctx) ir_assignment(param->clone(ctx, NULL)->as_rvalue(),
 204                                          new(ctx) ir_dereference_variable(parameters[i]),
 205                                          NULL);
 206          next_ir->insert_before(assign);
 207       }
 208
 209       ++i;
 210    }
 211
 212    delete [] parameters;
 213
 214    hash_table_dtor(ht);
 215 }
 216
 217
 218 ir_visitor_status
 219 ir_function_inlining_visitor::visit_enter(ir_expression *ir)
 220 {
 221    (void) ir;
 222    return visit_continue_with_parent;
 223 }
 224
 225
 226 ir_visitor_status
 227 ir_function_inlining_visitor::visit_enter(ir_return *ir)
 228 {
 229    (void) ir;
 230    return visit_continue_with_parent;
 231 }
 232
 233
 234 ir_visitor_status
 235 ir_function_inlining_visitor::visit_enter(ir_texture *ir)
 236 {
 237    (void) ir;
 238    return visit_continue_with_parent;
 239 }
 240
 241
 242 ir_visitor_status
 243 ir_function_inlining_visitor::visit_enter(ir_swizzle *ir)
 244 {
 245    (void) ir;
 246    return visit_continue_with_parent;
 247 }
 248
 249
 250 ir_visitor_status
 251 ir_function_inlining_visitor::visit_enter(ir_call *ir)
 252 {
 253    if (can_inline(ir)) {
 254       ir->generate_inline(ir);
 255       ir->remove();
 256       this->progress = true;
 257    }
 258
 259    return visit_continue;
 260 }
 261
 262
 263 /**
 264  * Replaces references to the "orig" variable with a clone of "repl."
 265  *
 266  * From the spec, opaque types can appear in the tree as function
 267  * (non-out) parameters and as the result of array indexing and
 268  * structure field selection.  In our builtin implementation, they
 269  * also appear in the sampler field of an ir_tex instruction.
 270  */
 271
 272 class ir_variable_replacement_visitor : public ir_hierarchical_visitor {
 273 public:
 274    ir_variable_replacement_visitor(ir_variable *orig, ir_dereference *repl)
 275    {
 276       this->orig = orig;
 277       this->repl = repl;
 278    }
 279
 280    virtual ~ir_variable_replacement_visitor()
 281    {
 282    }
 283
 284    virtual ir_visitor_status visit_leave(ir_call *);
 285    virtual ir_visitor_status visit_leave(ir_dereference_array *);
 286    virtual ir_visitor_status visit_leave(ir_dereference_record *);
 287    virtual ir_visitor_status visit_leave(ir_texture *);
 288
 289    void replace_deref(ir_dereference **deref);
 290    void replace_rvalue(ir_rvalue **rvalue);
 291
 292    ir_variable *orig;
 293    ir_dereference *repl;
 294 };
 295
 296 void
 297 ir_variable_replacement_visitor::replace_deref(ir_dereference **deref)
 298 {
 299    ir_dereference_variable *deref_var = (*deref)->as_dereference_variable();
 300    if (deref_var && deref_var->var == this->orig) {
 301       *deref = this->repl->clone(ralloc_parent(*deref), NULL);
 302    }
 303 }
 304
 305 void
 306 ir_variable_replacement_visitor::replace_rvalue(ir_rvalue **rvalue)
 307 {
 308    if (!*rvalue)
 309       return;
 310
 311    ir_dereference *deref = (*rvalue)->as_dereference();
 312
 313    if (!deref)
 314       return;
 315
 316    replace_deref(&deref);
 317    *rvalue = deref;
 318 }
 319
 320 ir_visitor_status
 321 ir_variable_replacement_visitor::visit_leave(ir_texture *ir)
 322 {
 323    replace_deref(&ir->sampler);
 324
 325    return visit_continue;
 326 }
 327
 328 ir_visitor_status
 329 ir_variable_replacement_visitor::visit_leave(ir_dereference_array *ir)
 330 {
 331    replace_rvalue(&ir->array);
 332    return visit_continue;
 333 }
 334
 335 ir_visitor_status
 336 ir_variable_replacement_visitor::visit_leave(ir_dereference_record *ir)
 337 {
 338    replace_rvalue(&ir->record);
 339    return visit_continue;
 340 }
 341
 342 ir_visitor_status
 343 ir_variable_replacement_visitor::visit_leave(ir_call *ir)
 344 {
 345    foreach_list_safe(n, &ir->actual_parameters) {
 346       ir_rvalue *param = (ir_rvalue *) n;
 347       ir_rvalue *new_param = param;
 348       replace_rvalue(&new_param);
 349
 350       if (new_param != param) {
 351          param->replace_with(new_param);
 352       }
 353    }
 354    return visit_continue;
 355 }
 356
 357 static void
 358 do_variable_replacement(exec_list *instructions,
 359                         ir_variable *orig,
 360                         ir_dereference *repl)
 361 {
 362    ir_variable_replacement_visitor v(orig, repl);
 363
 364    visit_list_elements(&v, instructions);
 365 }