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