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_return *);
  57    virtual ir_visitor_status visit_enter(ir_texture *);
  58    virtual ir_visitor_status visit_enter(ir_swizzle *);
  59
  60    bool progress;
  61 };
  62
  63
  64 bool
  65 do_function_inlining(exec_list *instructions)
  66 {
  67    ir_function_inlining_visitor v;
  68
  69    v.run(instructions);
  70
  71    return v.progress;
  72 }
  73
  74 static void
  75 replace_return_with_assignment(ir_instruction *ir, void *data)
  76 {
  77    void *ctx = ralloc_parent(ir);
  78    ir_dereference *orig_deref = (ir_dereference *) data;
  79    ir_return *ret = ir->as_return();
  80
  81    if (ret) {
  82       if (ret->value) {
  83          ir_rvalue *lhs = orig_deref->clone(ctx, NULL);
  84          ret->replace_with(new(ctx) ir_assignment(lhs, ret->value, NULL));
  85       } else {
  86          /* un-valued return has to be the last return, or we shouldn't
  87           * have reached here. (see can_inline()).
  88           */
  89          assert(ret->next->is_tail_sentinel());
  90          ret->remove();
  91       }
  92    }
  93 }
  94
  95 void
  96 ir_call::generate_inline(ir_instruction *next_ir)
  97 {
  98    void *ctx = ralloc_parent(this);
  99    ir_variable **parameters;
 100    int num_parameters;
 101    int i;
 102    struct hash_table *ht;
 103
 104    ht = hash_table_ctor(0, hash_table_pointer_hash, hash_table_pointer_compare);
 105
 106    num_parameters = 0;
 107    foreach_iter(exec_list_iterator, iter_sig, this->callee->parameters)
 108       num_parameters++;
 109
 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    exec_list_iterator sig_param_iter = this->callee->parameters.iterator();
 117    exec_list_iterator param_iter = this->actual_parameters.iterator();
 118    for (i = 0; i < num_parameters; i++) {
 119       ir_variable *sig_param = (ir_variable *) sig_param_iter.get();
 120       ir_rvalue *param = (ir_rvalue *) param_iter.get();
 121
 122       /* Generate a new variable for the parameter. */
 123       if (sig_param->type->base_type == GLSL_TYPE_SAMPLER) {
 124          /* For samplers, we want the inlined sampler references
 125           * referencing the passed in sampler variable, since that
 126           * will have the location information, which an assignment of
 127           * a sampler wouldn't.  Fix it up below.
 128           */
 129          parameters[i] = NULL;
 130       } else {
 131          parameters[i] = sig_param->clone(ctx, ht);
 132          parameters[i]->mode = ir_var_auto;
 133
 134          /* Remove the read-only decoration becuase we're going to write
 135           * directly to this variable.  If the cloned variable is left
 136           * read-only and the inlined function is inside a loop, the loop
 137           * analysis code will get confused.
 138           */
 139          parameters[i]->read_only = false;
 140          next_ir->insert_before(parameters[i]);
 141       }
 142
 143       /* Move the actual param into our param variable if it's an 'in' type. */
 144       if (parameters[i] && (sig_param->mode == ir_var_in ||
 145                             sig_param->mode == ir_var_const_in ||
 146                             sig_param->mode == ir_var_inout)) {
 147          ir_assignment *assign;
 148
 149          assign = new(ctx) ir_assignment(new(ctx) ir_dereference_variable(parameters[i]),
 150                                          param, NULL);
 151          next_ir->insert_before(assign);
 152       }
 153
 154       sig_param_iter.next();
 155       param_iter.next();
 156    }
 157
 158    exec_list new_instructions;
 159
 160    /* Generate the inlined body of the function to a new list */
 161    foreach_iter(exec_list_iterator, iter, callee->body) {
 162       ir_instruction *ir = (ir_instruction *)iter.get();
 163       ir_instruction *new_ir = ir->clone(ctx, ht);
 164
 165       new_instructions.push_tail(new_ir);
 166       visit_tree(new_ir, replace_return_with_assignment, this->return_deref);
 167    }
 168
 169    /* If any samplers were passed in, replace any deref of the sampler
 170     * with a deref of the sampler argument.
 171     */
 172    param_iter = this->actual_parameters.iterator();
 173    sig_param_iter = this->callee->parameters.iterator();
 174    for (i = 0; i < num_parameters; i++) {
 175       ir_instruction *const param = (ir_instruction *) param_iter.get();
 176       ir_variable *sig_param = (ir_variable *) sig_param_iter.get();
 177
 178       if (sig_param->type->base_type == GLSL_TYPE_SAMPLER) {
 179          ir_dereference *deref = param->as_dereference();
 180
 181          assert(deref);
 182          do_sampler_replacement(&new_instructions, sig_param, deref);
 183       }
 184       param_iter.next();
 185       sig_param_iter.next();
 186    }
 187
 188    /* Now push those new instructions in. */
 189    next_ir->insert_before(&new_instructions);
 190
 191    /* Copy back the value of any 'out' parameters from the function body
 192     * variables to our own.
 193     */
 194    i = 0;
 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       const ir_variable *const sig_param = (ir_variable *) sig_param_iter.get();
 200
 201       /* Move our param variable into the actual param if it's an 'out' type. */
 202       if (parameters[i] && (sig_param->mode == ir_var_out ||
 203                             sig_param->mode == ir_var_inout)) {
 204          ir_assignment *assign;
 205
 206          assign = new(ctx) ir_assignment(param->clone(ctx, NULL)->as_rvalue(),
 207                                          new(ctx) ir_dereference_variable(parameters[i]),
 208                                          NULL);
 209          next_ir->insert_before(assign);
 210       }
 211
 212       param_iter.next();
 213       sig_param_iter.next();
 214    }
 215
 216    delete [] parameters;
 217
 218    hash_table_dtor(ht);
 219 }
 220
 221
 222 ir_visitor_status
 223 ir_function_inlining_visitor::visit_enter(ir_expression *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_return *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_texture *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_swizzle *ir)
 248 {
 249    (void) ir;
 250    return visit_continue_with_parent;
 251 }
 252
 253
 254 ir_visitor_status
 255 ir_function_inlining_visitor::visit_enter(ir_call *ir)
 256 {
 257    if (can_inline(ir)) {
 258       ir->generate_inline(ir);
 259       ir->remove();
 260       this->progress = true;
 261    }
 262
 263    return visit_continue;
 264 }
 265
 266
 267 /**
 268  * Replaces references to the "sampler" variable with a clone of "deref."
 269  *
 270  * From the spec, samplers can appear in the tree as function
 271  * (non-out) parameters and as the result of array indexing and
 272  * structure field selection.  In our builtin implementation, they
 273  * also appear in the sampler field of an ir_tex instruction.
 274  */
 275
 276 class ir_sampler_replacement_visitor : public ir_hierarchical_visitor {
 277 public:
 278    ir_sampler_replacement_visitor(ir_variable *sampler, ir_dereference *deref)
 279    {
 280       this->sampler = sampler;
 281       this->deref = deref;
 282    }
 283
 284    virtual ~ir_sampler_replacement_visitor()
 285    {
 286    }
 287
 288    virtual ir_visitor_status visit_leave(ir_call *);
 289    virtual ir_visitor_status visit_leave(ir_dereference_array *);
 290    virtual ir_visitor_status visit_leave(ir_dereference_record *);
 291    virtual ir_visitor_status visit_leave(ir_texture *);
 292
 293    void replace_deref(ir_dereference **deref);
 294    void replace_rvalue(ir_rvalue **rvalue);
 295
 296    ir_variable *sampler;
 297    ir_dereference *deref;
 298 };
 299
 300 void
 301 ir_sampler_replacement_visitor::replace_deref(ir_dereference **deref)
 302 {
 303    ir_dereference_variable *deref_var = (*deref)->as_dereference_variable();
 304    if (deref_var && deref_var->var == this->sampler) {
 305       *deref = this->deref->clone(ralloc_parent(*deref), NULL);
 306    }
 307 }
 308
 309 void
 310 ir_sampler_replacement_visitor::replace_rvalue(ir_rvalue **rvalue)
 311 {
 312    if (!*rvalue)
 313       return;
 314
 315    ir_dereference *deref = (*rvalue)->as_dereference();
 316
 317    if (!deref)
 318       return;
 319
 320    replace_deref(&deref);
 321    *rvalue = deref;
 322 }
 323
 324 ir_visitor_status
 325 ir_sampler_replacement_visitor::visit_leave(ir_texture *ir)
 326 {
 327    replace_deref(&ir->sampler);
 328
 329    return visit_continue;
 330 }
 331
 332 ir_visitor_status
 333 ir_sampler_replacement_visitor::visit_leave(ir_dereference_array *ir)
 334 {
 335    replace_rvalue(&ir->array);
 336    return visit_continue;
 337 }
 338
 339 ir_visitor_status
 340 ir_sampler_replacement_visitor::visit_leave(ir_dereference_record *ir)
 341 {
 342    replace_rvalue(&ir->record);
 343    return visit_continue;
 344 }
 345
 346 ir_visitor_status
 347 ir_sampler_replacement_visitor::visit_leave(ir_call *ir)
 348 {
 349    foreach_iter(exec_list_iterator, iter, *ir) {
 350       ir_rvalue *param = (ir_rvalue *)iter.get();
 351       ir_rvalue *new_param = param;
 352       replace_rvalue(&new_param);
 353
 354       if (new_param != param) {
 355          param->replace_with(new_param);
 356       }
 357    }
 358    return visit_continue;
 359 }
 360
 361 static void
 362 do_sampler_replacement(exec_list *instructions,
 363                        ir_variable *sampler,
 364                        ir_dereference *deref)
 365 {
 366    ir_sampler_replacement_visitor v(sampler, deref);
 367
 368    visit_list_elements(&v, instructions);
 369 }