src/glsl/ir_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 ir_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 class ir_function_inlining_visitor : public ir_hierarchical_visitor {
  39 public:
  40    ir_function_inlining_visitor()
  41    {
  42       progress = false;
  43    }
  44
  45    virtual ~ir_function_inlining_visitor()
  46    {
  47       /* empty */
  48    }
  49
  50    virtual ir_visitor_status visit_enter(ir_expression *);
  51    virtual ir_visitor_status visit_enter(ir_call *);
  52    virtual ir_visitor_status visit_enter(ir_assignment *);
  53    virtual ir_visitor_status visit_enter(ir_return *);
  54    virtual ir_visitor_status visit_enter(ir_texture *);
  55    virtual ir_visitor_status visit_enter(ir_swizzle *);
  56
  57    bool progress;
  58 };
  59
  60
  61 bool
  62 automatic_inlining_predicate(ir_instruction *ir)
  63 {
  64    ir_call *call = ir->as_call();
  65
  66    if (call && can_inline(call))
  67       return true;
  68
  69    return false;
  70 }
  71
  72 bool
  73 do_function_inlining(exec_list *instructions)
  74 {
  75    ir_function_inlining_visitor v;
  76
  77    do_expression_flattening(instructions, automatic_inlining_predicate);
  78
  79    v.run(instructions);
  80
  81    return v.progress;
  82 }
  83
  84 static void
  85 replace_return_with_assignment(ir_instruction *ir, void *data)
  86 {
  87    void *ctx = talloc_parent(ir);
  88    ir_variable *retval = (ir_variable *)data;
  89    ir_return *ret = ir->as_return();
  90
  91    if (ret) {
  92       if (ret->value) {
  93          ir_rvalue *lhs = new(ctx) ir_dereference_variable(retval);
  94          ret->replace_with(new(ctx) ir_assignment(lhs, ret->value, NULL));
  95       } else {
  96          /* un-valued return has to be the last return, or we shouldn't
  97           * have reached here. (see can_inline()).
  98           */
  99          assert(!ret->next->is_tail_sentinel());
 100          ret->remove();
 101       }
 102    }
 103 }
 104
 105 ir_rvalue *
 106 ir_call::generate_inline(ir_instruction *next_ir)
 107 {
 108    void *ctx = talloc_parent(this);
 109    ir_variable **parameters;
 110    int num_parameters;
 111    int i;
 112    ir_variable *retval = NULL;
 113    struct hash_table *ht;
 114
 115    ht = hash_table_ctor(0, hash_table_pointer_hash, hash_table_pointer_compare);
 116
 117    num_parameters = 0;
 118    foreach_iter(exec_list_iterator, iter_sig, this->callee->parameters)
 119       num_parameters++;
 120
 121    parameters = new ir_variable *[num_parameters];
 122
 123    /* Generate storage for the return value. */
 124    if (this->callee->return_type) {
 125       retval = new(ctx) ir_variable(this->callee->return_type, "__retval",
 126                                     ir_var_auto);
 127       next_ir->insert_before(retval);
 128    }
 129
 130    /* Generate the declarations for the parameters to our inlined code,
 131     * and set up the mapping of real function body variables to ours.
 132     */
 133    i = 0;
 134    exec_list_iterator sig_param_iter = this->callee->parameters.iterator();
 135    exec_list_iterator param_iter = this->actual_parameters.iterator();
 136    for (i = 0; i < num_parameters; i++) {
 137       ir_variable *sig_param = (ir_variable *) sig_param_iter.get();
 138       ir_rvalue *param = (ir_rvalue *) param_iter.get();
 139
 140       /* Generate a new variable for the parameter. */
 141       if (sig_param->type->base_type == GLSL_TYPE_SAMPLER) {
 142          /* For samplers, we want the inlined sampler references
 143           * referencing the passed in sampler variable, since that
 144           * will have the location information, which an assignment of
 145           * a sampler wouldn't.
 146           */
 147          parameters[i] = NULL;
 148          hash_table_insert(ht, param->variable_referenced(), sig_param);
 149       } else {
 150          parameters[i] = sig_param->clone(ht);
 151          parameters[i]->mode = ir_var_auto;
 152          next_ir->insert_before(parameters[i]);
 153       }
 154
 155       /* Move the actual param into our param variable if it's an 'in' type. */
 156       if (parameters[i] && (sig_param->mode == ir_var_in ||
 157                             sig_param->mode == ir_var_inout)) {
 158          ir_assignment *assign;
 159
 160          assign = new(ctx) ir_assignment(new(ctx) ir_dereference_variable(parameters[i]),
 161                                          param, NULL);
 162          next_ir->insert_before(assign);
 163       }
 164
 165       sig_param_iter.next();
 166       param_iter.next();
 167    }
 168
 169    /* Generate the inlined body of the function. */
 170    foreach_iter(exec_list_iterator, iter, callee->body) {
 171       ir_instruction *ir = (ir_instruction *)iter.get();
 172       ir_instruction *new_ir = ir->clone(ht);
 173
 174       next_ir->insert_before(new_ir);
 175       visit_tree(new_ir, replace_return_with_assignment, retval);
 176    }
 177
 178    /* Copy back the value of any 'out' parameters from the function body
 179     * variables to our own.
 180     */
 181    i = 0;
 182    param_iter = this->actual_parameters.iterator();
 183    sig_param_iter = this->callee->parameters.iterator();
 184    for (i = 0; i < num_parameters; i++) {
 185       ir_instruction *const param = (ir_instruction *) param_iter.get();
 186       const ir_variable *const sig_param = (ir_variable *) sig_param_iter.get();
 187
 188       /* Move our param variable into the actual param if it's an 'out' type. */
 189       if (parameters[i] && (sig_param->mode == ir_var_out ||
 190                             sig_param->mode == ir_var_inout)) {
 191          ir_assignment *assign;
 192
 193          assign = new(ctx) ir_assignment(param->clone(NULL)->as_rvalue(),
 194                                          new(ctx) ir_dereference_variable(parameters[i]),
 195                                          NULL);
 196          next_ir->insert_before(assign);
 197       }
 198
 199       param_iter.next();
 200       sig_param_iter.next();
 201    }
 202
 203    delete [] parameters;
 204
 205    hash_table_dtor(ht);
 206
 207    if (retval)
 208       return new(ctx) ir_dereference_variable(retval);
 209    else
 210       return NULL;
 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       /* If the call was part of some tree, then it should have been
 251        * flattened out or we shouldn't have seen it because of a
 252        * visit_continue_with_parent in this visitor.
 253        */
 254       assert(ir == base_ir);
 255
 256       (void) ir->generate_inline(ir);
 257       ir->remove();
 258       this->progress = true;
 259    }
 260
 261    return visit_continue;
 262 }
 263
 264
 265 ir_visitor_status
 266 ir_function_inlining_visitor::visit_enter(ir_assignment *ir)
 267 {
 268    ir_call *call = ir->rhs->as_call();
 269    if (!call || !can_inline(call))
 270       return visit_continue;
 271
 272    /* generates the parameter setup, function body, and returns the return
 273     * value of the function
 274     */
 275    ir_rvalue *rhs = call->generate_inline(ir);
 276    assert(rhs);
 277
 278    ir->rhs = rhs;
 279    this->progress = true;
 280
 281    return visit_continue;
 282 }