src/glsl/lower_vec_index_to_cond_assign.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 lower_vec_index_to_cond_assign.cpp
  26  *
  27  * Turns indexing into vector types to a series of conditional moves
  28  * of each channel's swizzle into a temporary.
  29  *
  30  * Most GPUs don't have a native way to do this operation, and this
  31  * works around that.  For drivers using both this pass and
  32  * ir_vec_index_to_swizzle, there's a risk that this pass will happen
  33  * before sufficient constant folding to find that the array index is
  34  * constant.  However, we hope that other optimization passes,
  35  * particularly constant folding of assignment conditions and copy
  36  * propagation, will result in the same code in the end.
  37  */
  38
  39 #include "ir.h"
  40 #include "ir_visitor.h"
  41 #include "ir_optimization.h"
  42 #include "glsl_types.h"
  43
  44 /**
  45  * Visitor class for replacing expressions with ir_constant values.
  46  */
  47
  48 class ir_vec_index_to_cond_assign_visitor : public ir_hierarchical_visitor {
  49 public:
  50    ir_vec_index_to_cond_assign_visitor()
  51    {
  52       progress = false;
  53    }
  54
  55    ir_rvalue *convert_vec_index_to_cond_assign(ir_rvalue *val);
  56
  57    virtual ir_visitor_status visit_enter(ir_expression *);
  58    virtual ir_visitor_status visit_enter(ir_swizzle *);
  59    virtual ir_visitor_status visit_leave(ir_assignment *);
  60    virtual ir_visitor_status visit_enter(ir_return *);
  61    virtual ir_visitor_status visit_enter(ir_call *);
  62    virtual ir_visitor_status visit_enter(ir_if *);
  63
  64    bool progress;
  65 };
  66
  67 ir_rvalue *
  68 ir_vec_index_to_cond_assign_visitor::convert_vec_index_to_cond_assign(ir_rvalue *ir)
  69 {
  70    ir_dereference_array *orig_deref = ir->as_dereference_array();
  71    ir_assignment *assign, *value_assign;
  72    ir_variable *index, *var, *value;
  73    ir_dereference *deref, *deref_value;
  74    unsigned i;
  75
  76    if (!orig_deref)
  77       return ir;
  78
  79    if (orig_deref->array->type->is_matrix() ||
  80        orig_deref->array->type->is_array())
  81       return ir;
  82
  83    void *mem_ctx = ralloc_parent(ir);
  84
  85    assert(orig_deref->array_index->type->base_type == GLSL_TYPE_INT);
  86
  87    exec_list list;
  88
  89    /* Store the index to a temporary to avoid reusing its tree. */
  90    index = new(base_ir) ir_variable(glsl_type::int_type,
  91                                     "vec_index_tmp_i",
  92                                     ir_var_temporary);
  93    list.push_tail(index);
  94    deref = new(base_ir) ir_dereference_variable(index);
  95    assign = new(base_ir) ir_assignment(deref, orig_deref->array_index, NULL);
  96    list.push_tail(assign);
  97
  98    /* Store the value inside a temp, thus avoiding matrixes duplication */
  99    value = new(base_ir) ir_variable(orig_deref->array->type, "vec_value_tmp",
 100                                     ir_var_temporary);
 101    list.push_tail(value);
 102    deref_value = new(base_ir) ir_dereference_variable(value);
 103    value_assign = new(base_ir) ir_assignment(deref_value, orig_deref->array);
 104    list.push_tail(value_assign);
 105
 106    /* Temporary where we store whichever value we swizzle out. */
 107    var = new(base_ir) ir_variable(ir->type, "vec_index_tmp_v",
 108                                   ir_var_temporary);
 109    list.push_tail(var);
 110
 111    /* Generate a single comparison condition "mask" for all of the components
 112     * in the vector.
 113     */
 114    ir_rvalue *const cond_deref =
 115       compare_index_block(&list, index, 0,
 116                           orig_deref->array->type->vector_elements,
 117                           mem_ctx);
 118
 119    /* Generate a conditional move of each vector element to the temp. */
 120    for (i = 0; i < orig_deref->array->type->vector_elements; i++) {
 121       ir_rvalue *condition_swizzle =
 122          new(base_ir) ir_swizzle(cond_deref->clone(ir, NULL), i, 0, 0, 0, 1);
 123
 124       /* Just clone the rest of the deref chain when trying to get at the
 125        * underlying variable.
 126        */
 127       ir_rvalue *swizzle =
 128          new(base_ir) ir_swizzle(deref_value->clone(mem_ctx, NULL),
 129                                  i, 0, 0, 0, 1);
 130
 131       deref = new(base_ir) ir_dereference_variable(var);
 132       assign = new(base_ir) ir_assignment(deref, swizzle, condition_swizzle);
 133       list.push_tail(assign);
 134    }
 135
 136    /* Put all of the new instructions in the IR stream before the old
 137     * instruction.
 138     */
 139    base_ir->insert_before(&list);
 140
 141    this->progress = true;
 142    return new(base_ir) ir_dereference_variable(var);
 143 }
 144
 145 ir_visitor_status
 146 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_expression *ir)
 147 {
 148    unsigned int i;
 149
 150    for (i = 0; i < ir->get_num_operands(); i++) {
 151       ir->operands[i] = convert_vec_index_to_cond_assign(ir->operands[i]);
 152    }
 153
 154    return visit_continue;
 155 }
 156
 157 ir_visitor_status
 158 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_swizzle *ir)
 159 {
 160    /* Can't be hit from normal GLSL, since you can't swizzle a scalar (which
 161     * the result of indexing a vector is.  But maybe at some point we'll end up
 162     * using swizzling of scalars for vector construction.
 163     */
 164    ir->val = convert_vec_index_to_cond_assign(ir->val);
 165
 166    return visit_continue;
 167 }
 168
 169 ir_visitor_status
 170 ir_vec_index_to_cond_assign_visitor::visit_leave(ir_assignment *ir)
 171 {
 172    ir_variable *index, *var;
 173    ir_dereference_variable *deref;
 174    ir_assignment *assign;
 175    unsigned i;
 176
 177    ir->rhs = convert_vec_index_to_cond_assign(ir->rhs);
 178    if (ir->condition)
 179       ir->condition = convert_vec_index_to_cond_assign(ir->condition);
 180
 181    /* Last, handle the LHS */
 182    ir_dereference_array *orig_deref = ir->lhs->as_dereference_array();
 183
 184    if (!orig_deref ||
 185        orig_deref->array->type->is_matrix() ||
 186        orig_deref->array->type->is_array())
 187       return visit_continue;
 188
 189    void *mem_ctx = ralloc_parent(ir);
 190
 191    assert(orig_deref->array_index->type->base_type == GLSL_TYPE_INT);
 192
 193    exec_list list;
 194
 195    /* Store the index to a temporary to avoid reusing its tree. */
 196    index = new(ir) ir_variable(glsl_type::int_type, "vec_index_tmp_i",
 197                                ir_var_temporary);
 198    list.push_tail(index);
 199    deref = new(ir) ir_dereference_variable(index);
 200    assign = new(ir) ir_assignment(deref, orig_deref->array_index, NULL);
 201    list.push_tail(assign);
 202
 203    /* Store the RHS to a temporary to avoid reusing its tree. */
 204    var = new(ir) ir_variable(ir->rhs->type, "vec_index_tmp_v",
 205                              ir_var_temporary);
 206    list.push_tail(var);
 207    deref = new(ir) ir_dereference_variable(var);
 208    assign = new(ir) ir_assignment(deref, ir->rhs, NULL);
 209    list.push_tail(assign);
 210
 211    /* Generate a single comparison condition "mask" for all of the components
 212     * in the vector.
 213     */
 214    ir_rvalue *const cond_deref =
 215       compare_index_block(&list, index, 0,
 216                           orig_deref->array->type->vector_elements,
 217                           mem_ctx);
 218
 219    /* Generate a conditional move of each vector element to the temp. */
 220    for (i = 0; i < orig_deref->array->type->vector_elements; i++) {
 221       ir_rvalue *condition_swizzle =
 222          new(ir) ir_swizzle(cond_deref->clone(ir, NULL), i, 0, 0, 0, 1);
 223
 224
 225       /* Just clone the rest of the deref chain when trying to get at the
 226        * underlying variable.
 227        */
 228       ir_rvalue *swizzle =
 229          new(ir) ir_swizzle(orig_deref->array->clone(mem_ctx, NULL),
 230                             i, 0, 0, 0, 1);
 231
 232       deref = new(ir) ir_dereference_variable(var);
 233       assign = new(ir) ir_assignment(swizzle, deref, condition_swizzle);
 234       list.push_tail(assign);
 235    }
 236
 237    /* If the original assignment has a condition, respect that original
 238     * condition!  This is acomplished by wrapping the new conditional
 239     * assignments in an if-statement that uses the original condition.
 240     */
 241    if (ir->condition != NULL) {
 242       /* No need to clone the condition because the IR that it hangs on is
 243        * going to be removed from the instruction sequence.
 244        */
 245       ir_if *if_stmt = new(mem_ctx) ir_if(ir->condition);
 246
 247       list.move_nodes_to(&if_stmt->then_instructions);
 248       ir->insert_before(if_stmt);
 249    } else {
 250       ir->insert_before(&list);
 251    }
 252
 253    ir->remove();
 254
 255    this->progress = true;
 256
 257    return visit_continue;
 258 }
 259
 260 ir_visitor_status
 261 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_call *ir)
 262 {
 263    foreach_iter(exec_list_iterator, iter, *ir) {
 264       ir_rvalue *param = (ir_rvalue *)iter.get();
 265       ir_rvalue *new_param = convert_vec_index_to_cond_assign(param);
 266
 267       if (new_param != param) {
 268          param->replace_with(new_param);
 269       }
 270    }
 271
 272    return visit_continue;
 273 }
 274
 275 ir_visitor_status
 276 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_return *ir)
 277 {
 278    if (ir->value) {
 279       ir->value = convert_vec_index_to_cond_assign(ir->value);
 280    }
 281
 282    return visit_continue;
 283 }
 284
 285 ir_visitor_status
 286 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_if *ir)
 287 {
 288    ir->condition = convert_vec_index_to_cond_assign(ir->condition);
 289
 290    return visit_continue;
 291 }
 292
 293 bool
 294 do_vec_index_to_cond_assign(exec_list *instructions)
 295 {
 296    ir_vec_index_to_cond_assign_visitor v;
 297
 298    visit_list_elements(&v, instructions);
 299
 300    return v.progress;
 301 }