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;
  72    ir_variable *index, *var;
  73    ir_dereference *deref;
  74    ir_expression *condition;
  75    ir_swizzle *swizzle;
  76    int i;
  77
  78    if (!orig_deref)
  79       return ir;
  80
  81    if (orig_deref->array->type->is_matrix() ||
  82        orig_deref->array->type->is_array())
  83       return ir;
  84
  85    void *mem_ctx = ralloc_parent(ir);
  86
  87    assert(orig_deref->array_index->type->base_type == GLSL_TYPE_INT);
  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    base_ir->insert_before(index);
  94    deref = new(base_ir) ir_dereference_variable(index);
  95    assign = new(base_ir) ir_assignment(deref, orig_deref->array_index, NULL);
  96    base_ir->insert_before(assign);
  97
  98    /* Temporary where we store whichever value we swizzle out. */
  99    var = new(base_ir) ir_variable(ir->type, "vec_index_tmp_v",
 100                                   ir_var_temporary);
 101    base_ir->insert_before(var);
 102
 103    /* Generate a conditional move of each vector element to the temp. */
 104    for (i = 0; i < orig_deref->array->type->vector_elements; i++) {
 105       deref = new(base_ir) ir_dereference_variable(index);
 106       condition = new(base_ir) ir_expression(ir_binop_equal,
 107                                              glsl_type::bool_type,
 108                                              deref,
 109                                              new(base_ir) ir_constant(i));
 110
 111       /* Just clone the rest of the deref chain when trying to get at the
 112        * underlying variable.
 113        */
 114       swizzle = new(base_ir) ir_swizzle(orig_deref->array->clone(mem_ctx, NULL),
 115                                         i, 0, 0, 0, 1);
 116
 117       deref = new(base_ir) ir_dereference_variable(var);
 118       assign = new(base_ir) ir_assignment(deref, swizzle, condition);
 119       base_ir->insert_before(assign);
 120    }
 121
 122    this->progress = true;
 123    return new(base_ir) ir_dereference_variable(var);
 124 }
 125
 126 ir_visitor_status
 127 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_expression *ir)
 128 {
 129    unsigned int i;
 130
 131    for (i = 0; i < ir->get_num_operands(); i++) {
 132       ir->operands[i] = convert_vec_index_to_cond_assign(ir->operands[i]);
 133    }
 134
 135    return visit_continue;
 136 }
 137
 138 ir_visitor_status
 139 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_swizzle *ir)
 140 {
 141    /* Can't be hit from normal GLSL, since you can't swizzle a scalar (which
 142     * the result of indexing a vector is.  But maybe at some point we'll end up
 143     * using swizzling of scalars for vector construction.
 144     */
 145    ir->val = convert_vec_index_to_cond_assign(ir->val);
 146
 147    return visit_continue;
 148 }
 149
 150 ir_visitor_status
 151 ir_vec_index_to_cond_assign_visitor::visit_leave(ir_assignment *ir)
 152 {
 153    ir_variable *index, *var;
 154    ir_dereference_variable *deref;
 155    ir_assignment *assign;
 156    int i;
 157
 158    ir->rhs = convert_vec_index_to_cond_assign(ir->rhs);
 159    if (ir->condition)
 160       ir->condition = convert_vec_index_to_cond_assign(ir->condition);
 161
 162    /* Last, handle the LHS */
 163    ir_dereference_array *orig_deref = ir->lhs->as_dereference_array();
 164
 165    if (!orig_deref ||
 166        orig_deref->array->type->is_matrix() ||
 167        orig_deref->array->type->is_array())
 168       return visit_continue;
 169
 170    void *mem_ctx = ralloc_parent(ir);
 171
 172    assert(orig_deref->array_index->type->base_type == GLSL_TYPE_INT);
 173
 174    /* Store the index to a temporary to avoid reusing its tree. */
 175    index = new(ir) ir_variable(glsl_type::int_type, "vec_index_tmp_i",
 176                                ir_var_temporary);
 177    ir->insert_before(index);
 178    deref = new(ir) ir_dereference_variable(index);
 179    assign = new(ir) ir_assignment(deref, orig_deref->array_index, NULL);
 180    ir->insert_before(assign);
 181
 182    /* Store the RHS to a temporary to avoid reusing its tree. */
 183    var = new(ir) ir_variable(ir->rhs->type, "vec_index_tmp_v",
 184                              ir_var_temporary);
 185    ir->insert_before(var);
 186    deref = new(ir) ir_dereference_variable(var);
 187    assign = new(ir) ir_assignment(deref, ir->rhs, NULL);
 188    ir->insert_before(assign);
 189
 190    /* Generate a conditional move of each vector element to the temp. */
 191    for (i = 0; i < orig_deref->array->type->vector_elements; i++) {
 192       ir_rvalue *condition, *swizzle;
 193
 194       deref = new(ir) ir_dereference_variable(index);
 195       condition = new(ir) ir_expression(ir_binop_equal,
 196                                         glsl_type::bool_type,
 197                                         deref,
 198                                         new(ir) ir_constant(i));
 199
 200       /* Just clone the rest of the deref chain when trying to get at the
 201        * underlying variable.
 202        */
 203       swizzle = new(ir) ir_swizzle(orig_deref->array->clone(mem_ctx, NULL),
 204                                    i, 0, 0, 0, 1);
 205
 206       deref = new(ir) ir_dereference_variable(var);
 207       assign = new(ir) ir_assignment(swizzle, deref, condition);
 208       ir->insert_before(assign);
 209    }
 210    ir->remove();
 211
 212    this->progress = true;
 213
 214    return visit_continue;
 215 }
 216
 217 ir_visitor_status
 218 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_call *ir)
 219 {
 220    foreach_iter(exec_list_iterator, iter, *ir) {
 221       ir_rvalue *param = (ir_rvalue *)iter.get();
 222       ir_rvalue *new_param = convert_vec_index_to_cond_assign(param);
 223
 224       if (new_param != param) {
 225          param->replace_with(new_param);
 226       }
 227    }
 228
 229    return visit_continue;
 230 }
 231
 232 ir_visitor_status
 233 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_return *ir)
 234 {
 235    if (ir->value) {
 236       ir->value = convert_vec_index_to_cond_assign(ir->value);
 237    }
 238
 239    return visit_continue;
 240 }
 241
 242 ir_visitor_status
 243 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_if *ir)
 244 {
 245    ir->condition = convert_vec_index_to_cond_assign(ir->condition);
 246
 247    return visit_continue;
 248 }
 249
 250 bool
 251 do_vec_index_to_cond_assign(exec_list *instructions)
 252 {
 253    ir_vec_index_to_cond_assign_visitor v;
 254
 255    visit_list_elements(&v, instructions);
 256
 257    return v.progress;
 258 }