src/compiler/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 "compiler/glsl_types.h"
  43 #include "ir_builder.h"
  44
  45 using namespace ir_builder;
  46
  47 namespace {
  48
  49 /**
  50  * Visitor class for replacing expressions with ir_constant values.
  51  */
  52
  53 class ir_vec_index_to_cond_assign_visitor : public ir_hierarchical_visitor {
  54 public:
  55    ir_vec_index_to_cond_assign_visitor()
  56       : progress(false)
  57    {
  58       /* empty */
  59    }
  60
  61    ir_rvalue *convert_vec_index_to_cond_assign(void *mem_ctx,
  62                                                ir_rvalue *orig_vector,
  63                                                ir_rvalue *orig_index,
  64                                                const glsl_type *type);
  65
  66    ir_rvalue *convert_vector_extract_to_cond_assign(ir_rvalue *ir);
  67
  68    virtual ir_visitor_status visit_enter(ir_expression *);
  69    virtual ir_visitor_status visit_enter(ir_swizzle *);
  70    virtual ir_visitor_status visit_leave(ir_assignment *);
  71    virtual ir_visitor_status visit_enter(ir_return *);
  72    virtual ir_visitor_status visit_enter(ir_call *);
  73    virtual ir_visitor_status visit_enter(ir_if *);
  74
  75    bool progress;
  76 };
  77
  78 } /* anonymous namespace */
  79
  80 ir_rvalue *
  81 ir_vec_index_to_cond_assign_visitor::convert_vec_index_to_cond_assign(void *mem_ctx,
  82                                                                       ir_rvalue *orig_vector,
  83                                                                       ir_rvalue *orig_index,
  84                                                                       const glsl_type *type)
  85 {
  86    exec_list list;
  87    ir_factory body(&list, base_ir);
  88
  89    /* Store the index to a temporary to avoid reusing its tree. */
  90    assert(orig_index->type == glsl_type::int_type ||
  91           orig_index->type == glsl_type::uint_type);
  92    ir_variable *const index =
  93       body.make_temp(orig_index->type, "vec_index_tmp_i");
  94
  95    body.emit(assign(index, orig_index));
  96
  97    /* Store the value inside a temp, thus avoiding matrixes duplication */
  98    ir_variable *const value =
  99       body.make_temp(orig_vector->type, "vec_value_tmp");
 100
 101    body.emit(assign(value, orig_vector));
 102
 103
 104    /* Temporary where we store whichever value we swizzle out. */
 105    ir_variable *const var = body.make_temp(type, "vec_index_tmp_v");
 106
 107    /* Generate a single comparison condition "mask" for all of the components
 108     * in the vector.
 109     */
 110    ir_variable *const cond =
 111       compare_index_block(&list, index, 0,
 112                           orig_vector->type->vector_elements,
 113                           mem_ctx);
 114
 115    /* Generate a conditional move of each vector element to the temp. */
 116    for (unsigned i = 0; i < orig_vector->type->vector_elements; i++)
 117       body.emit(assign(var, swizzle(value, i, 1), swizzle(cond, i, 1)));
 118
 119    /* Put all of the new instructions in the IR stream before the old
 120     * instruction.
 121     */
 122    base_ir->insert_before(&list);
 123
 124    this->progress = true;
 125    return deref(var).val;
 126 }
 127
 128 ir_rvalue *
 129 ir_vec_index_to_cond_assign_visitor::convert_vector_extract_to_cond_assign(ir_rvalue *ir)
 130 {
 131    ir_expression *const expr = ir->as_expression();
 132
 133    if (expr == NULL || expr->operation != ir_binop_vector_extract)
 134       return ir;
 135
 136    return convert_vec_index_to_cond_assign(ralloc_parent(ir),
 137                                            expr->operands[0],
 138                                            expr->operands[1],
 139                                            ir->type);
 140 }
 141
 142 ir_visitor_status
 143 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_expression *ir)
 144 {
 145    for (unsigned i = 0; i < ir->num_operands; i++)
 146       ir->operands[i] = convert_vector_extract_to_cond_assign(ir->operands[i]);
 147
 148    return visit_continue;
 149 }
 150
 151 ir_visitor_status
 152 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_swizzle *ir)
 153 {
 154    /* Can't be hit from normal GLSL, since you can't swizzle a scalar (which
 155     * the result of indexing a vector is.  But maybe at some point we'll end up
 156     * using swizzling of scalars for vector construction.
 157     */
 158    ir->val = convert_vector_extract_to_cond_assign(ir->val);
 159
 160    return visit_continue;
 161 }
 162
 163 ir_visitor_status
 164 ir_vec_index_to_cond_assign_visitor::visit_leave(ir_assignment *ir)
 165 {
 166    ir->rhs = convert_vector_extract_to_cond_assign(ir->rhs);
 167
 168    if (ir->condition)
 169       ir->condition = convert_vector_extract_to_cond_assign(ir->condition);
 170
 171    return visit_continue;
 172 }
 173
 174 ir_visitor_status
 175 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_call *ir)
 176 {
 177    foreach_in_list_safe(ir_rvalue, param, &ir->actual_parameters) {
 178       ir_rvalue *new_param = convert_vector_extract_to_cond_assign(param);
 179
 180       if (new_param != param) {
 181          param->replace_with(new_param);
 182       }
 183    }
 184
 185    return visit_continue;
 186 }
 187
 188 ir_visitor_status
 189 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_return *ir)
 190 {
 191    if (ir->value)
 192       ir->value = convert_vector_extract_to_cond_assign(ir->value);
 193
 194    return visit_continue;
 195 }
 196
 197 ir_visitor_status
 198 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_if *ir)
 199 {
 200    ir->condition = convert_vector_extract_to_cond_assign(ir->condition);
 201
 202    return visit_continue;
 203 }
 204
 205 bool
 206 do_vec_index_to_cond_assign(exec_list *instructions)
 207 {
 208    ir_vec_index_to_cond_assign_visitor v;
 209
 210    visit_list_elements(&v, instructions);
 211
 212    return v.progress;
 213 }