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
  44 namespace {
  45
  46 /**
  47  * Visitor class for replacing expressions with ir_constant values.
  48  */
  49
  50 class ir_vec_index_to_cond_assign_visitor : public ir_hierarchical_visitor {
  51 public:
  52    ir_vec_index_to_cond_assign_visitor()
  53       : progress(false)
  54    {
  55       /* empty */
  56    }
  57
  58    ir_rvalue *convert_vec_index_to_cond_assign(void *mem_ctx,
  59                                                ir_rvalue *orig_vector,
  60                                                ir_rvalue *orig_index,
  61                                                const glsl_type *type);
  62
  63    ir_rvalue *convert_vector_extract_to_cond_assign(ir_rvalue *ir);
  64
  65    virtual ir_visitor_status visit_enter(ir_expression *);
  66    virtual ir_visitor_status visit_enter(ir_swizzle *);
  67    virtual ir_visitor_status visit_leave(ir_assignment *);
  68    virtual ir_visitor_status visit_enter(ir_return *);
  69    virtual ir_visitor_status visit_enter(ir_call *);
  70    virtual ir_visitor_status visit_enter(ir_if *);
  71
  72    bool progress;
  73 };
  74
  75 } /* anonymous namespace */
  76
  77 ir_rvalue *
  78 ir_vec_index_to_cond_assign_visitor::convert_vec_index_to_cond_assign(void *mem_ctx,
  79                                                                       ir_rvalue *orig_vector,
  80                                                                       ir_rvalue *orig_index,
  81                                                                       const glsl_type *type)
  82 {
  83    ir_assignment *assign, *value_assign;
  84    ir_variable *index, *var, *value;
  85    ir_dereference *deref, *deref_value;
  86    unsigned i;
  87
  88
  89    exec_list list;
  90
  91    /* Store the index to a temporary to avoid reusing its tree. */
  92    assert(orig_index->type == glsl_type::int_type ||
  93           orig_index->type == glsl_type::uint_type);
  94    index = new(base_ir) ir_variable(orig_index->type,
  95                                     "vec_index_tmp_i",
  96                                     ir_var_temporary);
  97    list.push_tail(index);
  98    deref = new(base_ir) ir_dereference_variable(index);
  99    assign = new(base_ir) ir_assignment(deref, orig_index, NULL);
 100    list.push_tail(assign);
 101
 102    /* Store the value inside a temp, thus avoiding matrixes duplication */
 103    value = new(base_ir) ir_variable(orig_vector->type, "vec_value_tmp",
 104                                     ir_var_temporary);
 105    list.push_tail(value);
 106    deref_value = new(base_ir) ir_dereference_variable(value);
 107    value_assign = new(base_ir) ir_assignment(deref_value, orig_vector);
 108    list.push_tail(value_assign);
 109
 110    /* Temporary where we store whichever value we swizzle out. */
 111    var = new(base_ir) ir_variable(type, "vec_index_tmp_v",
 112                                   ir_var_temporary);
 113    list.push_tail(var);
 114
 115    /* Generate a single comparison condition "mask" for all of the components
 116     * in the vector.
 117     */
 118    ir_variable *const cond =
 119       compare_index_block(&list, index, 0,
 120                           orig_vector->type->vector_elements,
 121                           mem_ctx);
 122
 123    /* Generate a conditional move of each vector element to the temp. */
 124    for (i = 0; i < orig_vector->type->vector_elements; i++) {
 125       ir_rvalue *condition_swizzle =
 126          new(base_ir) ir_swizzle(new(mem_ctx) ir_dereference_variable(cond),
 127                                  i, 0, 0, 0, 1);
 128
 129       /* Just clone the rest of the deref chain when trying to get at the
 130        * underlying variable.
 131        */
 132       ir_rvalue *swizzle =
 133          new(base_ir) ir_swizzle(deref_value->clone(mem_ctx, NULL),
 134                                  i, 0, 0, 0, 1);
 135
 136       deref = new(base_ir) ir_dereference_variable(var);
 137       assign = new(base_ir) ir_assignment(deref, swizzle, condition_swizzle);
 138       list.push_tail(assign);
 139    }
 140
 141    /* Put all of the new instructions in the IR stream before the old
 142     * instruction.
 143     */
 144    base_ir->insert_before(&list);
 145
 146    this->progress = true;
 147    return new(base_ir) ir_dereference_variable(var);
 148 }
 149
 150 ir_rvalue *
 151 ir_vec_index_to_cond_assign_visitor::convert_vector_extract_to_cond_assign(ir_rvalue *ir)
 152 {
 153    ir_expression *const expr = ir->as_expression();
 154
 155    if (expr == NULL || expr->operation != ir_binop_vector_extract)
 156       return ir;
 157
 158    return convert_vec_index_to_cond_assign(ralloc_parent(ir),
 159                                            expr->operands[0],
 160                                            expr->operands[1],
 161                                            ir->type);
 162 }
 163
 164 ir_visitor_status
 165 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_expression *ir)
 166 {
 167    for (unsigned i = 0; i < ir->num_operands; i++)
 168       ir->operands[i] = convert_vector_extract_to_cond_assign(ir->operands[i]);
 169
 170    return visit_continue;
 171 }
 172
 173 ir_visitor_status
 174 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_swizzle *ir)
 175 {
 176    /* Can't be hit from normal GLSL, since you can't swizzle a scalar (which
 177     * the result of indexing a vector is.  But maybe at some point we'll end up
 178     * using swizzling of scalars for vector construction.
 179     */
 180    ir->val = convert_vector_extract_to_cond_assign(ir->val);
 181
 182    return visit_continue;
 183 }
 184
 185 ir_visitor_status
 186 ir_vec_index_to_cond_assign_visitor::visit_leave(ir_assignment *ir)
 187 {
 188    ir->rhs = convert_vector_extract_to_cond_assign(ir->rhs);
 189
 190    if (ir->condition)
 191       ir->condition = convert_vector_extract_to_cond_assign(ir->condition);
 192
 193    return visit_continue;
 194 }
 195
 196 ir_visitor_status
 197 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_call *ir)
 198 {
 199    foreach_in_list_safe(ir_rvalue, param, &ir->actual_parameters) {
 200       ir_rvalue *new_param = convert_vector_extract_to_cond_assign(param);
 201
 202       if (new_param != param) {
 203          param->replace_with(new_param);
 204       }
 205    }
 206
 207    return visit_continue;
 208 }
 209
 210 ir_visitor_status
 211 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_return *ir)
 212 {
 213    if (ir->value)
 214       ir->value = convert_vector_extract_to_cond_assign(ir->value);
 215
 216    return visit_continue;
 217 }
 218
 219 ir_visitor_status
 220 ir_vec_index_to_cond_assign_visitor::visit_enter(ir_if *ir)
 221 {
 222    ir->condition = convert_vector_extract_to_cond_assign(ir->condition);
 223
 224    return visit_continue;
 225 }
 226
 227 bool
 228 do_vec_index_to_cond_assign(exec_list *instructions)
 229 {
 230    ir_vec_index_to_cond_assign_visitor v;
 231
 232    visit_list_elements(&v, instructions);
 233
 234    return v.progress;
 235 }