src/compiler/glsl/lower_vector_derefs.cpp

   1 /*
   2  * Copyright © 2013 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 #include "ir.h"
  24 #include "ir_builder.h"
  25 #include "ir_rvalue_visitor.h"
  26 #include "ir_optimization.h"
  27 #include "main/mtypes.h"
  28
  29 using namespace ir_builder;
  30
  31 namespace {
  32
  33 class vector_deref_visitor : public ir_rvalue_enter_visitor {
  34 public:
  35    vector_deref_visitor(void *mem_ctx, gl_shader_stage shader_stage)
  36       : progress(false), shader_stage(shader_stage),
  37         factory(&factory_instructions, mem_ctx)
  38    {
  39    }
  40
  41    virtual ~vector_deref_visitor()
  42    {
  43    }
  44
  45    virtual void handle_rvalue(ir_rvalue **rv);
  46    virtual ir_visitor_status visit_enter(ir_assignment *ir);
  47
  48    bool progress;
  49    gl_shader_stage shader_stage;
  50    exec_list factory_instructions;
  51    ir_factory factory;
  52 };
  53
  54 } /* anonymous namespace */
  55
  56 ir_visitor_status
  57 vector_deref_visitor::visit_enter(ir_assignment *ir)
  58 {
  59    if (!ir->lhs || ir->lhs->ir_type != ir_type_dereference_array)
  60       return ir_rvalue_enter_visitor::visit_enter(ir);
  61
  62    ir_dereference_array *const deref = (ir_dereference_array *) ir->lhs;
  63    if (!deref->array->type->is_vector())
  64       return ir_rvalue_enter_visitor::visit_enter(ir);
  65
  66    ir_rvalue *const new_lhs = deref->array;
  67
  68    void *mem_ctx = ralloc_parent(ir);
  69    ir_constant *old_index_constant =
  70       deref->array_index->constant_expression_value(mem_ctx);
  71    if (!old_index_constant) {
  72       if (shader_stage == MESA_SHADER_TESS_CTRL &&
  73           deref->variable_referenced()->data.mode == ir_var_shader_out) {
  74          /* Tessellation control shader outputs act as if they have memory
  75           * backing them and if we have writes from multiple threads
  76           * targeting the same vec4 (this can happen for patch outputs), the
  77           * load-vec-store pattern of ir_triop_vector_insert doesn't work.
  78           * Instead, we have to lower to a series of conditional write-masked
  79           * assignments.
  80           */
  81          ir_variable *const src_temp =
  82             factory.make_temp(ir->rhs->type, "scalar_tmp");
  83
  84          /* The newly created variable declaration goes before the assignment
  85           * because we're going to set it as the new LHS.
  86           */
  87          ir->insert_before(factory.instructions);
  88          ir->set_lhs(new(mem_ctx) ir_dereference_variable(src_temp));
  89
  90          ir_variable *const arr_index =
  91             factory.make_temp(deref->array_index->type, "index_tmp");
  92          factory.emit(assign(arr_index, deref->array_index));
  93
  94          for (unsigned i = 0; i < new_lhs->type->vector_elements; i++) {
  95             ir_constant *const cmp_index =
  96                ir_constant::zero(factory.mem_ctx, deref->array_index->type);
  97             cmp_index->value.u[0] = i;
  98
  99             ir_rvalue *const lhs_clone = new_lhs->clone(factory.mem_ctx, NULL);
 100             ir_dereference_variable *const src_temp_deref =
 101                new(mem_ctx) ir_dereference_variable(src_temp);
 102
 103             if (new_lhs->ir_type != ir_type_swizzle) {
 104                assert(lhs_clone->as_dereference());
 105                ir_assignment *cond_assign =
 106                   new(mem_ctx) ir_assignment(lhs_clone->as_dereference(),
 107                                              src_temp_deref,
 108                                              equal(arr_index, cmp_index),
 109                                              WRITEMASK_X << i);
 110                factory.emit(cond_assign);
 111             } else {
 112                ir_assignment *cond_assign =
 113                   new(mem_ctx) ir_assignment(swizzle(lhs_clone, i, 1),
 114                                              src_temp_deref,
 115                                              equal(arr_index, cmp_index));
 116                factory.emit(cond_assign);
 117             }
 118          }
 119          ir->insert_after(factory.instructions);
 120       } else {
 121          ir->rhs = new(mem_ctx) ir_expression(ir_triop_vector_insert,
 122                                               new_lhs->type,
 123                                               new_lhs->clone(mem_ctx, NULL),
 124                                               ir->rhs,
 125                                               deref->array_index);
 126          ir->write_mask = (1 << new_lhs->type->vector_elements) - 1;
 127          ir->set_lhs(new_lhs);
 128       }
 129    } else if (new_lhs->ir_type != ir_type_swizzle) {
 130       ir->set_lhs(new_lhs);
 131       ir->write_mask = 1 << old_index_constant->get_uint_component(0);
 132    } else {
 133       /* If the "new" LHS is a swizzle, use the set_lhs helper to instead
 134        * swizzle the RHS.
 135        */
 136       unsigned component[1] = { old_index_constant->get_uint_component(0) };
 137       ir->set_lhs(new(mem_ctx) ir_swizzle(new_lhs, component, 1));
 138    }
 139
 140    return ir_rvalue_enter_visitor::visit_enter(ir);
 141 }
 142
 143 void
 144 vector_deref_visitor::handle_rvalue(ir_rvalue **rv)
 145 {
 146    if (*rv == NULL || (*rv)->ir_type != ir_type_dereference_array)
 147       return;
 148
 149    ir_dereference_array *const deref = (ir_dereference_array *) *rv;
 150    if (!deref->array->type->is_vector())
 151       return;
 152
 153    void *mem_ctx = ralloc_parent(deref);
 154    *rv = new(mem_ctx) ir_expression(ir_binop_vector_extract,
 155                                     deref->array,
 156                                     deref->array_index);
 157 }
 158
 159 bool
 160 lower_vector_derefs(gl_linked_shader *shader)
 161 {
 162    vector_deref_visitor v(shader->ir, shader->Stage);
 163
 164    visit_list_elements(&v, shader->ir);
 165
 166    return v.progress;
 167 }