1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
|
/*
* Copyright © 2013 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
* DEALINGS IN THE SOFTWARE.
*/
#include "ir.h"
#include "ir_builder.h"
#include "ir_rvalue_visitor.h"
#include "ir_optimization.h"
#include "main/mtypes.h"
using namespace ir_builder;
namespace {
class vector_deref_visitor : public ir_rvalue_enter_visitor {
public:
vector_deref_visitor(void *mem_ctx, gl_shader_stage shader_stage)
: progress(false), shader_stage(shader_stage),
factory(&factory_instructions, mem_ctx)
{
}
virtual ~vector_deref_visitor()
{
}
virtual void handle_rvalue(ir_rvalue **rv);
virtual ir_visitor_status visit_enter(ir_assignment *ir);
bool progress;
gl_shader_stage shader_stage;
exec_list factory_instructions;
ir_factory factory;
};
} /* anonymous namespace */
ir_visitor_status
vector_deref_visitor::visit_enter(ir_assignment *ir)
{
if (!ir->lhs || ir->lhs->ir_type != ir_type_dereference_array)
return ir_rvalue_enter_visitor::visit_enter(ir);
ir_dereference_array *const deref = (ir_dereference_array *) ir->lhs;
if (!deref->array->type->is_vector())
return ir_rvalue_enter_visitor::visit_enter(ir);
/* SSBOs and shared variables are backed by memory and may be accessed by
* multiple threads simultaneously. It's not safe to lower a single
* component store to a load-vec-store because it may race with writes to
* other components.
*/
ir_variable *var = deref->variable_referenced();
if (var->data.mode == ir_var_shader_storage ||
var->data.mode == ir_var_shader_shared)
return ir_rvalue_enter_visitor::visit_enter(ir);
ir_rvalue *const new_lhs = deref->array;
void *mem_ctx = ralloc_parent(ir);
ir_constant *old_index_constant =
deref->array_index->constant_expression_value(mem_ctx);
if (!old_index_constant) {
if (shader_stage == MESA_SHADER_TESS_CTRL &&
deref->variable_referenced()->data.mode == ir_var_shader_out) {
/* Tessellation control shader outputs act as if they have memory
* backing them and if we have writes from multiple threads
* targeting the same vec4 (this can happen for patch outputs), the
* load-vec-store pattern of ir_triop_vector_insert doesn't work.
* Instead, we have to lower to a series of conditional write-masked
* assignments.
*/
ir_variable *const src_temp =
factory.make_temp(ir->rhs->type, "scalar_tmp");
/* The newly created variable declaration goes before the assignment
* because we're going to set it as the new LHS.
*/
ir->insert_before(factory.instructions);
ir->set_lhs(new(mem_ctx) ir_dereference_variable(src_temp));
ir_variable *const arr_index =
factory.make_temp(deref->array_index->type, "index_tmp");
factory.emit(assign(arr_index, deref->array_index));
for (unsigned i = 0; i < new_lhs->type->vector_elements; i++) {
ir_constant *const cmp_index =
ir_constant::zero(factory.mem_ctx, deref->array_index->type);
cmp_index->value.u[0] = i;
ir_rvalue *const lhs_clone = new_lhs->clone(factory.mem_ctx, NULL);
ir_dereference_variable *const src_temp_deref =
new(mem_ctx) ir_dereference_variable(src_temp);
if (new_lhs->ir_type != ir_type_swizzle) {
assert(lhs_clone->as_dereference());
ir_assignment *cond_assign =
new(mem_ctx) ir_assignment(lhs_clone->as_dereference(),
src_temp_deref,
equal(arr_index, cmp_index),
WRITEMASK_X << i);
factory.emit(cond_assign);
} else {
ir_assignment *cond_assign =
new(mem_ctx) ir_assignment(swizzle(lhs_clone, i, 1),
src_temp_deref,
equal(arr_index, cmp_index));
factory.emit(cond_assign);
}
}
ir->insert_after(factory.instructions);
} else {
ir->rhs = new(mem_ctx) ir_expression(ir_triop_vector_insert,
new_lhs->type,
new_lhs->clone(mem_ctx, NULL),
ir->rhs,
deref->array_index);
ir->write_mask = (1 << new_lhs->type->vector_elements) - 1;
ir->set_lhs(new_lhs);
}
} else {
unsigned index = old_index_constant->get_uint_component(0);
if (index >= new_lhs->type->vector_elements) {
/* Section 5.11 (Out-of-Bounds Accesses) of the GLSL 4.60 spec says:
*
* In the subsections described above for array, vector, matrix and
* structure accesses, any out-of-bounds access produced undefined
* behavior.... Out-of-bounds writes may be discarded or overwrite
* other variables of the active program.
*/
ir->remove();
return visit_continue;
}
if (new_lhs->ir_type != ir_type_swizzle) {
ir->set_lhs(new_lhs);
ir->write_mask = 1 << index;
} else {
/* If the "new" LHS is a swizzle, use the set_lhs helper to instead
* swizzle the RHS.
*/
unsigned component[1] = { index };
ir->set_lhs(new(mem_ctx) ir_swizzle(new_lhs, component, 1));
}
}
return ir_rvalue_enter_visitor::visit_enter(ir);
}
void
vector_deref_visitor::handle_rvalue(ir_rvalue **rv)
{
if (*rv == NULL || (*rv)->ir_type != ir_type_dereference_array)
return;
ir_dereference_array *const deref = (ir_dereference_array *) *rv;
if (!deref->array->type->is_vector())
return;
/* Back-ends need to be able to handle derefs on vectors for SSBOs, UBOs,
* and shared variables. They have to handle it for writes anyway so we
* may as well require it for reads.
*/
ir_variable *var = deref->variable_referenced();
if (var && (var->data.mode == ir_var_shader_storage ||
var->data.mode == ir_var_shader_shared ||
(var->data.mode == ir_var_uniform &&
var->get_interface_type())))
return;
void *mem_ctx = ralloc_parent(deref);
*rv = new(mem_ctx) ir_expression(ir_binop_vector_extract,
deref->array,
deref->array_index);
}
bool
lower_vector_derefs(gl_linked_shader *shader)
{
vector_deref_visitor v(shader->ir, shader->Stage);
visit_list_elements(&v, shader->ir);
return v.progress;
}
|
