1 /*
2 * Copyright © 2014 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_const_arrays_to_uniforms.cpp
26 *
27 * Lower constant arrays to uniform arrays.
28 *
29 * Some driver backends (such as i965 and nouveau) don't handle constant arrays
30 * gracefully, instead treating them as ordinary writable temporary arrays.
31 * Since arrays can be large, this often means spilling them to scratch memory,
32 * which usually involves a large number of instructions.
33 *
34 * This must be called prior to link_set_uniform_initializers(); we need the
35 * linker to process our new uniform's constant initializer.
36 *
37 * This should be called after optimizations, since those can result in
38 * splitting and removing arrays that are indexed by constant expressions.
39 */
40 #include "ir.h"
41 #include "ir_visitor.h"
42 #include "ir_rvalue_visitor.h"
43 #include "compiler/glsl_types.h"
44
45 namespace {
46 class lower_const_array_visitor : public ir_rvalue_visitor {
47 public:
lower_const_array_visitor(exec_list * insts,unsigned s,unsigned available_uni_components)48 lower_const_array_visitor(exec_list *insts, unsigned s,
49 unsigned available_uni_components)
50 {
51 instructions = insts;
52 stage = s;
53 const_count = 0;
54 free_uni_components = available_uni_components;
55 progress = false;
56 }
57
run()58 bool run()
59 {
60 visit_list_elements(this, instructions);
61 return progress;
62 }
63
64 ir_visitor_status visit_enter(ir_texture *);
65 void handle_rvalue(ir_rvalue **rvalue);
66
67 private:
68 exec_list *instructions;
69 unsigned stage;
70 unsigned const_count;
71 unsigned free_uni_components;
72 bool progress;
73 };
74
75 ir_visitor_status
visit_enter(ir_texture *)76 lower_const_array_visitor::visit_enter(ir_texture *)
77 {
78 return visit_continue_with_parent;
79 }
80
81 void
handle_rvalue(ir_rvalue ** rvalue)82 lower_const_array_visitor::handle_rvalue(ir_rvalue **rvalue)
83 {
84 if (!*rvalue)
85 return;
86
87 ir_constant *con = (*rvalue)->as_constant();
88 if (!con || !con->type->is_array())
89 return;
90
91 /* How many uniform component slots are required? */
92 unsigned component_slots = con->type->component_slots();
93
94 /* We would utilize more than is available, bail out. */
95 if (component_slots > free_uni_components)
96 return;
97
98 free_uni_components -= component_slots;
99
100 void *mem_ctx = ralloc_parent(con);
101
102 /* In the very unlikely event of 4294967295 constant arrays in a single
103 * shader, don't promote this to a uniform.
104 */
105 unsigned limit = ~0;
106 if (const_count == limit)
107 return;
108
109 char *uniform_name = ralloc_asprintf(mem_ctx, "constarray_%x_%u",
110 const_count, stage);
111 const_count++;
112
113 ir_variable *uni =
114 new(mem_ctx) ir_variable(con->type, uniform_name, ir_var_uniform);
115 uni->constant_initializer = con;
116 uni->constant_value = con;
117 uni->data.has_initializer = true;
118 uni->data.how_declared = ir_var_hidden;
119 uni->data.read_only = true;
120 /* Assume the whole thing is accessed. */
121 uni->data.max_array_access = uni->type->length - 1;
122 instructions->push_head(uni);
123
124 *rvalue = new(mem_ctx) ir_dereference_variable(uni);
125
126 progress = true;
127 }
128
129 } /* anonymous namespace */
130
131
132 static unsigned
count_uniforms(exec_list * instructions)133 count_uniforms(exec_list *instructions)
134 {
135 unsigned total = 0;
136
137 foreach_in_list(ir_instruction, node, instructions) {
138 ir_variable *const var = node->as_variable();
139
140 if (!var || var->data.mode != ir_var_uniform)
141 continue;
142
143 total += var->type->component_slots();
144 }
145 return total;
146 }
147
148 bool
lower_const_arrays_to_uniforms(exec_list * instructions,unsigned stage,unsigned max_uniform_components)149 lower_const_arrays_to_uniforms(exec_list *instructions, unsigned stage,
150 unsigned max_uniform_components)
151 {
152 unsigned uniform_components = count_uniforms(instructions);
153 unsigned free_uniform_slots = max_uniform_components - uniform_components;
154
155 lower_const_array_visitor v(instructions, stage, free_uniform_slots);
156 return v.run();
157 }
158