1/*
2 * Copyright © 2012 Vincent Lejeune
3 * Copyright © 2012 Intel Corporation
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
22 * DEALINGS IN THE SOFTWARE.
23 */
24
25#include "ir.h"
26#include "util/hash_table.h"
27
28/**
29 * \file lower_output_reads.cpp
30 *
31 * In GLSL, shader output variables (such as varyings) can be both read and
32 * written.  However, on some hardware, reading an output register causes
33 * trouble.
34 *
35 * This pass creates temporary shadow copies of every (used) shader output,
36 * and replaces all accesses to use those instead.  It also adds code to the
37 * main() function to copy the final values to the actual shader outputs.
38 */
39
40namespace {
41
42class output_read_remover : public ir_hierarchical_visitor {
43protected:
44   /**
45    * A hash table mapping from the original ir_variable shader outputs
46    * (ir_var_shader_out mode) to the new temporaries to be used instead.
47    */
48   hash_table *replacements;
49
50   unsigned stage;
51public:
52   output_read_remover(unsigned stage);
53   ~output_read_remover();
54   virtual ir_visitor_status visit(class ir_dereference_variable *);
55   virtual ir_visitor_status visit_leave(class ir_emit_vertex *);
56   virtual ir_visitor_status visit_leave(class ir_return *);
57   virtual ir_visitor_status visit_leave(class ir_function_signature *);
58};
59
60} /* anonymous namespace */
61
62/**
63 * Hash function for the output variables - computes the hash of the name.
64 * NOTE: We're using the name string to ensure that the hash doesn't depend
65 * on any random factors, otherwise the output_read_remover could produce
66 * the random order of the assignments.
67 *
68 * NOTE: If you want to reuse this function please take into account that
69 * generally the names of the variables are non-unique.
70 */
71static unsigned
72hash_table_var_hash(const void *key)
73{
74   const ir_variable * var = static_cast<const ir_variable *>(key);
75   return _mesa_hash_string(var->name);
76}
77
78output_read_remover::output_read_remover(unsigned stage)
79{
80   this->stage = stage;
81   replacements = _mesa_hash_table_create(NULL, hash_table_var_hash,
82                                          _mesa_key_pointer_equal);
83}
84
85output_read_remover::~output_read_remover()
86{
87   _mesa_hash_table_destroy(replacements, NULL);
88}
89
90ir_visitor_status
91output_read_remover::visit(ir_dereference_variable *ir)
92{
93   if (ir->var->data.mode != ir_var_shader_out || ir->var->data.fb_fetch_output)
94      return visit_continue;
95
96   hash_entry *entry = _mesa_hash_table_search(replacements, ir->var);
97   ir_variable *temp = entry ? (ir_variable *) entry->data : NULL;
98
99   /* If we don't have an existing temporary, create one. */
100   if (temp == NULL) {
101      void *var_ctx = ralloc_parent(ir->var);
102      temp = new(var_ctx) ir_variable(ir->var->type, ir->var->name,
103                                      ir_var_temporary);
104      /* copy flags which affect arithematical precision */
105      temp->data.invariant = ir->var->data.invariant;
106      temp->data.precise = ir->var->data.precise;
107      temp->data.precision = ir->var->data.precision;
108      _mesa_hash_table_insert(replacements, ir->var, temp);
109      ir->var->insert_after(temp);
110   }
111
112   /* Update the dereference to use the temporary */
113   ir->var = temp;
114
115   return visit_continue;
116}
117
118/**
119 * Create an assignment to copy a temporary value back to the actual output.
120 */
121static ir_assignment *
122copy(void *ctx, ir_variable *output, ir_variable *temp)
123{
124   ir_dereference_variable *lhs = new(ctx) ir_dereference_variable(output);
125   ir_dereference_variable *rhs = new(ctx) ir_dereference_variable(temp);
126   return new(ctx) ir_assignment(lhs, rhs);
127}
128
129/** Insert a copy-back assignment before a "return" statement or a call to
130 * EmitVertex().
131 */
132static void
133emit_return_copy(const void *key, void *data, void *closure)
134{
135   ir_return *ir = (ir_return *) closure;
136   ir->insert_before(copy(ir, (ir_variable *) key, (ir_variable *) data));
137}
138
139/** Insert a copy-back assignment at the end of the main() function */
140static void
141emit_main_copy(const void *key, void *data, void *closure)
142{
143   ir_function_signature *sig = (ir_function_signature *) closure;
144   sig->body.push_tail(copy(sig, (ir_variable *) key, (ir_variable *) data));
145}
146
147ir_visitor_status
148output_read_remover::visit_leave(ir_return *ir)
149{
150   hash_table_call_foreach(replacements, emit_return_copy, ir);
151   return visit_continue;
152}
153
154ir_visitor_status
155output_read_remover::visit_leave(ir_emit_vertex *ir)
156{
157   hash_table_call_foreach(replacements, emit_return_copy, ir);
158   return visit_continue;
159}
160
161ir_visitor_status
162output_read_remover::visit_leave(ir_function_signature *sig)
163{
164   if (strcmp(sig->function_name(), "main") != 0)
165      return visit_continue;
166
167   hash_table_call_foreach(replacements, emit_main_copy, sig);
168   return visit_continue;
169}
170
171void
172lower_output_reads(unsigned stage, exec_list *instructions)
173{
174   /* Due to the possible interactions between multiple tessellation control
175    * shader invocations, we leave output variables as-is.
176    */
177   if (stage == MESA_SHADER_TESS_CTRL)
178      return;
179
180   output_read_remover v(stage);
181   visit_list_elements(&v, instructions);
182}
183