1 /*
2 * Copyright © 2017 Ilia Mirkin
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_cs_derived.cpp
26 *
27 * For hardware that does not support the gl_GlobalInvocationID and
28 * gl_LocalInvocationIndex system values, replace them with fresh
29 * globals. Note that we can't rely on gl_WorkGroupSize or
30 * gl_LocalGroupSizeARB being available, since they may only have been defined
31 * in a non-main shader.
32 *
33 * [ This can happen if only a secondary shader has the layout(local_size_*)
34 * declaration. ]
35 *
36 * This is meant to be run post-linking.
37 */
38
39 #include "glsl_symbol_table.h"
40 #include "ir_hierarchical_visitor.h"
41 #include "ir.h"
42 #include "ir_builder.h"
43 #include "linker.h"
44 #include "program/prog_statevars.h"
45 #include "builtin_functions.h"
46 #include "main/mtypes.h"
47
48 using namespace ir_builder;
49
50 namespace {
51
52 class lower_cs_derived_visitor : public ir_hierarchical_visitor {
53 public:
lower_cs_derived_visitor(gl_linked_shader * shader)54 explicit lower_cs_derived_visitor(gl_linked_shader *shader)
55 : progress(false),
56 shader(shader),
57 local_size_variable(shader->Program->info.workgroup_size_variable),
58 gl_WorkGroupSize(NULL),
59 gl_WorkGroupID(NULL),
60 gl_LocalInvocationID(NULL),
61 gl_GlobalInvocationID(NULL),
62 gl_LocalInvocationIndex(NULL)
63 {
64 main_sig = _mesa_get_main_function_signature(shader->symbols);
65 assert(main_sig);
66 }
67
68 virtual ir_visitor_status visit(ir_dereference_variable *);
69
70 ir_variable *add_system_value(
71 int slot, const glsl_type *type, const char *name);
72 void find_sysvals();
73 void make_gl_GlobalInvocationID();
74 void make_gl_LocalInvocationIndex();
75
76 bool progress;
77
78 private:
79 gl_linked_shader *shader;
80 bool local_size_variable;
81 ir_function_signature *main_sig;
82
83 ir_rvalue *gl_WorkGroupSize;
84 ir_variable *gl_WorkGroupID;
85 ir_variable *gl_LocalInvocationID;
86
87 ir_variable *gl_GlobalInvocationID;
88 ir_variable *gl_LocalInvocationIndex;
89 };
90
91 } /* anonymous namespace */
92
93 ir_variable *
add_system_value(int slot,const glsl_type * type,const char * name)94 lower_cs_derived_visitor::add_system_value(
95 int slot, const glsl_type *type, const char *name)
96 {
97 ir_variable *var = new(shader) ir_variable(type, name, ir_var_system_value);
98 var->data.how_declared = ir_var_declared_implicitly;
99 var->data.read_only = true;
100 var->data.location = slot;
101 var->data.explicit_location = true;
102 var->data.explicit_index = 0;
103 shader->ir->push_head(var);
104
105 return var;
106 }
107
108 void
find_sysvals()109 lower_cs_derived_visitor::find_sysvals()
110 {
111 if (gl_WorkGroupSize != NULL)
112 return;
113
114 ir_variable *WorkGroupSize;
115 if (local_size_variable)
116 WorkGroupSize = shader->symbols->get_variable("gl_LocalGroupSizeARB");
117 else
118 WorkGroupSize = shader->symbols->get_variable("gl_WorkGroupSize");
119 if (WorkGroupSize)
120 gl_WorkGroupSize = new(shader) ir_dereference_variable(WorkGroupSize);
121 gl_WorkGroupID = shader->symbols->get_variable("gl_WorkGroupID");
122 gl_LocalInvocationID = shader->symbols->get_variable("gl_LocalInvocationID");
123
124 /*
125 * These may be missing due to either dead code elimination, or, in the
126 * case of the group size, due to the layout being declared in a non-main
127 * shader. Re-create them.
128 */
129
130 if (!gl_WorkGroupID)
131 gl_WorkGroupID = add_system_value(
132 SYSTEM_VALUE_WORKGROUP_ID, glsl_type::uvec3_type, "gl_WorkGroupID");
133 if (!gl_LocalInvocationID)
134 gl_LocalInvocationID = add_system_value(
135 SYSTEM_VALUE_LOCAL_INVOCATION_ID, glsl_type::uvec3_type,
136 "gl_LocalInvocationID");
137 if (!WorkGroupSize) {
138 if (local_size_variable) {
139 gl_WorkGroupSize = new(shader) ir_dereference_variable(
140 add_system_value(
141 SYSTEM_VALUE_WORKGROUP_SIZE, glsl_type::uvec3_type,
142 "gl_LocalGroupSizeARB"));
143 } else {
144 ir_constant_data data;
145 memset(&data, 0, sizeof(data));
146 for (int i = 0; i < 3; i++)
147 data.u[i] = shader->Program->info.workgroup_size[i];
148 gl_WorkGroupSize = new(shader) ir_constant(glsl_type::uvec3_type, &data);
149 }
150 }
151 }
152
153 void
make_gl_GlobalInvocationID()154 lower_cs_derived_visitor::make_gl_GlobalInvocationID()
155 {
156 if (gl_GlobalInvocationID != NULL)
157 return;
158
159 find_sysvals();
160
161 /* gl_GlobalInvocationID =
162 * gl_WorkGroupID * gl_WorkGroupSize + gl_LocalInvocationID
163 */
164 gl_GlobalInvocationID = new(shader) ir_variable(
165 glsl_type::uvec3_type, "__GlobalInvocationID", ir_var_temporary);
166 shader->ir->push_head(gl_GlobalInvocationID);
167
168 ir_instruction *inst =
169 assign(gl_GlobalInvocationID,
170 add(mul(gl_WorkGroupID, gl_WorkGroupSize->clone(shader, NULL)),
171 gl_LocalInvocationID));
172 main_sig->body.push_head(inst);
173 }
174
175 void
make_gl_LocalInvocationIndex()176 lower_cs_derived_visitor::make_gl_LocalInvocationIndex()
177 {
178 if (gl_LocalInvocationIndex != NULL)
179 return;
180
181 find_sysvals();
182
183 /* gl_LocalInvocationIndex =
184 * gl_LocalInvocationID.z * gl_WorkGroupSize.x * gl_WorkGroupSize.y +
185 * gl_LocalInvocationID.y * gl_WorkGroupSize.x +
186 * gl_LocalInvocationID.x;
187 */
188 gl_LocalInvocationIndex = new(shader)
189 ir_variable(glsl_type::uint_type, "__LocalInvocationIndex", ir_var_temporary);
190 shader->ir->push_head(gl_LocalInvocationIndex);
191
192 ir_expression *index_z =
193 mul(mul(swizzle_z(gl_LocalInvocationID), swizzle_x(gl_WorkGroupSize->clone(shader, NULL))),
194 swizzle_y(gl_WorkGroupSize->clone(shader, NULL)));
195 ir_expression *index_y =
196 mul(swizzle_y(gl_LocalInvocationID), swizzle_x(gl_WorkGroupSize->clone(shader, NULL)));
197 ir_expression *index_y_plus_z = add(index_y, index_z);
198 operand index_x(swizzle_x(gl_LocalInvocationID));
199 ir_expression *index_x_plus_y_plus_z = add(index_y_plus_z, index_x);
200 ir_instruction *inst =
201 assign(gl_LocalInvocationIndex, index_x_plus_y_plus_z);
202 main_sig->body.push_head(inst);
203 }
204
205 ir_visitor_status
visit(ir_dereference_variable * ir)206 lower_cs_derived_visitor::visit(ir_dereference_variable *ir)
207 {
208 if (ir->var->data.mode == ir_var_system_value &&
209 ir->var->data.location == SYSTEM_VALUE_GLOBAL_INVOCATION_ID) {
210 make_gl_GlobalInvocationID();
211 ir->var = gl_GlobalInvocationID;
212 progress = true;
213 }
214
215 if (ir->var->data.mode == ir_var_system_value &&
216 ir->var->data.location == SYSTEM_VALUE_LOCAL_INVOCATION_INDEX) {
217 make_gl_LocalInvocationIndex();
218 ir->var = gl_LocalInvocationIndex;
219 progress = true;
220 }
221
222 return visit_continue;
223 }
224
225 bool
lower_cs_derived(gl_linked_shader * shader)226 lower_cs_derived(gl_linked_shader *shader)
227 {
228 if (shader->Stage != MESA_SHADER_COMPUTE)
229 return false;
230
231 lower_cs_derived_visitor v(shader);
232 v.run(shader->ir);
233
234 return v.progress;
235 }
236