2 * Copyright (c) 2016 Intel Corporation
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:
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
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 DEALINGS
25 #include "compiler/nir/nir_builder.h"
27 struct lower_intrinsics_state
{
30 struct brw_stage_prog_data
*prog_data
;
31 struct brw_cs_prog_data
*cs_prog_data
;
33 nir_function_impl
*impl
;
36 bool cs_thread_id_used
;
40 read_thread_local_id(struct lower_intrinsics_state
*state
)
42 nir_builder
*b
= &state
->builder
;
43 nir_shader
*nir
= state
->nir
;
44 const unsigned *sizes
= nir
->info
.cs
.local_size
;
45 const unsigned group_size
= sizes
[0] * sizes
[1] * sizes
[2];
47 /* Some programs have local_size dimensions so small that the thread local
48 * ID will always be 0.
51 return nir_imm_int(b
, 0);
53 assert(state
->cs_prog_data
->thread_local_id_index
>= 0);
54 state
->cs_thread_id_used
= true;
55 const int id_index
= state
->cs_prog_data
->thread_local_id_index
;
57 nir_intrinsic_instr
*load
=
58 nir_intrinsic_instr_create(nir
, nir_intrinsic_load_uniform
);
59 load
->num_components
= 1;
60 load
->src
[0] = nir_src_for_ssa(nir_imm_int(b
, 0));
61 nir_ssa_dest_init(&load
->instr
, &load
->dest
, 1, 32, NULL
);
62 nir_intrinsic_set_base(load
, id_index
* sizeof(uint32_t));
63 nir_intrinsic_set_range(load
, sizeof(uint32_t));
64 nir_builder_instr_insert(b
, &load
->instr
);
65 return &load
->dest
.ssa
;
69 lower_cs_intrinsics_convert_block(struct lower_intrinsics_state
*state
,
72 bool progress
= false;
73 nir_builder
*b
= &state
->builder
;
74 nir_shader
*nir
= state
->nir
;
76 nir_foreach_instr_safe(instr
, block
) {
77 if (instr
->type
!= nir_instr_type_intrinsic
)
80 nir_intrinsic_instr
*intrinsic
= nir_instr_as_intrinsic(instr
);
82 b
->cursor
= nir_after_instr(&intrinsic
->instr
);
85 switch (intrinsic
->intrinsic
) {
86 case nir_intrinsic_load_local_invocation_index
: {
87 assert(nir
->stage
== MESA_SHADER_COMPUTE
);
88 /* We construct the local invocation index from:
90 * gl_LocalInvocationIndex =
91 * cs_thread_local_id + subgroup_invocation;
93 nir_ssa_def
*thread_local_id
= read_thread_local_id(state
);
94 nir_ssa_def
*channel
= nir_load_subgroup_invocation(b
);
95 sysval
= nir_iadd(b
, channel
, thread_local_id
);
99 case nir_intrinsic_load_local_invocation_id
: {
100 assert(nir
->stage
== MESA_SHADER_COMPUTE
);
101 /* We lower gl_LocalInvocationID from gl_LocalInvocationIndex based
104 * gl_LocalInvocationID.x =
105 * gl_LocalInvocationIndex % gl_WorkGroupSize.x;
106 * gl_LocalInvocationID.y =
107 * (gl_LocalInvocationIndex / gl_WorkGroupSize.x) %
108 * gl_WorkGroupSize.y;
109 * gl_LocalInvocationID.z =
110 * (gl_LocalInvocationIndex /
111 * (gl_WorkGroupSize.x * gl_WorkGroupSize.y)) %
112 * gl_WorkGroupSize.z;
114 unsigned *size
= nir
->info
.cs
.local_size
;
116 nir_ssa_def
*local_index
= nir_load_local_invocation_index(b
);
118 nir_const_value uvec3
;
120 uvec3
.u32
[1] = size
[0];
121 uvec3
.u32
[2] = size
[0] * size
[1];
122 nir_ssa_def
*div_val
= nir_build_imm(b
, 3, 32, uvec3
);
123 uvec3
.u32
[0] = size
[0];
124 uvec3
.u32
[1] = size
[1];
125 uvec3
.u32
[2] = size
[2];
126 nir_ssa_def
*mod_val
= nir_build_imm(b
, 3, 32, uvec3
);
128 sysval
= nir_umod(b
, nir_udiv(b
, local_index
, div_val
), mod_val
);
136 nir_ssa_def_rewrite_uses(&intrinsic
->dest
.ssa
, nir_src_for_ssa(sysval
));
137 nir_instr_remove(&intrinsic
->instr
);
139 state
->progress
= true;
146 lower_cs_intrinsics_convert_impl(struct lower_intrinsics_state
*state
)
148 nir_builder_init(&state
->builder
, state
->impl
);
150 nir_foreach_block(block
, state
->impl
) {
151 lower_cs_intrinsics_convert_block(state
, block
);
154 nir_metadata_preserve(state
->impl
,
155 nir_metadata_block_index
| nir_metadata_dominance
);
159 brw_nir_lower_intrinsics(nir_shader
*nir
, struct brw_stage_prog_data
*prog_data
)
161 /* Currently we only lower intrinsics for compute shaders */
162 if (nir
->stage
!= MESA_SHADER_COMPUTE
)
165 bool progress
= false;
166 struct lower_intrinsics_state state
;
167 memset(&state
, 0, sizeof(state
));
169 state
.prog_data
= prog_data
;
172 state
.progress
= false;
173 nir_foreach_function(function
, nir
) {
174 if (function
->impl
) {
175 state
.impl
= function
->impl
;
176 lower_cs_intrinsics_convert_impl(&state
);
179 progress
|= state
.progress
;
180 } while (state
.progress
);
182 if (nir
->stage
== MESA_SHADER_COMPUTE
&& !state
.cs_thread_id_used
)
183 state
.cs_prog_data
->thread_local_id_index
= -1;