2 * Copyright © 2015 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 "nir/nir_builder.h"
27 struct apply_dynamic_offsets_state
{
31 const struct anv_pipeline_layout
*layout
;
33 uint32_t indices_start
;
37 apply_dynamic_offsets_block(nir_block
*block
, void *void_state
)
39 struct apply_dynamic_offsets_state
*state
= void_state
;
40 struct anv_descriptor_set_layout
*set_layout
;
42 nir_builder
*b
= &state
->builder
;
44 nir_foreach_instr_safe(block
, instr
) {
45 if (instr
->type
!= nir_instr_type_intrinsic
)
48 nir_intrinsic_instr
*intrin
= nir_instr_as_intrinsic(instr
);
50 unsigned block_idx_src
;
51 switch (intrin
->intrinsic
) {
52 case nir_intrinsic_load_ubo
:
53 case nir_intrinsic_load_ssbo
:
56 case nir_intrinsic_store_ssbo
:
60 continue; /* the loop */
63 nir_instr
*res_instr
= intrin
->src
[block_idx_src
].ssa
->parent_instr
;
64 assert(res_instr
->type
== nir_instr_type_intrinsic
);
65 nir_intrinsic_instr
*res_intrin
= nir_instr_as_intrinsic(res_instr
);
66 assert(res_intrin
->intrinsic
== nir_intrinsic_vulkan_resource_index
);
68 unsigned set
= res_intrin
->const_index
[0];
69 unsigned binding
= res_intrin
->const_index
[1];
71 set_layout
= state
->layout
->set
[set
].layout
;
72 if (set_layout
->binding
[binding
].dynamic_offset_index
< 0)
75 b
->cursor
= nir_before_instr(&intrin
->instr
);
77 /* First, we need to generate the uniform load for the buffer offset */
78 uint32_t index
= state
->layout
->set
[set
].dynamic_offset_start
+
79 set_layout
->binding
[binding
].dynamic_offset_index
;
80 uint32_t array_size
= set_layout
->binding
[binding
].array_size
;
82 nir_intrinsic_instr
*offset_load
=
83 nir_intrinsic_instr_create(state
->shader
, nir_intrinsic_load_uniform
);
84 offset_load
->num_components
= 2;
85 nir_intrinsic_set_base(offset_load
, state
->indices_start
+ index
* 8);
86 nir_intrinsic_set_range(offset_load
, array_size
* 8);
87 offset_load
->src
[0] = nir_src_for_ssa(nir_imul(b
, res_intrin
->src
[0].ssa
,
90 nir_ssa_dest_init(&offset_load
->instr
, &offset_load
->dest
, 2, 32, NULL
);
91 nir_builder_instr_insert(b
, &offset_load
->instr
);
93 nir_src
*offset_src
= nir_get_io_offset_src(intrin
);
94 nir_ssa_def
*new_offset
= nir_iadd(b
, offset_src
->ssa
,
95 &offset_load
->dest
.ssa
);
97 /* In order to avoid out-of-bounds access, we predicate */
98 nir_ssa_def
*pred
= nir_uge(b
, nir_channel(b
, &offset_load
->dest
.ssa
, 1),
100 nir_if
*if_stmt
= nir_if_create(b
->shader
);
101 if_stmt
->condition
= nir_src_for_ssa(pred
);
102 nir_cf_node_insert(b
->cursor
, &if_stmt
->cf_node
);
104 nir_instr_remove(&intrin
->instr
);
105 *offset_src
= nir_src_for_ssa(new_offset
);
106 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, &intrin
->instr
);
108 if (intrin
->intrinsic
!= nir_intrinsic_store_ssbo
) {
109 /* It's a load, we need a phi node */
110 nir_phi_instr
*phi
= nir_phi_instr_create(b
->shader
);
111 nir_ssa_dest_init(&phi
->instr
, &phi
->dest
,
112 intrin
->num_components
,
113 intrin
->dest
.ssa
.bit_size
, NULL
);
115 nir_phi_src
*src1
= ralloc(phi
, nir_phi_src
);
116 struct exec_node
*tnode
= exec_list_get_tail(&if_stmt
->then_list
);
117 src1
->pred
= exec_node_data(nir_block
, tnode
, cf_node
.node
);
118 src1
->src
= nir_src_for_ssa(&intrin
->dest
.ssa
);
119 exec_list_push_tail(&phi
->srcs
, &src1
->node
);
121 b
->cursor
= nir_after_cf_list(&if_stmt
->else_list
);
122 nir_const_value zero_val
= { .u32
= { 0, 0, 0, 0 } };
123 nir_ssa_def
*zero
= nir_build_imm(b
, intrin
->num_components
,
124 intrin
->dest
.ssa
.bit_size
, zero_val
);
126 nir_phi_src
*src2
= ralloc(phi
, nir_phi_src
);
127 struct exec_node
*enode
= exec_list_get_tail(&if_stmt
->else_list
);
128 src2
->pred
= exec_node_data(nir_block
, enode
, cf_node
.node
);
129 src2
->src
= nir_src_for_ssa(zero
);
130 exec_list_push_tail(&phi
->srcs
, &src2
->node
);
132 assert(intrin
->dest
.is_ssa
);
133 nir_ssa_def_rewrite_uses(&intrin
->dest
.ssa
,
134 nir_src_for_ssa(&phi
->dest
.ssa
));
136 nir_instr_insert_after_cf(&if_stmt
->cf_node
, &phi
->instr
);
144 anv_nir_apply_dynamic_offsets(struct anv_pipeline
*pipeline
,
146 struct brw_stage_prog_data
*prog_data
)
148 struct apply_dynamic_offsets_state state
= {
150 .layout
= pipeline
->layout
,
151 .indices_start
= shader
->num_uniforms
,
154 if (!state
.layout
|| !state
.layout
->stage
[shader
->stage
].has_dynamic_offsets
)
157 nir_foreach_function(shader
, function
) {
158 if (function
->impl
) {
159 nir_builder_init(&state
.builder
, function
->impl
);
160 nir_foreach_block_call(function
->impl
, apply_dynamic_offsets_block
,
162 nir_metadata_preserve(function
->impl
, nir_metadata_block_index
|
163 nir_metadata_dominance
);
167 struct anv_push_constants
*null_data
= NULL
;
168 for (unsigned i
= 0; i
< MAX_DYNAMIC_BUFFERS
; i
++) {
169 prog_data
->param
[i
* 2 + shader
->num_uniforms
/ 4] =
170 (const union gl_constant_value
*)&null_data
->dynamic
[i
].offset
;
171 prog_data
->param
[i
* 2 + 1 + shader
->num_uniforms
/ 4] =
172 (const union gl_constant_value
*)&null_data
->dynamic
[i
].range
;
175 shader
->num_uniforms
+= MAX_DYNAMIC_BUFFERS
* 8;