2 * Copyright © 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 "nir_builder.h"
28 emit_load_store(nir_builder
*b
, nir_intrinsic_instr
*orig_instr
,
29 nir_deref_var
*deref
, nir_deref
*tail
,
30 nir_ssa_def
**dest
, nir_ssa_def
*src
);
33 emit_indirect_load_store(nir_builder
*b
, nir_intrinsic_instr
*orig_instr
,
34 nir_deref_var
*deref
, nir_deref
*arr_parent
,
36 nir_ssa_def
**dest
, nir_ssa_def
*src
)
38 assert(arr_parent
->child
&&
39 arr_parent
->child
->deref_type
== nir_deref_type_array
);
40 nir_deref_array
*arr
= nir_deref_as_array(arr_parent
->child
);
41 assert(arr
->deref_array_type
== nir_deref_array_type_indirect
);
42 assert(arr
->indirect
.is_ssa
);
45 if (start
== end
- 1) {
46 /* Base case. Just emit the load/store op */
47 nir_deref_array direct
= *arr
;
48 direct
.deref_array_type
= nir_deref_array_type_direct
;
49 direct
.base_offset
+= start
;
50 direct
.indirect
= NIR_SRC_INIT
;
52 arr_parent
->child
= &direct
.deref
;
53 emit_load_store(b
, orig_instr
, deref
, &arr
->deref
, dest
, src
);
54 arr_parent
->child
= &arr
->deref
;
56 int mid
= start
+ (end
- start
) / 2;
58 nir_ssa_def
*then_dest
, *else_dest
;
60 nir_if
*if_stmt
= nir_if_create(b
->shader
);
61 if_stmt
->condition
= nir_src_for_ssa(nir_ilt(b
, arr
->indirect
.ssa
,
62 nir_imm_int(b
, mid
)));
63 nir_cf_node_insert(b
->cursor
, &if_stmt
->cf_node
);
65 b
->cursor
= nir_after_cf_list(&if_stmt
->then_list
);
66 emit_indirect_load_store(b
, orig_instr
, deref
, arr_parent
,
67 start
, mid
, &then_dest
, src
);
69 b
->cursor
= nir_after_cf_list(&if_stmt
->else_list
);
70 emit_indirect_load_store(b
, orig_instr
, deref
, arr_parent
,
71 mid
, end
, &else_dest
, src
);
73 b
->cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
76 /* We're a load. We need to insert a phi node */
77 nir_phi_instr
*phi
= nir_phi_instr_create(b
->shader
);
78 unsigned bit_size
= then_dest
->bit_size
;
79 nir_ssa_dest_init(&phi
->instr
, &phi
->dest
,
80 then_dest
->num_components
, bit_size
, NULL
);
82 nir_phi_src
*src0
= ralloc(phi
, nir_phi_src
);
83 src0
->pred
= nir_cf_node_as_block(nir_if_last_then_node(if_stmt
));
84 src0
->src
= nir_src_for_ssa(then_dest
);
85 exec_list_push_tail(&phi
->srcs
, &src0
->node
);
87 nir_phi_src
*src1
= ralloc(phi
, nir_phi_src
);
88 src1
->pred
= nir_cf_node_as_block(nir_if_last_else_node(if_stmt
));
89 src1
->src
= nir_src_for_ssa(else_dest
);
90 exec_list_push_tail(&phi
->srcs
, &src1
->node
);
92 nir_builder_instr_insert(b
, &phi
->instr
);
93 *dest
= &phi
->dest
.ssa
;
99 emit_load_store(nir_builder
*b
, nir_intrinsic_instr
*orig_instr
,
100 nir_deref_var
*deref
, nir_deref
*tail
,
101 nir_ssa_def
**dest
, nir_ssa_def
*src
)
103 for (; tail
->child
; tail
= tail
->child
) {
104 if (tail
->child
->deref_type
!= nir_deref_type_array
)
107 nir_deref_array
*arr
= nir_deref_as_array(tail
->child
);
108 if (arr
->deref_array_type
!= nir_deref_array_type_indirect
)
111 int length
= glsl_get_length(tail
->type
);
113 emit_indirect_load_store(b
, orig_instr
, deref
, tail
, -arr
->base_offset
,
114 length
- arr
->base_offset
, dest
, src
);
118 assert(tail
&& tail
->child
== NULL
);
120 /* We reached the end of the deref chain. Emit the instruction */
123 /* This is a load instruction */
124 nir_intrinsic_instr
*load
=
125 nir_intrinsic_instr_create(b
->shader
, nir_intrinsic_load_var
);
126 load
->num_components
= orig_instr
->num_components
;
128 nir_deref_as_var(nir_copy_deref(load
, &deref
->deref
));
129 unsigned bit_size
= orig_instr
->dest
.ssa
.bit_size
;
130 nir_ssa_dest_init(&load
->instr
, &load
->dest
,
131 load
->num_components
, bit_size
, NULL
);
132 nir_builder_instr_insert(b
, &load
->instr
);
133 *dest
= &load
->dest
.ssa
;
135 /* This is a store instruction */
136 nir_intrinsic_instr
*store
=
137 nir_intrinsic_instr_create(b
->shader
, nir_intrinsic_store_var
);
138 store
->num_components
= orig_instr
->num_components
;
139 nir_intrinsic_set_write_mask(store
, nir_intrinsic_write_mask(orig_instr
));
140 store
->variables
[0] =
141 nir_deref_as_var(nir_copy_deref(store
, &deref
->deref
));
142 store
->src
[0] = nir_src_for_ssa(src
);
143 nir_builder_instr_insert(b
, &store
->instr
);
148 deref_has_indirect(nir_deref_var
*deref
)
150 for (nir_deref
*tail
= deref
->deref
.child
; tail
; tail
= tail
->child
) {
151 if (tail
->deref_type
!= nir_deref_type_array
)
154 nir_deref_array
*arr
= nir_deref_as_array(tail
);
155 if (arr
->deref_array_type
== nir_deref_array_type_indirect
)
163 lower_indirect_block(nir_block
*block
, nir_builder
*b
,
164 nir_variable_mode modes
)
166 bool progress
= false;
168 nir_foreach_instr_safe(instr
, block
) {
169 if (instr
->type
!= nir_instr_type_intrinsic
)
172 nir_intrinsic_instr
*intrin
= nir_instr_as_intrinsic(instr
);
173 if (intrin
->intrinsic
!= nir_intrinsic_load_var
&&
174 intrin
->intrinsic
!= nir_intrinsic_store_var
)
177 if (!deref_has_indirect(intrin
->variables
[0]))
180 /* Only lower variables whose mode is in the mask */
181 if (!(modes
& intrin
->variables
[0]->var
->data
.mode
))
184 b
->cursor
= nir_before_instr(&intrin
->instr
);
186 if (intrin
->intrinsic
== nir_intrinsic_load_var
) {
188 emit_load_store(b
, intrin
, intrin
->variables
[0],
189 &intrin
->variables
[0]->deref
, &result
, NULL
);
190 nir_ssa_def_rewrite_uses(&intrin
->dest
.ssa
, nir_src_for_ssa(result
));
192 assert(intrin
->src
[0].is_ssa
);
193 emit_load_store(b
, intrin
, intrin
->variables
[0],
194 &intrin
->variables
[0]->deref
, NULL
, intrin
->src
[0].ssa
);
196 nir_instr_remove(&intrin
->instr
);
204 lower_indirects_impl(nir_function_impl
*impl
, nir_variable_mode modes
)
207 nir_builder_init(&builder
, impl
);
208 bool progress
= false;
210 nir_foreach_block_safe(block
, impl
) {
211 progress
|= lower_indirect_block(block
, &builder
, modes
);
215 nir_metadata_preserve(impl
, nir_metadata_none
);
220 /** Lowers indirect variable loads/stores to direct loads/stores.
222 * The pass works by replacing any indirect load or store with an if-ladder
223 * that does a binary search on the array index.
226 nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
)
228 bool progress
= false;
230 nir_foreach_function(function
, shader
) {
232 progress
= lower_indirects_impl(function
->impl
, modes
) || progress
;