2 * Copyright © 2014 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
24 * Jason Ekstrand (jason@jlekstrand.net)
28 #include "nir_constant_expressions.h"
32 * Implements SSA-based constant folding.
36 constant_fold_alu_instr(nir_alu_instr
*instr
, void *mem_ctx
)
38 nir_const_value src
[NIR_MAX_VEC_COMPONENTS
][NIR_MAX_VEC_COMPONENTS
];
40 if (!instr
->dest
.dest
.is_ssa
)
43 /* In the case that any outputs/inputs have unsized types, then we need to
44 * guess the bit-size. In this case, the validator ensures that all
45 * bit-sizes match so we can just take the bit-size from first
46 * output/input with an unsized type. If all the outputs/inputs are sized
47 * then we don't need to guess the bit-size at all because the code we
48 * generate for constant opcodes in this case already knows the sizes of
49 * the types involved and does not need the provided bit-size for anything
50 * (although it still requires to receive a valid bit-size).
52 unsigned bit_size
= 0;
53 if (!nir_alu_type_get_type_size(nir_op_infos
[instr
->op
].output_type
))
54 bit_size
= instr
->dest
.dest
.ssa
.bit_size
;
56 for (unsigned i
= 0; i
< nir_op_infos
[instr
->op
].num_inputs
; i
++) {
57 if (!instr
->src
[i
].src
.is_ssa
)
61 !nir_alu_type_get_type_size(nir_op_infos
[instr
->op
].input_types
[i
]))
62 bit_size
= instr
->src
[i
].src
.ssa
->bit_size
;
64 nir_instr
*src_instr
= instr
->src
[i
].src
.ssa
->parent_instr
;
66 if (src_instr
->type
!= nir_instr_type_load_const
)
68 nir_load_const_instr
* load_const
= nir_instr_as_load_const(src_instr
);
70 for (unsigned j
= 0; j
< nir_ssa_alu_instr_src_components(instr
, i
);
72 src
[i
][j
] = load_const
->value
[instr
->src
[i
].swizzle
[j
]];
75 /* We shouldn't have any source modifiers in the optimization loop. */
76 assert(!instr
->src
[i
].abs
&& !instr
->src
[i
].negate
);
82 /* We shouldn't have any saturate modifiers in the optimization loop. */
83 assert(!instr
->dest
.saturate
);
85 nir_const_value dest
[NIR_MAX_VEC_COMPONENTS
];
86 nir_const_value
*srcs
[NIR_MAX_VEC_COMPONENTS
];
87 memset(dest
, 0, sizeof(dest
));
88 for (unsigned i
= 0; i
< nir_op_infos
[instr
->op
].num_inputs
; ++i
)
90 nir_eval_const_opcode(instr
->op
, dest
, instr
->dest
.dest
.ssa
.num_components
,
93 nir_load_const_instr
*new_instr
=
94 nir_load_const_instr_create(mem_ctx
,
95 instr
->dest
.dest
.ssa
.num_components
,
96 instr
->dest
.dest
.ssa
.bit_size
);
98 memcpy(new_instr
->value
, dest
, sizeof(*new_instr
->value
) * new_instr
->def
.num_components
);
100 nir_instr_insert_before(&instr
->instr
, &new_instr
->instr
);
102 nir_ssa_def_rewrite_uses(&instr
->dest
.dest
.ssa
,
103 nir_src_for_ssa(&new_instr
->def
));
105 nir_instr_remove(&instr
->instr
);
112 constant_fold_intrinsic_instr(nir_intrinsic_instr
*instr
)
114 bool progress
= false;
116 if ((instr
->intrinsic
== nir_intrinsic_demote_if
||
117 instr
->intrinsic
== nir_intrinsic_discard_if
) &&
118 nir_src_is_const(instr
->src
[0])) {
119 if (nir_src_as_bool(instr
->src
[0])) {
120 /* This method of getting a nir_shader * from a nir_instr is
121 * admittedly gross, but given the rarity of hitting this case I think
122 * it's preferable to plumbing an otherwise unused nir_shader *
123 * parameter through four functions to get here.
125 nir_cf_node
*cf_node
= &instr
->instr
.block
->cf_node
;
126 nir_function_impl
*impl
= nir_cf_node_get_function(cf_node
);
127 nir_shader
*shader
= impl
->function
->shader
;
129 nir_intrinsic_op op
= instr
->intrinsic
== nir_intrinsic_discard_if
?
130 nir_intrinsic_discard
:
131 nir_intrinsic_demote
;
132 nir_intrinsic_instr
*new_instr
= nir_intrinsic_instr_create(shader
, op
);
133 nir_instr_insert_before(&instr
->instr
, &new_instr
->instr
);
134 nir_instr_remove(&instr
->instr
);
137 /* We're not discarding, just delete the instruction */
138 nir_instr_remove(&instr
->instr
);
147 constant_fold_block(nir_block
*block
, void *mem_ctx
)
149 bool progress
= false;
151 nir_foreach_instr_safe(instr
, block
) {
152 switch (instr
->type
) {
153 case nir_instr_type_alu
:
154 progress
|= constant_fold_alu_instr(nir_instr_as_alu(instr
), mem_ctx
);
156 case nir_instr_type_intrinsic
:
158 constant_fold_intrinsic_instr(nir_instr_as_intrinsic(instr
));
161 /* Don't know how to constant fold */
170 nir_opt_constant_folding_impl(nir_function_impl
*impl
)
172 void *mem_ctx
= ralloc_parent(impl
);
173 bool progress
= false;
175 nir_foreach_block(block
, impl
) {
176 progress
|= constant_fold_block(block
, mem_ctx
);
180 nir_metadata_preserve(impl
, nir_metadata_block_index
|
181 nir_metadata_dominance
);
184 impl
->valid_metadata
&= ~nir_metadata_not_properly_reset
;
192 nir_opt_constant_folding(nir_shader
*shader
)
194 bool progress
= false;
196 nir_foreach_function(function
, shader
) {
198 progress
|= nir_opt_constant_folding_impl(function
->impl
);