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.
35 struct constant_fold_state
{
37 nir_function_impl
*impl
;
42 constant_fold_alu_instr(nir_alu_instr
*instr
, void *mem_ctx
)
44 nir_const_value src
[4];
46 if (!instr
->dest
.dest
.is_ssa
)
49 /* In the case that any outputs/inputs have unsized types, then we need to
50 * guess the bit-size. In this case, the validator ensures that all
51 * bit-sizes match so we can just take the bit-size from first
52 * output/input with an unsized type. If all the outputs/inputs are sized
53 * then we don't need to guess the bit-size at all because the code we
54 * generate for constant opcodes in this case already knows the sizes of
55 * the types involved and does not need the provided bit-size for anything
56 * (although it still requires to receive a valid bit-size).
58 unsigned bit_size
= 0;
59 if (!nir_alu_type_get_type_size(nir_op_infos
[instr
->op
].output_type
))
60 bit_size
= instr
->dest
.dest
.ssa
.bit_size
;
62 for (unsigned i
= 0; i
< nir_op_infos
[instr
->op
].num_inputs
; i
++) {
63 if (!instr
->src
[i
].src
.is_ssa
)
67 !nir_alu_type_get_type_size(nir_op_infos
[instr
->op
].input_sizes
[i
])) {
68 bit_size
= instr
->src
[i
].src
.ssa
->bit_size
;
71 nir_instr
*src_instr
= instr
->src
[i
].src
.ssa
->parent_instr
;
73 if (src_instr
->type
!= nir_instr_type_load_const
)
75 nir_load_const_instr
* load_const
= nir_instr_as_load_const(src_instr
);
77 for (unsigned j
= 0; j
< nir_ssa_alu_instr_src_components(instr
, i
);
79 if (load_const
->def
.bit_size
== 64)
80 src
[i
].u64
[j
] = load_const
->value
.u64
[instr
->src
[i
].swizzle
[j
]];
82 src
[i
].u32
[j
] = load_const
->value
.u32
[instr
->src
[i
].swizzle
[j
]];
85 /* We shouldn't have any source modifiers in the optimization loop. */
86 assert(!instr
->src
[i
].abs
&& !instr
->src
[i
].negate
);
92 /* We shouldn't have any saturate modifiers in the optimization loop. */
93 assert(!instr
->dest
.saturate
);
95 nir_const_value dest
=
96 nir_eval_const_opcode(instr
->op
, instr
->dest
.dest
.ssa
.num_components
,
99 nir_load_const_instr
*new_instr
=
100 nir_load_const_instr_create(mem_ctx
,
101 instr
->dest
.dest
.ssa
.num_components
,
102 instr
->dest
.dest
.ssa
.bit_size
);
104 new_instr
->value
= dest
;
106 nir_instr_insert_before(&instr
->instr
, &new_instr
->instr
);
108 nir_ssa_def_rewrite_uses(&instr
->dest
.dest
.ssa
,
109 nir_src_for_ssa(&new_instr
->def
));
111 nir_instr_remove(&instr
->instr
);
118 constant_fold_deref(nir_instr
*instr
, nir_deref_var
*deref
)
120 bool progress
= false;
122 for (nir_deref
*tail
= deref
->deref
.child
; tail
; tail
= tail
->child
) {
123 if (tail
->deref_type
!= nir_deref_type_array
)
126 nir_deref_array
*arr
= nir_deref_as_array(tail
);
128 if (arr
->deref_array_type
== nir_deref_array_type_indirect
&&
129 arr
->indirect
.is_ssa
&&
130 arr
->indirect
.ssa
->parent_instr
->type
== nir_instr_type_load_const
) {
131 nir_load_const_instr
*indirect
=
132 nir_instr_as_load_const(arr
->indirect
.ssa
->parent_instr
);
134 arr
->base_offset
+= indirect
->value
.u32
[0];
136 /* Clear out the source */
137 nir_instr_rewrite_src(instr
, &arr
->indirect
, nir_src_for_ssa(NULL
));
139 arr
->deref_array_type
= nir_deref_array_type_direct
;
149 constant_fold_intrinsic_instr(nir_intrinsic_instr
*instr
)
151 bool progress
= false;
153 unsigned num_vars
= nir_intrinsic_infos
[instr
->intrinsic
].num_variables
;
154 for (unsigned i
= 0; i
< num_vars
; i
++) {
155 progress
|= constant_fold_deref(&instr
->instr
, instr
->variables
[i
]);
158 if (instr
->intrinsic
== nir_intrinsic_discard_if
) {
159 nir_const_value
*src_val
= nir_src_as_const_value(instr
->src
[0]);
160 if (src_val
&& src_val
->u32
[0] == 0) {
161 nir_instr_remove(&instr
->instr
);
170 constant_fold_tex_instr(nir_tex_instr
*instr
)
172 bool progress
= false;
175 progress
|= constant_fold_deref(&instr
->instr
, instr
->texture
);
178 progress
|= constant_fold_deref(&instr
->instr
, instr
->sampler
);
184 constant_fold_block(nir_block
*block
, void *mem_ctx
)
186 bool progress
= false;
188 nir_foreach_instr_safe(instr
, block
) {
189 switch (instr
->type
) {
190 case nir_instr_type_alu
:
191 progress
|= constant_fold_alu_instr(nir_instr_as_alu(instr
), mem_ctx
);
193 case nir_instr_type_intrinsic
:
195 constant_fold_intrinsic_instr(nir_instr_as_intrinsic(instr
));
197 case nir_instr_type_tex
:
198 progress
|= constant_fold_tex_instr(nir_instr_as_tex(instr
));
201 /* Don't know how to constant fold */
210 nir_opt_constant_folding_impl(nir_function_impl
*impl
)
212 void *mem_ctx
= ralloc_parent(impl
);
213 bool progress
= false;
215 nir_foreach_block(block
, impl
) {
216 progress
|= constant_fold_block(block
, mem_ctx
);
220 nir_metadata_preserve(impl
, nir_metadata_block_index
|
221 nir_metadata_dominance
);
227 nir_opt_constant_folding(nir_shader
*shader
)
229 bool progress
= false;
231 nir_foreach_function(function
, shader
) {
233 progress
|= nir_opt_constant_folding_impl(function
->impl
);