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 * Connor Abbott (cwabbott0@gmail.com)
29 #include <main/imports.h>
32 * SSA-based copy propagation
35 static bool is_move(nir_alu_instr
*instr
)
37 assert(instr
->src
[0].src
.is_ssa
);
39 if (instr
->op
!= nir_op_fmov
&&
40 instr
->op
!= nir_op_imov
)
43 if (instr
->dest
.saturate
)
46 /* we handle modifiers in a separate pass */
48 if (instr
->src
[0].abs
|| instr
->src
[0].negate
)
55 static bool is_vec(nir_alu_instr
*instr
)
57 for (unsigned i
= 0; i
< nir_op_infos
[instr
->op
].num_inputs
; i
++) {
58 assert(instr
->src
[i
].src
.is_ssa
);
60 /* we handle modifiers in a separate pass */
61 if (instr
->src
[i
].abs
|| instr
->src
[i
].negate
)
65 return instr
->op
== nir_op_vec2
||
66 instr
->op
== nir_op_vec3
||
67 instr
->op
== nir_op_vec4
;
71 is_swizzleless_move(nir_alu_instr
*instr
)
74 for (unsigned i
= 0; i
< 4; i
++) {
75 if (!((instr
->dest
.write_mask
>> i
) & 1))
77 if (instr
->src
[0].swizzle
[i
] != i
)
81 } else if (is_vec(instr
)) {
82 nir_ssa_def
*def
= NULL
;
83 for (unsigned i
= 0; i
< nir_op_infos
[instr
->op
].num_inputs
; i
++) {
84 if (instr
->src
[i
].swizzle
[0] != i
)
88 def
= instr
->src
[i
].src
.ssa
;
89 } else if (instr
->src
[i
].src
.ssa
!= def
) {
100 copy_prop_src(nir_src
*src
, nir_instr
*parent_instr
, nir_if
*parent_if
,
101 unsigned num_components
)
105 nir_instr
*src_instr
= src
->ssa
->parent_instr
;
106 nir_ssa_def
*copy_def
;
107 if (src_instr
->type
== nir_instr_type_alu
) {
108 nir_alu_instr
*alu_instr
= nir_instr_as_alu(src_instr
);
109 if (!is_swizzleless_move(alu_instr
))
112 if (alu_instr
->src
[0].src
.ssa
->num_components
!= num_components
)
115 copy_def
= alu_instr
->src
[0].src
.ssa
;
121 nir_instr_rewrite_src(parent_instr
, src
, nir_src_for_ssa(copy_def
));
123 assert(src
== &parent_if
->condition
);
124 nir_if_rewrite_condition(parent_if
, nir_src_for_ssa(copy_def
));
131 copy_prop_alu_src(nir_alu_instr
*parent_alu_instr
, unsigned index
)
133 nir_alu_src
*src
= &parent_alu_instr
->src
[index
];
134 assert(src
->src
.is_ssa
);
136 nir_instr
*src_instr
= src
->src
.ssa
->parent_instr
;
137 if (src_instr
->type
!= nir_instr_type_alu
)
140 nir_alu_instr
*alu_instr
= nir_instr_as_alu(src_instr
);
141 if (!is_move(alu_instr
) && !is_vec(alu_instr
))
145 unsigned new_swizzle
[4] = {0, 0, 0, 0};
147 if (alu_instr
->op
== nir_op_fmov
||
148 alu_instr
->op
== nir_op_imov
) {
149 for (unsigned i
= 0; i
< 4; i
++)
150 new_swizzle
[i
] = alu_instr
->src
[0].swizzle
[src
->swizzle
[i
]];
151 def
= alu_instr
->src
[0].src
.ssa
;
155 for (unsigned i
= 0; i
< 4; i
++) {
156 if (!nir_alu_instr_channel_used(parent_alu_instr
, index
, i
))
159 nir_ssa_def
*new_def
= alu_instr
->src
[src
->swizzle
[i
]].src
.ssa
;
166 new_swizzle
[i
] = alu_instr
->src
[src
->swizzle
[i
]].swizzle
[0];
170 for (unsigned i
= 0; i
< 4; i
++)
171 src
->swizzle
[i
] = new_swizzle
[i
];
173 nir_instr_rewrite_src(&parent_alu_instr
->instr
, &src
->src
,
174 nir_src_for_ssa(def
));
180 copy_prop_instr(nir_instr
*instr
)
182 bool progress
= false;
183 switch (instr
->type
) {
184 case nir_instr_type_alu
: {
185 nir_alu_instr
*alu_instr
= nir_instr_as_alu(instr
);
187 for (unsigned i
= 0; i
< nir_op_infos
[alu_instr
->op
].num_inputs
; i
++)
188 while (copy_prop_alu_src(alu_instr
, i
))
194 case nir_instr_type_deref
: {
195 nir_deref_instr
*deref
= nir_instr_as_deref(instr
);
197 if (deref
->deref_type
!= nir_deref_type_var
) {
198 assert(deref
->dest
.is_ssa
);
199 const unsigned comps
= deref
->dest
.ssa
.num_components
;
200 while (copy_prop_src(&deref
->parent
, instr
, NULL
, comps
))
204 if (deref
->deref_type
== nir_deref_type_array
||
205 deref
->deref_type
== nir_deref_type_ptr_as_array
) {
206 while (copy_prop_src(&deref
->arr
.index
, instr
, NULL
, 1))
213 case nir_instr_type_tex
: {
214 nir_tex_instr
*tex
= nir_instr_as_tex(instr
);
215 for (unsigned i
= 0; i
< tex
->num_srcs
; i
++) {
216 unsigned num_components
= nir_tex_instr_src_size(tex
, i
);
217 while (copy_prop_src(&tex
->src
[i
].src
, instr
, NULL
, num_components
))
224 case nir_instr_type_intrinsic
: {
225 nir_intrinsic_instr
*intrin
= nir_instr_as_intrinsic(instr
);
227 i
< nir_intrinsic_infos
[intrin
->intrinsic
].num_srcs
; i
++) {
228 unsigned num_components
= nir_intrinsic_src_components(intrin
, i
);
230 while (copy_prop_src(&intrin
->src
[i
], instr
, NULL
, num_components
))
237 case nir_instr_type_phi
: {
238 nir_phi_instr
*phi
= nir_instr_as_phi(instr
);
239 assert(phi
->dest
.is_ssa
);
240 unsigned num_components
= phi
->dest
.ssa
.num_components
;
241 nir_foreach_phi_src(src
, phi
) {
242 while (copy_prop_src(&src
->src
, instr
, NULL
, num_components
))
255 copy_prop_if(nir_if
*if_stmt
)
257 return copy_prop_src(&if_stmt
->condition
, NULL
, if_stmt
, 1);
261 nir_copy_prop_impl(nir_function_impl
*impl
)
263 bool progress
= false;
265 nir_foreach_block(block
, impl
) {
266 nir_foreach_instr(instr
, block
) {
267 if (copy_prop_instr(instr
))
271 nir_if
*if_stmt
= nir_block_get_following_if(block
);
272 if (if_stmt
&& copy_prop_if(if_stmt
))
277 nir_metadata_preserve(impl
, nir_metadata_block_index
|
278 nir_metadata_dominance
);
281 impl
->valid_metadata
&= ~nir_metadata_not_properly_reset
;
289 nir_copy_prop(nir_shader
*shader
)
291 bool progress
= false;
293 nir_foreach_function(function
, shader
) {
294 if (function
->impl
&& nir_copy_prop_impl(function
->impl
))