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)
31 * Implements a simple pass that lowers vecN instructions to a series of
32 * moves with partial writes.
36 src_matches_dest_reg(nir_dest
*dest
, nir_src
*src
)
38 if (dest
->is_ssa
|| src
->is_ssa
)
41 return (dest
->reg
.reg
== src
->reg
.reg
&&
42 dest
->reg
.base_offset
== src
->reg
.base_offset
&&
43 !dest
->reg
.indirect
&&
48 * For a given starting writemask channel and corresponding source index in
49 * the vec instruction, insert a MOV to the vec instruction's dest of all the
50 * writemask channels that get read from the same src reg.
52 * Returns the writemask of our MOV, so the parent loop calling this knows
53 * which ones have been processed.
56 insert_mov(nir_alu_instr
*vec
, unsigned start_channel
,
57 unsigned start_src_idx
, void *mem_ctx
)
59 unsigned src_idx
= start_src_idx
;
60 assert(src_idx
< nir_op_infos
[vec
->op
].num_inputs
);
62 nir_alu_instr
*mov
= nir_alu_instr_create(mem_ctx
, nir_op_imov
);
63 nir_alu_src_copy(&mov
->src
[0], &vec
->src
[src_idx
], mem_ctx
);
64 nir_alu_dest_copy(&mov
->dest
, &vec
->dest
, mem_ctx
);
66 mov
->dest
.write_mask
= (1u << start_channel
);
67 mov
->src
[0].swizzle
[start_channel
] = vec
->src
[src_idx
].swizzle
[0];
70 for (unsigned i
= start_channel
+ 1; i
< 4; i
++) {
71 if (!(vec
->dest
.write_mask
& (1 << i
)))
74 if (nir_srcs_equal(vec
->src
[src_idx
].src
, vec
->src
[start_src_idx
].src
)) {
75 mov
->dest
.write_mask
|= (1 << i
);
76 mov
->src
[0].swizzle
[i
] = vec
->src
[src_idx
].swizzle
[0];
81 nir_instr_insert_before(&vec
->instr
, &mov
->instr
);
83 return mov
->dest
.write_mask
;
87 lower_vec_to_movs_block(nir_block
*block
, void *mem_ctx
)
89 nir_foreach_instr_safe(block
, instr
) {
90 if (instr
->type
!= nir_instr_type_alu
)
93 nir_alu_instr
*vec
= (nir_alu_instr
*)instr
;
101 continue; /* The loop */
104 /* Since we insert multiple MOVs, we have to be non-SSA. */
105 assert(!vec
->dest
.dest
.is_ssa
);
107 unsigned finished_write_mask
= 0;
109 /* First, emit a MOV for all the src channels that are in the
110 * destination reg, in case other values we're populating in the dest
111 * might overwrite them.
113 for (unsigned i
= 0, src_idx
= 0; i
< 4; i
++) {
114 if (!(vec
->dest
.write_mask
& (1 << i
)))
117 if (src_matches_dest_reg(&vec
->dest
.dest
, &vec
->src
[src_idx
].src
)) {
118 finished_write_mask
|= insert_mov(vec
, i
, src_idx
, mem_ctx
);
124 /* Now, emit MOVs for all the other src channels. */
125 for (unsigned i
= 0, src_idx
= 0; i
< 4; i
++) {
126 if (!(vec
->dest
.write_mask
& (1 << i
)))
129 if (!(finished_write_mask
& (1 << i
)))
130 finished_write_mask
|= insert_mov(vec
, i
, src_idx
, mem_ctx
);
135 nir_instr_remove(&vec
->instr
);
143 nir_lower_vec_to_movs_impl(nir_function_impl
*impl
)
145 nir_foreach_block(impl
, lower_vec_to_movs_block
, ralloc_parent(impl
));
149 nir_lower_vec_to_movs(nir_shader
*shader
)
151 nir_foreach_overload(shader
, overload
) {
153 nir_lower_vec_to_movs_impl(overload
->impl
);