1 /*
2 * Copyright © 2015 Intel Corporation
3 *
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:
10 *
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
13 * Software.
14 *
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
21 * IN THE SOFTWARE.
22 *
23 */
24
25 #include "nir.h"
26 #include "nir_builder.h"
27
28 /*
29 * lowers:
30 *
31 * packDouble2x32(foo) -> packDouble2x32Split(foo.x, foo.y)
32 * unpackDouble2x32(foo) -> vec2(unpackDouble2x32_x(foo), unpackDouble2x32_y(foo))
33 * packInt2x32(foo) -> packInt2x32Split(foo.x, foo.y)
34 * unpackInt2x32(foo) -> vec2(unpackInt2x32_x(foo), unpackInt2x32_y(foo))
35 */
36
37 static nir_def *
lower_pack_64_from_32(nir_builder * b,nir_def * src)38 lower_pack_64_from_32(nir_builder *b, nir_def *src)
39 {
40 return nir_pack_64_2x32_split(b, nir_channel(b, src, 0),
41 nir_channel(b, src, 1));
42 }
43
44 static nir_def *
lower_unpack_64_to_32(nir_builder * b,nir_def * src)45 lower_unpack_64_to_32(nir_builder *b, nir_def *src)
46 {
47 return nir_vec2(b, nir_unpack_64_2x32_split_x(b, src),
48 nir_unpack_64_2x32_split_y(b, src));
49 }
50
51 static nir_def *
lower_pack_32_from_16(nir_builder * b,nir_def * src)52 lower_pack_32_from_16(nir_builder *b, nir_def *src)
53 {
54 return nir_pack_32_2x16_split(b, nir_channel(b, src, 0),
55 nir_channel(b, src, 1));
56 }
57
58 static nir_def *
lower_unpack_32_to_16(nir_builder * b,nir_def * src)59 lower_unpack_32_to_16(nir_builder *b, nir_def *src)
60 {
61 return nir_vec2(b, nir_unpack_32_2x16_split_x(b, src),
62 nir_unpack_32_2x16_split_y(b, src));
63 }
64
65 static nir_def *
lower_pack_64_from_16(nir_builder * b,nir_def * src)66 lower_pack_64_from_16(nir_builder *b, nir_def *src)
67 {
68 nir_def *xy = nir_pack_32_2x16_split(b, nir_channel(b, src, 0),
69 nir_channel(b, src, 1));
70
71 nir_def *zw = nir_pack_32_2x16_split(b, nir_channel(b, src, 2),
72 nir_channel(b, src, 3));
73
74 return nir_pack_64_2x32_split(b, xy, zw);
75 }
76
77 static nir_def *
lower_unpack_64_to_16(nir_builder * b,nir_def * src)78 lower_unpack_64_to_16(nir_builder *b, nir_def *src)
79 {
80 nir_def *xy = nir_unpack_64_2x32_split_x(b, src);
81 nir_def *zw = nir_unpack_64_2x32_split_y(b, src);
82
83 return nir_vec4(b, nir_unpack_32_2x16_split_x(b, xy),
84 nir_unpack_32_2x16_split_y(b, xy),
85 nir_unpack_32_2x16_split_x(b, zw),
86 nir_unpack_32_2x16_split_y(b, zw));
87 }
88
89 static nir_def *
lower_pack_32_from_8(nir_builder * b,nir_def * src)90 lower_pack_32_from_8(nir_builder *b, nir_def *src)
91 {
92 if (b->shader->options->has_pack_32_4x8) {
93 return nir_pack_32_4x8_split(b,
94 nir_channel(b, src, 0),
95 nir_channel(b, src, 1),
96 nir_channel(b, src, 2),
97 nir_channel(b, src, 3));
98 } else {
99 nir_def *src32 = nir_u2u32(b, src);
100
101 return nir_ior(b,
102 nir_ior(b,
103 nir_channel(b, src32, 0) ,
104 nir_ishl_imm(b, nir_channel(b, src32, 1), 8)),
105 nir_ior(b,
106 nir_ishl_imm(b, nir_channel(b, src32, 2), 16),
107 nir_ishl_imm(b, nir_channel(b, src32, 3), 24)));
108 }
109 }
110
111 static nir_def *
lower_unpack_32_to_8(nir_builder * b,nir_def * src)112 lower_unpack_32_to_8(nir_builder *b, nir_def *src)
113 {
114 /* Some drivers call nir_lower_pack after the last time nir_opt_algebraic
115 * is called. To prevent issues there, don't generate byte extraction
116 * instructions when the lowering flag is set.
117 */
118 if (b->shader->options->lower_extract_byte) {
119 return nir_vec4(b, nir_u2u8(b, src ),
120 nir_u2u8(b, nir_ushr_imm(b, src, 8)),
121 nir_u2u8(b, nir_ushr_imm(b, src, 16)),
122 nir_u2u8(b, nir_ushr_imm(b, src, 24)));
123 } else {
124 return nir_vec4(b, nir_u2u8(b, nir_extract_u8_imm(b, src, 0)),
125 nir_u2u8(b, nir_extract_u8_imm(b, src, 1)),
126 nir_u2u8(b, nir_extract_u8_imm(b, src, 2)),
127 nir_u2u8(b, nir_extract_u8_imm(b, src, 3)));
128 }
129 }
130
131 static bool
lower_pack_instr(nir_builder * b,nir_alu_instr * alu_instr,void * data)132 lower_pack_instr(nir_builder *b, nir_alu_instr *alu_instr, void *data)
133 {
134 nir_lower_packing_op op;
135 switch (alu_instr->op) {
136 case nir_op_pack_64_2x32:
137 op = nir_lower_packing_op_pack_64_2x32;
138 break;
139 case nir_op_unpack_64_2x32:
140 op = nir_lower_packing_op_unpack_64_2x32;
141 break;
142 case nir_op_pack_64_4x16:
143 op = nir_lower_packing_op_pack_64_4x16;
144 break;
145 case nir_op_unpack_64_4x16:
146 op = nir_lower_packing_op_unpack_64_4x16;
147 break;
148 case nir_op_pack_32_2x16:
149 op = nir_lower_packing_op_pack_32_2x16;
150 break;
151 case nir_op_unpack_32_2x16:
152 op = nir_lower_packing_op_unpack_32_2x16;
153 break;
154 case nir_op_pack_32_4x8:
155 op = nir_lower_packing_op_pack_32_4x8;
156 break;
157 case nir_op_unpack_32_4x8:
158 op = nir_lower_packing_op_unpack_32_4x8;
159 break;
160 default:
161 return false;
162 }
163
164 if (b->shader->options->skip_lower_packing_ops & BITFIELD_BIT(op))
165 return false;
166
167 b->cursor = nir_before_instr(&alu_instr->instr);
168
169 typedef nir_def *(*lower_func_t)(nir_builder *b, nir_def *src);
170 static const lower_func_t lower_funcs[nir_lower_packing_num_ops] = {
171 [nir_lower_packing_op_pack_64_2x32] = lower_pack_64_from_32,
172 [nir_lower_packing_op_unpack_64_2x32] = lower_unpack_64_to_32,
173 [nir_lower_packing_op_pack_64_4x16] = lower_pack_64_from_16,
174 [nir_lower_packing_op_unpack_64_4x16] = lower_unpack_64_to_16,
175 [nir_lower_packing_op_pack_32_2x16] = lower_pack_32_from_16,
176 [nir_lower_packing_op_unpack_32_2x16] = lower_unpack_32_to_16,
177 [nir_lower_packing_op_pack_32_4x8] = lower_pack_32_from_8,
178 [nir_lower_packing_op_unpack_32_4x8] = lower_unpack_32_to_8,
179 };
180
181 nir_def *src = nir_ssa_for_alu_src(b, alu_instr, 0);
182 nir_def *dest = lower_funcs[op](b, src);
183 nir_def_replace(&alu_instr->def, dest);
184
185 return true;
186 }
187
188 bool
nir_lower_pack(nir_shader * shader)189 nir_lower_pack(nir_shader *shader)
190 {
191 return nir_shader_alu_pass(shader, lower_pack_instr,
192 nir_metadata_control_flow, NULL);
193 }
194