Skip to content
This repository was archived by the owner on Dec 22, 2021. It is now read-only.

Commit 9a169ae

Browse files
Jakob Stoklund Olesenstoklund
Jakob Stoklund Olesen
authored and
stoklund
committed
Add a separate document for the binary encoding of SIMD.
1 parent 5bd1db5 commit 9a169ae

File tree

2 files changed

+171
-0
lines changed

2 files changed

+171
-0
lines changed

‎proposals/simd/BinarySIMD.md‎

Lines changed: 169 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,169 @@
1+
# Binary encoding of SIMD
2+
3+
This document describes the binary encoding of the SIMD value type and
4+
instructions.
5+
6+
## SIMD value type
7+
8+
The `v128` value type is encoded as 0x7b:
9+
10+
```
11+
valtype ::= ...
12+
| 0x7B => v128
13+
```
14+
15+
## SIMD instruction encodings
16+
17+
All SIMD instructions are encoded as a 0xfd prefix byte followed by a
18+
SIMD-specific opcode in LEB128 format:
19+
20+
```
21+
instr ::= ...
22+
| 0xFB simdop:varuint32 ...
23+
```
24+
25+
Some SIMD instructions have additional immediate operands following `simdop`.
26+
The `v8x16.shuffle` instruction has 16 bytes after `simdop`.
27+
28+
| Instruction | `simdop` | Immediate operands |
29+
| --------------------------|---------:|--------------------|
30+
| `v128.const` | 0 | - |
31+
| `v128.load` | 1 | m:memarg |
32+
| `v128.store` | 2 | m:memarg |
33+
| `i8x16.splat` | 3 | - |
34+
| `i16x8.splat` | 4 | - |
35+
| `i32x4.splat` | 5 | - |
36+
| `i64x2.splat` | 6 | - |
37+
| `f32x4.splat` | 7 | - |
38+
| `f64x2.splat` | 8 | - |
39+
| `i8x16.extract_lane_s` | 9 | i:LaneIdx16 |
40+
| `i8x16.extract_lane_u` | 10 | i:LaneIdx16 |
41+
| `i16x8.extract_lane_s` | 11 | i:LaneIdx8 |
42+
| `i16x8.extract_lane_u` | 12 | i:LaneIdx8 |
43+
| `i32x4.extract_lane` | 13 | i:LaneIdx4 |
44+
| `i64x2.extract_lane` | 14 | i:LaneIdx2 |
45+
| `f32x4.extract_lane` | 15 | i:LaneIdx4 |
46+
| `f64x2.extract_lane` | 16 | i:LaneIdx2 |
47+
| `i8x16.replace_lane` | 17 | i:LaneIdx16 |
48+
| `i16x8.replace_lane` | 18 | i:LaneIdx8 |
49+
| `i32x4.replace_lane` | 19 | i:LaneIdx4 |
50+
| `i64x2.replace_lane` | 20 | i:LaneIdx2 |
51+
| `f32x4.replace_lane` | 21 | i:LaneIdx4 |
52+
| `f64x2.replace_lane` | 22 | i:LaneIdx2 |
53+
| `v8x16.shuffle` | 23 | s:LaneIdx32[16] |
54+
| `i8x16.add` | 24 | - |
55+
| `i16x8.add` | 25 | - |
56+
| `i32x4.add` | 26 | - |
57+
| `i64x2.add` | 27 | - |
58+
| `i8x16.sub` | 28 | - |
59+
| `i16x8.sub` | 29 | - |
60+
| `i32x4.sub` | 30 | - |
61+
| `i64x2.sub` | 31 | - |
62+
| `i8x16.mul` | 32 | - |
63+
| `i16x8.mul` | 33 | - |
64+
| `i32x4.mul` | 34 | - |
65+
| `i8x16.neg` | 35 | - |
66+
| `i16x8.neg` | 36 | - |
67+
| `i32x4.neg` | 37 | - |
68+
| `i64x2.neg` | 38 | - |
69+
| `i8x16.add_saturate_s` | 39 | - |
70+
| `i8x16.add_saturate_u` | 40 | - |
71+
| `i16x8.add_saturate_s` | 41 | - |
72+
| `i16x8.add_saturate_u` | 42 | - |
73+
| `i8x16.sub_saturate_s` | 43 | - |
74+
| `i8x16.sub_saturate_u` | 44 | - |
75+
| `i16x8.sub_saturate_s` | 45 | - |
76+
| `i16x8.sub_saturate_u` | 46 | - |
77+
| `i8x16.shl` | 47 | - |
78+
| `i16x8.shl` | 48 | - |
79+
| `i32x4.shl` | 49 | - |
80+
| `i64x2.shl` | 50 | - |
81+
| `i8x16.shr_s` | 51 | - |
82+
| `i8x16.shr_u` | 52 | - |
83+
| `i16x8.shr_s` | 53 | - |
84+
| `i16x8.shr_u` | 54 | - |
85+
| `i32x4.shr_s` | 55 | - |
86+
| `i32x4.shr_u` | 56 | - |
87+
| `i64x2.shr_s` | 57 | - |
88+
| `i64x2.shr_u` | 58 | - |
89+
| `v128.and` | 59 | - |
90+
| `v128.or` | 60 | - |
91+
| `v128.xor` | 61 | - |
92+
| `v128.not` | 62 | - |
93+
| `v128.bitselect` | 63 | - |
94+
| `i8x16.any_true` | 64 | - |
95+
| `i16x8.any_true` | 65 | - |
96+
| `i32x4.any_true` | 66 | - |
97+
| `i64x2.any_true` | 67 | - |
98+
| `i8x16.all_true` | 68 | - |
99+
| `i16x8.all_true` | 69 | - |
100+
| `i32x4.all_true` | 70 | - |
101+
| `i64x2.all_true` | 71 | - |
102+
| `i8x16.eq` | 72 | - |
103+
| `i16x8.eq` | 73 | - |
104+
| `i32x4.eq` | 74 | - |
105+
| `f32x4.eq` | 75 | - |
106+
| `f64x2.eq` | 76 | - |
107+
| `i8x16.ne` | 77 | - |
108+
| `i16x8.ne` | 78 | - |
109+
| `i32x4.ne` | 79 | - |
110+
| `f32x4.ne` | 80 | - |
111+
| `f64x2.ne` | 81 | - |
112+
| `i8x16.lt_s` | 82 | - |
113+
| `i8x16.lt_u` | 83 | - |
114+
| `i16x8.lt_s` | 84 | - |
115+
| `i16x8.lt_u` | 85 | - |
116+
| `i32x4.lt_s` | 86 | - |
117+
| `i32x4.lt_u` | 87 | - |
118+
| `f32x4.lt` | 88 | - |
119+
| `f64x2.lt` | 89 | - |
120+
| `i8x16.le_s` | 90 | - |
121+
| `i8x16.le_u` | 91 | - |
122+
| `i16x8.le_s` | 92 | - |
123+
| `i16x8.le_u` | 93 | - |
124+
| `i32x4.le_s` | 94 | - |
125+
| `i32x4.le_u` | 95 | - |
126+
| `f32x4.le` | 96 | - |
127+
| `f64x2.le` | 97 | - |
128+
| `i8x16.gt_s` | 98 | - |
129+
| `i8x16.gt_u` | 99 | - |
130+
| `i16x8.gt_s` | 100 | - |
131+
| `i16x8.gt_u` | 101 | - |
132+
| `i32x4.gt_s` | 102 | - |
133+
| `i32x4.gt_u` | 103 | - |
134+
| `f32x4.gt` | 104 | - |
135+
| `f64x2.gt` | 105 | - |
136+
| `i8x16.ge_s` | 106 | - |
137+
| `i8x16.ge_u` | 107 | - |
138+
| `i16x8.ge_s` | 108 | - |
139+
| `i16x8.ge_u` | 109 | - |
140+
| `i32x4.ge_s` | 110 | - |
141+
| `i32x4.ge_u` | 111 | - |
142+
| `f32x4.ge` | 112 | - |
143+
| `f64x2.ge` | 113 | - |
144+
| `f32x4.neg` | 114 | - |
145+
| `f64x2.neg` | 115 | - |
146+
| `f32x4.abs` | 116 | - |
147+
| `f64x2.abs` | 117 | - |
148+
| `f32x4.min` | 118 | - |
149+
| `f64x2.min` | 119 | - |
150+
| `f32x4.max` | 120 | - |
151+
| `f64x2.max` | 121 | - |
152+
| `f32x4.add` | 122 | - |
153+
| `f64x2.add` | 123 | - |
154+
| `f32x4.sub` | 124 | - |
155+
| `f64x2.sub` | 125 | - |
156+
| `f32x4.div` | 126 | - |
157+
| `f64x2.div` | 127 | - |
158+
| `f32x4.mul` | 128 | - |
159+
| `f64x2.mul` | 129 | - |
160+
| `f32x4.sqrt` | 130 | - |
161+
| `f64x2.sqrt` | 131 | - |
162+
| `f32x4.convert_s/i32x4` | 132 | - |
163+
| `f32x4.convert_u/i32x4` | 133 | - |
164+
| `f64x2.convert_s/i64x2` | 134 | - |
165+
| `f64x2.convert_u/i64x2` | 135 | - |
166+
| `i32x4.trunc_s/f32x4:sat` | 136 | - |
167+
| `i32x4.trunc_u/f32x4:sat` | 137 | - |
168+
| `i64x2.trunc_s/f64x2:sat` | 138 | - |
169+
| `i64x2.trunc_u/f64x2:sat` | 139 | - |

‎proposals/simd/SIMD.md‎

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -4,6 +4,8 @@ This specification describes a 128-bit packed *Single Instruction Multiple
44
Data* (SIMD) extension to WebAssembly that can be implemented efficiently on
55
current popular instruction set architectures.
66

7+
See also [The binary encoding of SIMD instructions](BinarySIMD.md).
8+
79
# Types
810

911
WebAssembly is extended with a new `v128` value type and a number of new kinds

0 commit comments

Comments
 (0)