1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
|
//@ compile-flags: -Copt-level=3 -C no-prepopulate-passes
//@ only-64bit (so I don't need to worry about usize)
//@ revisions: aarch64 x86_64
//@ [aarch64] only-aarch64
//@ [aarch64] compile-flags: -C target-feature=+neon
//@ [x86_64] only-x86_64
//@ [x86_64] compile-flags: -C target-feature=+sse2
#![crate_type = "lib"]
#![feature(core_intrinsics)]
#![feature(portable_simd)]
use std::intrinsics::transmute;
use std::simd::{Simd, f32x4, f64x2, i32x4, i64x2};
type PtrX2 = Simd<*const (), 2>;
// These tests use the "C" ABI so that the vectors in question aren't passed and
// returned though memory (as they are in the "Rust" ABI), which greatly
// simplifies seeing the difference between the in-operand cases vs the ones
// that fallback to just using the `LocalKind::Memory` path.
// CHECK-LABEL: <2 x i64> @mixed_int(<4 x i32> %v)
#[no_mangle]
pub extern "C" fn mixed_int(v: i32x4) -> i64x2 {
// CHECK-NOT: alloca
// CHECK: %[[RET:.+]] = bitcast <4 x i32> %v to <2 x i64>
// CHECK: ret <2 x i64> %[[RET]]
unsafe { transmute(v) }
}
// CHECK-LABEL: <2 x double> @mixed_float(<4 x float> %v)
#[no_mangle]
pub extern "C" fn mixed_float(v: f32x4) -> f64x2 {
// CHECK-NOT: alloca
// CHECK: %[[RET:.+]] = bitcast <4 x float> %v to <2 x double>
// CHECK: ret <2 x double> %[[RET]]
unsafe { transmute(v) }
}
// CHECK-LABEL: <4 x i32> @float_int_same_lanes(<4 x float> %v)
#[no_mangle]
pub extern "C" fn float_int_same_lanes(v: f32x4) -> i32x4 {
// CHECK-NOT: alloca
// CHECK: %[[RET:.+]] = bitcast <4 x float> %v to <4 x i32>
// CHECK: ret <4 x i32> %[[RET]]
unsafe { transmute(v) }
}
// CHECK-LABEL: <2 x double> @int_float_same_lanes(<2 x i64> %v)
#[no_mangle]
pub extern "C" fn int_float_same_lanes(v: i64x2) -> f64x2 {
// CHECK-NOT: alloca
// CHECK: %[[RET:.+]] = bitcast <2 x i64> %v to <2 x double>
// CHECK: ret <2 x double> %[[RET]]
unsafe { transmute(v) }
}
// CHECK-LABEL: <2 x i64> @float_int_widen(<4 x float> %v)
#[no_mangle]
pub extern "C" fn float_int_widen(v: f32x4) -> i64x2 {
// CHECK-NOT: alloca
// CHECK: %[[RET:.+]] = bitcast <4 x float> %v to <2 x i64>
// CHECK: ret <2 x i64> %[[RET]]
unsafe { transmute(v) }
}
// CHECK-LABEL: <2 x double> @int_float_widen(<4 x i32> %v)
#[no_mangle]
pub extern "C" fn int_float_widen(v: i32x4) -> f64x2 {
// CHECK-NOT: alloca
// CHECK: %[[RET:.+]] = bitcast <4 x i32> %v to <2 x double>
// CHECK: ret <2 x double> %[[RET]]
unsafe { transmute(v) }
}
// CHECK-LABEL: <4 x i32> @float_int_narrow(<2 x double> %v)
#[no_mangle]
pub extern "C" fn float_int_narrow(v: f64x2) -> i32x4 {
// CHECK-NOT: alloca
// CHECK: %[[RET:.+]] = bitcast <2 x double> %v to <4 x i32>
// CHECK: ret <4 x i32> %[[RET]]
unsafe { transmute(v) }
}
// CHECK-LABEL: <4 x float> @int_float_narrow(<2 x i64> %v)
#[no_mangle]
pub extern "C" fn int_float_narrow(v: i64x2) -> f32x4 {
// CHECK-NOT: alloca
// CHECK: %[[RET:.+]] = bitcast <2 x i64> %v to <4 x float>
// CHECK: ret <4 x float> %[[RET]]
unsafe { transmute(v) }
}
// CHECK-LABEL: <2 x ptr> @float_ptr_same_lanes(<2 x double> %v)
#[no_mangle]
pub extern "C" fn float_ptr_same_lanes(v: f64x2) -> PtrX2 {
// CHECK-NOT: alloca
// CHECK: %[[TEMP:.+]] = alloca [16 x i8]
// CHECK-NOT: alloca
// CHECK: call void @llvm.lifetime.start.p0(i64 16, ptr %[[TEMP]])
// CHECK: store <2 x double> %v, ptr %[[TEMP]]
// CHECK: %[[RET:.+]] = load <2 x ptr>, ptr %[[TEMP]]
// CHECK: call void @llvm.lifetime.end.p0(i64 16, ptr %[[TEMP]])
// CHECK: ret <2 x ptr> %[[RET]]
unsafe { transmute(v) }
}
// CHECK-LABEL: <2 x double> @ptr_float_same_lanes(<2 x ptr> %v)
#[no_mangle]
pub extern "C" fn ptr_float_same_lanes(v: PtrX2) -> f64x2 {
// CHECK-NOT: alloca
// CHECK: %[[TEMP:.+]] = alloca [16 x i8]
// CHECK-NOT: alloca
// CHECK: call void @llvm.lifetime.start.p0(i64 16, ptr %[[TEMP]])
// CHECK: store <2 x ptr> %v, ptr %[[TEMP]]
// CHECK: %[[RET:.+]] = load <2 x double>, ptr %[[TEMP]]
// CHECK: call void @llvm.lifetime.end.p0(i64 16, ptr %[[TEMP]])
// CHECK: ret <2 x double> %[[RET]]
unsafe { transmute(v) }
}
// CHECK-LABEL: <2 x ptr> @int_ptr_same_lanes(<2 x i64> %v)
#[no_mangle]
pub extern "C" fn int_ptr_same_lanes(v: i64x2) -> PtrX2 {
// CHECK-NOT: alloca
// CHECK: %[[TEMP:.+]] = alloca [16 x i8]
// CHECK-NOT: alloca
// CHECK: call void @llvm.lifetime.start.p0(i64 16, ptr %[[TEMP]])
// CHECK: store <2 x i64> %v, ptr %[[TEMP]]
// CHECK: %[[RET:.+]] = load <2 x ptr>, ptr %[[TEMP]]
// CHECK: call void @llvm.lifetime.end.p0(i64 16, ptr %[[TEMP]])
// CHECK: ret <2 x ptr> %[[RET]]
unsafe { transmute(v) }
}
// CHECK-LABEL: <2 x i64> @ptr_int_same_lanes(<2 x ptr> %v)
#[no_mangle]
pub extern "C" fn ptr_int_same_lanes(v: PtrX2) -> i64x2 {
// CHECK-NOT: alloca
// CHECK: %[[TEMP:.+]] = alloca [16 x i8]
// CHECK-NOT: alloca
// CHECK: call void @llvm.lifetime.start.p0(i64 16, ptr %[[TEMP]])
// CHECK: store <2 x ptr> %v, ptr %[[TEMP]]
// CHECK: %[[RET:.+]] = load <2 x i64>, ptr %[[TEMP]]
// CHECK: call void @llvm.lifetime.end.p0(i64 16, ptr %[[TEMP]])
// CHECK: ret <2 x i64> %[[RET]]
unsafe { transmute(v) }
}
// CHECK-LABEL: <2 x ptr> @float_ptr_widen(<4 x float> %v)
#[no_mangle]
pub extern "C" fn float_ptr_widen(v: f32x4) -> PtrX2 {
// CHECK-NOT: alloca
// CHECK: %[[TEMP:.+]] = alloca [16 x i8]
// CHECK-NOT: alloca
// CHECK: call void @llvm.lifetime.start.p0(i64 16, ptr %[[TEMP]])
// CHECK: store <4 x float> %v, ptr %[[TEMP]]
// CHECK: %[[RET:.+]] = load <2 x ptr>, ptr %[[TEMP]]
// CHECK: call void @llvm.lifetime.end.p0(i64 16, ptr %[[TEMP]])
// CHECK: ret <2 x ptr> %[[RET]]
unsafe { transmute(v) }
}
// CHECK-LABEL: <2 x ptr> @int_ptr_widen(<4 x i32> %v)
#[no_mangle]
pub extern "C" fn int_ptr_widen(v: i32x4) -> PtrX2 {
// CHECK-NOT: alloca
// CHECK: %[[TEMP:.+]] = alloca [16 x i8]
// CHECK-NOT: alloca
// CHECK: call void @llvm.lifetime.start.p0(i64 16, ptr %[[TEMP]])
// CHECK: store <4 x i32> %v, ptr %[[TEMP]]
// CHECK: %[[RET:.+]] = load <2 x ptr>, ptr %[[TEMP]]
// CHECK: call void @llvm.lifetime.end.p0(i64 16, ptr %[[TEMP]])
// CHECK: ret <2 x ptr> %[[RET]]
unsafe { transmute(v) }
}
|
