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
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple armv8a-none-none-eabihf -mattr=fullfp16 < %s | FileCheck %s
define <4 x half> @fptrunc_vector_f32_f16(<4 x float> %a) {
; CHECK-LABEL: fptrunc_vector_f32_f16:
; CHECK: @ %bb.0: @ %bb
; CHECK-NEXT: vcvt.f16.f32 d0, q0
; CHECK-NEXT: bx lr
bb:
%z = fptrunc <4 x float> %a to <4 x half>
ret <4 x half> %z
}
define <4 x half> @fptrunc_vector_f64_f16(<4 x double> %a) {
; CHECK-LABEL: fptrunc_vector_f64_f16:
; CHECK: @ %bb.0: @ %bb
; CHECK-NEXT: vcvtb.f16.f64 s0, d0
; CHECK-NEXT: vcvtb.f16.f64 s8, d1
; CHECK-NEXT: vmov r1, s0
; CHECK-NEXT: vcvtb.f16.f64 s2, d2
; CHECK-NEXT: vmov r0, s8
; CHECK-NEXT: vmov.16 d0[0], r1
; CHECK-NEXT: vmov.16 d0[1], r0
; CHECK-NEXT: vmov r0, s2
; CHECK-NEXT: vcvtb.f16.f64 s2, d3
; CHECK-NEXT: vmov.16 d0[2], r0
; CHECK-NEXT: vmov r0, s2
; CHECK-NEXT: vmov.16 d0[3], r0
; CHECK-NEXT: bx lr
bb:
%z = fptrunc <4 x double> %a to <4 x half>
ret <4 x half> %z
}
define <4 x float> @fpext_vector_f16_f32(<4 x half> %a) {
; CHECK-LABEL: fpext_vector_f16_f32:
; CHECK: @ %bb.0: @ %bb
; CHECK-NEXT: vcvt.f32.f16 q0, d0
; CHECK-NEXT: bx lr
bb:
%z = fpext <4 x half> %a to <4 x float>
ret <4 x float> %z
}
define <4 x double> @fpext_vector_f16_f64(<4 x half> %a) {
; CHECK-LABEL: fpext_vector_f16_f64:
; CHECK: @ %bb.0: @ %bb
; CHECK-NEXT: vmovx.f16 s4, s0
; CHECK-NEXT: vmovx.f16 s2, s1
; CHECK-NEXT: vcvtb.f64.f16 d17, s4
; CHECK-NEXT: vcvtb.f64.f16 d3, s2
; CHECK-NEXT: vcvtb.f64.f16 d16, s0
; CHECK-NEXT: vcvtb.f64.f16 d2, s1
; CHECK-NEXT: vorr q0, q8, q8
; CHECK-NEXT: bx lr
bb:
%z = fpext <4 x half> %a to <4 x double>
ret <4 x double> %z
}
|
