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
|
; RUN: llc -mtriple=arm-eabi -mattr=+neon %s -o - | FileCheck %s
; PR12540: ARM backend lowering of FP_ROUND v2f64 to v2f32.
define <2 x float> @vtrunc(<2 x double> %a) {
; CHECK: vcvt.f32.f64 [[S0:s[0-9]+]], [[D0:d[0-9]+]]
; CHECK: vcvt.f32.f64 [[S1:s[0-9]+]], [[D1:d[0-9]+]]
%vt = fptrunc <2 x double> %a to <2 x float>
ret <2 x float> %vt
}
define <2 x double> @vextend(<2 x float> %a) {
; CHECK: vcvt.f64.f32 [[D0:d[0-9]+]], [[S0:s[0-9]+]]
; CHECK: vcvt.f64.f32 [[D1:d[0-9]+]], [[S1:s[0-9]+]]
%ve = fpext <2 x float> %a to <2 x double>
ret <2 x double> %ve
}
; We used to generate vmovs between scalar and vfp/neon registers.
; CHECK: vsitofp_double
define void @vsitofp_double(ptr %loadaddr,
ptr %storeaddr) {
%v0 = load <2 x i32>, ptr %loadaddr
; CHECK: vldr
; CHECK-NEXT: vcvt.f64.s32
; CHECK-NEXT: vcvt.f64.s32
; CHECK-NEXT: vst
%r = sitofp <2 x i32> %v0 to <2 x double>
store <2 x double> %r, ptr %storeaddr
ret void
}
; CHECK: vuitofp_double
define void @vuitofp_double(ptr %loadaddr,
ptr %storeaddr) {
%v0 = load <2 x i32>, ptr %loadaddr
; CHECK: vldr
; CHECK-NEXT: vcvt.f64.u32
; CHECK-NEXT: vcvt.f64.u32
; CHECK-NEXT: vst
%r = uitofp <2 x i32> %v0 to <2 x double>
store <2 x double> %r, ptr %storeaddr
ret void
}
|
