• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: llc < %s -mtriple=arm64-eabi -aarch64-neon-syntax=apple | FileCheck %s
2; RUN: llc < %s -O0 -fast-isel -mtriple=arm64-eabi -aarch64-neon-syntax=apple | FileCheck %s
3
4define <2 x double> @test_vcvt_f64_f32(<2 x float> %x) nounwind readnone ssp {
5; CHECK-LABEL: test_vcvt_f64_f32:
6  %vcvt1.i = fpext <2 x float> %x to <2 x double>
7; CHECK: fcvtl	v0.2d, v0.2s
8  ret <2 x double> %vcvt1.i
9; CHECK: ret
10}
11
12define <2 x double> @test_vcvt_high_f64_f32(<4 x float> %x) nounwind readnone ssp {
13; CHECK-LABEL: test_vcvt_high_f64_f32:
14  %cvt_in = shufflevector <4 x float> %x, <4 x float> undef, <2 x i32> <i32 2, i32 3>
15  %vcvt1.i = fpext <2 x float> %cvt_in to <2 x double>
16; CHECK: fcvtl2	v0.2d, v0.4s
17  ret <2 x double> %vcvt1.i
18; CHECK: ret
19}
20
21define <2 x float> @test_vcvt_f32_f64(<2 x double> %v) nounwind readnone ssp {
22; CHECK-LABEL: test_vcvt_f32_f64:
23  %vcvt1.i = fptrunc <2 x double> %v to <2 x float>
24; CHECK: fcvtn
25  ret <2 x float> %vcvt1.i
26; CHECK: ret
27}
28
29define <4 x float> @test_vcvt_high_f32_f64(<2 x float> %x, <2 x double> %v) nounwind readnone ssp {
30; CHECK-LABEL: test_vcvt_high_f32_f64:
31
32  %cvt = fptrunc <2 x double> %v to <2 x float>
33  %vcvt2.i = shufflevector <2 x float> %x, <2 x float> %cvt, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
34; CHECK: fcvtn2
35  ret <4 x float> %vcvt2.i
36; CHECK: ret
37}
38
39define <2 x float> @test_vcvtx_f32_f64(<2 x double> %v) nounwind readnone ssp {
40; CHECK-LABEL: test_vcvtx_f32_f64:
41  %vcvtx1.i = tail call <2 x float> @llvm.aarch64.neon.fcvtxn.v2f32.v2f64(<2 x double> %v) nounwind
42; CHECK: fcvtxn
43  ret <2 x float> %vcvtx1.i
44; CHECK: ret
45}
46
47define <4 x float> @test_vcvtx_high_f32_f64(<2 x float> %x, <2 x double> %v) nounwind readnone ssp {
48; CHECK-LABEL: test_vcvtx_high_f32_f64:
49  %vcvtx2.i = tail call <2 x float> @llvm.aarch64.neon.fcvtxn.v2f32.v2f64(<2 x double> %v) nounwind
50  %res = shufflevector <2 x float> %x, <2 x float> %vcvtx2.i, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
51; CHECK: fcvtxn2
52  ret <4 x float> %res
53; CHECK: ret
54}
55
56
57declare <2 x double> @llvm.aarch64.neon.vcvthighfp2df(<4 x float>) nounwind readnone
58declare <2 x double> @llvm.aarch64.neon.vcvtfp2df(<2 x float>) nounwind readnone
59
60declare <2 x float> @llvm.aarch64.neon.vcvtdf2fp(<2 x double>) nounwind readnone
61declare <4 x float> @llvm.aarch64.neon.vcvthighdf2fp(<2 x float>, <2 x double>) nounwind readnone
62
63declare <2 x float> @llvm.aarch64.neon.fcvtxn.v2f32.v2f64(<2 x double>) nounwind readnone
64
65define i16 @to_half(float %in) {
66; CHECK-LABEL: to_half:
67; CHECK: fcvt h[[HALFVAL:[0-9]+]], s0
68; CHECK: fmov {{w[0-9]+}}, {{s[0-9]+}}
69  %res = call i16 @llvm.convert.to.fp16.f32(float %in)
70  ret i16 %res
71}
72
73define float @from_half(i16 %in) {
74; CHECK-LABEL: from_half:
75; CHECK: fmov {{s[0-9]+}}, {{w[0-9]+}}
76; CHECK: fcvt s0, {{h[0-9]+}}
77  %res = call float @llvm.convert.from.fp16.f32(i16 %in)
78  ret float %res
79}
80
81declare float @llvm.convert.from.fp16.f32(i16) #1
82declare i16 @llvm.convert.to.fp16.f32(float) #1
83