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
|
; RUN: llc -mtriple=armv7-none-linux-gnueabi -mcpu=cortex-a9 -mattr=+neon,+neonfp -float-abi=hard < %s | FileCheck %s
define <2 x float> @test_vmovs_via_vext_lane0to0(float %arg, <2 x float> %in) {
; CHECK: test_vmovs_via_vext_lane0to0:
%vec = insertelement <2 x float> %in, float %arg, i32 0
%res = fadd <2 x float> %vec, %vec
; CHECK: vext.32 d1, d1, d0, #1
; CHECK: vext.32 d1, d1, d1, #1
; CHECK: vadd.f32 {{d[0-9]+}}, d1, d1
ret <2 x float> %res
}
define <2 x float> @test_vmovs_via_vext_lane0to1(float %arg, <2 x float> %in) {
; CHECK: test_vmovs_via_vext_lane0to1:
%vec = insertelement <2 x float> %in, float %arg, i32 1
%res = fadd <2 x float> %vec, %vec
; CHECK: vext.32 d1, d1, d1, #1
; CHECK: vext.32 d1, d1, d0, #1
; CHECK: vadd.f32 {{d[0-9]+}}, d1, d1
ret <2 x float> %res
}
define <2 x float> @test_vmovs_via_vext_lane1to0(float, float %arg, <2 x float> %in) {
; CHECK: test_vmovs_via_vext_lane1to0:
%vec = insertelement <2 x float> %in, float %arg, i32 0
%res = fadd <2 x float> %vec, %vec
; CHECK: vext.32 d1, d1, d1, #1
; CHECK: vext.32 d1, d0, d1, #1
; CHECK: vadd.f32 {{d[0-9]+}}, d1, d1
ret <2 x float> %res
}
define <2 x float> @test_vmovs_via_vext_lane1to1(float, float %arg, <2 x float> %in) {
; CHECK: test_vmovs_via_vext_lane1to1:
%vec = insertelement <2 x float> %in, float %arg, i32 1
%res = fadd <2 x float> %vec, %vec
; CHECK: vext.32 d1, d0, d1, #1
; CHECK: vext.32 d1, d1, d1, #1
; CHECK: vadd.f32 {{d[0-9]+}}, d1, d1
ret <2 x float> %res
}
define float @test_vmovs_via_vdup(float, float %ret, float %lhs, float %rhs) {
; CHECK: test_vmovs_via_vdup:
; Do an operation (which will end up NEON because of +neonfp) to convince the
; execution-domain pass that NEON is a good thing to use.
%res = fadd float %ret, %ret
; It makes sense for LLVM to do the addition in d0 here, because it's going
; to be returned. This means it will want a "vmov s0, s1":
; CHECK: vdup.32 d0, d0[1]
ret float %res
}
|