1 ; RUN: llc < %s -march=arm -mattr=+neon > %t
2 ; RUN: grep {vceq\\.f32} %t | count 1
3 ; RUN: grep {vcgt\\.f32} %t | count 9
4 ; RUN: grep {vcge\\.f32} %t | count 5
5 ; RUN: grep vorr %t | count 4
6 ; RUN: grep vmvn %t | count 7
8 ; This tests fcmp operations that do not map directly to NEON instructions.
10 ; une is implemented with VCEQ/VMVN
11 define <2 x i32> @vcunef32(<2 x float>* %A, <2 x float>* %B) nounwind {
12 %tmp1 = load <2 x float>* %A
13 %tmp2 = load <2 x float>* %B
14 %tmp3 = fcmp une <2 x float> %tmp1, %tmp2
15 %tmp4 = sext <2 x i1> %tmp3 to <2 x i32>
19 ; olt is implemented with VCGT
20 define <2 x i32> @vcoltf32(<2 x float>* %A, <2 x float>* %B) nounwind {
21 %tmp1 = load <2 x float>* %A
22 %tmp2 = load <2 x float>* %B
23 %tmp3 = fcmp olt <2 x float> %tmp1, %tmp2
24 %tmp4 = sext <2 x i1> %tmp3 to <2 x i32>
28 ; ole is implemented with VCGE
29 define <2 x i32> @vcolef32(<2 x float>* %A, <2 x float>* %B) nounwind {
30 %tmp1 = load <2 x float>* %A
31 %tmp2 = load <2 x float>* %B
32 %tmp3 = fcmp ole <2 x float> %tmp1, %tmp2
33 %tmp4 = sext <2 x i1> %tmp3 to <2 x i32>
37 ; uge is implemented with VCGT/VMVN
38 define <2 x i32> @vcugef32(<2 x float>* %A, <2 x float>* %B) nounwind {
39 %tmp1 = load <2 x float>* %A
40 %tmp2 = load <2 x float>* %B
41 %tmp3 = fcmp uge <2 x float> %tmp1, %tmp2
42 %tmp4 = sext <2 x i1> %tmp3 to <2 x i32>
46 ; ule is implemented with VCGT/VMVN
47 define <2 x i32> @vculef32(<2 x float>* %A, <2 x float>* %B) nounwind {
48 %tmp1 = load <2 x float>* %A
49 %tmp2 = load <2 x float>* %B
50 %tmp3 = fcmp ule <2 x float> %tmp1, %tmp2
51 %tmp4 = sext <2 x i1> %tmp3 to <2 x i32>
55 ; ugt is implemented with VCGE/VMVN
56 define <2 x i32> @vcugtf32(<2 x float>* %A, <2 x float>* %B) nounwind {
57 %tmp1 = load <2 x float>* %A
58 %tmp2 = load <2 x float>* %B
59 %tmp3 = fcmp ugt <2 x float> %tmp1, %tmp2
60 %tmp4 = sext <2 x i1> %tmp3 to <2 x i32>
64 ; ult is implemented with VCGE/VMVN
65 define <2 x i32> @vcultf32(<2 x float>* %A, <2 x float>* %B) nounwind {
66 %tmp1 = load <2 x float>* %A
67 %tmp2 = load <2 x float>* %B
68 %tmp3 = fcmp ult <2 x float> %tmp1, %tmp2
69 %tmp4 = sext <2 x i1> %tmp3 to <2 x i32>
73 ; ueq is implemented with VCGT/VCGT/VORR/VMVN
74 define <2 x i32> @vcueqf32(<2 x float>* %A, <2 x float>* %B) nounwind {
75 %tmp1 = load <2 x float>* %A
76 %tmp2 = load <2 x float>* %B
77 %tmp3 = fcmp ueq <2 x float> %tmp1, %tmp2
78 %tmp4 = sext <2 x i1> %tmp3 to <2 x i32>
82 ; one is implemented with VCGT/VCGT/VORR
83 define <2 x i32> @vconef32(<2 x float>* %A, <2 x float>* %B) nounwind {
84 %tmp1 = load <2 x float>* %A
85 %tmp2 = load <2 x float>* %B
86 %tmp3 = fcmp one <2 x float> %tmp1, %tmp2
87 %tmp4 = sext <2 x i1> %tmp3 to <2 x i32>
91 ; uno is implemented with VCGT/VCGE/VORR/VMVN
92 define <2 x i32> @vcunof32(<2 x float>* %A, <2 x float>* %B) nounwind {
93 %tmp1 = load <2 x float>* %A
94 %tmp2 = load <2 x float>* %B
95 %tmp3 = fcmp uno <2 x float> %tmp1, %tmp2
96 %tmp4 = sext <2 x i1> %tmp3 to <2 x i32>
100 ; ord is implemented with VCGT/VCGE/VORR
101 define <2 x i32> @vcordf32(<2 x float>* %A, <2 x float>* %B) nounwind {
102 %tmp1 = load <2 x float>* %A
103 %tmp2 = load <2 x float>* %B
104 %tmp3 = fcmp ord <2 x float> %tmp1, %tmp2
105 %tmp4 = sext <2 x i1> %tmp3 to <2 x i32>