1 ; RUN: llc < %s -march=arm -mattr=+neon | FileCheck %s
3 ; This tests icmp operations that do not map directly to NEON instructions.
4 ; Not-equal (ne) operations are implemented by VCEQ/VMVN. Less-than (lt/ult)
5 ; and less-than-or-equal (le/ule) are implemented by swapping the arguments
6 ; to VCGT and VCGE. Test all the operand types for not-equal but only sample
7 ; the other operations.
9 define <8 x i8> @vcnei8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
13 %tmp1 = load <8 x i8>* %A
14 %tmp2 = load <8 x i8>* %B
15 %tmp3 = icmp ne <8 x i8> %tmp1, %tmp2
16 %tmp4 = sext <8 x i1> %tmp3 to <8 x i8>
20 define <4 x i16> @vcnei16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
24 %tmp1 = load <4 x i16>* %A
25 %tmp2 = load <4 x i16>* %B
26 %tmp3 = icmp ne <4 x i16> %tmp1, %tmp2
27 %tmp4 = sext <4 x i1> %tmp3 to <4 x i16>
31 define <2 x i32> @vcnei32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
35 %tmp1 = load <2 x i32>* %A
36 %tmp2 = load <2 x i32>* %B
37 %tmp3 = icmp ne <2 x i32> %tmp1, %tmp2
38 %tmp4 = sext <2 x i1> %tmp3 to <2 x i32>
42 define <16 x i8> @vcneQi8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
46 %tmp1 = load <16 x i8>* %A
47 %tmp2 = load <16 x i8>* %B
48 %tmp3 = icmp ne <16 x i8> %tmp1, %tmp2
49 %tmp4 = sext <16 x i1> %tmp3 to <16 x i8>
53 define <8 x i16> @vcneQi16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
57 %tmp1 = load <8 x i16>* %A
58 %tmp2 = load <8 x i16>* %B
59 %tmp3 = icmp ne <8 x i16> %tmp1, %tmp2
60 %tmp4 = sext <8 x i1> %tmp3 to <8 x i16>
64 define <4 x i32> @vcneQi32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
68 %tmp1 = load <4 x i32>* %A
69 %tmp2 = load <4 x i32>* %B
70 %tmp3 = icmp ne <4 x i32> %tmp1, %tmp2
71 %tmp4 = sext <4 x i1> %tmp3 to <4 x i32>
75 define <16 x i8> @vcltQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
78 %tmp1 = load <16 x i8>* %A
79 %tmp2 = load <16 x i8>* %B
80 %tmp3 = icmp slt <16 x i8> %tmp1, %tmp2
81 %tmp4 = sext <16 x i1> %tmp3 to <16 x i8>
85 define <4 x i16> @vcles16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
88 %tmp1 = load <4 x i16>* %A
89 %tmp2 = load <4 x i16>* %B
90 %tmp3 = icmp sle <4 x i16> %tmp1, %tmp2
91 %tmp4 = sext <4 x i1> %tmp3 to <4 x i16>
95 define <4 x i16> @vcltu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
98 %tmp1 = load <4 x i16>* %A
99 %tmp2 = load <4 x i16>* %B
100 %tmp3 = icmp ult <4 x i16> %tmp1, %tmp2
101 %tmp4 = sext <4 x i1> %tmp3 to <4 x i16>
105 define <4 x i32> @vcleQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
108 %tmp1 = load <4 x i32>* %A
109 %tmp2 = load <4 x i32>* %B
110 %tmp3 = icmp ule <4 x i32> %tmp1, %tmp2
111 %tmp4 = sext <4 x i1> %tmp3 to <4 x i32>