1 ; RUN: opt < %s -cost-model -analyze -mtriple=thumbv7-apple-ios6.0.0 -march=arm -mcpu=cortex-a8 | FileCheck %s --check-prefix=COST
2 ; RUN: llc -mtriple=arm-eabi -mattr=+neon %s -o - | FileCheck %s
3 ; Make sure that ARM backend with NEON handles vselect.
5 define void @vmax_v4i32(<4 x i32>* %m, <4 x i32> %a, <4 x i32> %b) {
6 ; CHECK: vmax.s32 {{q[0-9]+}}, {{q[0-9]+}}, {{q[0-9]+}}
7 %cmpres = icmp sgt <4 x i32> %a, %b
8 %maxres = select <4 x i1> %cmpres, <4 x i32> %a, <4 x i32> %b
9 store <4 x i32> %maxres, <4 x i32>* %m
13 ; We adjusted the cost model of the following selects. When we improve code
14 ; lowering we also need to adjust the cost.
15 %T0_10 = type <16 x i16>
16 %T1_10 = type <16 x i1>
17 ; CHECK-LABEL: func_blend10:
18 define void @func_blend10(%T0_10* %loadaddr, %T0_10* %loadaddr2,
19 %T1_10* %blend, %T0_10* %storeaddr) {
20 %v0 = load %T0_10, %T0_10* %loadaddr
21 %v1 = load %T0_10, %T0_10* %loadaddr2
22 %c = icmp slt %T0_10 %v0, %v1
26 ; COST: cost of 40 {{.*}} select
27 %r = select %T1_10 %c, %T0_10 %v0, %T0_10 %v1
28 store %T0_10 %r, %T0_10* %storeaddr
31 %T0_14 = type <8 x i32>
32 %T1_14 = type <8 x i1>
33 ; CHECK-LABEL: func_blend14:
34 define void @func_blend14(%T0_14* %loadaddr, %T0_14* %loadaddr2,
35 %T1_14* %blend, %T0_14* %storeaddr) {
36 %v0 = load %T0_14, %T0_14* %loadaddr
37 %v1 = load %T0_14, %T0_14* %loadaddr2
38 %c = icmp slt %T0_14 %v0, %v1
42 ; COST: cost of 41 {{.*}} select
43 %r = select %T1_14 %c, %T0_14 %v0, %T0_14 %v1
44 store %T0_14 %r, %T0_14* %storeaddr
47 %T0_15 = type <16 x i32>
48 %T1_15 = type <16 x i1>
49 ; CHECK-LABEL: func_blend15:
50 define void @func_blend15(%T0_15* %loadaddr, %T0_15* %loadaddr2,
51 %T1_15* %blend, %T0_15* %storeaddr) {
54 %v0 = load %T0_15, %T0_15* %loadaddr
55 %v1 = load %T0_15, %T0_15* %loadaddr2
56 %c = icmp slt %T0_15 %v0, %v1
58 ; COST: cost of 82 {{.*}} select
59 %r = select %T1_15 %c, %T0_15 %v0, %T0_15 %v1
60 store %T0_15 %r, %T0_15* %storeaddr
63 %T0_18 = type <4 x i64>
64 %T1_18 = type <4 x i1>
65 ; CHECK-LABEL: func_blend18:
66 define void @func_blend18(%T0_18* %loadaddr, %T0_18* %loadaddr2,
67 %T1_18* %blend, %T0_18* %storeaddr) {
70 %v0 = load %T0_18, %T0_18* %loadaddr
71 %v1 = load %T0_18, %T0_18* %loadaddr2
72 %c = icmp slt %T0_18 %v0, %v1
74 ; COST: cost of 19 {{.*}} select
75 %r = select %T1_18 %c, %T0_18 %v0, %T0_18 %v1
76 store %T0_18 %r, %T0_18* %storeaddr
79 %T0_19 = type <8 x i64>
80 %T1_19 = type <8 x i1>
81 ; CHECK-LABEL: func_blend19:
82 define void @func_blend19(%T0_19* %loadaddr, %T0_19* %loadaddr2,
83 %T1_19* %blend, %T0_19* %storeaddr) {
88 %v0 = load %T0_19, %T0_19* %loadaddr
89 %v1 = load %T0_19, %T0_19* %loadaddr2
90 %c = icmp slt %T0_19 %v0, %v1
92 ; COST: cost of 50 {{.*}} select
93 %r = select %T1_19 %c, %T0_19 %v0, %T0_19 %v1
94 store %T0_19 %r, %T0_19* %storeaddr
97 %T0_20 = type <16 x i64>
98 %T1_20 = type <16 x i1>
99 ; CHECK-LABEL: func_blend20:
100 define void @func_blend20(%T0_20* %loadaddr, %T0_20* %loadaddr2,
101 %T1_20* %blend, %T0_20* %storeaddr) {
110 %v0 = load %T0_20, %T0_20* %loadaddr
111 %v1 = load %T0_20, %T0_20* %loadaddr2
112 %c = icmp slt %T0_20 %v0, %v1
114 ; COST: cost of 100 {{.*}} select
115 %r = select %T1_20 %c, %T0_20 %v0, %T0_20 %v1
116 store %T0_20 %r, %T0_20* %storeaddr