1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=aarch64-- | FileCheck %s --check-prefixes=CHECK,CHECK-SD
3 ; RUN: llc < %s -mtriple=aarch64-- -global-isel | FileCheck %s --check-prefixes=CHECK,CHECK-GI
5 declare i4 @llvm.ssub.sat.i4(i4, i4)
6 declare i8 @llvm.ssub.sat.i8(i8, i8)
7 declare i16 @llvm.ssub.sat.i16(i16, i16)
8 declare i32 @llvm.ssub.sat.i32(i32, i32)
9 declare i64 @llvm.ssub.sat.i64(i64, i64)
10 declare <4 x i32> @llvm.ssub.sat.v4i32(<4 x i32>, <4 x i32>)
12 define i32 @func(i32 %x, i32 %y) nounwind {
13 ; CHECK-SD-LABEL: func:
15 ; CHECK-SD-NEXT: subs w8, w0, w1
16 ; CHECK-SD-NEXT: asr w9, w8, #31
17 ; CHECK-SD-NEXT: eor w9, w9, #0x80000000
18 ; CHECK-SD-NEXT: csel w0, w9, w8, vs
21 ; CHECK-GI-LABEL: func:
23 ; CHECK-GI-NEXT: mov w8, #-2147483648 // =0x80000000
24 ; CHECK-GI-NEXT: subs w9, w0, w1
25 ; CHECK-GI-NEXT: cset w10, vs
26 ; CHECK-GI-NEXT: add w8, w8, w9, asr #31
27 ; CHECK-GI-NEXT: tst w10, #0x1
28 ; CHECK-GI-NEXT: csel w0, w8, w9, ne
30 %tmp = call i32 @llvm.ssub.sat.i32(i32 %x, i32 %y);
34 define i64 @func2(i64 %x, i64 %y) nounwind {
35 ; CHECK-SD-LABEL: func2:
37 ; CHECK-SD-NEXT: subs x8, x0, x1
38 ; CHECK-SD-NEXT: asr x9, x8, #63
39 ; CHECK-SD-NEXT: eor x9, x9, #0x8000000000000000
40 ; CHECK-SD-NEXT: csel x0, x9, x8, vs
43 ; CHECK-GI-LABEL: func2:
45 ; CHECK-GI-NEXT: mov x8, #-9223372036854775808 // =0x8000000000000000
46 ; CHECK-GI-NEXT: subs x9, x0, x1
47 ; CHECK-GI-NEXT: cset w10, vs
48 ; CHECK-GI-NEXT: add x8, x8, x9, asr #63
49 ; CHECK-GI-NEXT: tst w10, #0x1
50 ; CHECK-GI-NEXT: csel x0, x8, x9, ne
52 %tmp = call i64 @llvm.ssub.sat.i64(i64 %x, i64 %y);
56 define i16 @func16(i16 %x, i16 %y) nounwind {
57 ; CHECK-SD-LABEL: func16:
59 ; CHECK-SD-NEXT: sxth w8, w0
60 ; CHECK-SD-NEXT: mov w9, #32767 // =0x7fff
61 ; CHECK-SD-NEXT: sub w8, w8, w1, sxth
62 ; CHECK-SD-NEXT: cmp w8, w9
63 ; CHECK-SD-NEXT: csel w8, w8, w9, lt
64 ; CHECK-SD-NEXT: mov w9, #-32768 // =0xffff8000
65 ; CHECK-SD-NEXT: cmn w8, #8, lsl #12 // =32768
66 ; CHECK-SD-NEXT: csel w0, w8, w9, gt
69 ; CHECK-GI-LABEL: func16:
71 ; CHECK-GI-NEXT: sxth w8, w0
72 ; CHECK-GI-NEXT: sub w8, w8, w1, sxth
73 ; CHECK-GI-NEXT: sxth w9, w8
74 ; CHECK-GI-NEXT: sbfx w10, w8, #15, #1
75 ; CHECK-GI-NEXT: sub w10, w10, #8, lsl #12 // =32768
76 ; CHECK-GI-NEXT: cmp w8, w9
77 ; CHECK-GI-NEXT: csel w0, w10, w8, ne
79 %tmp = call i16 @llvm.ssub.sat.i16(i16 %x, i16 %y);
83 define i8 @func8(i8 %x, i8 %y) nounwind {
84 ; CHECK-SD-LABEL: func8:
86 ; CHECK-SD-NEXT: sxtb w9, w0
87 ; CHECK-SD-NEXT: mov w8, #127 // =0x7f
88 ; CHECK-SD-NEXT: sub w9, w9, w1, sxtb
89 ; CHECK-SD-NEXT: cmp w9, #127
90 ; CHECK-SD-NEXT: csel w8, w9, w8, lt
91 ; CHECK-SD-NEXT: mov w9, #-128 // =0xffffff80
92 ; CHECK-SD-NEXT: cmn w8, #128
93 ; CHECK-SD-NEXT: csel w0, w8, w9, gt
96 ; CHECK-GI-LABEL: func8:
98 ; CHECK-GI-NEXT: sxtb w8, w0
99 ; CHECK-GI-NEXT: sub w8, w8, w1, sxtb
100 ; CHECK-GI-NEXT: sxtb w9, w8
101 ; CHECK-GI-NEXT: sbfx w10, w8, #7, #1
102 ; CHECK-GI-NEXT: sub w10, w10, #128
103 ; CHECK-GI-NEXT: cmp w8, w9
104 ; CHECK-GI-NEXT: csel w0, w10, w8, ne
106 %tmp = call i8 @llvm.ssub.sat.i8(i8 %x, i8 %y);
110 define i4 @func3(i4 %x, i4 %y) nounwind {
111 ; CHECK-SD-LABEL: func3:
112 ; CHECK-SD: // %bb.0:
113 ; CHECK-SD-NEXT: lsl w9, w1, #28
114 ; CHECK-SD-NEXT: sbfx w10, w0, #0, #4
115 ; CHECK-SD-NEXT: mov w8, #7 // =0x7
116 ; CHECK-SD-NEXT: sub w9, w10, w9, asr #28
117 ; CHECK-SD-NEXT: cmp w9, #7
118 ; CHECK-SD-NEXT: csel w8, w9, w8, lt
119 ; CHECK-SD-NEXT: mov w9, #-8 // =0xfffffff8
120 ; CHECK-SD-NEXT: cmn w8, #8
121 ; CHECK-SD-NEXT: csel w0, w8, w9, gt
124 ; CHECK-GI-LABEL: func3:
125 ; CHECK-GI: // %bb.0:
126 ; CHECK-GI-NEXT: sbfx w8, w0, #0, #4
127 ; CHECK-GI-NEXT: sbfx w9, w1, #0, #4
128 ; CHECK-GI-NEXT: sub w8, w8, w9
129 ; CHECK-GI-NEXT: sbfx w9, w8, #0, #4
130 ; CHECK-GI-NEXT: asr w10, w9, #3
131 ; CHECK-GI-NEXT: cmp w8, w9
132 ; CHECK-GI-NEXT: add w10, w10, #8
133 ; CHECK-GI-NEXT: csel w0, w10, w8, ne
135 %tmp = call i4 @llvm.ssub.sat.i4(i4 %x, i4 %y);
139 define <4 x i32> @vec(<4 x i32> %x, <4 x i32> %y) nounwind {
142 ; CHECK-NEXT: sqsub v0.4s, v0.4s, v1.4s
144 %tmp = call <4 x i32> @llvm.ssub.sat.v4i32(<4 x i32> %x, <4 x i32> %y);