[llvm-project.git] / llvm / test / CodeGen / AArch64 / sve-streaming-mode-fixed-length-reductions.ll
| blob | 688537704a6f73a5182dd5a3364e403e325e39b1 |
1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc -mattr=+sve < %s | FileCheck %s
3 ; RUN: llc -mattr=+dotprod,+sve < %s | FileCheck %s -check-prefix=DOT
4 ; RUN: llc -mattr=+dotprod,+sve -force-streaming-compatible < %s | FileCheck %s --check-prefix=STREAMING-SVE
5 ; RUN: llc -mattr=+dotprod,+sme -force-streaming < %s | FileCheck %s --check-prefix=STREAMING-SVE
7 target triple = "aarch64-unknown-linux-gnu"
9 define i32 @reduce_uaddv_v16i8(<32 x i8> %a) {
10 ; CHECK-LABEL: reduce_uaddv_v16i8:
11 ; CHECK: // %bb.0:
12 ; CHECK-NEXT: ushll2 v2.8h, v1.16b, #0
13 ; CHECK-NEXT: ushll2 v3.8h, v0.16b, #0
14 ; CHECK-NEXT: ushll v1.8h, v1.8b, #0
15 ; CHECK-NEXT: ushll v0.8h, v0.8b, #0
16 ; CHECK-NEXT: uaddl2 v4.4s, v3.8h, v2.8h
17 ; CHECK-NEXT: uaddl v2.4s, v3.4h, v2.4h
18 ; CHECK-NEXT: uaddl2 v5.4s, v0.8h, v1.8h
19 ; CHECK-NEXT: uaddl v0.4s, v0.4h, v1.4h
20 ; CHECK-NEXT: add v1.4s, v5.4s, v4.4s
21 ; CHECK-NEXT: add v0.4s, v0.4s, v2.4s
22 ; CHECK-NEXT: add v0.4s, v0.4s, v1.4s
23 ; CHECK-NEXT: addv s0, v0.4s
24 ; CHECK-NEXT: fmov w0, s0
25 ; CHECK-NEXT: ret
26 ;
27 ; DOT-LABEL: reduce_uaddv_v16i8:
28 ; DOT: // %bb.0:
29 ; DOT-NEXT: movi v2.16b, #1
30 ; DOT-NEXT: movi v3.2d, #0000000000000000
31 ; DOT-NEXT: udot v3.4s, v1.16b, v2.16b
32 ; DOT-NEXT: udot v3.4s, v0.16b, v2.16b
33 ; DOT-NEXT: addv s0, v3.4s
34 ; DOT-NEXT: fmov w0, s0
35 ; DOT-NEXT: ret
36 ;
37 ; STREAMING-SVE-LABEL: reduce_uaddv_v16i8:
38 ; STREAMING-SVE: // %bb.0:
39 ; STREAMING-SVE-NEXT: // kill: def $q1 killed $q1 def $z1
40 ; STREAMING-SVE-NEXT: uunpklo z2.h, z1.b
41 ; STREAMING-SVE-NEXT: // kill: def $q0 killed $q0 def $z0
42 ; STREAMING-SVE-NEXT: uunpklo z3.h, z0.b
43 ; STREAMING-SVE-NEXT: ptrue p0.s, vl4
44 ; STREAMING-SVE-NEXT: ext z1.b, z1.b, z1.b, #8
45 ; STREAMING-SVE-NEXT: ext z0.b, z0.b, z0.b, #8
46 ; STREAMING-SVE-NEXT: uunpklo z1.h, z1.b
47 ; STREAMING-SVE-NEXT: uunpklo z0.h, z0.b
48 ; STREAMING-SVE-NEXT: uunpklo z4.s, z2.h
49 ; STREAMING-SVE-NEXT: ext z2.b, z2.b, z2.b, #8
50 ; STREAMING-SVE-NEXT: uunpklo z6.s, z3.h
51 ; STREAMING-SVE-NEXT: ext z3.b, z3.b, z3.b, #8
52 ; STREAMING-SVE-NEXT: mov z5.d, z1.d
53 ; STREAMING-SVE-NEXT: uunpklo z7.s, z0.h
54 ; STREAMING-SVE-NEXT: ext z0.b, z0.b, z0.b, #8
55 ; STREAMING-SVE-NEXT: uunpklo z2.s, z2.h
56 ; STREAMING-SVE-NEXT: uunpklo z3.s, z3.h
57 ; STREAMING-SVE-NEXT: add z4.s, z6.s, z4.s
58 ; STREAMING-SVE-NEXT: ext z5.b, z5.b, z1.b, #8
59 ; STREAMING-SVE-NEXT: uunpklo z1.s, z1.h
60 ; STREAMING-SVE-NEXT: uunpklo z0.s, z0.h
61 ; STREAMING-SVE-NEXT: add z2.s, z3.s, z2.s
62 ; STREAMING-SVE-NEXT: uunpklo z5.s, z5.h
63 ; STREAMING-SVE-NEXT: add z1.s, z7.s, z1.s
64 ; STREAMING-SVE-NEXT: add z0.s, z0.s, z5.s
65 ; STREAMING-SVE-NEXT: add z1.s, z4.s, z1.s
66 ; STREAMING-SVE-NEXT: add z0.s, z2.s, z0.s
67 ; STREAMING-SVE-NEXT: add z0.s, z1.s, z0.s
68 ; STREAMING-SVE-NEXT: uaddv d0, p0, z0.s
69 ; STREAMING-SVE-NEXT: fmov w0, s0
70 ; STREAMING-SVE-NEXT: ret
71 %1 = zext <32 x i8> %a to <32 x i32>
72 %2 = call i32 @llvm.vector.reduce.add.v32i32(<32 x i32> %1)
73 ret i32 %2
74 }
76 define i32 @reduce_saddv_v16i8(<32 x i8> %a) {
77 ; CHECK-LABEL: reduce_saddv_v16i8:
78 ; CHECK: // %bb.0:
79 ; CHECK-NEXT: sshll2 v2.8h, v1.16b, #0
80 ; CHECK-NEXT: sshll2 v3.8h, v0.16b, #0
81 ; CHECK-NEXT: sshll v1.8h, v1.8b, #0
82 ; CHECK-NEXT: sshll v0.8h, v0.8b, #0
83 ; CHECK-NEXT: saddl2 v4.4s, v3.8h, v2.8h
84 ; CHECK-NEXT: saddl v2.4s, v3.4h, v2.4h
85 ; CHECK-NEXT: saddl2 v5.4s, v0.8h, v1.8h
86 ; CHECK-NEXT: saddl v0.4s, v0.4h, v1.4h
87 ; CHECK-NEXT: add v1.4s, v5.4s, v4.4s
88 ; CHECK-NEXT: add v0.4s, v0.4s, v2.4s
89 ; CHECK-NEXT: add v0.4s, v0.4s, v1.4s
90 ; CHECK-NEXT: addv s0, v0.4s
91 ; CHECK-NEXT: fmov w0, s0
92 ; CHECK-NEXT: ret
93 ;
94 ; DOT-LABEL: reduce_saddv_v16i8:
95 ; DOT: // %bb.0:
96 ; DOT-NEXT: movi v2.16b, #1
97 ; DOT-NEXT: movi v3.2d, #0000000000000000
98 ; DOT-NEXT: sdot v3.4s, v1.16b, v2.16b
99 ; DOT-NEXT: sdot v3.4s, v0.16b, v2.16b
100 ; DOT-NEXT: addv s0, v3.4s
101 ; DOT-NEXT: fmov w0, s0
102 ; DOT-NEXT: ret
103 ;
104 ; STREAMING-SVE-LABEL: reduce_saddv_v16i8:
105 ; STREAMING-SVE: // %bb.0:
106 ; STREAMING-SVE-NEXT: // kill: def $q1 killed $q1 def $z1
107 ; STREAMING-SVE-NEXT: sunpklo z2.h, z1.b
108 ; STREAMING-SVE-NEXT: // kill: def $q0 killed $q0 def $z0
109 ; STREAMING-SVE-NEXT: sunpklo z3.h, z0.b
110 ; STREAMING-SVE-NEXT: ptrue p0.s, vl4
111 ; STREAMING-SVE-NEXT: ext z1.b, z1.b, z1.b, #8
112 ; STREAMING-SVE-NEXT: ext z0.b, z0.b, z0.b, #8
113 ; STREAMING-SVE-NEXT: sunpklo z1.h, z1.b
114 ; STREAMING-SVE-NEXT: sunpklo z0.h, z0.b
115 ; STREAMING-SVE-NEXT: sunpklo z4.s, z2.h
116 ; STREAMING-SVE-NEXT: ext z2.b, z2.b, z2.b, #8
117 ; STREAMING-SVE-NEXT: sunpklo z6.s, z3.h
118 ; STREAMING-SVE-NEXT: ext z3.b, z3.b, z3.b, #8
119 ; STREAMING-SVE-NEXT: mov z5.d, z1.d
120 ; STREAMING-SVE-NEXT: sunpklo z7.s, z0.h
121 ; STREAMING-SVE-NEXT: ext z0.b, z0.b, z0.b, #8
122 ; STREAMING-SVE-NEXT: sunpklo z2.s, z2.h
123 ; STREAMING-SVE-NEXT: sunpklo z3.s, z3.h
124 ; STREAMING-SVE-NEXT: add z4.s, z6.s, z4.s
125 ; STREAMING-SVE-NEXT: ext z5.b, z5.b, z1.b, #8
126 ; STREAMING-SVE-NEXT: sunpklo z1.s, z1.h
127 ; STREAMING-SVE-NEXT: sunpklo z0.s, z0.h
128 ; STREAMING-SVE-NEXT: add z2.s, z3.s, z2.s
129 ; STREAMING-SVE-NEXT: sunpklo z5.s, z5.h
130 ; STREAMING-SVE-NEXT: add z1.s, z7.s, z1.s
131 ; STREAMING-SVE-NEXT: add z0.s, z0.s, z5.s
132 ; STREAMING-SVE-NEXT: add z1.s, z4.s, z1.s
133 ; STREAMING-SVE-NEXT: add z0.s, z2.s, z0.s
134 ; STREAMING-SVE-NEXT: add z0.s, z1.s, z0.s
135 ; STREAMING-SVE-NEXT: uaddv d0, p0, z0.s
136 ; STREAMING-SVE-NEXT: fmov w0, s0
137 ; STREAMING-SVE-NEXT: ret
138 %1 = sext <32 x i8> %a to <32 x i32>
139 %2 = call i32 @llvm.vector.reduce.add.v32i32(<32 x i32> %1)
140 ret i32 %2
141 }