xref: /llvm-project/llvm/test/CodeGen/AArch64/sve-vecreduce-dot.ll (revision c2bd5c25b3634e55089d34afe922aa38eee743e2)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=aarch64 -mattr=+sve,+dotprod < %s | FileCheck %s
3
4define i32 @test(<vscale x 32 x i8> %bin.rdx, <vscale x 32 x i8> %bin.rdx2)  {
5; CHECK-LABEL: test:
6; CHECK:       // %bb.0:
7; CHECK-NEXT:    sunpklo z4.h, z2.b
8; CHECK-NEXT:    sunpklo z5.h, z0.b
9; CHECK-NEXT:    sunpkhi z0.h, z0.b
10; CHECK-NEXT:    sunpkhi z2.h, z2.b
11; CHECK-NEXT:    sunpklo z6.h, z1.b
12; CHECK-NEXT:    sunpkhi z1.h, z1.b
13; CHECK-NEXT:    sunpklo z7.h, z3.b
14; CHECK-NEXT:    sunpkhi z3.h, z3.b
15; CHECK-NEXT:    ptrue p0.s
16; CHECK-NEXT:    sunpkhi z24.s, z5.h
17; CHECK-NEXT:    sunpklo z5.s, z5.h
18; CHECK-NEXT:    sunpklo z25.s, z4.h
19; CHECK-NEXT:    sunpklo z26.s, z0.h
20; CHECK-NEXT:    sunpkhi z4.s, z4.h
21; CHECK-NEXT:    sunpklo z27.s, z2.h
22; CHECK-NEXT:    sunpkhi z0.s, z0.h
23; CHECK-NEXT:    sunpkhi z2.s, z2.h
24; CHECK-NEXT:    sunpklo z28.s, z6.h
25; CHECK-NEXT:    sunpkhi z6.s, z6.h
26; CHECK-NEXT:    mul z5.s, p0/m, z5.s, z25.s
27; CHECK-NEXT:    sunpkhi z25.s, z1.h
28; CHECK-NEXT:    sunpklo z1.s, z1.h
29; CHECK-NEXT:    mul z26.s, p0/m, z26.s, z27.s
30; CHECK-NEXT:    mul z4.s, p0/m, z4.s, z24.s
31; CHECK-NEXT:    sunpkhi z24.s, z3.h
32; CHECK-NEXT:    mul z0.s, p0/m, z0.s, z2.s
33; CHECK-NEXT:    sunpkhi z2.s, z7.h
34; CHECK-NEXT:    sunpklo z7.s, z7.h
35; CHECK-NEXT:    sunpklo z3.s, z3.h
36; CHECK-NEXT:    mla z0.s, p0/m, z25.s, z24.s
37; CHECK-NEXT:    mad z2.s, p0/m, z6.s, z4.s
38; CHECK-NEXT:    mad z1.s, p0/m, z3.s, z26.s
39; CHECK-NEXT:    movprfx z3, z5
40; CHECK-NEXT:    mla z3.s, p0/m, z28.s, z7.s
41; CHECK-NEXT:    add z0.s, z2.s, z0.s
42; CHECK-NEXT:    add z1.s, z3.s, z1.s
43; CHECK-NEXT:    add z0.s, z1.s, z0.s
44; CHECK-NEXT:    uaddv d0, p0, z0.s
45; CHECK-NEXT:    fmov w0, s0
46; CHECK-NEXT:    ret
47  %a = sext <vscale x 32 x i8> %bin.rdx to <vscale x 32 x i32>
48  %b = sext <vscale x 32 x i8> %bin.rdx2 to <vscale x 32 x i32>
49  %c = mul <vscale x 32 x i32> %a, %b
50  %r = call i32 @llvm.vector.reduce.add.nxv32i32(<vscale x 32 x i32> %c)
51  ret i32 %r
52}
53declare i32 @llvm.vector.reduce.add.nxv32i32(<vscale x 32 x i32> )
54