brintos

brintos / llvm-project-archived public Read only

0
0
Text · 2.0 KiB · 0472d5c Raw
53 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -mtriple=aarch64 -mattr=+sve,+dotprod < %s | FileCheck %s3 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.b8; CHECK-NEXT:    sunpklo z5.h, z0.b9; CHECK-NEXT:    sunpkhi z0.h, z0.b10; CHECK-NEXT:    sunpkhi z2.h, z2.b11; CHECK-NEXT:    sunpklo z6.h, z1.b12; CHECK-NEXT:    sunpkhi z1.h, z1.b13; CHECK-NEXT:    sunpklo z7.h, z3.b14; CHECK-NEXT:    sunpkhi z3.h, z3.b15; CHECK-NEXT:    ptrue p0.s16; CHECK-NEXT:    sunpkhi z24.s, z5.h17; CHECK-NEXT:    sunpklo z5.s, z5.h18; CHECK-NEXT:    sunpklo z25.s, z4.h19; CHECK-NEXT:    sunpklo z26.s, z0.h20; CHECK-NEXT:    sunpkhi z4.s, z4.h21; CHECK-NEXT:    sunpklo z27.s, z2.h22; CHECK-NEXT:    sunpkhi z0.s, z0.h23; CHECK-NEXT:    sunpkhi z2.s, z2.h24; CHECK-NEXT:    sunpklo z28.s, z6.h25; CHECK-NEXT:    sunpkhi z6.s, z6.h26; CHECK-NEXT:    mul z5.s, p0/m, z5.s, z25.s27; CHECK-NEXT:    sunpkhi z25.s, z1.h28; CHECK-NEXT:    sunpklo z1.s, z1.h29; CHECK-NEXT:    mul z26.s, p0/m, z26.s, z27.s30; CHECK-NEXT:    mul z4.s, p0/m, z4.s, z24.s31; CHECK-NEXT:    sunpkhi z24.s, z3.h32; CHECK-NEXT:    mul z0.s, p0/m, z0.s, z2.s33; CHECK-NEXT:    sunpkhi z2.s, z7.h34; CHECK-NEXT:    sunpklo z7.s, z7.h35; CHECK-NEXT:    sunpklo z3.s, z3.h36; CHECK-NEXT:    mla z0.s, p0/m, z25.s, z24.s37; CHECK-NEXT:    mad z2.s, p0/m, z6.s, z4.s38; CHECK-NEXT:    mad z1.s, p0/m, z3.s, z26.s39; CHECK-NEXT:    mla z5.s, p0/m, z28.s, z7.s40; CHECK-NEXT:    add z0.s, z2.s, z0.s41; CHECK-NEXT:    add z1.s, z5.s, z1.s42; CHECK-NEXT:    add z0.s, z1.s, z0.s43; CHECK-NEXT:    uaddv d0, p0, z0.s44; CHECK-NEXT:    fmov w0, s045; CHECK-NEXT:    ret46  %a = sext <vscale x 32 x i8> %bin.rdx to <vscale x 32 x i32>47  %b = sext <vscale x 32 x i8> %bin.rdx2 to <vscale x 32 x i32>48  %c = mul <vscale x 32 x i32> %a, %b49  %r = call i32 @llvm.vector.reduce.add.nxv32i32(<vscale x 32 x i32> %c)50  ret i32 %r51}52declare i32 @llvm.vector.reduce.add.nxv32i32(<vscale x 32 x i32> )53