12 lines · plain
1; RUN: llc -mtriple=r600 -mcpu=redwood < %s2 3declare float @llvm.r600.dot4(<4 x float>, <4 x float>) nounwind readnone4 5define amdgpu_kernel void @test_dp4(ptr addrspace(1) %out, ptr addrspace(1) %a, ptr addrspace(1) %b) nounwind {6 %src0 = load <4 x float>, ptr addrspace(1) %a, align 167 %src1 = load <4 x float>, ptr addrspace(1) %b, align 168 %dp4 = call float @llvm.r600.dot4(<4 x float> %src0, <4 x float> %src1) nounwind readnone9 store float %dp4, ptr addrspace(1) %out, align 410 ret void11}12