module.ll

define amdgpu_kernel void @test( i32 %N , float* %out , float* %in ) {
entrypoint:
  %0 = call i32 @llvm.amdgcn.workitem.id.x()
  %1 = icmp sge i32 %0, %N
  br i1 %1, label %L0, label %L1

L0:
  ret void

L1:
  ;
  ; read float from %in
  ;
  %2 = getelementptr float, float* %in, i32 %0
  %3 = load float, float* %2, align 4
  ;
  ; put float onto stack
  ;
  %4 = alloca float, align 4, addrspace(5)
  %5 = getelementptr float, float addrspace(5)* %4, i32 0
  store float %3, float addrspace(5)* %5, align 4
  ;
  ; read float back from stack
  ;
  %6 = getelementptr float, float addrspace(5)* %4, i32 0
  %7 = load float, float addrspace(5)* %6, align 4
  ;
  ; store float into %out
  ;
  %8 = getelementptr float, float* %out, i32 %0
  store float %7, float* %8, align 4
  ret void
}

declare i32 @llvm.amdgcn.workitem.id.x()