// vfnmadd: vd[i] = -(vd[i] * vs1[i]) - vs2[i]
VI_VFP_VV_LOOP
({
  vd = f32_mulAdd(f32(vd.v ^ F32_SIGN), vs1, f32(vs2.v ^ F32_SIGN));
})