if(WITH_GPU)
  nv_library(math_cuda SRCS math.cu DEPS eigen_function dense_tensor convert_utils kernel_context kernel_factory pten_transpose_cuda)
  nv_library(linalg_cuda SRCS linalg.cu DEPS eigen_function dense_tensor kernel_context kernel_factory)
  nv_library(utils_cuda SRCS utils.cu DEPS dense_tensor kernel_context kernel_factory memory convert_utils)
  nv_library(manipulation_cuda SRCS manipulation.cu DEPS dense_tensor kernel_context kernel_factory utils_cuda unary)
  nv_library(scale_kernel_cuda SRCS scale_kernel.cu DEPS dense_tensor kernel_context kernel_factory eigen_function)
  nv_library(full_kernel_cuda SRCS full_kernel.cu DEPS dense_tensor kernel_context kernel_factory eigen_function)
  nv_library(conj_kernel_cuda SRCS conj_kernel.cu DEPS dense_tensor kernel_context kernel_factory)
elseif(WITH_ROCM)
  hip_library(math_cuda SRCS math.cu DEPS eigen_function dense_tensor convert_utils kernel_context kernel_factory pten_transpose_cuda)
  hip_library(linalg_cuda SRCS linalg.cu DEPS eigen_function dense_tensor kernel_context kernel_factory)
  hip_library(utils_cuda SRCS utils.cu DEPS dense_tensor kernel_context kernel_factory memory convert_utils)
  hip_library(manipulation_cuda SRCS manipulation.cu DEPS dense_tensor kernel_context kernel_factory utils_cuda unary)
  hip_library(scale_kernel_cuda SRCS scale_kernel.cu DEPS dense_tensor kernel_context kernel_factory eigen_function)
  hip_library(full_kernel_cuda SRCS full_kernel.cu DEPS dense_tensor kernel_context kernel_factory eigen_function)
  hip_library(conj_kernel_cuda SRCS conj_kernel.cu DEPS dense_tensor kernel_context kernel_factory)
endif()