experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[ROCm] Enabling vectorization/packet support for some cwise ops on the ROCm platform

Browse Source
This commit is contained in:
Deven Desai 2020-01-06 21:53:49 +00:00
parent ebc6577281
commit 09e2eaf342
2 changed files with 0 additions and 56 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

44
tensorflow/core/kernels/cwise_ops.h
View File

@ -177,11 +177,7 @@ template <typename T>
struct functor_traits<div_no_nan_op<T>> {
enum {
Cost = functor_traits<scalar_quotient_op<T>>::Cost + NumTraits<T>::AddCost,
#if TENSORFLOW_USE_ROCM
PacketAccess = false,
#else
PacketAccess = true,
#endif // TENSORFLOW_USE_ROCM
};
};
@ -194,11 +190,7 @@ template <typename T>
struct functor_traits<mul_no_nan_op<T>> {
enum {
Cost = functor_traits<scalar_product_op<T>>::Cost + NumTraits<T>::AddCost,
#if TENSORFLOW_USE_ROCM
PacketAccess = false,
#else
PacketAccess = true,
#endif // TENSORFLOW_USE_ROCM
};
};
@ -257,11 +249,7 @@ struct functor_traits<
scalar_left<Tout, Tin, Binary, is_scalar_in_host_memory>> {
enum {
Cost = functor_traits<Binary>::Cost,
#if TENSORFLOW_USE_ROCM
PacketAccess = false,
#else
PacketAccess = functor_traits<Binary>::PacketAccess,
#endif // TENSORFLOW_USE_ROCM
};
};
@ -312,11 +300,7 @@ struct functor_traits<
scalar_right<Tout, Tin, Binary, is_scalar_in_host_memory>> {
enum {
Cost = functor_traits<Binary>::Cost,
#if TENSORFLOW_USE_ROCM
PacketAccess = false,
#else
PacketAccess = functor_traits<Binary>::PacketAccess,
#endif // TENSORFLOW_USE_ROCM
};
};
@ -458,11 +442,7 @@ struct functor_traits<google_floor_div<Scalar>> {
Cost = 2 * Eigen::internal::scalar_div_cost<
Scalar, packet_traits<Scalar>::HasDiv>::value +
NumTraits<Scalar>::AddCost,
#if TENSORFLOW_USE_ROCM
PacketAccess = false,
#else
PacketAccess = packet_traits<Scalar>::HasDiv
#endif
};
};
@ -485,12 +465,8 @@ struct functor_traits<google_floor_div_real<Scalar>> {
Cost = 2 * Eigen::internal::scalar_div_cost<
Scalar, packet_traits<Scalar>::HasDiv>::value +
2 * NumTraits<Scalar>::AddCost,
#if TENSORFLOW_USE_ROCM
PacketAccess = false,
#else
PacketAccess =
packet_traits<Scalar>::HasDiv && packet_traits<Scalar>::HasFloor
#endif
};
};
@ -596,13 +572,9 @@ struct functor_traits<scalar_round_half_to_even_op<Scalar>> {
enum {
Cost = Eigen::NumTraits<Scalar>::IsInteger ? 0
: 4 * NumTraits<Scalar>::AddCost,
#if TENSORFLOW_USE_ROCM
PacketAccess = false,
#else
PacketAccess = packet_traits<Scalar>::HasFloor &&
packet_traits<Scalar>::HasAdd &&
packet_traits<Scalar>::HasMul,
#endif
};
};
@ -638,11 +610,7 @@ template <typename Scalar, bool IsInteger>
struct functor_traits<scalar_round_up_op<Scalar, IsInteger>> {
enum {
Cost = IsInteger ? 0 : 4 * NumTraits<Scalar>::AddCost,
#if TENSORFLOW_USE_ROCM
PacketAccess = false,
#else
PacketAccess = IsInteger || packet_traits<Scalar>::HasFloor
#endif
};
};
@ -695,11 +663,7 @@ struct functor_traits<xlogy_op<Scalar>> {
enum {
Cost = functor_traits<scalar_log_op<Scalar>>::Cost +
Eigen::NumTraits<Scalar>::MulCost,
#if TENSORFLOW_USE_ROCM
PacketAccess = false,
#else
PacketAccess = functor_traits<scalar_log_op<Scalar>>::PacketAccess
#endif
};
};
@ -730,11 +694,7 @@ struct functor_traits<xdivy_op<Scalar>> {
Eigen::NumTraits<Scalar>::AddCost +
Eigen::internal::scalar_div_cost<Scalar,
packet_traits<Scalar>::HasDiv>::value,
#if TENSORFLOW_USE_ROCM
PacketAccess = false,
#else
PacketAccess = packet_traits<Scalar>::HasDiv
#endif
};
};
@ -760,11 +720,7 @@ template <typename T>
struct functor_traits<scalar_erfinv_op<T>> {
enum {
Cost = functor_traits<scalar_ndtri_op<T>>::Cost + NumTraits<T>::AddCost,
#if TENSORFLOW_USE_ROCM
PacketAccess = false,
#else
PacketAccess = packet_traits<T>::HasNdtri,
#endif
};
};

12
tensorflow/core/kernels/cwise_ops_gradients.h
View File

@ -94,11 +94,7 @@ template <typename T>
struct functor_traits<scalar_inverse_gradient_op<T>> {
enum {
Cost = NumTraits<T>::AddCost + 2 * NumTraits<T>::MulCost,
#if TENSORFLOW_USE_ROCM
PacketAccess = false,
#else
PacketAccess = packet_traits<T>::HasMul,
#endif
};
};
@ -127,11 +123,7 @@ struct scalar_sqrt_gradient_op {
template <typename T>
struct functor_traits<scalar_sqrt_gradient_op<T>> {
enum {
#if TENSORFLOW_USE_ROCM
PacketAccess = false,
#else
PacketAccess = packet_traits<T>::HasMul & packet_traits<T>::HasDiv,
#endif
Cost = NumTraits<T>::MulCost + scalar_div_cost<T, PacketAccess>::value,
};
};
@ -166,11 +158,7 @@ template <typename T>
struct functor_traits<scalar_rsqrt_gradient_op<T>> {
enum {
Cost = 4 * NumTraits<T>::MulCost,
#if TENSORFLOW_USE_ROCM
PacketAccess = false,
#else
PacketAccess = packet_traits<T>::HasMul,
#endif
};
};