Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[PHI decoupling] remove "paddle/fluid/operators/amp/fp16_type_traits.h" in phi #48051

Merged
merged 3 commits into from
Nov 18, 2022
Merged

[PHI decoupling] remove "paddle/fluid/operators/amp/fp16_type_traits.h" in phi #48051

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
d3a30f3
rm "paddle/fluid/operators/amp/fp16_type_traits.h" in phi
huangjiyi Nov 16, 2022
fe712a9
Merge branch 'develop' into decouple_fp16_type_utils_h
huangjiyi Nov 17, 2022
e93ca58
Merge remote-tracking branch 'upstream/develop' into decouple_fp16_ty…
huangjiyi Nov 17, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion paddle/fluid/operators/group_norm_op.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -324,7 +324,7 @@ class GroupNormKernel<phi::GPUContext, T> : public framework::OpKernel<T> {
dim3 grid(group_size, groups, x_dims[0]);
dim3 threads(block_size, 1, 1);
if (data_layout == DataLayout::kNCHW) {
using AccT = typename details::MPTypeTrait<T>::Type;
using AccT = typename phi::dtype::MPTypeTrait<T>::Type;
constexpr int vec_size = sizeof(float4) / sizeof(T);
int size = group_size * imsize;
const int max_num_threads = 1024;
Expand Down
2 changes: 1 addition & 1 deletion paddle/fluid/operators/uniform_random_op.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -165,7 +165,7 @@ void UniformRandom(const framework::ExecutionContext& context,

if (seed == 0) {
// Use global Generator seed
using MT = typename details::MPTypeTrait<T>::Type;
using MT = typename phi::dtype::MPTypeTrait<T>::Type;
phi::funcs::uniform_distribution<MT> dist;
phi::funcs::uniform_real_transform<MT> trans(min, max);
phi::funcs::distribution_and_transform<T>(dev_cxt, tensor, dist, trans);
Expand Down
8 changes: 4 additions & 4 deletions paddle/phi/kernels/funcs/functors.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,7 @@ limitations under the License. */

#pragma once

#include "paddle/fluid/operators/amp/fp16_type_traits.h"
#include "paddle/phi/common/amp_type_traits.h"
#include "paddle/phi/kernels/funcs/math.h"

namespace phi {
Expand All @@ -38,7 +38,7 @@ struct AddGradFunctor {

template <typename T>
struct ScaleFunctor {
using MT = typename paddle::operators::details::MPTypeTrait<T>::Type;
using MT = typename phi::dtype::MPTypeTrait<T>::Type;
explicit ScaleFunctor(const MT coeff) : coeff_(coeff) {}

inline HOSTDEVICE T operator()(T ele) {
Expand Down Expand Up @@ -125,7 +125,7 @@ struct SigmoidGradFunctor {

template <typename T>
struct GeluFunctor {
using MT = typename paddle::operators::details::MPTypeTrait<T>::Type;
using MT = typename phi::dtype::MPTypeTrait<T>::Type;
inline HOSTDEVICE T operator()(T x) {
// this function is tanh approximation of gelu
// actual gelu is:
Expand All @@ -141,7 +141,7 @@ struct GeluFunctor {

template <typename T>
struct GeluGradFunctor {
using MT = typename paddle::operators::details::MPTypeTrait<T>::Type;
using MT = typename phi::dtype::MPTypeTrait<T>::Type;
inline HOSTDEVICE T UseX(T x) {
MT mx = static_cast<MT>(x);
MT tanh_out =
Expand Down
4 changes: 2 additions & 2 deletions paddle/phi/kernels/gpu/norm_grad_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,8 +22,8 @@
#include <hipcub/hipcub.hpp>
namespace cub = hipcub;
#endif
#include "paddle/fluid/operators/amp/fp16_type_traits.h"
#include "paddle/phi/backends/gpu/gpu_context.h"
#include "paddle/phi/common/amp_type_traits.h"
#include "paddle/phi/common/bfloat16.h"
#include "paddle/phi/core/kernel_registry.h"
#include "paddle/phi/kernels/funcs/common_shape.h"
Expand All @@ -38,7 +38,7 @@ __global__ void NormalizeGradient(const T* x,
const int axis_n,
const int post,
T* x_grad) {
using MT = typename paddle::operators::details::MPTypeTrait<T>::Type;
using MT = typename phi::dtype::MPTypeTrait<T>::Type;
typedef cub::BlockReduce<MT, BlockDim> BlockReduce;
__shared__ typename BlockReduce::TempStorage temp_storage_sum;
int num = pre * post;
Expand Down
4 changes: 2 additions & 2 deletions paddle/phi/kernels/gpu/norm_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,8 +22,8 @@
#include <hipcub/hipcub.hpp>
namespace cub = hipcub;
#endif
#include "paddle/fluid/operators/amp/fp16_type_traits.h"
#include "paddle/phi/backends/gpu/gpu_context.h"
#include "paddle/phi/common/amp_type_traits.h"
#include "paddle/phi/common/float16.h"
#include "paddle/phi/core/kernel_registry.h"
#include "paddle/phi/kernels/funcs/common_shape.h"
Expand All @@ -46,7 +46,7 @@ __global__ void Normalize(const T* x,
const T eps,
T* y,
T* out_norm) {
using MT = typename paddle::operators::details::MPTypeTrait<T>::Type;
using MT = typename phi::dtype::MPTypeTrait<T>::Type;
typedef cub::BlockReduce<MT, BlockDim> BlockReduce;
__shared__ typename BlockReduce::TempStorage temp_storage;
int num = pre * post;
Expand Down
6 changes: 3 additions & 3 deletions paddle/phi/kernels/gpu/sgd_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,10 +15,10 @@
#include "paddle/phi/kernels/sgd_kernel.h"

#include "paddle/fluid/framework/mixed_vector.h"
#include "paddle/fluid/operators/amp/fp16_type_traits.h"
#include "paddle/phi/backends/gpu/gpu_context.h"
#include "paddle/phi/backends/gpu/gpu_helper.h"
#include "paddle/phi/backends/gpu/gpu_primitives.h"
#include "paddle/phi/common/amp_type_traits.h"
#include "paddle/phi/core/kernel_registry.h"

namespace phi {
Expand Down Expand Up @@ -72,7 +72,7 @@ void SGDDenseKernel(const Context& dev_ctx,
bool multi_precision,
DenseTensor* param_out,
DenseTensor* master_param_out) {
using MPDType = typename paddle::operators::details::MPTypeTrait<T>::Type;
using MPDType = typename phi::dtype::MPTypeTrait<T>::Type;
// do check here
// if (multi_precision) {
// bool has_master =
Expand Down Expand Up @@ -109,7 +109,7 @@ void SGDDenseParamSparseGradKernel(
bool multi_precision,
DenseTensor* param_out,
DenseTensor* master_param_out) {
using MPDType = typename paddle::operators::details::MPTypeTrait<T>::Type;
using MPDType = typename phi::dtype::MPTypeTrait<T>::Type;
// do some check here
// if (multi_precision) {
// bool has_master =
Expand Down
4 changes: 2 additions & 2 deletions paddle/phi/kernels/primitive/functor_primitives.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,7 @@

#pragma once

#include "paddle/fluid/operators/amp/fp16_type_traits.h"
#include "paddle/phi/common/amp_type_traits.h"
#include "paddle/phi/common/float16.h"
#include "paddle/phi/core/enforce.h"
#include "paddle/phi/kernels/funcs/eigen/extensions.h"
Expand Down Expand Up @@ -79,7 +79,7 @@ struct IdentityFunctor {
template <typename Tx, typename Ty = Tx>
struct DivideFunctor {
private:
using MPType = typename ::paddle::operators::details::MPTypeTrait<Tx>::Type;
using MPType = typename ::phi::dtype::MPTypeTrait<Tx>::Type;

public:
HOSTDEVICE inline DivideFunctor() { n_inv = static_cast<MPType>(1.0f); }
Expand Down