wehub-resource-sync
665 lines
36 KiB
Plaintext
665 lines
36 KiB
Plaintext
// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 (the "License");
|
|
// you may not use this file except in compliance with the License.
|
|
// You may obtain a copy of the License at
|
|
//
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
//
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
// See the License for the specific language governing permissions and
|
|
// limitations under the License.
|
|
|
|
#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
|
|
|
|
#include "paddle/phi/kernels/activation_kernel.h"
|
|
#include "paddle/common/flags.h"
|
|
#include "paddle/phi/backends/gpu/gpu_context.h"
|
|
#include "paddle/phi/core/kernel_registry.h"
|
|
#include "paddle/phi/kernels/abs_kernel.h"
|
|
#include "paddle/phi/kernels/contiguous_kernel.h"
|
|
#include "paddle/phi/kernels/funcs/activation_functor.h"
|
|
#include "paddle/phi/kernels/funcs/broadcast_function.h"
|
|
#include "paddle/phi/kernels/funcs/complex_functors.h"
|
|
#include "paddle/phi/kernels/funcs/dense_tensor_iterator.h"
|
|
#include "paddle/phi/kernels/funcs/elementwise_base.h"
|
|
#include "paddle/phi/kernels/funcs/index_elementwise.cu.h"
|
|
#include "paddle/phi/kernels/selu_kernel.h"
|
|
#include "paddle/phi/kernels/stride/elementwise_stride_base.cu.h"
|
|
|
|
#if defined(__NVCC__) || defined(__HIPCC__) || defined(__xpu__)
|
|
#include "paddle/phi/kernels/funcs/dims_simplifier.h"
|
|
#endif
|
|
COMMON_DECLARE_bool(use_stride_kernel);
|
|
COMMON_DECLARE_bool(use_stride_compute_kernel);
|
|
COMMON_DECLARE_bool(force_stride_compute_contig_out);
|
|
|
|
namespace phi {
|
|
#define DEFINE_CUDA_ACTIVATION_STRIDE_OP(name, functor_class) \
|
|
template <typename T, typename Context> \
|
|
void name##StrideKernel( \
|
|
const Context &dev_ctx, const DenseTensor &x, DenseTensor *out) { \
|
|
if (!FLAGS_use_stride_kernel) { \
|
|
PADDLE_THROW(common::errors::Fatal( \
|
|
"FLAGS_use_stride_kernel is closed. Strided kernel " \
|
|
"be called, something wrong has happened!")); \
|
|
} \
|
|
DenseTensor x_; \
|
|
bool zero_size = false; \
|
|
if (x.numel() == 0) { \
|
|
zero_size = true; \
|
|
} \
|
|
if (!FLAGS_use_stride_compute_kernel || zero_size) { \
|
|
if (!x.meta().is_contiguous()) { \
|
|
x_ = Tensor2Contiguous<Context>(dev_ctx, x); \
|
|
} else { \
|
|
x_ = x; \
|
|
} \
|
|
} else { \
|
|
x_ = x; \
|
|
} \
|
|
if (x_.meta().is_contiguous()) { \
|
|
auto meta = out->meta(); \
|
|
meta.strides = meta.calc_strides(out->dims()); \
|
|
out->set_meta(meta); \
|
|
phi::name##Kernel<T, Context>(dev_ctx, x_, out); \
|
|
return; \
|
|
} \
|
|
if (!FLAGS_use_stride_compute_kernel) { \
|
|
PADDLE_THROW( \
|
|
common::errors::Fatal("FLAGS_use_stride_compute_kernel is closed. " \
|
|
"Kernel using DenseTensorIterator " \
|
|
"be called, something wrong has happened!")); \
|
|
} \
|
|
if (FLAGS_force_stride_compute_contig_out) { \
|
|
auto meta = out->meta(); \
|
|
meta.strides = meta.calc_strides(out->dims()); \
|
|
out->set_meta(meta); \
|
|
} \
|
|
\
|
|
LaunchUnaryElementwiseStrideKernel<T, Context>( \
|
|
dev_ctx, x_, funcs::functor_class<T>(), out); \
|
|
}
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Cos, CudaCosFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Sin, CudaSinFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Tan, CudaTanFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Acos, CudaAcosFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Asin, CudaAsinFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Atan, CudaAtanFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Sinh, CudaSinhFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Cosh, CudaCoshFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Asinh, CudaAsinhFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Acosh, CudaAcoshFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Atanh, CudaAtanhFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Relu, CudaReluFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Tanh, CudaTanhFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Silu, CudaSiluFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Reciprocal, CudaReciprocalFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Square, CudaSquareFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Sqrt, CudaSqrtFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Rsqrt, CudaRsqrtFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Softsign, CudaSoftsignFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Sigmoid, CudaSigmoidFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(LogSigmoid, CudaLogSigmoidFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Floor, CudaFloorFunctor)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_OP(Ceil, CudaCeilFunctor)
|
|
#undef DEFINE_CUDA_ACTIVATION_STRIDE_OP
|
|
#define DEFINE_CUDA_ACTIVATION_WITH_INT_IN_FLOAT_OUT_STRIDE_OP(name, \
|
|
functor_class) \
|
|
template <typename T, typename Context> \
|
|
void name##StrideKernel( \
|
|
const Context &dev_ctx, const DenseTensor &x, DenseTensor *out) { \
|
|
if (!FLAGS_use_stride_kernel) { \
|
|
PADDLE_THROW(common::errors::Fatal( \
|
|
"FLAGS_use_stride_kernel is closed. Strided kernel " \
|
|
"be called, something wrong has happened!")); \
|
|
} \
|
|
DenseTensor x_; \
|
|
bool zero_size = false; \
|
|
if (x.numel() == 0) { \
|
|
zero_size = true; \
|
|
} \
|
|
if (!FLAGS_use_stride_compute_kernel || zero_size) { \
|
|
if (!x.meta().is_contiguous()) { \
|
|
x_ = Tensor2Contiguous<Context>(dev_ctx, x); \
|
|
} else { \
|
|
x_ = x; \
|
|
} \
|
|
} else { \
|
|
x_ = x; \
|
|
} \
|
|
if (x_.meta().is_contiguous()) { \
|
|
auto meta = out->meta(); \
|
|
meta.strides = meta.calc_strides(out->dims()); \
|
|
out->set_meta(meta); \
|
|
phi::name##Kernel<T, Context>(dev_ctx, x_, out); \
|
|
return; \
|
|
} \
|
|
if (!FLAGS_use_stride_compute_kernel) { \
|
|
PADDLE_THROW( \
|
|
common::errors::Fatal("FLAGS_use_stride_compute_kernel is closed. " \
|
|
"Kernel using DenseTensorIterator " \
|
|
"be called, something wrong has happened!")); \
|
|
} \
|
|
if (FLAGS_force_stride_compute_contig_out) { \
|
|
auto meta = out->meta(); \
|
|
meta.strides = meta.calc_strides(out->dims()); \
|
|
out->set_meta(meta); \
|
|
} \
|
|
using U = \
|
|
typename std::conditional_t<std::is_integral<T>::value, float, T>; \
|
|
LaunchUnaryElementwiseStrideKernel<U, Context>( \
|
|
dev_ctx, x_, funcs::functor_class<T>(), out); \
|
|
}
|
|
DEFINE_CUDA_ACTIVATION_WITH_INT_IN_FLOAT_OUT_STRIDE_OP(Log, CudaLogFunctor)
|
|
DEFINE_CUDA_ACTIVATION_WITH_INT_IN_FLOAT_OUT_STRIDE_OP(Log2, CudaLog2Functor)
|
|
DEFINE_CUDA_ACTIVATION_WITH_INT_IN_FLOAT_OUT_STRIDE_OP(Log10, CudaLog10Functor)
|
|
DEFINE_CUDA_ACTIVATION_WITH_INT_IN_FLOAT_OUT_STRIDE_OP(Log1p, CudaLog1pFunctor)
|
|
DEFINE_CUDA_ACTIVATION_WITH_INT_IN_FLOAT_OUT_STRIDE_OP(Exp, CudaExpFunctor)
|
|
DEFINE_CUDA_ACTIVATION_WITH_INT_IN_FLOAT_OUT_STRIDE_OP(Expm1, CudaExpm1Functor)
|
|
#undef DEFINE_CUDA_ACTIVATION_WITH_INT_IN_FLOAT_OUT_STRIDE_OP
|
|
|
|
#define DEFINE_CUDA_ACTIVATION_STRIDE_WITH_ONE_ATTRS( \
|
|
name, functor_class, attr) \
|
|
template <typename T, typename Context> \
|
|
void name##StrideKernel(const Context &dev_ctx, \
|
|
const DenseTensor &x, \
|
|
float attr, \
|
|
DenseTensor *out) { \
|
|
if (!FLAGS_use_stride_kernel) { \
|
|
PADDLE_THROW(common::errors::Fatal( \
|
|
"FLAGS_use_stride_kernel is closed. Strided kernel " \
|
|
"be called, something wrong has happened!")); \
|
|
} \
|
|
DenseTensor x_; \
|
|
bool zero_size = false; \
|
|
if (x.numel() == 0) { \
|
|
zero_size = true; \
|
|
} \
|
|
if (!FLAGS_use_stride_compute_kernel || zero_size) { \
|
|
if (!x.meta().is_contiguous()) { \
|
|
x_ = Tensor2Contiguous<Context>(dev_ctx, x); \
|
|
} else { \
|
|
x_ = x; \
|
|
} \
|
|
} else { \
|
|
x_ = x; \
|
|
} \
|
|
if (x_.meta().is_contiguous()) { \
|
|
auto meta = out->meta(); \
|
|
meta.strides = meta.calc_strides(out->dims()); \
|
|
out->set_meta(meta); \
|
|
phi::name##Kernel<T, Context>(dev_ctx, x_, attr, out); \
|
|
return; \
|
|
} \
|
|
if (!FLAGS_use_stride_compute_kernel) { \
|
|
PADDLE_THROW( \
|
|
common::errors::Fatal("FLAGS_use_stride_compute_kernel is closed. " \
|
|
"Kernel using DenseTensorIterator " \
|
|
"be called, something wrong has happened!")); \
|
|
} \
|
|
if (FLAGS_force_stride_compute_contig_out) { \
|
|
auto meta = out->meta(); \
|
|
meta.strides = meta.calc_strides(out->dims()); \
|
|
out->set_meta(meta); \
|
|
} \
|
|
\
|
|
funcs::functor_class<T> functor; \
|
|
auto attrs = functor.GetAttrs(); \
|
|
*(attrs[0].second) = attr; \
|
|
LaunchUnaryElementwiseStrideKernel<T, Context>(dev_ctx, x_, functor, out); \
|
|
}
|
|
|
|
#define DEFINE_CUDA_ACTIVATION_STRIDE_WITH_ONE_DOUBLE_ATTRS( \
|
|
name, functor_class, attr) \
|
|
template <typename T, typename Context> \
|
|
void name##StrideKernel(const Context &dev_ctx, \
|
|
const DenseTensor &x, \
|
|
double attr, \
|
|
DenseTensor *out) { \
|
|
if (!FLAGS_use_stride_kernel) { \
|
|
PADDLE_THROW(common::errors::Fatal( \
|
|
"FLAGS_use_stride_kernel is closed. Strided kernel " \
|
|
"be called, something wrong has happened!")); \
|
|
} \
|
|
DenseTensor x_; \
|
|
bool zero_size = false; \
|
|
if (x.numel() == 0) { \
|
|
zero_size = true; \
|
|
} \
|
|
if (!FLAGS_use_stride_compute_kernel || zero_size) { \
|
|
if (!x.meta().is_contiguous()) { \
|
|
x_ = Tensor2Contiguous<Context>(dev_ctx, x); \
|
|
} else { \
|
|
x_ = x; \
|
|
} \
|
|
} else { \
|
|
x_ = x; \
|
|
} \
|
|
if (x_.meta().is_contiguous()) { \
|
|
auto meta = out->meta(); \
|
|
meta.strides = meta.calc_strides(out->dims()); \
|
|
out->set_meta(meta); \
|
|
phi::name##Kernel<T, Context>(dev_ctx, x_, attr, out); \
|
|
return; \
|
|
} \
|
|
if (!FLAGS_use_stride_compute_kernel) { \
|
|
PADDLE_THROW( \
|
|
common::errors::Fatal("FLAGS_use_stride_compute_kernel is closed. " \
|
|
"Kernel using DenseTensorIterator " \
|
|
"be called, something wrong has happened!")); \
|
|
} \
|
|
funcs::functor_class<T> functor; \
|
|
auto attrs = functor.GetAttrs(); \
|
|
*(attrs[0].second) = attr; \
|
|
LaunchUnaryElementwiseStrideKernel<T, Context>(dev_ctx, x_, functor, out); \
|
|
}
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_WITH_ONE_DOUBLE_ATTRS(LeakyRelu,
|
|
CudaLeakyReluFunctor,
|
|
alpha)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_WITH_ONE_ATTRS(HardShrink,
|
|
CudaHardShrinkFunctor,
|
|
threshold)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_WITH_ONE_ATTRS(SoftShrink,
|
|
CudaSoftShrinkFunctor,
|
|
lambda)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_WITH_ONE_ATTRS(Elu, CudaELUFunctor, alpha)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_WITH_ONE_ATTRS(Celu, CudaCELUFunctor, alpha)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_WITH_ONE_ATTRS(Mish, CudaMishFunctor, threshold)
|
|
#undef DEFINE_CUDA_ACTIVATION_STRIDE_WITH_ONE_ATTRS
|
|
|
|
#define DEFINE_CUDA_ACTIVATION_STRIDE_WITH_TWO_ATTRS( \
|
|
name, functor_class, attr1, attr2) \
|
|
template <typename T, typename Context> \
|
|
void name##StrideKernel(const Context &dev_ctx, \
|
|
const DenseTensor &x, \
|
|
float attr1, \
|
|
float attr2, \
|
|
DenseTensor *out) { \
|
|
if (!FLAGS_use_stride_kernel) { \
|
|
PADDLE_THROW(common::errors::Fatal( \
|
|
"FLAGS_use_stride_kernel is closed. Strided kernel " \
|
|
"be called, something wrong has happened!")); \
|
|
} \
|
|
DenseTensor x_; \
|
|
bool zero_size = false; \
|
|
if (x.numel() == 0) { \
|
|
zero_size = true; \
|
|
} \
|
|
if (!FLAGS_use_stride_compute_kernel || zero_size) { \
|
|
if (!x.meta().is_contiguous()) { \
|
|
x_ = Tensor2Contiguous<Context>(dev_ctx, x); \
|
|
} else { \
|
|
x_ = x; \
|
|
} \
|
|
} else { \
|
|
x_ = x; \
|
|
} \
|
|
if (x_.meta().is_contiguous()) { \
|
|
auto meta = out->meta(); \
|
|
meta.strides = meta.calc_strides(out->dims()); \
|
|
out->set_meta(meta); \
|
|
phi::name##Kernel<T, Context>(dev_ctx, x_, attr1, attr2, out); \
|
|
return; \
|
|
} \
|
|
if (!FLAGS_use_stride_compute_kernel) { \
|
|
PADDLE_THROW( \
|
|
common::errors::Fatal("FLAGS_use_stride_compute_kernel is closed. " \
|
|
"Kernel using DenseTensorIterator " \
|
|
"be called, something wrong has happened!")); \
|
|
} \
|
|
if (FLAGS_force_stride_compute_contig_out) { \
|
|
auto meta = out->meta(); \
|
|
meta.strides = meta.calc_strides(out->dims()); \
|
|
out->set_meta(meta); \
|
|
} \
|
|
\
|
|
funcs::functor_class<T> functor; \
|
|
auto attrs = functor.GetAttrs(); \
|
|
*(attrs[0].second) = attr1; \
|
|
*(attrs[1].second) = attr2; \
|
|
LaunchUnaryElementwiseStrideKernel<T, Context>(dev_ctx, x_, functor, out); \
|
|
}
|
|
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_WITH_TWO_ATTRS(HardTanh,
|
|
CudaHardTanhFunctor,
|
|
t_min,
|
|
t_max)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_WITH_TWO_ATTRS(HardSigmoid,
|
|
CudaHardSigmoidFunctor,
|
|
slope,
|
|
offset)
|
|
DEFINE_CUDA_ACTIVATION_STRIDE_WITH_TWO_ATTRS(Selu,
|
|
CudaSeluFunctor,
|
|
scale,
|
|
alpha)
|
|
#undef DEFINE_CUDA_ACTIVATION_STRIDE_WITH_ONE_ATTRS
|
|
|
|
template <typename T, typename Context>
|
|
void SoftplusStrideKernel(const Context &dev_ctx,
|
|
const DenseTensor &x,
|
|
double beta,
|
|
double threshold,
|
|
DenseTensor *out) {
|
|
if (!FLAGS_use_stride_kernel) {
|
|
PADDLE_THROW(common::errors::Fatal(
|
|
"FLAGS_use_stride_kernel is closed. Strided kernel be called, "
|
|
"something wrong has happened!"));
|
|
}
|
|
DenseTensor x_;
|
|
bool zero_size = false;
|
|
if (x.numel() == 0) {
|
|
zero_size = true;
|
|
}
|
|
if (!FLAGS_use_stride_compute_kernel || zero_size) {
|
|
if (!x.meta().is_contiguous()) {
|
|
x_ = Tensor2Contiguous<Context>(dev_ctx, x);
|
|
} else {
|
|
x_ = x;
|
|
}
|
|
} else {
|
|
x_ = x;
|
|
}
|
|
if (x_.meta().is_contiguous()) {
|
|
auto meta = out->meta();
|
|
meta.strides = meta.calc_strides(out->dims());
|
|
out->set_meta(meta);
|
|
phi::SoftplusKernel<T, Context>(dev_ctx, x_, beta, threshold, out);
|
|
return;
|
|
}
|
|
if (!FLAGS_use_stride_compute_kernel) {
|
|
PADDLE_THROW(common::errors::Fatal(
|
|
"FLAGS_use_stride_compute_kernel is closed. Kernel using "
|
|
"DenseTensorIterator be called, something wrong has happened!"));
|
|
}
|
|
funcs::CudaSoftplusFunctor<T> functor;
|
|
auto attrs = functor.GetAttrs();
|
|
*(attrs[0].second) = beta;
|
|
*(attrs[1].second) = threshold;
|
|
LaunchUnaryElementwiseStrideKernel<T, Context>(dev_ctx, x_, functor, out);
|
|
}
|
|
|
|
template <typename T, typename Context>
|
|
void RoundStrideKernel(const Context &dev_ctx,
|
|
const DenseTensor &x,
|
|
const int decimals,
|
|
DenseTensor *out) {
|
|
if (!FLAGS_use_stride_kernel) {
|
|
PADDLE_THROW(common::errors::Fatal(
|
|
"FLAGS_use_stride_kernel is closed. Strided kernel "
|
|
"be called, something wrong has happened!"));
|
|
}
|
|
DenseTensor x_;
|
|
bool zero_size = false;
|
|
if (x.numel() == 0) {
|
|
zero_size = true;
|
|
}
|
|
if (!FLAGS_use_stride_compute_kernel || zero_size) {
|
|
if (!x.meta().is_contiguous()) {
|
|
x_ = Tensor2Contiguous<Context>(dev_ctx, x);
|
|
} else {
|
|
x_ = x;
|
|
}
|
|
} else {
|
|
x_ = x;
|
|
}
|
|
if (x_.meta().is_contiguous()) {
|
|
auto meta = out->meta();
|
|
meta.strides = meta.calc_strides(out->dims());
|
|
out->set_meta(meta);
|
|
phi::RoundKernel<T, Context>(dev_ctx, x_, decimals, out);
|
|
return;
|
|
}
|
|
if (!FLAGS_use_stride_compute_kernel) {
|
|
PADDLE_THROW(
|
|
common::errors::Fatal("FLAGS_use_stride_compute_kernel is closed. "
|
|
"Kernel using DenseTensorIterator "
|
|
"be called, something wrong has happened!"));
|
|
}
|
|
|
|
if (FLAGS_force_stride_compute_contig_out) {
|
|
auto meta = out->meta();
|
|
meta.strides = meta.calc_strides(out->dims());
|
|
out->set_meta(meta);
|
|
}
|
|
|
|
funcs::CudaRoundFunctor<T> functor;
|
|
auto attrs = functor.GetAttrs();
|
|
*(attrs[0].second) = decimals;
|
|
LaunchUnaryElementwiseStrideKernel<T, Context>(dev_ctx, x_, functor, out);
|
|
}
|
|
template <typename T, typename Context>
|
|
void HardSwishStrideKernel(const Context &dev_ctx,
|
|
const DenseTensor &x,
|
|
DenseTensor *out) {
|
|
if (!FLAGS_use_stride_kernel) {
|
|
PADDLE_THROW(common::errors::Fatal(
|
|
"FLAGS_use_stride_kernel is closed. Strided kernel "
|
|
"be called, something wrong has happened!"));
|
|
}
|
|
DenseTensor x_;
|
|
bool zero_size = false;
|
|
if (x.numel() == 0) {
|
|
zero_size = true;
|
|
}
|
|
if (!FLAGS_use_stride_compute_kernel || zero_size) {
|
|
if (!x.meta().is_contiguous()) {
|
|
x_ = Tensor2Contiguous<Context>(dev_ctx, x);
|
|
} else {
|
|
x_ = x;
|
|
}
|
|
} else {
|
|
x_ = x;
|
|
}
|
|
if (x_.meta().is_contiguous()) {
|
|
auto meta = out->meta();
|
|
meta.strides = meta.calc_strides(out->dims());
|
|
out->set_meta(meta);
|
|
phi::HardSwishKernel<T, Context>(dev_ctx, x_, out);
|
|
return;
|
|
}
|
|
if (!FLAGS_use_stride_compute_kernel) {
|
|
PADDLE_THROW(
|
|
common::errors::Fatal("FLAGS_use_stride_compute_kernel is closed. "
|
|
"Kernel using DenseTensorIterator "
|
|
"be called, something wrong has happened!"));
|
|
}
|
|
|
|
if (FLAGS_force_stride_compute_contig_out) {
|
|
auto meta = out->meta();
|
|
meta.strides = meta.calc_strides(out->dims());
|
|
out->set_meta(meta);
|
|
}
|
|
|
|
funcs::CudaHardSwishFunctor<T> functor;
|
|
float threshold = 6;
|
|
float scale = 6;
|
|
float offset = 3;
|
|
auto attrs = functor.GetAttrs();
|
|
*(attrs[0].second) = threshold;
|
|
*(attrs[1].second) = scale;
|
|
*(attrs[2].second) = offset;
|
|
LaunchUnaryElementwiseStrideKernel<T, Context>(dev_ctx, x_, functor, out);
|
|
}
|
|
template <typename T, typename Enable = void>
|
|
struct CudaAbsFunctor;
|
|
template <typename T>
|
|
struct CudaAbsFunctor<T, funcs::Complex<T, phi::dtype::Real<T>>> {
|
|
__device__ __forceinline__ phi::dtype::Real<T> operator()(const T x) const {
|
|
return abs(x);
|
|
}
|
|
};
|
|
template <typename T>
|
|
struct CudaAbsFunctor<
|
|
T,
|
|
std::enable_if_t<std::is_same<T, phi::dtype::Real<T>>::value &&
|
|
std::is_same<T, phi::bfloat16>::value>> {
|
|
__device__ __forceinline__ T operator()(const T x) const { return abs(x); }
|
|
};
|
|
template <typename T>
|
|
struct CudaAbsFunctor<
|
|
T,
|
|
std::enable_if_t<std::is_same<T, phi::dtype::Real<T>>::value &&
|
|
!std::is_same<T, phi::bfloat16>::value>> {
|
|
__device__ __forceinline__ T operator()(const T x) const {
|
|
return std::abs(x);
|
|
}
|
|
};
|
|
template <typename T, typename Context>
|
|
void AbsStrideKernel(const Context &dev_ctx,
|
|
const DenseTensor &x,
|
|
DenseTensor *out) {
|
|
if (!FLAGS_use_stride_kernel) {
|
|
PADDLE_THROW(common::errors::Fatal(
|
|
"FLAGS_use_stride_kernel is closed. Strided kernel "
|
|
"be called, something wrong has happened!"));
|
|
}
|
|
DenseTensor x_;
|
|
bool zero_size = false;
|
|
if (x.numel() == 0) {
|
|
zero_size = true;
|
|
}
|
|
if (!FLAGS_use_stride_compute_kernel || zero_size) {
|
|
if (!x.meta().is_contiguous()) {
|
|
x_ = Tensor2Contiguous<Context>(dev_ctx, x);
|
|
} else {
|
|
x_ = x;
|
|
}
|
|
} else {
|
|
x_ = x;
|
|
}
|
|
if (x_.meta().is_contiguous()) {
|
|
auto meta = out->meta();
|
|
meta.strides = meta.calc_strides(out->dims());
|
|
out->set_meta(meta);
|
|
phi::AbsKernel<T, Context>(dev_ctx, x_, out);
|
|
return;
|
|
}
|
|
if (!FLAGS_use_stride_compute_kernel) {
|
|
PADDLE_THROW(
|
|
common::errors::Fatal("FLAGS_use_stride_compute_kernel is closed. "
|
|
"Kernel using DenseTensorIterator "
|
|
"be called, something wrong has happened!"));
|
|
}
|
|
|
|
if (FLAGS_force_stride_compute_contig_out) {
|
|
auto meta = out->meta();
|
|
meta.strides = meta.calc_strides(out->dims());
|
|
out->set_meta(meta);
|
|
}
|
|
|
|
auto functor = CudaAbsFunctor<T>();
|
|
LaunchUnaryElementwiseStrideKernel<phi::dtype::Real<T>, Context>(
|
|
dev_ctx, x_, functor, out);
|
|
}
|
|
} // namespace phi
|
|
PD_REGISTER_KERNEL(abs,
|
|
GPU,
|
|
STRIDED,
|
|
phi::AbsStrideKernel,
|
|
float,
|
|
double,
|
|
int,
|
|
int64_t,
|
|
phi::float16,
|
|
phi::bfloat16,
|
|
phi::complex64,
|
|
phi::complex128) {
|
|
kernel->OutputAt(0).SetDataType(phi::dtype::ToReal(kernel_key.dtype()));
|
|
}
|
|
#define REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(cos, func) \
|
|
PD_REGISTER_KERNEL(cos, \
|
|
GPU, \
|
|
STRIDED, \
|
|
phi::func, \
|
|
float, \
|
|
double, \
|
|
phi::float16, \
|
|
phi::bfloat16, \
|
|
phi::complex64, \
|
|
phi::complex128) {}
|
|
|
|
#define REGISTER_ACTIVATION_MATH_STRIDE_KERNEL(exp, func) \
|
|
PD_REGISTER_KERNEL(exp, \
|
|
GPU, \
|
|
STRIDED, \
|
|
phi::func, \
|
|
float, \
|
|
double, \
|
|
int, \
|
|
int64_t, \
|
|
phi::float16, \
|
|
phi::bfloat16, \
|
|
phi::complex64, \
|
|
phi::complex128) {}
|
|
|
|
#define REGISTER_ACTIVATION_FLOOR_STRIDE_KERNEL(floor, func) \
|
|
PD_REGISTER_KERNEL(floor, \
|
|
GPU, \
|
|
STRIDED, \
|
|
phi::func, \
|
|
float, \
|
|
double, \
|
|
uint8_t, \
|
|
int8_t, \
|
|
int16_t, \
|
|
int, \
|
|
int64_t, \
|
|
phi::float16, \
|
|
phi::bfloat16) {}
|
|
|
|
#define REGISTER_ACTIVATION_STRIDE_KERNEL(leaky_relu, func) \
|
|
PD_REGISTER_KERNEL(leaky_relu, \
|
|
GPU, \
|
|
STRIDED, \
|
|
phi::func, \
|
|
float, \
|
|
double, \
|
|
phi::float16, \
|
|
phi::bfloat16) {}
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(cos, CosStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(sin, SinStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(tan, TanStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(acos, AcosStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(asin, AsinStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(atan, AtanStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(sinh, SinhStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(cosh, CoshStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(asinh, AsinhStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(acosh, AcoshStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(atanh, AtanhStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(tanh, TanhStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL(hardtanh, HardTanhStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL(leaky_relu, LeakyReluStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL(mish, MishStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(silu, SiluStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(softplus, SoftplusStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(softsign, SoftsignStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(sigmoid, SigmoidStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(logsigmoid,
|
|
LogSigmoidStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL(hard_shrink, HardShrinkStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL(softshrink, SoftShrinkStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL(celu, CeluStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL(elu, EluStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL(hardsigmoid, HardSigmoidStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL(selu, SeluStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(hardswish, HardSwishStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(reciprocal,
|
|
ReciprocalStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL_WITH_COMPLEX(sqrt, SqrtStrideKernel)
|
|
REGISTER_ACTIVATION_STRIDE_KERNEL(rsqrt, RsqrtStrideKernel)
|
|
REGISTER_ACTIVATION_MATH_STRIDE_KERNEL(square, SquareStrideKernel)
|
|
REGISTER_ACTIVATION_MATH_STRIDE_KERNEL(log, LogStrideKernel)
|
|
REGISTER_ACTIVATION_MATH_STRIDE_KERNEL(log2, Log2StrideKernel)
|
|
REGISTER_ACTIVATION_MATH_STRIDE_KERNEL(log10, Log10StrideKernel)
|
|
REGISTER_ACTIVATION_MATH_STRIDE_KERNEL(log1p, Log1pStrideKernel)
|
|
REGISTER_ACTIVATION_MATH_STRIDE_KERNEL(exp, ExpStrideKernel)
|
|
REGISTER_ACTIVATION_MATH_STRIDE_KERNEL(expm1, Expm1StrideKernel)
|
|
REGISTER_ACTIVATION_MATH_STRIDE_KERNEL(round, RoundStrideKernel)
|
|
REGISTER_ACTIVATION_FLOOR_STRIDE_KERNEL(floor, FloorStrideKernel)
|
|
REGISTER_ACTIVATION_FLOOR_STRIDE_KERNEL(ceil, CeilStrideKernel)
|
|
#endif
|