activation_mkldnn_op.cc

/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.

   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. */

#include "paddle/fluid/operators/activation_op.h"
#include "paddle/fluid/operators/mkldnn/softplus_mkldnn_op.h"
#include "paddle/fluid/platform/mkldnn_reuse.h"

namespace pten {
class DenseTensor;
}  // namespace pten

namespace paddle {
namespace framework {}  // namespace framework
namespace platform {
class MKLDNNDeviceContext;
}  // namespace platform
}  // namespace paddle

namespace paddle {
namespace operators {

using framework::DataLayout;
using framework::Tensor;
using dnnl::memory;
using dnnl::primitive;
using dnnl::stream;
using platform::GetMKLDNNFormat;
using platform::MKLDNNDeviceContext;
using platform::to_void_cast;

template <typename Functor>
class MKLDNNActivationKernel
    : public framework::OpKernel<typename Functor::ELEMENT_TYPE> {
 public:
  void Compute(const framework::ExecutionContext &ctx) const override {
    const auto *x = ctx.Input<Tensor>("X");
    PADDLE_ENFORCE_EQ(
        x->layout(), DataLayout::kMKLDNN,
        platform::errors::InvalidArgument("Wrong layout set for X tensor"));
    PADDLE_ENFORCE_NE(
        x->format(), MKLDNNMemoryFormat::undef,
        platform::errors::InvalidArgument("Wrong format set for X tensor"));

    Functor functor;
    functor(ctx);
  }
};

template <typename Functor>
class MKLDNNActivationGradKernel
    : public framework::OpKernel<typename Functor::ELEMENT_TYPE> {
 public:
  void Compute(const framework::ExecutionContext &ctx) const override {
    const auto *diff_y = ctx.Input<Tensor>(framework::GradVarName("Out"));
    PADDLE_ENFORCE_EQ(diff_y->layout(), DataLayout::kMKLDNN,
                      platform::errors::InvalidArgument(
                          "Wrong layout set for Input OutGrad tensor"));
    PADDLE_ENFORCE_NE(diff_y->format(), MKLDNNMemoryFormat::undef,
                      platform::errors::InvalidArgument(
                          "Wrong format set for Input OutGrad tensor"));

    Functor functor;
    functor(ctx);
  }
};

template <typename T>
void eltwise_forward(const framework::ExecutionContext &ctx,
                     dnnl::algorithm algorithm) {
  PADDLE_ENFORCE_EQ(platform::is_cpu_place(ctx.GetPlace()), true,
                    paddle::platform::errors::PreconditionNotMet(
                        "Operator DNNL eletwise_forward must use CPUPlace"));
  auto &dev_ctx = ctx.template device_context<MKLDNNDeviceContext>();
  const auto &mkldnn_engine = dev_ctx.GetEngine();

  const auto *x = ctx.Input<Tensor>("X");
  auto *out = ctx.Output<Tensor>("Out");

  bool is_inplaced = x->IsSharedBufferWith(*out);

  platform::ActivationMKLDNNHandler<T> handler(algorithm, ctx, mkldnn_engine,
                                               ctx.GetPlace(), x);

  auto src_memory_p = handler.AcquireSrcMemory(x);
  std::shared_ptr<dnnl::memory> dst_memory_p = nullptr;
  if (is_inplaced) {
    dst_memory_p = src_memory_p;
    out->mutable_data<T>(ctx.GetPlace());
  } else {
    dst_memory_p = handler.AcquireDstMemory(out);
  }
  auto activation_p = handler.AcquireForwardPrimitive();

  auto &astream = paddle::platform::MKLDNNDeviceContext::tls().get_stream();
  activation_p->execute(
      astream, {{DNNL_ARG_FROM, *src_memory_p}, {DNNL_ARG_TO, *dst_memory_p}});
  astream.wait();

  out->set_layout(DataLayout::kMKLDNN);
  out->set_format(GetMKLDNNFormat(*dst_memory_p));
}

template <typename T>
void eltwise_grad(const framework::ExecutionContext &ctx,
                  dnnl::algorithm algorithm) {
  auto &dev_ctx = ctx.template device_context<MKLDNNDeviceContext>();
  const auto &mkldnn_engine = dev_ctx.GetEngine();

  const auto *x = ctx.Input<Tensor>("X");
  const auto *dout = ctx.Input<Tensor>(framework::GradVarName("Out"));
  auto *dx = ctx.Output<Tensor>(framework::GradVarName("X"));

  platform::ActivationMKLDNNHandler<T> handler(algorithm, ctx, mkldnn_engine,
                                               ctx.GetPlace(), x, dout);

  auto src_memory_p = handler.AcquireBackwardSrcMemory(x);
  auto diff_dst_memory_p = handler.AcquireDiffDstMemory(dout);
  auto diff_src_memory_p = handler.AcquireDiffSrcMemory(dx);
  auto activation_backward_p = handler.AcquireBackwardPrimitive();

  auto &astream = paddle::platform::MKLDNNDeviceContext::tls().get_stream();
  activation_backward_p->execute(astream,
                                 {{DNNL_ARG_SRC, *src_memory_p},
                                  {DNNL_ARG_DIFF_DST, *diff_dst_memory_p},
                                  {DNNL_ARG_DIFF_SRC, *diff_src_memory_p}});
  astream.wait();

  dx->set_layout(DataLayout::kMKLDNN);
  dx->set_format(GetMKLDNNFormat(*diff_src_memory_p));
}

template <typename T>
void eltwise_grad_use_out(const framework::ExecutionContext &ctx,
                          dnnl::algorithm algorithm) {
  auto &dev_ctx = ctx.template device_context<MKLDNNDeviceContext>();
  const auto &mkldnn_engine = dev_ctx.GetEngine();

  const auto *out = ctx.Input<Tensor>("Out");
  const auto *dout = ctx.Input<Tensor>(framework::GradVarName("Out"));
  auto *dx = ctx.Output<Tensor>(framework::GradVarName("X"));

  platform::ActivationMKLDNNHandler<T> handler(algorithm, ctx, mkldnn_engine,
                                               ctx.GetPlace(), out, dout);

  auto dst_memory_p = handler.AcquireBackwardSrcMemory(out);
  auto diff_dst_memory_p = handler.AcquireDiffDstMemory(dout);
  auto diff_src_memory_p = handler.AcquireDiffSrcMemory(dx);
  auto activation_backward_p = handler.AcquireBackwardPrimitive();

  auto &astream = paddle::platform::MKLDNNDeviceContext::tls().get_stream();
  activation_backward_p->execute(astream,
                                 {{DNNL_ARG_DST, *dst_memory_p},
                                  {DNNL_ARG_DIFF_DST, *diff_dst_memory_p},
                                  {DNNL_ARG_DIFF_SRC, *diff_src_memory_p}});
  astream.wait();

  dx->set_layout(DataLayout::kMKLDNN);
  dx->set_format(GetMKLDNNFormat(*diff_src_memory_p));
}

template <typename T, dnnl::algorithm algorithm>
struct MKLDNNActivationFunc : public BaseActivationFunctor<T> {
  void operator()(const framework::ExecutionContext &ctx) const {
    eltwise_forward<T>(ctx, algorithm);
  }
};

template <typename T, dnnl::algorithm algorithm>
struct MKLDNNActivationGradFunc : public BaseActivationFunctor<T> {
  void operator()(const framework::ExecutionContext &ctx) const {
    eltwise_grad<T>(ctx, algorithm);
  }
};

template <typename T, dnnl::algorithm algorithm>
struct MKLDNNActivationGradUseOutFunc : public BaseActivationFunctor<T> {
  void operator()(const framework::ExecutionContext &ctx) const {
    eltwise_grad_use_out<T>(ctx, algorithm);
  }
};

template <typename T>
struct GeluMKLDNNFunctor : public BaseActivationFunctor<T> {
  void operator()(const framework::ExecutionContext &ctx) const {
    const bool approximate = ctx.Attr<bool>("approximate");
    if (approximate) {
      eltwise_forward<T>(ctx, dnnl::algorithm::eltwise_gelu_tanh);
    } else {
      eltwise_forward<T>(ctx, dnnl::algorithm::eltwise_gelu_erf);
    }
  }
};

template <typename T>
struct GeluMKLDNNGradFunctor : public BaseActivationFunctor<T> {
  void operator()(const framework::ExecutionContext &ctx) const {
    const bool approximate = ctx.Attr<bool>("approximate");
    if (approximate) {
      eltwise_grad<T>(ctx, dnnl::algorithm::eltwise_gelu_tanh);
    } else {
      eltwise_grad<T>(ctx, dnnl::algorithm::eltwise_gelu_erf);
    }
  }
};

template <typename T>
struct SoftplusMKLDNNFunctor : public BaseActivationFunctor<T> {
  void operator()(const framework::ExecutionContext &ctx) const {
    custom_softplus_eltwise_forward<T>(ctx);
  }
};

template <typename T>
using ReluMKLDNNFunctor =
    MKLDNNActivationFunc<T, dnnl::algorithm::eltwise_relu>;

template <typename T>
using Relu6MKLDNNFunctor =
    MKLDNNActivationFunc<T, dnnl::algorithm::eltwise_bounded_relu>;

template <typename T>
using SwishMKLDNNFunctor =
    MKLDNNActivationFunc<T, dnnl::algorithm::eltwise_swish>;

template <typename T>
using HardSwishMKLDNNFunctor =
    MKLDNNActivationFunc<T, dnnl::algorithm::eltwise_hardswish>;

template <typename T>
using SigmoidMKLDNNFunctor =
    MKLDNNActivationFunc<T, dnnl::algorithm::eltwise_logistic>;

template <typename T>
using TanhMKLDNNFunctor =
    MKLDNNActivationFunc<T, dnnl::algorithm::eltwise_tanh>;

template <typename T>
using SqrtMKLDNNFunctor =
    MKLDNNActivationFunc<T, dnnl::algorithm::eltwise_sqrt>;

template <typename T>
using AbsMKLDNNFunctor = MKLDNNActivationFunc<T, dnnl::algorithm::eltwise_abs>;

template <typename T>
using EluMKLDNNFunctor = MKLDNNActivationFunc<T, dnnl::algorithm::eltwise_elu>;

template <typename T>
using ExpMKLDNNFunctor = MKLDNNActivationFunc<T, dnnl::algorithm::eltwise_exp>;

template <typename T>
using ReluMKLDNNGradFunctor =
    MKLDNNActivationGradFunc<T, dnnl::algorithm::eltwise_relu>;

template <typename T>
using Relu6MKLDNNGradFunctor =
    MKLDNNActivationGradFunc<T, dnnl::algorithm::eltwise_bounded_relu>;

template <typename T>
using SwishMKLDNNGradFunctor =
    MKLDNNActivationGradFunc<T, dnnl::algorithm::eltwise_swish>;

template <typename T>
using HardSwishMKLDNNGradFunctor =
    MKLDNNActivationGradFunc<T, dnnl::algorithm::eltwise_hardswish>;

template <typename T>
using SigmoidMKLDNNGradUseOutFunctor = MKLDNNActivationGradUseOutFunc<
    T, dnnl::algorithm::eltwise_logistic_use_dst_for_bwd>;

template <typename T>
using TanhMKLDNNGradUseOutFunctor = MKLDNNActivationGradUseOutFunc<
    T, dnnl::algorithm::eltwise_tanh_use_dst_for_bwd>;

template <typename T>
using SqrtMKLDNNGradUseOutFunctor = MKLDNNActivationGradUseOutFunc<
    T, dnnl::algorithm::eltwise_sqrt_use_dst_for_bwd>;

template <typename T>
using AbsMKLDNNGradFunctor =
    MKLDNNActivationGradFunc<T, dnnl::algorithm::eltwise_abs>;

template <typename T>
using EluMKLDNNGradUseOutFunctor = MKLDNNActivationGradUseOutFunc<
    T, dnnl::algorithm::eltwise_elu_use_dst_for_bwd>;

template <typename T>
using ExpMKLDNNGradUseOutFunctor = MKLDNNActivationGradUseOutFunc<
    T, dnnl::algorithm::eltwise_exp_use_dst_for_bwd>;

}  // namespace operators
}  // namespace paddle

namespace ops = paddle::operators;

#define REGISTER_ACTIVATION_MKLDNN_KERNEL(act_type, functor, grad_functor) \
  REGISTER_OP_KERNEL(act_type, MKLDNN, ::paddle::platform::CPUPlace,       \
                     ops::MKLDNNActivationKernel<ops::functor<float>>);    \
  REGISTER_OP_KERNEL(                                                      \
      act_type##_grad, MKLDNN, ::paddle::platform::CPUPlace,               \
      ops::MKLDNNActivationGradKernel<ops::grad_functor<float>>);

#define REGISTER_ACTIVATION_MKLDNN_BF16_KERNEL(act_type, functor,             \
                                               grad_functor)                  \
  REGISTER_OP_KERNEL(                                                         \
      act_type, MKLDNN, ::paddle::platform::CPUPlace,                         \
      ops::MKLDNNActivationKernel<ops::functor<float>>,                       \
      ops::MKLDNNActivationKernel<ops::functor<paddle::platform::bfloat16>>); \
  REGISTER_OP_KERNEL(                                                         \
      act_type##_grad, MKLDNN, ::paddle::platform::CPUPlace,                  \
      ops::MKLDNNActivationGradKernel<ops::grad_functor<float>>,              \
      ops::MKLDNNActivationGradKernel<                                        \
          ops::grad_functor<paddle::platform::bfloat16>>);

#define FOR_EACH_MKLDNN_KERNEL_FUNCTOR(__macro)                            \
  __macro(relu6, Relu6MKLDNNFunctor, Relu6MKLDNNGradFunctor);              \
  __macro(leaky_relu, ReluMKLDNNFunctor, ReluMKLDNNGradFunctor);           \
  __macro(swish, SwishMKLDNNFunctor, SwishMKLDNNGradFunctor);              \
  __macro(hard_swish, HardSwishMKLDNNFunctor, HardSwishMKLDNNGradFunctor); \
  __macro(tanh, TanhMKLDNNFunctor, TanhMKLDNNGradUseOutFunctor);           \
  __macro(abs, AbsMKLDNNFunctor, AbsMKLDNNGradFunctor);                    \
  __macro(elu, EluMKLDNNFunctor, EluMKLDNNGradUseOutFunctor);              \
  __macro(exp, ExpMKLDNNFunctor, ExpMKLDNNGradUseOutFunctor);

FOR_EACH_MKLDNN_KERNEL_FUNCTOR(REGISTER_ACTIVATION_MKLDNN_KERNEL);
REGISTER_ACTIVATION_MKLDNN_BF16_KERNEL(relu, ReluMKLDNNFunctor,
                                       ReluMKLDNNGradFunctor);
REGISTER_ACTIVATION_MKLDNN_BF16_KERNEL(gelu, GeluMKLDNNFunctor,
                                       GeluMKLDNNGradFunctor);
REGISTER_ACTIVATION_MKLDNN_BF16_KERNEL(sigmoid, SigmoidMKLDNNFunctor,
                                       SigmoidMKLDNNGradUseOutFunctor);
REGISTER_ACTIVATION_MKLDNN_BF16_KERNEL(sqrt, SqrtMKLDNNFunctor,
                                       SqrtMKLDNNGradUseOutFunctor);

namespace ops = paddle::operators;
REGISTER_OP_KERNEL(
    softplus, MKLDNN, paddle::platform::CPUPlace,
    ops::MKLDNNActivationKernel<ops::SoftplusMKLDNNFunctor<float>>);